阅读说明：本技术 伪正交冗余葡萄糖传感器、系统和方法 (Pseudo-orthogonal redundant glucose sensors, systems, and methods ) 是由 安德烈亚·瓦尔萨乌斯基 迈克尔·E·米勒 于 2018-10-15 设计创作，主要内容包括：一种伪正交冗余葡萄糖传感器装置可以包含一个或多个基于过氧化物的电化学葡萄糖传感器和一个或多个基于氧的电化学传感器。所述一个或多个基于过氧化物的电化学葡萄糖传感器可以作为一个或多个传统的基于过氧化物的传感器操作,其可以包含具有作为催化剂的葡萄糖氧化酶的化学堆叠。所述一个或多个基于氧的电化学传感器可以用于测量氧,以及通过计算两个工作电极之间的氧差来测量葡萄糖。在本发明的实施例中,基于氧的传感器之一可以直接用作诊断,以确定每个基于过氧化物的葡萄糖传感器是否正常工作,以及确定使用哪种感测模式。由于内部基于氧的参考,所述葡萄糖传感器装置提供耐氧葡萄糖感测,以及接近正交的冗余。(A pseudo-orthogonal redundant glucose sensor device may include one or more peroxide-based electrochemical glucose sensors and one or more oxygen-based electrochemical sensors. The one or more peroxide-based electrochemical glucose sensors may operate as one or more conventional peroxide-based sensors, which may comprise a chemical stack having glucose oxidase as a catalyst. The one or more oxygen-based electrochemical sensors can be used to measure oxygen and to measure glucose by calculating the difference in oxygen between the two working electrodes. In an embodiment of the present invention, one of the oxygen-based sensors may be used directly as a diagnostic to determine whether each peroxide-based glucose sensor is functioning properly, and which sensing mode to use. The glucose sensor device provides oxygen-tolerant glucose sensing, as well as near-orthogonal redundancy due to an internal oxygen-based reference.)

1. A continuous glucose monitoring system, comprising:

a pseudo-orthogonally redundant glucose sensor apparatus for determining a glucose concentration in a body of a user, the sensor apparatus comprising:

a peroxide-based electrochemical glucose sensor;

a first oxygen-based electrochemical sensor; and

a second oxygen-based electrochemical sensor; and

sensor electronics, the sensor electronics including at least one physical microprocessor configured to:

(a) receiving a peroxide-based output signal from the peroxide-based glucose sensor, the peroxide-based output signal indicative of a glucose level in the user's body;

(b) receive a first oxygen-based output signal from the first oxygen-based sensor and a second oxygen-based output signal from the second oxygen-based sensor;

(c) calculating a single oxygen-based signal based on the first and second oxygen-based output signals, wherein the single oxygen-based signal is indicative of the glucose level in the user; and is

(d) Fusing the peroxide-based output signal and the single oxygen-based signal to calculate a single fused sensor glucose value for the blood glucose level in the user.

2. The system of claim 1, wherein the sensor device is implanted or subcutaneously disposed within the user's body.

3. The system of claim 1, further comprising a transmitter, wherein the transmitter is worn on a body of a user.

4. The system of claim 3, further comprising a handheld monitor.

5. The system of claim 4, further comprising an insulin pump.

6. The system of claim 5, wherein the glucose monitoring system is a closed loop system.

7. The system of claim 1, wherein the peroxide-based electrochemical glucose sensor is carried on a first flexure and the first and second oxygen-based electrochemical sensors are carried on a second flexure.

8. The system of claim 1, wherein the first oxygen-based sensor comprises glucose oxidase (GOx) as a catalyst and operates at a negative potential.

9. The system of claim 8, wherein the second oxygen-based sensor does not contain GOx and operates at a negative potential.

10. The system of claim 9, wherein the microprocessor calculates the single oxygen-based signal by calculating a difference between the first oxygen-based output signal and the second oxygen-based output signal.

11. The system of claim 1, further comprising a second peroxide-based electrochemical glucose sensor, wherein the first and second peroxide-based electrochemical glucose sensors are redundant glucose sensors.

12. The system of claim 1, wherein the microprocessor periodically calculates the single oxygen-based signal and compares the single oxygen-based signal to the peroxide-based output signal to diagnose whether the peroxide-based glucose sensor is functioning properly.

13. The system of claim 12, wherein the microprocessor determines that the peroxide-based glucose sensor is functioning properly if a difference between the single oxygen-based signal and the peroxide-based output signal exceeds a threshold.

14. A continuous glucose monitoring system, comprising:

a pseudo-orthogonally redundant glucose sensor device for determining glucose levels in a user's body, the sensor device comprising:

a first peroxide-based electrochemical glucose sensor; and

an oxygen-based electrochemical sensor; and

sensor electronics, the sensor electronics including at least one physical microprocessor configured to:

(a) receiving a first peroxide-based output signal from the first peroxide-based electrochemical glucose sensor, the first peroxide-based output signal indicative of a glucose level in the user's body;

(b) receiving an output signal from the oxygen-based electrochemical sensor, the output signal being indicative of a measured oxygen level in the user's body;

(c) determining whether the measured oxygen level in the user's body is at or above a calculated threshold oxygen level; and is

(d) Calculating the glucose level in the user based on the first peroxide-based output signal if the measured oxygen level is at or above the threshold oxygen level.

15. The system of claim 14, wherein if the microprocessor determines that the measured oxygen level is below the threshold oxygen level, the microprocessor inverts the potential of the first peroxide-based electrochemical glucose sensor to convert the first peroxide-based electrochemical glucose sensor to a second oxygen-based sensor and calculates the glucose level in the user's body based on respective output signals of the first and second oxygen-based electrochemical sensors.

16. The system of claim 15, wherein the microprocessor calculates the glucose level in the user by calculating a difference between the output signal of the first oxygen-based electrochemical sensor and the output signal of the second oxygen-based electrochemical sensor.

17. The system of claim 14, further comprising a second peroxide-based electrochemically redundant glucose sensor.

18. The system of claim 17, wherein the microprocessor receives a second peroxide-based output signal from the second peroxide-based electrochemical sensor, the second peroxide-based output signal indicative of the glucose level in the user's body, and wherein, if the measured oxygen level is at or above the threshold oxygen level, the microprocessor fuses the first and second peroxide-based output signals to calculate a single fused sensor glucose value for the blood glucose level in the user's body.

19. The system of claim 14 wherein the first peroxide-based electrochemical glucose sensor is carried on a first flexure and the oxygen-based electrochemical sensor is carried on a second flexure.

20. The system of claim 14, wherein the glucose monitoring system is a closed loop system.

Technical Field

Embodiments of the invention generally relate to sensor technology, including sensors for sensing various physiological parameters (e.g., glucose concentration). More particularly, embodiments of the present invention relate to pseudo-orthogonal redundant glucose sensors, devices and sensor systems, including closed-loop insulin infusion systems, as well as fusion algorithms, Electrochemical Impedance Spectroscopy (EIS) and Application Specific Integrated Circuits (ASICs) for implementing the use of such pseudo-orthogonal redundant glucose sensors, devices, sensor systems and methods.

Background

A subject (e.g., a patient) and medical personnel desire to monitor readings of a physiological condition in the subject. Illustratively, the subject wishes to monitor blood glucose levels in the subject on a continuous basis. Currently, a patient can measure his/her BG using a Blood Glucose (BG) measuring device (i.e., a blood glucose meter) such as a strip meter (test strip meter), a continuous glucose measuring system (or continuous glucose monitor), or hospital hemacue. BG measurement devices measure BG levels in patients using various methods, such as a blood sample of the patient, a sensor in contact with a bodily fluid, an optical sensor, an enzymatic sensor, or a fluorescence/fluorescence quenching sensor. When the BG measurement device generates a BG measurement value, the measurement value is displayed on the BG measurement device.

Infusion pump devices and systems for delivering or dispensing prescribed medications, such as insulin, to a patient are relatively well known in the medical arts. In one form, such devices include a relatively compact pump housing adapted to receive a syringe or reservoir carrying a medicament for administration to a patient via an infusion tube and associated catheter or infuser. The programmable control may operate the infusion pump continuously or at periodic intervals to achieve tightly controlled and precise drug delivery over an extended period of time. Such infusion pumps are used to administer insulin and other drugs, U.S. patent No. 4,562,751; 4,678,408 No; 4,685,903 No; 5,080,653 No; and 5,097,122, which are incorporated herein by reference.

In diabetic individuals, each individual's body has a baseline insulin demand, which can typically be maintained by continuous or continuous administration of basal amounts of insulin to the patient using an infusion pump. However, when additional glucose is present (i.e., exceeds a basal level) in a diabetic individual (e.g., when the individual eats), the amount and time of insulin to be administered must be determined in order to adequately account for the additional glucose while avoiding the infusion of too much insulin. Typically, large doses of insulin are administered to compensate for meals (i.e., large dose meals). Diabetics typically determine the amount of insulin they may need based on the carbohydrate content of the diet to meet expected dietary needs.

Over the years, a variety of electrochemical glucose sensors have been developed for obtaining an indication of blood glucose levels in diabetic patients. Such readings are useful for monitoring and/or adjusting a treatment regimen, which typically involves periodic administration of insulin to the patient. In general, small, flexible electrochemical sensors can be used to obtain periodic readings over long periods of time. In one form, the flexible subcutaneous sensor is constructed in accordance with thin film shielding techniques. In commonly assigned U.S. patent No. 5,390,671; U.S. Pat. No. 5,391,250; 5,482,473; and 5,586,553, which are incorporated herein by reference.

These electrochemical sensors have been applied to telemetry characteristic monitoring systems. As described in commonly assigned U.S. patent No. 6,809,653 ("the' 653 patent"), the entire contents of which are incorporated herein by reference, a telemetry system includes a remotely located data receiving device, a sensor for generating a signal indicative of a characteristic of a user, and a transmitter device for processing the signal received from the sensor and wirelessly transmitting the processed signal to the remotely located data receiving device. The data receiving device may be a feature monitor, a data receiver (providing data to another device), an RF programmer, a drug delivery device (e.g., an infusion pump), etc.

Current continuous glucose measurement systems include subcutaneous (or short-term) sensors and implantable (or long-term) sensors. For each short-term sensor and long-term sensor, the patient must wait a certain amount of time for the continuous glucose sensor to stabilize and provide an accurate reading. In many continuous glucose sensors, the subject must wait three hours for the continuous glucose sensor to stabilize before using any glucose measurements. This can be inconvenient to the patient and in some cases can result in the patient not using a continuous glucose measurement system.

In addition, when the glucose sensor is first inserted into the skin or under the skin of a patient, the glucose sensor cannot operate in a steady state. The electrical readings of the sensor represent the patient's glucose level, varying over a wide range of readings. In the past, sensor stabilization typically required several hours. One technique for sensor stabilization is detailed, for example, in the' 653 patent, where the initialization process for sensor stabilization can be reduced to about one hour. A high voltage (e.g., 1.0-1.2 volts) may be applied for 1-2 minutes to allow the sensor to stabilize, and then a low voltage (e.g., between 0.5-0.6 volts) may be applied for the remainder of the initialization process (e.g., around 58 minutes).

It is also desirable to allow the electrodes of the sensor to be sufficiently "wetted" or hydrated prior to use of the electrodes of the sensor. If the electrodes of the sensor are not sufficiently hydrated, the result may be inaccurate readings of the patient's physiological condition. A user of a current blood glucose sensor may be instructed not to power the sensor immediately. These blood glucose sensors may not work in an optimal or efficient manner if used prematurely.

Most of the prior art Continuous Glucose Monitoring (CGM) is ancillary, meaning that readings provided by CGM devices (including, for example, implantable or subcutaneous sensors) cannot be used to make clinical decisions without reference values. The reference values, in turn, must be obtained from the fingertip using, for example, a BG meter. The reference value is required because the amount of information provided by the sensor/sensing assembly is limited. In particular, the sensing component can only provide the raw sensor value (i.e., sensor current or Isig) and the back voltage for processing. Thus, during analysis, if the raw sensor signal appears abnormal (e.g., if the signal is decreasing), the only way that sensor failure can be distinguished from physiological changes in the user/patient (i.e., changes in glucose level within the body) may be to acquire a reference glucose value through the fingertip. It is well known that the reference fingertip is also used to calibrate the sensor.

The art has sought ways to eliminate or at least minimize the number of fingertips necessary for calibrating and evaluating sensor health. However, in view of the number and complexity of the numerous sensor failure modes, no satisfactory solution has been found. At best, diagnostics based on direct assessment of Isig or based on comparison of two isigs have been developed. In either case, because Isig tracks glucose levels in the body, Isig is not analyte-independent by definition. Thus, Isig itself is not a reliable source of information for sensor diagnostics, nor is it a reliable predictor of continuous sensor performance.

To date, another limitation in the art exists in the lack of sensor electronics that are capable of not only operating the sensor, but also performing real-time sensor and electrode diagnostics while managing the power supply to the sensor, and so on for redundant electrodes, redundant sensors, complementary sensors, and redundant and complementary sensors. Indeed, the concept of electrode redundancy has existed for some time. However, in the past, obtaining more than one reading at a time using electrode redundancy (and/or complementary and redundant electrodes) and assessing the relative health of the redundant electrodes, the overall reliability of the sensor, and the frequency (if any) required to calibrate the reference values has met with little success.

Furthermore, even if redundant sensing electrodes are used, the number is typically limited to only two. Again, this is due in part to the lack of advanced electronics to run, evaluate, and manage multiple independent working electrodes (e.g., up to 5 or more) in real time. Yet another reason is that the view of using redundant electrodes to obtain "independent" sensor signals is limited, for which two redundant electrodes are sufficient. As mentioned above, although this is one function of utilizing redundant electrodes, it is not the only one.

Disclosure of Invention

According to one embodiment of the present invention, a continuous glucose monitoring system comprises a pseudo-orthogonal redundant glucose sensor device for determining a glucose concentration in a body of a user, wherein the sensor device comprises a peroxide-based electrochemical glucose sensor, a first oxygen-based electrochemical sensor, and a second oxygen-based electrochemical sensor; and sensor electronics, wherein the sensor electronics include at least one physical microprocessor configured to: (a) receiving a peroxide-based output signal from the peroxide-based glucose sensor, the peroxide-based output signal indicative of a glucose level in a user's body; (b) receiving a first oxygen-based output signal from the first oxygen-based sensor and a second oxygen-based output signal from the second oxygen-based sensor; (c) calculating a single oxygen-based signal based on the first and second oxygen-based output signals, wherein the single oxygen-based signal is indicative of a glucose level in the user; and (d) a single fused sensor glucose value that fuses the peroxide-based output signal and the single oxygen-based signal to calculate a blood glucose level in the body of the user.

According to another embodiment of the present invention, a continuous glucose monitoring system includes a pseudo-orthogonal redundant glucose sensor device for determining a glucose level in a body of a user, wherein the sensor device includes a first peroxide-based electrochemical glucose sensor and an oxygen-based electrochemical sensor; and sensor electronics, wherein the sensor electronics include at least one physical microprocessor configured to: (a) receiving a first peroxide-based output signal from the first peroxide-based electrochemical glucose sensor, the first peroxide-based output signal indicative of a glucose level in the user's body; (b) receiving an output signal from the oxygen-based electrochemical sensor, the output signal indicative of a measured oxygen level in the body of the user; (c) determining whether the measured oxygen level in the user's body is equal to or above a calculated threshold oxygen level; (d) calculating a glucose level in the user from the first peroxide-based output signal if the measured oxygen level is at or above the threshold oxygen level.

Drawings

Embodiments of the present invention will be described in detail with reference to the drawings, wherein like reference numerals designate corresponding parts in the drawings.

FIG. 1 is a perspective view of a subcutaneous sensor insertion device and a block diagram of sensor electronics in accordance with an embodiment of the present invention.

Fig. 2A shows a substrate having two sides, wherein a first side comprises an electrode arrangement and a second side comprises electronic circuitry.

FIG. 2B shows an overall block diagram of electronic circuitry for sensing sensor output.

FIG. 3 shows a block diagram of sensor electronics and a sensor including multiple electrodes, in accordance with an embodiment of the invention.

FIG. 4 shows an alternative embodiment of the invention, including a sensor and sensor electronics according to an embodiment of the invention.

FIG. 5 shows an electronic block diagram of a sensor electrode and a voltage applied to the sensor electrode according to an embodiment of the invention.

Fig. 6A illustrates a method of applying pulses during a settling time range to reduce the settling time range, in accordance with an embodiment of the present invention.

FIG. 6B illustrates a method of stabilizing a sensor according to an embodiment of the invention.

FIG. 6C illustrates the use of feedback in stabilizing a sensor according to an embodiment of the present invention.

FIG. 7 illustrates the effect of stabilizing a sensor according to an embodiment of the present invention.

FIG. 8A shows a block diagram of sensor electronics and a sensor including a voltage generation device, according to an embodiment of the invention.

Fig. 8B shows a voltage generation device for implementing this embodiment of the present invention.

FIG. 8C shows a voltage generation device that generates two voltage values according to an embodiment of the invention.

Fig. 8D shows a voltage generating device having three voltage generating systems according to an embodiment of the present invention.

FIG. 9A shows sensor electronics including a microcontroller for generating voltage pulses according to an embodiment of the invention.

FIG. 9B illustrates sensor electronics including an analysis module according to an embodiment of the invention.

FIG. 10 shows a block diagram of a sensor system including hydration electronics, in accordance with an embodiment of the present invention.

FIG. 11 illustrates an embodiment of the present invention that includes a mechanical switch to assist in determining hydration time.

FIG. 12 illustrates a method of detecting hydration according to an embodiment of the present invention.

FIG. 13A illustrates a method of hydrating a sensor according to an embodiment of the invention.

FIG. 13B illustrates an additional method for verifying sensor hydration according to an embodiment of the invention.

14A, 14B, and 14C illustrate a method of combining hydration of a sensor with stabilization of the sensor, according to an embodiment of the invention.

FIG. 15A illustrates an EIS-based analysis of system response to the application of a periodic alternating signal, in accordance with an embodiment of the invention.

Fig. 15B shows a known circuit model for electrochemical impedance spectroscopy.

FIG. 16A shows an example of a Nyquist curve for an embodiment of the invention in which an AC voltage plus a DC voltage (DC bias) is applied to the working electrode for a selected spectrum from 0.1Hz to 1000 Mhz.

Fig. 16B illustrates another example of a nyquist curve having a linear fit for relatively low frequencies and an intercept approaching the real impedance value at relatively high frequencies.

FIGS. 16C and 16D show the infinite and finite response of the glucose sensor to a sinusoidal operating potential.

Figure 16E shows a bode plot of amplitude according to an embodiment of the present invention.

Fig. 16F shows a bode plot of the phase according to an embodiment of the invention.

FIG. 17 illustrates a Nyquist plot of sensor impedance as a function of sensor aging according to an embodiment of the present invention.

FIG. 18 illustrates a method of applying EIS techniques in stabilizing and detecting sensor aging according to an embodiment of the invention.

FIG. 19 illustrates a schedule for performing an EIS process according to an embodiment of the invention.

FIG. 20 illustrates a method of using an EIS procedure to detect and repair a sensor in conjunction with remedial actions, according to an embodiment of the invention.

Fig. 21A and 21B illustrate an example of sensor remedial action according to an embodiment of the present invention.

Fig. 22 shows the nyquist curve for a normally operating sensor, where the nyquist slope increases gradually and the intercept decreases gradually as the sensor wear time progresses.

FIG. 23A shows the raw current signal (Isig) from two redundant working electrodes, and the real impedance of each electrode at 1kHz, according to an embodiment of the invention.

Fig. 23B shows a nyquist curve for the first working electrode (WE1) of fig. 23A.

Fig. 23C shows a nyquist curve of the second working electrode (WE2) of fig. 23A.

FIG. 24 shows an example of signal dips for two redundant working electrodes, and the real impedance of each electrode at 1kHz, according to an embodiment of the invention.

FIG. 25A illustrates the substantially glucose-independent dependence of real impedance, imaginary impedance, and phase at relatively high frequencies for a normally operating glucose sensor according to an embodiment of the invention.

FIG. 25B shows an example of different levels of glucose dependence of real impedance at relatively lower frequencies according to an embodiment of the invention.

FIG. 25C shows an example of different levels of glucose dependence of the phase at relatively lower frequencies, according to an embodiment of the invention.

FIG. 26 shows trends for 1kHz real impedance, 1kHz imaginary impedance, and relatively high frequency phase when a glucose sensor loses sensitivity due to hypoxia at the sensor insertion site, in accordance with an embodiment of the invention.

FIG. 27 shows Isig and phase for simulating hypoxia in vitro at different glucose concentrations, according to an embodiment of the invention.

Fig. 28A-28C illustrate examples of sensitivity loss due to oxygen deficit that occurs with redundant working electrodes WE1 and WE2 and EIS-based parameters of the electrodes, according to embodiments of the invention.

Figure 28D shows EIS-induced spikes in the raw Isig of the example of figures 28A-28C.

Fig. 29 shows an example of sensitivity loss due to hypoxia caused by occlusion according to an embodiment of the present invention.

Fig. 30A-30C show examples of sensitivity loss due to biological contamination occurring with redundant working electrodes WE1 and WE2 and EIS-based parameters of the electrodes, according to embodiments of the invention.

Figure 30D shows EIS-induced spikes in the raw Isig of the example of figures 30A-30C.

FIG. 31 illustrates a diagnostic process for sensor fault detection according to an embodiment of the present invention.

Fig. 32A and 32B illustrate another diagnostic process for sensor fault detection according to an embodiment of the present invention.

FIG. 33A shows a top-level flow diagram relating to a current (Isig) -based fusion algorithm, according to an embodiment of the invention.

FIG. 33B shows a top-level flow diagram relating to a Sensor Glucose (SG) -based fusion algorithm, according to an embodiment of the invention.

FIG. 34 shows details of the Sensor Glucose (SG) -based fusion algorithm of FIG. 33B, according to an embodiment of the invention.

FIG. 35 shows details of the current (Isig) -based fusion algorithm of FIG. 33A, according to an embodiment of the present invention.

FIG. 36 is a graphical representation of calibration of a sensor at steady state according to an embodiment of the invention.

FIG. 37 is an illustration of calibration of a sensor in transition, according to an embodiment of the invention.

FIG. 38A is a graphical representation of EIS based dynamic slope (with slope adjustment) for sensor calibration according to an embodiment of the present invention.

FIG. 38B illustrates an EIS assisted sensor calibration flow diagram involving low start-up detection according to an embodiment of the invention.

FIG. 39 shows sensor current (Isig) and 1kHz impedance magnitude for in vitro simulation of interferents in close proximity to the sensor, in accordance with an embodiment of the invention.

Figures 40A and 40B show baud curves for the simulated phase and impedance shown in figure 39, respectively.

Fig. 40C shows the simulated nyquist curve shown in fig. 39.

FIG. 41 shows another in vitro simulation with an interferent according to an embodiment of the invention.

42A and 42B illustrate ASIC block diagrams according to embodiments of the present invention.

FIG. 43 shows a potentiostat configuration for a sensor with redundant working electrodes, in accordance with an embodiment of the invention.

Fig. 44 shows an equivalent AC inter-electrode circuit of a sensor having the potentiostat configuration shown in fig. 43.

FIG. 45 shows some major blocks of EIS circuitry in an analog front-end IC circuit of a glucose sensor according to an embodiment of the invention.

Fig. 46A-46F show simulations of signals of the EIS circuitry shown in fig. 45 for a 0 degree phase current multiplied by a 0 degree phase.

FIGS. 47A-47F show simulations of signals of the EIS circuitry shown in FIG. 45 for 0 degree phase current multiplied by 90 degree phase.

FIG. 48 shows a circuit model according to an embodiment of the invention.

49A-49C show diagrams of circuit models in accordance with alternative embodiments of the present invention.

Fig. 50A is a nyquist curve covering an equivalent circuit simulation according to an embodiment of the present invention.

Fig. 50B is an enlarged view of the high-frequency portion of fig. 50A.

Fig. 51 shows a nyquist curve according to an embodiment of the present invention, in which Cdl is continuously increased in the direction of arrow a.

Fig. 52 shows a nyquist curve according to an embodiment of the invention, wherein α increases continuously in the direction of arrow a.

Fig. 53 shows a nyquist curve according to an embodiment of the present invention, in which Rp increases continuously in the direction of arrow a.

FIG. 54 shows a Nyquist curve for an embodiment of the invention where the Walberg admittance is increasing in the direction of arrow A.

Fig. 55 shows a nyquist curve according to an embodiment of the present invention, where λ is increasing in the direction of arrow a.

FIG. 56 illustrates the effect of film capacitance on the Nyquist curve according to an embodiment of the invention.

Fig. 57 shows a nyquist curve in which the film resistance is continuously increased in the direction of arrow a according to an embodiment of the present invention.

Fig. 58 shows a nyquist curve according to an embodiment of the present invention, in which Rsol increases continuously in the direction of arrow a.

FIGS. 59A-59C illustrate changes in EIS parameters associated with circuit elements during startup and calibration according to embodiments of the invention.

60A-60C illustrate variations in different sets of EIS parameters associated with circuit elements during startup and calibration in accordance with embodiments of the present invention.

61A-61C illustrate variations in another different set of EIS parameters associated with a circuit element during startup and calibration according to embodiments of the invention.

FIG. 62 illustrates EIS responses of a plurality of electrodes according to an embodiment of the invention.

FIG. 63 is a Nyquist plot illustrating the effect of Isig calibration by glucose increase according to an embodiment of the present invention.

FIG. 64 shows oxygen (V) according to an embodiment of the invention_cntr) The effect of the response on the nyquist curve.

FIG. 65 illustrates the shift of the Nyquist curve due to temperature change, according to an embodiment of the present invention.

FIG. 66 shows the relationship between Isig and blood glucose according to an embodiment of the present invention.

67A-67B illustrate sensor drift according to embodiments of the present invention.

Figure 68 shows an increase in membrane resistance during a sensitivity loss according to an embodiment of the invention.

FIG. 69 illustrates a drop in Warburg admittance during a sensitivity loss, according to an embodiment of the invention.

FIG. 70 shows a calibration curve according to an embodiment of the invention.

Figure 71 shows a higher frequency semicircle that becomes visible on the nyquist curve according to an embodiment of the present invention.

FIGS. 72A and 72B show V according to an embodiment of the invention_cntrTrack and Cdl decrease.

FIG. 73 illustrates a slope of a change in a calibration curve according to an embodiment of the invention.

Fig. 74 shows the varying length of the nyquist curve according to an embodiment of the present invention.

Fig. 75 shows an enlarged view of the low and high frequency regions of the nyquist curve of fig. 74.

FIGS. 76A and 76B illustrate membrane resistance increase, according to embodiments of the invention,Cdl and V_cntrThe combined effect of the rail reduction.

FIG. 77 shows relative Cdl values for two working electrodes according to an embodiment of the invention.

FIG. 78 shows relative Rp values for two working electrodes according to an embodiment of the invention.

FIG. 79 illustrates the combined effect of changing EIS parameters of a calibration curve according to an embodiment of the invention.

Fig. 80 shows that the length of the nyquist curve is longer in the low frequency region where there is a sensitivity loss, according to an embodiment of the present invention.

FIG. 81 is a flow diagram of sensor self-calibration based on sensitivity change detection, according to an embodiment of the invention.

Fig. 82 illustrates horizontal shift in the nyquist curve caused by sensitivity loss according to an embodiment of the present invention.

FIG. 83 illustrates a method of developing a heuristic EIS metric based on Nyquist curves, in accordance with embodiments of the present invention.

Fig. 84 shows a relationship between Rm and a calibration factor according to an embodiment of the present invention.

FIG. 85 shows a relationship between Rm and normalized Isig according to an embodiment of the invention.

FIG. 86 shows an Isig curve for different glucose levels over time, in accordance with an embodiment of the present invention.

FIG. 87 shows Cdl curves for different glucose levels over time, according to an embodiment of the present invention.

FIG. 88 illustrates a second inflection point of the curve of FIG. 86 in accordance with an embodiment of the present invention.

Fig. 89 illustrates a second inflection point corresponding to Rm of the peak in fig. 88, according to an embodiment of the present invention.

FIG. 90 shows a graphical representation of the relationship between Calibration Factor (CF) and Rmem + Rsol, in accordance with an embodiment of the present invention.

Fig. 91A is a graph showing in vivo results for a MARD over all effective BGs for approximately the first 8 hours of sensor life, according to an embodiment of the invention.

Fig. 91B is a graph illustrating the median ARD number for all valid BGs over about the first 8 hours of sensor life, according to an embodiment of the invention.

Fig. 92A-92C illustrate calibration factor adjustment according to embodiments of the present invention.

Fig. 93A-93C illustrate calibration factor adjustment according to embodiments of the present invention.

Fig. 94A-94C illustrate calibration factor adjustment according to embodiments of the present invention.

FIG. 95 shows an illustrative example of initial decay in Cdl according to an embodiment of the invention.

FIG. 96 illustrates the effect of the removal of the non-faradaic current on Isig, according to an embodiment of the present invention.

FIG. 97A shows a calibration factor prior to removal of non-faradaic current of two working electrodes, in accordance with an embodiment of the present invention.

FIG. 97B shows calibration factors after removal of non-faradaic current of two working electrodes, in accordance with an embodiment of the invention.

FIGS. 98A and 98B illustrate the effect on MARD of the removal of the non-faradaic current according to an embodiment of the present invention.

FIG. 99 is a graphical representation of double layer capacitance over time in accordance with an embodiment of the present invention.

The graph 100 shows the movement of the occurrence of Rmem + Rsol and the occurrence of high frequency semicircles during a sensitivity loss according to an embodiment of the present invention.

FIG. 101A shows a flow diagram for detection of sensitivity loss using combinational logic, according to an embodiment of the present invention.

FIG. 101B shows a flow diagram for detection of sensitivity loss using combinational logic, in accordance with another embodiment of the present invention.

FIG. 102 shows an illustrative method of distinguishing between new and used sensors using the Nyquist slope as a marker in accordance with an embodiment of the invention.

103A-103C show illustrative examples of Nyquist curves having different lengths for different sensor configurations, in accordance with embodiments of the invention.

Fig. 104 shows the length of the nyquist curve over time for the sensors of fig. 103A-103C.

FIG. 105 shows a flow diagram for blanking sensor data or terminating a sensor, according to an embodiment of the invention.

FIG. 106 shows a flow diagram of sensor termination according to an embodiment of the invention.

Fig. 107 shows a flow chart of the detection of a dip in a signal according to an embodiment of the invention.

FIG. 108A shows Isig and V according to an embodiment of the invention_cntrFig. 108B shows the change in glucose over time according to an embodiment of the invention.

FIG. 109A shows a calibration ratio over time, and FIG. 109B shows a glucose over time, according to an embodiment of the invention.

Fig. 110A and 110B illustrate the trend of the calibration factor over time, according to an embodiment of the present invention.

Fig. 111 shows a flow chart of First Day Calibration (FDC) according to an embodiment of the invention.

FIG. 112 illustrates a flow diagram for EIS based calibration according to an embodiment of the invention.

Fig. 113 shows a flowchart of a conventional calibration method.

FIG. 114 shows a calibration flow diagram according to an embodiment of the invention.

FIG. 115 illustrates a calibration flow diagram according to other embodiments of the invention.

FIG. 116 illustrates a calibration flow diagram according to yet other embodiments of the invention.

FIG. 117 shows a calibration flow diagram according to other embodiments of the invention.

FIG. 118 shows a table of comparison MARD values calculated based on an embodiment of the present invention.

FIG. 119 shows a flow diagram for computing raw fusion weights, according to an embodiment of the invention.

Fig. 120 shows a Sensor Glucose (SG) fusion logic diagram according to an embodiment of the invention.

FIG. 121A shows a glucose sensor stack according to an embodiment of the invention.

FIG. 121B shows an oxygen sensor stack according to an embodiment of the invention.

FIG. 122 illustrates a dual flexure system according to an embodiment of the present invention.

Detailed Description

In the following description, reference is made to the accompanying drawings, which form a part hereof and which illustrate several embodiments of the present invention. It is to be understood that other embodiments may be utilized and structural and operational changes may be made without departing from the scope of the present invention.

The present invention is described below with reference to flowchart illustrations of methods, systems, apparatus, devices, programming products, and computer program products. It will be understood that each block of the flowchart illustrations, and combinations of blocks in the flowchart illustrations, can be implemented by programming instructions, including computer program instructions (such as any of the menu screens described in the figures). These computer program instructions may be loaded onto a computer or other programmable data processing apparatus (e.g., a controller, microcontroller, or processor in sensor electronics) to produce a machine, such that the instructions which execute on the computer or other programmable data processing apparatus create instructions for implementing the functions specified in the flowchart block or blocks. These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. The computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks and/or menu presented herein. The programming instructions may also be stored in and/or implemented by electronic circuitry, including Integrated Circuits (ICs) and Application Specific Integrated Circuits (ASICs) used in conjunction with the sensor devices, apparatus, and systems.

FIG. 1 is a perspective view of a subcutaneous sensor insertion device and a block diagram of sensor electronics according to an embodiment of the present invention. As shown in fig. 1, a subcutaneous sensor set 10 is provided for subcutaneous placement of a movable portion (e.g., see fig. 2) or the like of a flexible sensor 12 at a selected location within a user's body. The subcutaneous or transcutaneous portion of the sensor set 10 includes a hollow slotted insertion needle 14 and a cannula 16. The needle 14 is used to facilitate quick and easy subcutaneous placement of the cannula 16 at a subcutaneous insertion site. Inside the cannula 16 is a sensing portion 18 of the sensor 12 for exposing one or more sensor electrodes 20 to the body fluid of the user through a window 22 formed in the cannula 16. In one embodiment of the present invention, the one or more sensor electrodes 20 may comprise a counter electrode, a reference electrode, and one or more working electrodes. After insertion, the insertion needle 14 is withdrawn to leave the cannula 16 with the sensing portion 18 and the sensor electrode 20 in the selected insertion position.

In certain embodiments, subcutaneous sensor set 10 facilitates precise placement of flexible thin film electrochemical sensors 12 for monitoring a particular blood parameter type indicative of a user's condition. The sensor 12 monitors glucose levels in the body and may be used in conjunction with, for example, U.S. patent No. 4,562,751; 4,678,408 No; an external or implantable automatic or semi-automatic drug infusion pump as described in 4,685,903 or 4,573,994 is used in combination to control insulin delivery to a diabetic patient.

Particular embodiments of the flexible electrochemical sensor 12 are constructed in accordance with thin film shielding techniques to include an elongated thin film conductor embedded or encapsulated between a selected insulating material (e.g., a polyimide film or sheet of polyimide) and the membrane. When the sensing portion 18 (or active portion) of the sensor 12 is placed subcutaneously in an insertion site, the sensor electrodes 20 at the distal end of the sensing portion 18 are exposed through one of the insulating layers for direct contact with the patient's blood or other bodily fluids. The sensing portion 18 is connected to a connection portion 24 which terminates in a conductive contact pad or the like, which is also exposed through one of the insulating layers. In alternative embodiments, other types of implantable sensors may be used, such as chemical-based sensors, optical-based sensors, and the like.

As is known in the art, the connection portion 24 and contact pads are typically adapted for direct electrical connection to a suitable monitor or sensor electronics 100 to monitor the condition of the user in response to signals obtained from the sensor electrodes 20. Further description of flexible film sensors of this general type may be found, for example, in U.S. Pat. No. 5,391,250, which is incorporated herein by reference. The connection portion 24 may be conveniently electrically connected to the monitor or sensor electronics 100 or through a connector block 28 (or the like) as shown and described, for example, in U.S. patent No. 5,482,473, which is also incorporated herein by reference. Thus, the subcutaneous sensor set 10 may be configured or formed to work with wired or wireless feature monitoring systems in accordance with embodiments of the present invention.

Sensor electrodes 20 may be used for a variety of sensing applications and may be configured in a variety of ways. For example, the sensor electrode 20 may be used in physiological parameter sensing applications where a type of biomolecule is used as a catalyst. For example, the sensor electrode 20 may be used in a glucose and oxygen sensor having glucose oxidase enzyme (GOx) that catalyzes a reaction with the sensor electrode 20. This reaction produces gluconic acid (C)₆H₁₂O₇) And hydrogen peroxide (H)₂O₂) Proportional to the amount of glucose present. Thus, as will be described in further detail below, this type of sensor may be referred to as a "peroxide-based sensor" because the glucose measurement is achieved by measuring the amount of hydrogen peroxide produced.

The sensor electrodes 20, along with biomolecules or some other catalyst, may be placed within the human body in a vascular or non-vascular environment. For example, the sensor electrodes 20 and biomolecules may be placed in a vein and subjected to blood flow, or may be placed in the subcutaneous or peritoneal region of the human body.

The monitor 100 may also be referred to as the sensor electronics 100. The monitor 100 may include a power supply 110, a sensor interface 122, processing electronics 124, and data formatting electronics 128. The monitor 100 may be coupled to the sensor suite 10 with a cable 102 through a connector of the connector block 28 electrically coupled to the connection portion 24. In an alternative embodiment, the cable may be omitted. In this embodiment of the invention, the monitor 100 may comprise a suitable connector for direct connection to the connection portion 104 of the sensor suite 10. The sensor suite 10 may be modified such that the connector portion 104 is located at a different location (e.g., at the top of the sensor suite) to facilitate placement of the monitor 100 on the sensor suite.

In embodiments of the present invention, sensor interface 122, processing electronics 124, and data formatting electronics 128 are formed as separate semiconductor chips, however, alternative embodiments may combine various semiconductor chips into a single or multiple custom semiconductor chips. The sensor interface 122 is connected to the cable 102, which is connected to the sensor group 10.

The power source 110 may be a battery. The battery may comprise three series of silver oxide 357 cells. In alternative embodiments, different battery chemistries may be used, such as lithium-based chemistries, alkaline batteries, nickel metal hydride, etc., and different numbers of batteries may be used. The monitor 100 provides power to the sensor suite through the cable 102 and cable connector 104 by a power supply 110. In one embodiment of the invention, the power is a voltage provided to the sensor group 10. In one embodiment of the invention, the power is a current provided to the sensor group 10. In one embodiment of the invention, the power is a voltage that is provided to the sensor group 10 at a particular voltage.

Fig. 2A and 2B illustrate an implantable sensor and electronics for driving the implantable sensor according to an embodiment of the invention. Fig. 2A shows a substrate 220 having two sides, wherein a first side 222 comprises an electrode arrangement and a second side 224 comprises electronic circuitry. As shown in fig. 2A, the first side 222 of the substrate includes two counter electrode working electrode pairs 240, 242, 244, 246 on opposite sides of a reference electrode 248. The second side 224 of the substrate includes electronic circuitry. As shown, the electronic circuitry may be enclosed in a hermetically sealed enclosure 226, providing a protective enclosure for the electronic circuitry. This allows the sensor substrate 220 to be inserted into a vascular environment or other environment that may subject the electronic circuitry to fluids. By sealing the electronic circuitry in a hermetically sealed enclosure 226, the electronic circuitry can operate without risk of being shorted by surrounding fluids. Also shown in fig. 2A is a pad 228 to which input and output lines of the electronic circuitry may be connected. The electronic circuitry itself may be fabricated in a variety of ways. According to one embodiment of the invention, the electronic circuitry may be fabricated as an integrated circuit using techniques common in the industry.

FIG. 2B shows an overall block diagram of an electronic circuit for sensing sensor output according to an embodiment of the invention. At least one pair of sensor electrodes 310 may interface to a data converter 312, the output of which may interface to a counter 314. Counter 314 may be controlled by control logic 316. The output of counter 314 may be connected to line interface 318. Line interface 318 may be connected to input and output lines 320 and may also be connected to control logic 316. Input and output lines 320 may also be connected to a power rectifier 322.

Sensor electrodes 310 may be used for a variety of sensing applications and may be configured in a variety of ways. For example, sensor electrode 310 may be used in physiological parameter sensing applications where a type of biomolecule is used as a catalyst. For example, the sensor electrode 310 may be used in a glucose and oxygen sensor having glucose oxidase enzyme (GOx) that catalyzes a reaction with the sensor electrode 310. The sensor electrodes 310, along with biomolecules or some other catalyst, may be placed within the human body in a vascular or non-vascular environment. For example, the sensor electrodes 310 and biomolecules may be placed in a vein and subjected to blood flow.

FIG. 3 shows a block diagram of sensor electronics and a sensor including multiple electrodes, in accordance with an embodiment of the invention. The sensor group or sensor system 350 includes sensors 355 and sensor electronics 360. Sensor 355 includes counter electrode 365, reference electrode 370, and working electrode 375. The sensor electronics 360 includes a power supply 380, a regulator 385, a signal processor 390, a measurement processor 395, and a display/transmission module 397. Power supply 380 provides power (in the form of voltage, current, or voltage containing current) to regulator 385. The regulator 385 transmits the regulated voltage to the sensor 355. In one embodiment of the invention, the regulator 385 transmits a voltage to the counter electrode 365 of the sensor 355.

The sensor 355 generates a sensor signal indicative of the concentration of the physiological characteristic being measured. For example, the sensor signal may be indicative of a blood glucose reading. In one embodiment of the invention using a subcutaneous sensor, the sensor signal may be indicative of the level of hydrogen peroxide in the subject. In one embodiment of the invention using a blood or cranial sensor, the amount of oxygen is measured by the sensor and is represented by the sensor signal. In one embodiment of the invention using an implantable or long-term sensor, the sensor signal may be indicative of the oxygen level in the subject. The sensor signal may be measured at working electrode 375. In one embodiment of the invention, the sensor signal may be a current measured at the working electrode. In one embodiment of the invention, the sensor signal may be a voltage measured at the working electrode.

After measuring the sensor signal at the sensor 355 (e.g., working electrode), the signal processor 390 receives the sensor signal (e.g., measured current or voltage). The signal processor 390 processes the sensor signals and generates processed sensor signals. The measurement processor 395 receives the processed sensor signal and calibrates the processed sensor signal with the reference value. In one embodiment of the invention, the reference value is stored in a reference memory and provided to the measurement processor 395. The measurement processor 395 generates sensor measurements. The sensor measurements may be stored in a measurement memory (not shown). The sensor measurements may be sent to a display/transmission device for display on a display in the housing with the sensor electronics, or transmitted to an external device.

The sensor electronics 360 may be a monitor that includes a display for displaying a reading of the physiological characteristic. The sensor electronics 360 may also be installed in a desktop computer, pager, television including communication capabilities, laptop computer, server, network computer, Personal Digital Assistant (PDA), portable telephone including computer functionality, infusion pump including a display, glucose sensor including a display, and/or infusion pump/glucose sensor combination. The sensor electronics 360 may be housed in a cellular phone, a smart phone, a network device, a home network device, and/or other devices connected to a home network.

Fig. 4 shows an alternative embodiment that includes a sensor and sensor electronics. The sensor group or sensor system 400 includes sensor electronics 360 and sensors 355. The sensor includes a counter electrode 365, a reference electrode 370, and a working electrode 375. The sensor electronics 360 includes a microcontroller 410 and a digital-to-analog converter (DAC) 420. The sensor electronics 360 may also include a current-to-frequency converter (I/F converter) 430.

The microcontroller 410 includes software program code or programmable logic that, when executed, causes the microcontroller 410 to transmit a signal to the DAC420, wherein the signal represents a voltage level or voltage value to be applied to the sensor 355. The DAC420 receives the signal and generates a voltage value at a level indicated by the microcontroller 410. In embodiments of the present invention, microcontroller 410 may change the representation of the voltage level of the signal frequently or infrequently. Illustratively, the signal from the microcontroller 410 may instruct the DAC420 to apply a first voltage value for one second and a second voltage value for two seconds.

The sensor 355 may receive a voltage level or value. In one embodiment of the invention, the counter electrode 365 may receive the output of an operational amplifier having as inputs the reference voltage and the voltage value from the DAC 420. Application of the voltage level causes the sensor 355 to generate a sensor signal indicative of the concentration of the physiological characteristic being measured. In one embodiment of the invention, the microcontroller 410 may measure the sensor signal (e.g., current value) from the working electrode. Illustratively, the sensor signal measurement circuit 431 may measure the sensor signal. In one embodiment of the present invention, the sensor signal measurement circuit 431 may include a resistor, and a current may flow through the resistor to measure the value of the sensor signal. In one embodiment of the present invention, the sensor signal may be a current level signal and the sensor signal measurement circuit 431 may be a current-to-frequency (I/F) converter 430. The current-to-frequency converter 430 may measure the sensor signal from the current reading, convert the sensor signal to a frequency-based sensor signal, and transmit the frequency-based sensor signal to the microcontroller 410. In embodiments of the present invention, the microcontroller 410 is able to more easily receive frequency-based sensor signals than non-frequency-based sensor signals. The microcontroller 410 receives the frequency-based or non-frequency-based sensor signal and determines a physiological characteristic value (e.g., blood glucose level) of the subject. The microcontroller 410 may contain program code that, when executed or run, is capable of receiving the sensor signal and converting the sensor signal into a physiological characteristic value. In one embodiment, the microcontroller 410 may convert the sensor signal to a blood glucose level. In some embodiments, microcontroller 410 may utilize measurements stored in internal memory to determine the subject's blood glucose level. In some embodiments, the microcontroller 410 may utilize measurements stored in a memory external to the microcontroller 410 to assist in determining the subject's blood glucose level.

After the microcontroller 410 determines the physiological characteristic value, the microcontroller 410 may store the measured values of the physiological characteristic value for a plurality of time periods. For example, blood glucose values may be sent from the sensor to the microcontroller 410 every one or five seconds, and the microcontroller may save the sensor's measurements of BG readings for five or ten minutes. The microcontroller 410 can transmit the measured value of the physiological characteristic value to a display on the sensor electronics 360. For example, the sensor electronics 360 may be a monitor that includes a display that provides a blood glucose reading to the subject. In one embodiment, the microcontroller 410 may communicate the measured value of the physiological characteristic value to an output interface of the microcontroller 410. The output interface of the microcontroller 410 may transmit the measured value of the physiological characteristic value (e.g., blood glucose value) to an external device (e.g., an infusion pump, a combination infusion pump/glucose meter, a computer, a personal digital assistant, a pager, a network appliance, a server, a cellular telephone, or any computing device).

FIG. 5 shows an electronic block diagram of a sensor electrode and a voltage applied to the sensor electrode according to one embodiment. In the embodiment shown in FIG. 5, an operational amplifier 530 or other servo control device may be connected to the sensor electrode 510 via a circuit/electrode interface 538. Operational amplifier 530, which utilizes feedback through the sensor electrode, attempts to maintain a specified voltage (which the DAC may wish to apply) between reference electrode 532 and working electrode 534 by adjusting the voltage at counter electrode 536. Current can then flow from the counter electrode 536 to the working electrode 534. Such current may be measured to determine the electrochemical reaction between the sensor electrode 510 and the biomolecules of the sensor that have been placed in the vicinity of the sensor electrode 510 and used as a catalyst. The circuitry disclosed in fig. 5 may be used for long-term or implantable sensors, or may be used for short-term or subcutaneous sensors.

In long-term sensor embodiments where glucose oxidase (GOx) is used as the catalyst in the sensor, current can flow from the counter electrode 536 to the working electrode 534 only if oxygen is present near the enzyme and sensor electrode 510. Illustratively, if the voltage set at the reference electrode 532 is held at about 0.5 volts, the amount of current flowing from the counter electrode 536 to the working electrode 534 has a fairly linear unit slope relationship with the amount of enzyme and oxygen present in the region surrounding the electrodes. Thus, by maintaining reference electrode 532 at about 0.5 volts and using this region of the current-voltage curve to vary the blood oxygen level, the accuracy of determining the amount of oxygen in the blood can be improved. Different embodiments may utilize different sensors with biomolecules other than glucose oxidase and therefore voltages other than 0.5 volts may be set at the reference electrode.

As described above, during initial implantation or insertion of the sensor 510, the sensor 510 may provide inaccurate readings due to adjustments made to the sensor by the subject and electrochemical byproducts caused by the catalyst used in the sensor. Many sensors require a period of stability in order for sensor 510 to provide an accurate reading of a physiological parameter of a subject. During the stabilization period, the sensor 510 does not provide an accurate blood glucose measurement. Users and manufacturers of sensors may wish to improve the stability time frame of the sensor so that the sensor can be used quickly after insertion into the subject's body or subcutaneous layer of the subject.

In previous sensor electrode systems, the stabilization period or time range may be in the range of one hour to three hours. To reduce the settling period or time frame and increase the timeliness of the sensor accuracy, the sensor (or electrodes of the sensor) may be subjected to multiple pulses rather than applying one pulse followed by another voltage. Fig. 6A illustrates one method of applying pulses during a settling time range to reduce the settling time range. In this embodiment, the voltage application means applies 600 a first voltage to the electrodes for a first time or a first period of time. In one embodiment, the first voltage may be a DC constant voltage. This causes an anode current to be generated. In an alternative embodiment, a digital-to-analog converter or another voltage source may supply a voltage to the electrodes for a first period of time. Anodic current means that electrons are pushed towards the electrode where the voltage is applied. In some embodiments, the applying means may apply a current rather than a voltage. In embodiments where a voltage is applied to the sensor, the voltage regulator may wait (i.e., not apply a voltage) for a second time, time range, or period 605 after applying the first voltage to the electrodes. In other words, the voltage applying device waits until the second period of time elapses. The absence of an applied voltage generates a cathodic current that causes the non-voltage applied electrodes to pick up electrons. The applying 610 of the first voltage to the electrode for the first period of time and then the not applying the voltage for the second period of time is repeated over a plurality of iterations. This may be referred to as an anode and cathode cycle. In one embodiment, the total number of iterations of the stabilization method is three, i.e. three voltages are applied for a first period of time, followed by a second period of time, respectively, without voltages. In one embodiment, the first voltage may be 1.07 volts. In further embodiments, the first voltage may be 0.535 volts, or may be about 0.7 volts.

Repeated application and non-application of voltage causes the sensor (and electrodes) to undergo anode-cathode cycling. The anode-cathode cycle results in a reduction in electrochemical by-products that are generated by the patient's body in response to insertion of the sensor or implantation of the sensor. Electrochemical by-products result in background currents, which lead to inaccurate measurements of physiological parameters of a subject. Under certain operating conditions, electrochemical by-products can be eliminated. Under other operating conditions, electrochemical byproducts may be reduced or significantly reduced. Successful stabilization allows for anode-cathode cycling to reach equilibrium, significant reduction in electrochemical by-products, and minimization of background current.

In one embodiment, the first voltage applied to the sensor electrode may be a positive voltage. In an alternative embodiment, the applied first voltage may be a negative voltage. Further, the first voltage may be applied to the working electrode. In some embodiments, the first voltage may be applied to the counter electrode or the reference electrode.

In some embodiments, the duration of the voltage pulse and the duration of the non-application of voltage may be equal, for example, three minutes at a time. In other embodiments, the duration of the voltage application or voltage pulse may be different values, e.g., the first time and the second time may be different. In one embodiment, the first time period may be five minutes and the waiting time period may be two minutes. In one variation, the first time period may be two minutes and the waiting period (or second time period) may be five minutes. In other words, the duration of applying the first voltage may be two minutes, and no voltage may be applied for five minutes. This temporal atmosphere is intended to be illustrative only and not limiting. For example, the first time range may be two, three, five, or ten minutes, and the second time range may be five minutes, ten minutes, twenty minutes, or the like. The time ranges (e.g., the first time and the second time) may depend on the unique characteristics of the different electrodes, sensors, and/or patient physiological characteristics.

In conjunction with the foregoing, more or less than three pulses may be utilized to stabilize the glucose sensor. In other words, the number of iterations may be greater than 3 or less than 3. For example, four voltage pulses (e.g., no voltage after a high voltage) may be applied to one of the electrodes, or six voltage pulses may be applied to one of the electrodes.

Illustratively, three consecutive pulses of 1.07 volts (followed by corresponding latencies) may be sufficient for a subcutaneously implanted sensor. In one embodiment, three 0.7 volt continuous voltage pulses may be used. For sensors implanted in blood or cranial fluid (e.g., long-term or permanent sensors), the three consecutive pulses may have higher or lower voltage values (whether negative or positive). Further, more than three pulses (e.g., five, eight, twelve) may be utilized to generate an anode-cathode cycle between anode and cathode currents in any subcutaneous, blood, or cranial fluid sensor.

FIG. 6B illustrates a method of stabilizing a sensor according to an embodiment of the invention. In the embodiment shown in fig. 6B, the voltage application device may apply 630 a first voltage to the sensor for a first time to initiate an anodic cycle at the electrodes of the sensor. The voltage applying means may be a DC power supply, a digital-to-analog converter or a voltage regulator. After the first period of time has elapsed, a second voltage is applied 635 to the sensor for a second time to initiate a cathodic cycle at the electrodes of the sensor. Illustratively, as shown in the method of FIG. 6A, a different voltage (different than the first voltage) is applied to the sensor during the second time range, rather than no voltage being applied. In an embodiment of the present invention, applying the first voltage for the first time and applying the second voltage for the second time is repeated 640 for a plurality of iterations. In certain embodiments, applying the first voltage for the first time and applying the second voltage for the second time may each be applied within a stable time range (e.g., 10 minutes, 15 minutes, or 20 minutes) rather than multiple iterations. The stabilization time range is the entire time range of the stabilization sequence, e.g., until the sensor (and electrodes) are stabilized. The benefits of this stabilization method are faster sensor operation, less background current (in other words, some background current is suppressed), and better glucose response.

In one embodiment, the first voltage may be 0.535 volts applied for five minutes, the second voltage may be 1.070 volts applied for two minutes, the first voltage of 0.535 volts applied for five minutes, the second voltage of 1.070 volts applied for two minutes, the first voltage of 0.535 volts applied for five minutes, and the second voltage of 1.070 volts applied for two minutes. In other words, in this embodiment, there are three iterations of the voltage pulse scheme. The second time range of the pulsing method may vary, for example the time range for applying the second voltage may extend from two minutes to five, ten, fifteen or twenty minutes. Further, after applying three iterations in this embodiment of the invention, a nominal operating voltage of 0.535 volts may be applied.

1.070 volts and 0.535 volts are illustrative values. Other voltage values may be selected based on various factors. These factors may include the type of enzyme used in the sensor, the membrane used in the sensor, the duty cycle of the sensor, the length of the pulse, and/or the amplitude of the pulse. Under certain operating conditions, the first voltage may be in the range of 1.00 to 1.09 volts and the second voltage may be in the range of 0.510 to 0.565 volts. In other operating embodiments, the range including the first voltage and the second voltage may have a higher range, such as 0.3 volts, 0.6 volts, 0.9 volts, depending on the voltage sensitivity of the electrodes in the sensor. Under other operating conditions, the voltage may be in the range of 0.8 volts to 1.34 volts, while other voltages may be in the range of 0.335 to 0.735. Under other operating conditions, the range of higher voltages may be less than the range of lower voltages. Illustratively, the higher voltage may be in the range of 0.9 to 1.09 volts and the lower voltage may be in the range of 0.235 to 0.835 volts.

In one embodiment, the first voltage and the second voltage may be positive voltages or, alternatively, may be negative voltages in other embodiments. In another embodiment, the first voltage may be positive and the second voltage may be negative, or alternatively, the first voltage may be negative and the second voltage may be positive. The first voltage may be a different voltage level for each iteration. Further, the first voltage may be a d.c. constant voltage. Further, the first voltage may be a ramp voltage, a sinusoidal voltage, a stepped voltage, or other commonly used voltage waveforms. In one embodiment, the second voltage may be a d.c. constant voltage, a ramp voltage, a sinusoidal voltage, a stepped voltage, or other commonly used voltage waveform. In an alternative embodiment, the first voltage or the second voltage may be an AC signal that relies on a DC waveform. In general, the first voltage may be one type of voltage (e.g., a ramp voltage), the second voltage may be a second type of voltage (e.g., a sinusoidal voltage), and the first voltage (or the second voltage) may have a different waveform for each iteration. For example, if there are three cycles in the stabilization method, in the first cycle, the first voltage may be a ramp voltage, in the second cycle, the first voltage may be a constant voltage, and in the third cycle, the first voltage may be a sinusoidal voltage.

In one embodiment, the duration of the first time range and the duration of the second time range may have the same value, or alternatively, the duration of the first time range and the duration of the second time range may have different values. For example, the duration of the first time range may be two minutes, the duration of the second time range may be five minutes, and the number of iterations may be three. As described above, the stabilization method may involve multiple iterations. In various embodiments, the duration of each first time range may vary and the duration of each second time range may vary during different iterations of the stabilization method. Illustratively, during the first iteration of the anode-cathode cycle, the first time range may be 2 minutes and the second time range may be 5 minutes. During the second iteration, the first time range may be 1 minute and the second time range may be 3 minutes. During the third iteration, the first time range may be 3 minutes and the second time range may be 10 minutes.

In one embodiment, a first voltage of 0.535 volts is applied to the electrodes in the sensor for two minutes to initiate the anodic cycle, and then a second voltage of 1.07 volts is applied to the electrodes for five minutes to initiate the cathodic cycle. The first voltage of 0.535 volts was then applied again for two minutes to initiate the anodic cycle and a second voltage of 1.07 volts was applied to the sensor for five minutes. In the third iteration, 0.535 volts was applied for two minutes to initiate the anodic cycle, followed by 1.07 volts for five minutes. For example, when the sensor provides a reading of a physiological characteristic of the subject, the voltage applied to the sensor over the actual operating time range of the sensor is 0.535.

Shorter duration voltage pulses may be used in the embodiments of fig. 6A and 6B. A shorter duration voltage pulse may be used to apply the first voltage, the second voltage, or both. In one embodiment, the amplitude of the shorter duration voltage pulses of the first voltage is-1.07 volts and the amplitude of the shorter duration voltage pulses of the second voltage is about half the high amplitude, e.g., -0.535 volts. Alternatively, the amplitude of the shorter duration pulses of the first voltage may be 0.535 volts and the amplitude of the shorter duration pulses of the second voltage may be 1.07 volts.

In embodiments utilizing short duration pulses, the voltage may not be applied continuously throughout the first time period. Alternatively, the voltage application means may transmit a plurality of short duration pulses during the first time period. In other words, during the first time period, a plurality of small width voltage pulses or short duration voltage pulses may be applied to the electrodes of the sensor. Each small-width or short-duration pulse may have a width of a few milliseconds. Illustratively, the pulse width may be 30 milliseconds, 50 milliseconds, 70 milliseconds, or 200 milliseconds. These values are intended to be illustrative, not limiting. In one embodiment (such as the one shown in fig. 6A), these short duration pulses are applied to the sensor (electrodes) for a first period of time, and then no voltage is applied for a second period of time.

Each short duration pulse may have the same duration during the first time period. For example, each short duration voltage pulse may have a time width of 50 milliseconds, and each pulse delay between pulses may be 950 milliseconds. In this example, if two minutes is the measurement time of the first time range, 120 short duration voltage pulses may be applied to the sensor. Alternatively, each short-duration voltage pulse may have a different duration. In various embodiments, each short-duration voltage pulse may have the same amplitude value, or may have a different amplitude value. By utilizing short duration voltage pulses rather than continuously applying a voltage to the sensor, the same anodic and cathodic cycling can occur, and the sensor (e.g., electrodes) experience less total energy or charge over time. Using a short duration voltage pulse uses less power than applying a continuous voltage to the electrodes because less energy is applied to the sensor (and therefore to the electrodes).

FIG. 6C illustrates the use of feedback in a stabilization sensor according to one embodiment. The sensor system may contain a feedback mechanism for determining whether additional pulses are needed to stabilize the sensor. In one embodiment, the sensor signal generated by the electrode (e.g., working electrode) may be analyzed to determine whether the sensor signal is stable. A first voltage is applied 630 to the electrodes for a first time range to initiate an anodic cycle. A second voltage 635 is applied to the electrodes for a second time range to initiate the cathodic cycle. In embodiments of the invention, the analysis module may analyze the sensor signal (e.g., the current emitted by the sensor signal, the resistance at a particular point in the sensor, the impedance at a particular node in the sensor) and determine whether a threshold measurement 637 has been reached (e.g., by comparing to the threshold measurement to determine whether the sensor is providing an accurate reading). If the sensor reading is determined to be accurate, indicating that the electrode (and thus the sensor) has reached a plateau 642, no further application of the first voltage and/or the second voltage will occur. If stabilization is not achieved, additional anode/cathode cycles can be initiated by applying 630 a first voltage to the electrodes for a first period of time, and then applying 635 a second voltage to the electrodes for a second period of time.

In some embodiments, the analysis module may be used after three anodic/cathodic cycles of applying the first voltage and the second voltage to the electrodes of the sensor. However, as shown in fig. 6C, after the first voltage and the second voltage are applied once, an analysis module may be employed.

The analysis module may be used to measure the voltage emitted upon introduction of a current across one or both electrodes. The analysis module may monitor the voltage level at the electrode or at the receive level. In one embodiment, if the voltage level is above a certain threshold, this may mean that the sensor is stable. In one embodiment, if the voltage level drops below a threshold level, this may indicate that the sensor is stable and ready to provide a reading. In one embodiment, the current may be directed to an electrode or through a pair of electrodes. The analysis module may monitor the level of current emitted from the electrode. In this embodiment, the analysis module can monitor the current if the current differs from the sensor signal current by an order of magnitude. If the current is above or below the current threshold, this may indicate that the sensor is stable.

In one embodiment of the invention, the analysis module may measure the impedance between two electrodes of the sensor. The analysis module may compare the impedance to a threshold or target impedance value and may stabilize the sensor (and thus the sensor signal) if the measured impedance is below the target or threshold impedance. In one embodiment, the analysis module may measure the resistance between two electrodes of the sensor. In this embodiment of the invention, if the analysis module compares the resistance to a threshold or target resistance value and the measured resistance value is less than the threshold or target resistance value, the analysis module may determine that the sensor is stable and may utilize the sensor signal.

FIG. 7 illustrates the effect of stabilizing a sensor according to an embodiment of the present invention. Line 705 represents the glucose sensor reading of the glucose sensor, where the previous single pulse stabilization method was utilized. Line 710 represents a glucose reading for a glucose sensor in which three voltage pulses are applied (e.g., 3 voltage pulses of 2 minutes duration followed by 5 minutes of no voltage applied, respectively). The x-axis 715 represents an amount of time. Points 720, 725, 730 and 735 represent measured glucose readings taken with a fingertip and then entered into a glucose meter. As shown, the previous single pulse stabilization method takes about 1 hour and 30 minutes to stabilize to the desired glucose reading, e.g., 100 units. In contrast, the three-pulse stabilization method takes only about 15 minutes to stabilize the glucose sensor and greatly improves the stabilization time range.

FIG. 8A shows a block diagram of sensor electronics and a sensor including a voltage generating device. The voltage generating or applying device 810 comprises an electronic device, logic, or circuit that generates voltage pulses. The sensor electronics 360 may also include an input device 820 for receiving reference values and other useful data. In one embodiment, the sensor electronics may include a measurement memory 830 for storing sensor measurements. In this embodiment, power supply 380 may provide power to the sensor electronics. The power supply 380 may provide power to a regulator 385 that provides a regulated voltage to the voltage generating or applying device 810. Connection 811 indicates that in the illustrated embodiment, the connection couples or connects sensor 355 to sensor electronics 360.

In the embodiment shown in fig. 8A, the voltage generating or applying device 810 provides a voltage (e.g., a first voltage or a second voltage) to an input of the operational amplifier 840. The voltage generating or applying device 810 can also provide a voltage to the working electrode 375 of the sensor 355. The other input of the operational amplifier 840 is coupled to the reference electrode 370 of the sensor. The application of voltage from the voltage generation or application device 810 to the operational amplifier 840 drives the voltage measured at the counter electrode 365 to approach or equal the voltage applied at the working electrode 375. In one embodiment, a voltage generation or application device 810 can be used to apply a desired voltage between the counter electrode and the working electrode. This can be achieved by applying a fixed voltage directly to the counter electrode.

In one embodiment as shown in fig. 6A and 6B, the voltage generating device 810 generates a first voltage to be applied to the sensor during a first time range. The voltage generating device 810 transmits the first voltage to an operational amplifier 840 that drives the voltage at the counter electrode 365 of the sensor 355 to the first voltage. In some embodiments, the voltage generating device 810 can also transmit the first voltage directly to the counter electrode 365 of the sensor 355. In the embodiment shown in fig. 6A, the voltage generating device 810 then does not transmit the first voltage to the sensor 355 for the second time range. In other words, the voltage generating device 810 is turned off or disconnected. The voltage generating device 810 may be programmed to cycle through a number of iterations or a stable time range (e.g., twenty minutes) between applying the first voltage and not applying the voltage. Fig. 8B shows a voltage generation device for implementing this embodiment of the present invention. The voltage regulator 385 passes the regulated voltage to the voltage generating device 810. The control circuit 860 controls the closing and opening of the switch 850. If switch 850 is closed, a voltage is applied. If switch 850 is open, the voltage is not applied. The timer 865 provides a signal to the control circuit 860 to instruct the control circuit 860 to open and close the switch 850. The control circuit 860 contains logic that can instruct the circuit to open and close the switch 850 multiple times (to match the necessary iterations). In one embodiment, timer 865 may also transmit a stabilization signal to identify that the stabilization sequence has completed, i.e., that the stabilization time range has elapsed.

In one embodiment, the voltage generating device generates the first voltage for a first time range and generates the second voltage for a second time range. FIG. 8C shows a voltage generation device for generating two voltage values to implement this embodiment. In this embodiment, a two position switch 870 is used. Illustratively, if the first switch position 871 is turned on or closed by the timer 865 indicating the control circuit 860, the voltage generating device 810 generates the first voltage for a first time range. After the first voltage has been applied for the first time range, the timer sends a signal to the control circuit 860 indicating that the first time range has elapsed, and the control circuit 860 directs the switch 870 to move to the second position 872. When the switch 870 is in the second position 872, the regulated voltage is directed to a voltage step-down/buck converter 880 to reduce the regulated voltage to a smaller value. The smaller value is then delivered to the operational amplifier 840 for a second time range. After the timer 865 has signaled to the control circuit 860 that the second time range has elapsed, the control circuit 860 moves the switch 870 back to the first position. This will continue until the required number of iterations has been completed or a settling time range has elapsed. In one embodiment of the present invention, the sensor transmits the sensor signal 350 to the signal processor 390 after the sensor settling time range has elapsed.

Fig. 8D shows a voltage application device 810 for performing more complex voltage application to the sensor. The voltage applying means 810 may include a control means 860, a switch 890, a sinusoidal voltage generating means 891, a ramp voltage generating means 892, and a constant voltage generating means 893. In other embodiments, the voltage application may generate an AC wave on top of a DC signal or other various voltage pulse waveforms. In the embodiment shown in fig. 8D, the control device 860 may move the switch to one of three voltage generating systems 891 (sine), 892 (ramp), 893 (constant DC). This causes each voltage generation system to generate an identified voltage waveform. Under certain operating conditions (e.g., where three sinusoidal pulses are to be applied), the control device 860 may cause the switch 890 to connect the voltage from the voltage regulator 385 to the sinusoidal voltage generator 891 so that the voltage application device 810 generates a sinusoidal voltage. Under other operating conditions, for example, when a ramp voltage is applied to the sensor as a first voltage of a first of the three pulses, a sine wave voltage is applied to the sensor as a first voltage of a second of the three pulses, and a constant DC voltage is applied to the sensor as a first voltage of a third of the three pulses, the control device 860 may cause the switch 890 to move between connecting the voltage from the voltage generating or applying device 810 to the ramp voltage generating system 892, then to the sine voltage generating system 891, and then to the constant DC voltage generating system 893 during a first time range of the anode/cathode cycle. In this embodiment, control 860 may also direct or control switches to connect some of the voltage generating subsystems to the voltage from regulator 385 during a second time frame (e.g., during application of a second voltage).

FIG. 9A shows sensor electronics including a microcontroller for generating voltage pulses. Advanced sensor electronics may include microcontroller 410 (see fig. 4), digital-to-analog converter (DAC)420, operational amplifier 840, and sensor signal measurement circuit 431. In one embodiment, the sensor signal measurement circuit may be a current-to-frequency (I/F) converter 430. In the embodiment shown in fig. 9A, software or programmable logic in the microcontroller 410 provides instructions to transmit a signal to the DAC420, which in turn instructs the DAC420 to output a particular voltage to the operational amplifier 840. Microcontroller 410 may also be instructed to output a particular voltage to working electrode 375 as shown by line 911 in fig. 9A. As described above, applying a particular voltage to operational amplifier 840 and working electrode 375 may drive the voltage measured at the counter electrode to a particular voltage magnitude. In other words, the microcontroller 410 outputs a signal representative of a voltage or voltage waveform to be applied to the sensor 355 (e.g., an operational amplifier 840 coupled to the sensor 355). In an alternative embodiment, the fixed voltage may be set by applying a voltage directly from DAC420 between the reference electrode and working electrode 375. Similar results can be obtained by applying a voltage to each electrode (the difference being equal to the fixed voltage applied between the reference and working electrodes). Further, the fixed voltage may be set by applying a voltage between the reference electrode and the counter electrode. Under certain operating conditions, the microcontroller 410 may generate a pulse of a particular amplitude that the DAC420 understands to represent that a voltage of a particular amplitude is to be applied to the sensor. After the first time range, the microcontroller 410 outputs (by program or programmable logic) a second signal indicating that the DAC420 is not outputting a voltage (for the sensor electronics 360 operating according to the method described in fig. 6A) or outputting a second voltage (for the sensor electronics 360 operating according to the method described in fig. 6B). After the second time range has elapsed, the microcontroller 410 then repeats the cycle of sending a signal indicating the first voltage to be applied (for the first time range) and then sending a signal indicating that either no voltage is applied or the second voltage is to be applied (for the second time range).

Under other operating conditions, the microcontroller 410 may generate a signal to the DAC420 indicating the DAC output ramp voltage. Under other operating conditions, the microcontroller 410 may generate a signal to the DAC420 that indicates the voltage at which the DAC420 outputs an analog sinusoidal voltage. These signals may be incorporated into any of the pulsing methods discussed in the preceding paragraphs or earlier in this application. In one embodiment, the microcontroller 410 may generate instructions and/or pulse sequences that the DAC420 receives and interprets as meaning that a particular pulse sequence is to be applied. For example, the microcontroller 410 may transmit a sequence of instructions (via signals and/or pulses) that instruct the DAC420 to generate a constant voltage for a first iteration of a first time range, a ramp voltage for a first iteration of a second time range, a sinusoidal voltage for a second iteration of the first time range, and a square wave having two values for a second iteration of the second time range.

Microcontroller 410 may contain programmable logic or programs for continuing such a loop over a stable time range or over multiple iterations. Illustratively, the microcontroller 410 may contain counting logic for identifying when the first time range or the second time range has elapsed. Additionally, microcontroller 410 may contain counting logic for identifying that the settling time range has elapsed. After any of the foregoing time ranges have elapsed, the counting logic may instruct the microcontroller to send a new signal or stop transmitting signals to the DAC 420.

The use of microcontroller 410 allows for multiple voltage amplitudes to be applied in multiple sequences over multiple durations. In one embodiment of the present invention, microcontroller 410 may contain control logic or programming to instruct digital-to-analog converter 420 to transmit a voltage pulse of approximately 1.0 volt in amplitude for a first time period of 1 minute, then transmit a voltage pulse of approximately 0.5 volts for a second time period of 4 minutes, and repeat the cycle four times. In one embodiment, the microcontroller 420 may be programmed to transmit a signal that causes the DAC420 to apply a voltage pulse of the same magnitude for each first voltage in each iteration. The microcontroller 410 may be programmed to transmit a signal that causes the DAC to apply a voltage pulse of different amplitude for each first voltage in each iteration. In this embodiment, the microcontroller 410 may also be programmed to transmit a signal that causes the DAC420 to apply a voltage pulse of different magnitude for each second voltage in each iteration. Illustratively, the microcontroller 410 may be programmed to transmit signals to cause the DAC420 to apply a first voltage pulse of about 1.0 volt in a first iteration, a second voltage pulse of about 0.5 volt in the first iteration, a first voltage of 0.7 volt and a second voltage of 0.4 volt in a second iteration, and a first voltage of 1.2 volt and a second voltage of 0.8 volt in a third iteration.

The microcontroller 410 may also be programmed to instruct the DAC420 to provide a plurality of short-duration voltage pulses for a first time range. In this embodiment of the invention, rather than applying a voltage for the entire first time range (e.g., two minutes), multiple pulses of shorter duration may be applied to the sensor. In this embodiment, the microcontroller 410 may also be programmed to instruct the DAC420 to provide a plurality of short duration voltage pulses to the sensor for a second time range. Illustratively, the microcontroller 410 may send a signal that causes the DAC to apply a plurality of short duration voltage pulses, where the short duration is 50 milliseconds or 100 milliseconds. Between these short duration pulses, the DAC may apply no voltage, or the DAC may apply a minimum voltage. The microcontroller may cause the DAC420 to apply the short duration voltage pulses for a first time range (e.g., two minutes). The microcontroller 410 may then send a signal for a second time range that causes the DAC to apply no voltage or a short duration voltage pulse to the sensor at the magnitude of the second voltage, which may be 0.75 volts, for example, and the second time range may be 5 minutes. In one embodiment, the microcontroller 410 may send a signal to the DAC420 that causes the DAC420 to apply a different magnitude of voltage for each short duration pulse within the first time range and/or the second time range. In one embodiment, the microcontroller 410 may send a signal to the DAC420 that causes the DAC420 to apply a pattern of voltage amplitudes to the short-duration voltage pulses for either the first time range or the second time range. For example, the microcontroller may transmit one or more signals that instruct the DAC420 to apply thirty 20 millisecond pulses to the sensor during the first time range. Each of the thirty 20 millisecond pulses may have the same amplitude or may have different amplitudes. In this embodiment, the microcontroller 410 may instruct the DAC420 to apply the short duration pulse during the second time range, or may instruct the DAC420 to apply another voltage waveform during the second time range.

Although the disclosures in fig. 6-8 disclose applying a voltage, a current may also be applied to the sensor to initiate the stabilization process. Illustratively, in the embodiment shown in fig. 6B, a first current may be applied during a first time range to initiate an anodic or cathodic response, and a second current may be applied during a second time range to initiate an opposite anodic or cathodic response. The application of the first and second currents may last for a plurality of iterations or may continue within a stable time range. In one embodiment, the first current may be applied during a first time range and the first voltage may be applied during a second time range. In other words, one of the anodic or cathodic cycles may be triggered by a current applied to the sensor and the other of the anodic or cathodic cycles may be triggered by a voltage applied to the sensor. As described above, the applied current may be a constant current, a ramp current, a step pulse current, or a sinusoidal current. Under certain operating conditions, the current may be applied as a series of short duration pulses during a first time range.

FIG. 9B shows a sensor and sensor electronics with feedback during a stabilization period using an analysis module according to an embodiment of the invention. Fig. 9B incorporates an analysis module 950 into the sensor electronics 360. The analysis module 950 uses feedback from the sensors to determine whether the sensors are stable. In one embodiment, the microcontroller 410 may contain instructions or commands that control the DAC420 such that the DAC420 applies a voltage or current to a portion of the sensor 355. Fig. 9B illustrates that a voltage or current can be applied between reference electrode 370 and working electrode 375. However, the voltage or current may be applied between the electrodes or directly to one of the electrodes, and the invention should not be limited by the embodiment shown in FIG. 9B. Dashed line 955 shows the application of a voltage or current. The analysis module 950 may measure voltage, current, resistance, or impedance in the sensor 355. Fig. 9B illustrates the measurement being taken at working electrode 375, but this should not limit the invention as other embodiments may measure voltage, current, resistance or impedance between the electrodes of the sensor or directly at reference electrode 370 or counter electrode 365. The analysis module 950 may receive the measured voltage, current, resistance, or impedance and may compare the measured value to a stored value (e.g., a threshold). Dashed line 956 represents the analysis module 950 reading or measuring a voltage, current, resistance or impedance. Under certain operating conditions, if the measured voltage, current, resistance or impedance is above a threshold, the sensor is stable and the sensor signal provides an accurate reading of the patient's physiological condition. Under other operating conditions, the sensor stabilizes if the measured voltage, current, resistance or impedance is below a threshold. Under other operating conditions, the analysis module 950 may verify that the measured voltage, current, resistance, or impedance is stable over a particular time range (e.g., one or two minutes). This may indicate that the sensor 355 is stable and that the sensor signal is emitting an accurate measurement of a physiological parameter (e.g., blood glucose level) of the subject. After the analysis module 950 has determined that the sensor is stable and the sensor signal is providing an accurate measurement, the analysis module 950 may transmit a signal to the microcontroller 410 indicating that the sensor is stable and the microcontroller 410 may begin using the sensor signal or receiving the sensor signal from the sensor 355 (e.g., a sensor stable signal). This is indicated by dashed line 957.

FIG. 10 shows a block diagram of a sensor system including hydration electronics. The sensor system includes a connector 1010, a sensor 1012, and a monitor or sensor electronics 1025. Sensor 1012 includes electrodes 1020 and connecting portion 1024. In one embodiment, sensor 1012 may be connected to sensor electronics 1025 through connector 1010 and a cable. In other embodiments, sensor 1012 may be directly connected to sensor electronics 1025. In some embodiments, sensor 1012 may be incorporated into the same physical device as sensor electronics 1025. The monitor or sensor electronics 1025 may include a power supply 1030, a regulator 1035, a signal processor 1040, a measurement processor 1045, and a processor 1050. The monitor or sensor electronics 1025 may also include hydration detection circuitry 1060. Hydration detection circuitry 1060 interfaces with sensor 1012 to determine whether electrodes 1020 of sensor 1012 are sufficiently hydrated. If the electrode 1020 is not sufficiently hydrated, the electrode 1020 may not provide an accurate glucose reading, and therefore it is important to know when the electrode 1020 is sufficiently hydrated. Once the electrodes 1020 are sufficiently hydrated, accurate glucose readings can be obtained.

In the embodiment shown in FIG. 10, the hydration detection circuitry 1060 may include a delay or timer module 1065 and a connection detection module 1070. In embodiments utilizing short-term or subcutaneous sensors, the sensor electronics or monitor 1025 is connected to the sensor 1012 after the sensor 1012 has been inserted into the subcutaneous tissue. The connection detection module 1070 identifies that the sensor electronics 1025 has been connected to the sensor 1012 and sends a signal to the timer module 1065. This is illustrated in FIG. 10 by arrow 1084, which represents detector 1083 detecting a connection and sending a signal to connection detection module 1070 indicating that sensor 1012 has been connected to sensor electronics 1025. In embodiments using an implantable or long-term sensor, the connection detection module 1070 identifies that the implantable sensor has been inserted into the body. The timer module 1065 receives the connect signal and waits for a set or determined hydration time. Illustratively, the hydration time may be two minutes, five minutes, ten minutes, or 20 minutes. These examples are intended to be illustrative, not limiting. The time range is not necessarily a fixed number of minutes and may include any number of seconds. In one embodiment, after the timer module 1065 has waited for the set hydration time, the timer module 1065 may notify the processor 1050 that the sensor 1012 is hydrated by sending a hydration signal, as shown by line 1086.

In this embodiment, the processor 1050 may receive the hydration signal and begin utilizing the sensor signal (e.g., sensor measurements) only after the hydration signal has been received. In another embodiment, hydration detection circuitry 1060 may be coupled between the sensor (sensor electrode 1020) and the signal processor 1040. In this embodiment, the hydration detection circuitry 1060 may prevent sensor signals from being sent to the signal processor 1040 until the timer module 1065 has notified the hydration detection circuitry 1060 that the set hydration time has elapsed. This is illustrated by the dashed lines labeled with reference numerals 1080 and 1081. Illustratively, the timer module 1065 may transmit a connection signal to a switch (or transistor) to turn on the switch and advance the sensor signal to the signal processor 1040. In an alternative embodiment, the timer module 1065 may transmit a connection signal to turn on the switch 1088 (or close the switch 1088) in the hydration detection circuit 1060 to allow the voltage from the regulator 1035 to be applied to the sensor 1012 after the hydration time has elapsed. In other words, in this embodiment, the voltage from regulator 1035 is not applied to sensor 1012 until after the hydration time has elapsed.

FIG. 11 illustrates an embodiment that includes a mechanical switch to assist in determining hydration time. In one embodiment, a single housing may contain sensor assembly 1120 and sensor electronics 1125. In another embodiment, sensor assembly 1120 may be located in one housing and sensor electronics 1125 may be located in a separate housing, but sensor assembly 1120 and sensor electronics 1125 may be connected together. In this embodiment, the connection detection mechanism 1160 may be a mechanical switch. The mechanical switch may detect that sensor 1120 is physically connected to sensor electronics 1125. Timer circuit 1135 may also be activated when mechanical switch 1160 detects that sensor 1120 is connected to sensor electronics 1125. In other words, the mechanical switch may be closed and a signal may be communicated to the timer circuit 1135. Once the hydration time has elapsed, the timer circuit 1135 transmits a signal to the switch 1140 to allow the regulator 1035 to apply a voltage to the sensor 1120. In other words, the voltage is not applied until the hydration time has elapsed. In one embodiment, once the hydration time has elapsed, the current may replace the voltage applied to the sensor. In an alternative embodiment, power may first be applied to sensor 1120 when mechanical switch 1160 identifies that sensor 1120 has been physically connected to sensor electronics 1125. The power sent to the sensor 1120 causes a sensor signal to be output from the working electrode in the sensor 1120. The sensor signals may be measured and sent to the processor 1175. Processor 1175 may include a counter input. Under certain operating conditions, after a set hydration time has elapsed since the sensor signal was input to the processor 1175, the processor 1175 may begin processing the sensor signal as an accurate measurement of glucose in the subject. In other words, processor 1170 has received the sensor signal from potentiostat circuit 1170 for a certain time, but does not process the signal until an instruction is received from the processor's counter input identifying that the hydration time has elapsed. In one embodiment, potentiostat circuit 1170 may contain a current-to-frequency converter 1180. In this embodiment, the current-to-frequency converter 1180 may receive the sensor signal as a current value and may convert the current value to a frequency value that is easier for the processor 1175 to process.

The mechanical switch 1160 may also notify the processor 1175 when the sensor 1120 has been disconnected from the sensor electronics 1125. This is indicated by dashed line 1176 in fig. 11. This may cause processor 1170 to power down or otherwise reduce power to the various components, chips, and/or circuits of sensor electronics 1125. If sensor 1120 is not connected, the battery or power source may be drained if the components or circuitry of sensor electronics 1125 are in a powered state. Thus, if the mechanical switch 1160 detects that the sensor 1120 has been disconnected from the sensor electronics 1125, the mechanical switch may indicate this to the processor 1175, and the processor 1175 may power down or reduce the power provided to one or more of the electronic circuits, chips, or components of the sensor electronics 1125.

FIG. 12 illustrates an electrical method of detecting hydration according to one embodiment of the present invention. In one embodiment, an electrical detection mechanism for detecting sensor connection may be utilized. In this embodimentThe hydration detection electronics 1250 may include an AC power source 1255 and a detection circuit 1260. Hydration detection electronics 1250 can be located in the sensor electronics 1225. Sensor 1220 may include a counter electrode 1221, a reference electrode 1222, and a working electrode 1223. As shown in FIG. 12, an AC power source 1255 is coupled to a voltage setting device 1275, a reference electrode 1222 and a detection circuit 1260. In this embodiment, an AC signal from an AC power source is applied to the reference electrode connection, as shown by dashed line 1291 in fig. 12. The AC signal may be coupled to sensor 1220 through an impedance, and if sensor 1220 is connected to sensor electronics 1225, the coupled signal is significantly attenuated. Thus, a low level AC signal is present at the input of the detection circuit 1260. This may also be referred to as a highly attenuated signal or a signal having a high attenuation level. Under certain operating conditions, the voltage level of the AC signal may be V_applied*(C_coupling)/(C_coupling+C_sensor). If the detection circuit 1260 detects that a high level AC signal (low attenuation signal) is present at the input of the detection circuit 1260, no interrupt is sent to the microcontroller 410 because the sensor 1220 is not sufficiently hydrated or activated. For example, the input to the detection circuit 1260 can be a comparator. If sensor 1220 is sufficiently hydrated (or wetted), an effective capacitance (e.g., capacitance C in FIG. 12) is formed between the counter electrode and the reference electrode_r-c) And an effective capacitance is formed between the reference electrode and the working electrode (e.g., capacitance C in fig. 12)_w-r). In other words, the effective capacitance is related to the capacitance formed between the two nodes and does not represent that the actual capacitor is placed in the circuit between the two electrodes. In one embodiment, the AC signal from AC power supply 1255 is driven by capacitor C_r-cAnd C_w-rSubstantially attenuated, and the detection circuit 1260 detects the presence of a low-level or highly attenuated alternating current signal from the AC source 1255 at the input of the detection circuit 1260. This embodiment is important because the number of connections to the sensor is reduced by the existing connection between the sensor 1120 and the sensor electronics 1125. In other words, the mechanical switch disclosed in fig. 11 requires a switch and related connection between sensor 1120 and sensor electronics 1125. GetEliminating mechanical switches is advantageous because the sensor 1120 continues to shrink in size, and the elimination of components helps achieve this size reduction. In alternative embodiments, the AC signal may be applied to a different electrode (e.g., a counter electrode or a working electrode), and the present invention may operate in a similar manner.

As described above, after the detection circuit 1260 has detected that a low-level AC signal is present at the input of the detection circuit 1260, the detection circuit 1260 may later detect that a high-level AC signal with low attenuation is present at the input. This indicates that the sensor 1220 has been disconnected from the sensor electronics 1225 or that the sensor is not functioning properly. If the sensor has been disconnected from the sensor electronics 1225, the AC power source may be coupled to the input of the detection circuit 1260 with little or low attenuation. As described above, the detection circuit 1260 may generate an interrupt to the microcontroller. The interrupt may be received by the microcontroller, and the microcontroller may reduce or eliminate power provided to one or more of the components or circuits in the sensor electronics 1225. This may be referred to as a second interrupt. Again, this helps to reduce power consumption of sensor electronics 1225, particularly when sensor 1220 is not connected to sensor electronics 1225.

In an alternative embodiment, an AC signal may be applied to the reference electrode 1222, as shown by reference numeral 1291, and the impedance measurement device 1277 may measure the impedance of the regions in the sensor 1220. Illustratively, the region may be a region between the reference electrode and the working electrode, as shown by dashed line 1292 in fig. 12. Under certain operating conditions, the impedance measurement device 1277 may transmit a signal to the detection circuit 1260 if the measured impedance has dropped below an impedance threshold or other set criteria. This indicates that the sensor is fully hydrated. Under other operating conditions, once the impedance is above the impedance threshold, the impedance measurement device 1277 may transmit a signal to the detection circuit 1260. The detection circuit 1260 then transmits an interrupt to the microcontroller 410. In another embodiment, the impedance measurement device 1277 may transmit an interrupt or signal directly to the microcontroller.

In an alternative embodiment, the AC power supply 1255 may be replaced with a DC power supply. If a DC power supply is used, a resistance measurement element may be used in place of the impedance measurement element 1277. In embodiments utilizing a resistance measurement element, once the resistance drops below a resistance threshold or set criteria, the resistance measurement element may transmit a signal to the detection circuit 1260 (represented by dashed line 1293) or directly to the microcontroller indicating that the sensor is sufficiently hydrated and that power may be provided to the sensor.

In the embodiment shown in FIG. 12, if the detection circuit 1260 detects a low level or highly attenuated AC signal from the AC power source, an interrupt is generated to the microcontroller 410. The interruption indicates that the sensor has been sufficiently hydrated. In this embodiment, in response to an interrupt, microcontroller 410 generates a signal that is transmitted to digital-to-analog converter 420 to instruct or cause digital-to-analog converter 420 to apply a voltage or current to sensor 1220. Any of the different pulse sequences or short duration pulses described above in fig. 6A, 6B, or 6C or related text describing the application of the pulses may be applied to the sensor 1220. Illustratively, the voltage from the DAC420 may be applied to an operational amplifier 1275, the output of which is applied to the counter electrode 1221 of the sensor 1220. This causes the sensor (e.g., the working electrode 1223 of the sensor) to generate a sensor signal. Because the sensor is sufficiently hydrated, as identified by the interruption, the sensor signal generated at the working electrode 1223 accurately measures glucose. The sensor signal is measured by a sensor signal measuring device 431, and the sensor signal measuring device 431 transmits the sensor signal to the microcontroller 410, where a parameter of a physiological condition of the subject is measured. The generation of an interrupt indicates that the sensor is sufficiently hydrated and that sensor 1220 is now providing an accurate glucose measurement. In this embodiment, the hydration period may depend on the type and/or manufacturer of the sensor and the response of the sensor to the insertion or implantation into the subject. Illustratively, one sensor 1220 may have a hydration time of five minutes, and one sensor 1220 may have a hydration time of one, two, three, six, or 20 minutes. Also, any amount of time may be an acceptable amount of hydration time for the sensor, but a lesser amount of time is preferred.

If sensor 1220 is already connected, but is not sufficiently hydrated or wetted, then effective capacitance C_r-cAnd C_w-rThe electrodes in sensor 1120 are dry before insertion, and because the electrodes are dry, there is no good electrical path (or conductive path) between the two electrodes, so the detection circuit 1260 can still detect high level AC signals or low attenuation AC signals, and no interruption is generated.

FIG. 13A illustrates a method of hydrating a sensor according to an embodiment of the invention. In one embodiment, the sensor may be physically connected 1310 to the sensor electronics. After connection, in one embodiment, a timer or counter may be started to count the hydration time 1320. After the hydration time has elapsed, a signal may be transmitted 1330 to a subsystem in the sensor electronics to initiate application of a voltage to the sensor. As described above, in one embodiment, the microcontroller may receive a signal and instruct the DAC to apply a voltage to the sensor, or in another embodiment of the invention, the switch may receive a signal that allows the regulator to apply a voltage to the sensor. Hydration times may be five, two, ten minutes and may vary depending on the subject and the type of sensor.

In an alternative embodiment, an AC signal (e.g., a low voltage AC signal) may be applied 1340 to the sensor (e.g., a reference electrode of the sensor) after the sensor is connected to the sensor electronics. The AC signal may be applied because connecting the sensor to the sensor electronics allows the AC signal to be applied to the sensor. After the AC signal is applied, an effective capacitance 1350 is formed between the electrode in the sensor to which the voltage is applied and the other two electrodes. The detection circuit determines 1360 the level of the AC signal present at the input of the detection circuit. If a low level AC signal (or highly attenuated AC signal) is present at the input of the detection circuit, the detection circuit generates 1370 an interrupt and sends it to the microcontroller, since the effective capacitance forms a good electrical conduit between the electrodes and causes the AC signal to attenuate.

The microcontroller receives the interrupt generated by the detection circuit and transmits 1380 a signal to the digital-to-analog converter that indicates or causes the digital-to-analog converter to apply a voltage to an electrode (e.g., a counter electrode) of the sensor. Applying a voltage to the electrodes of the sensor causes the sensor to produce or generate a sensor signal 1390. The sensor signal measuring device 431 measures the generated sensor signal and transmits the sensor signal to the microcontroller. The microcontroller receives 1395 the sensor signal from the sensor signal measurement device coupled to the working electrode and processes the sensor signal to extract a measurement of a physiological characteristic of the subject or patient.

FIG. 13B illustrates an additional method for verifying hydration of a sensor, according to an embodiment of the invention. In the embodiment shown in FIG. 13B, the sensor is physically connected 1310 to the sensor electronics. An AC signal is applied 1341 to an electrode in the sensor, such as a reference electrode. Alternatively, in another embodiment, a DC signal is applied 1341 to the electrodes in the sensor. If an AC signal is applied, the impedance measurement element measures 1351 the impedance at a point within the sensor. Alternatively, if a DC signal is applied, the resistance measurement element measures 1351 the resistance at a point within the sensor. If the resistance or impedance is below the resistance threshold or impedance threshold (or other set criteria), respectively, the impedance (or resistance) measuring element will 1361 transmit to the detection circuit (or allow a signal to be transmitted to the detection circuit), and the detection circuit will interrupt transmission to the microcontroller, thereby identifying that the sensor is hydrated. Reference numerals 1380, 1390, and 1395 are the same in fig. 13A and 13B because they represent the same action.

The microcontroller receives the interrupt and transmits 1380 a signal to the digital to analog converter to apply a voltage to the sensor. In an alternative embodiment, the digital-to-analog converter may apply a current to the sensor, as described above. A sensor (e.g., a working electrode) generates 1390 a sensor signal indicative of a physiological parameter of the patient. The microcontroller receives 1395 the sensor signal from a sensor signal measurement device that measures the sensor signal at an electrode (e.g., a working electrode) in the sensor. The microcontroller processes the sensor signal to extract a measurement of a physiological characteristic of the subject or patient, such as the patient's blood glucose level.

Fig. 14A and 14B illustrate a method of combining sensor hydration with sensor stabilization, according to an embodiment of the invention. In the embodiment of the invention shown in fig. 14A, the sensor is connected 1405 to the sensor electronics. An AC signal is applied 1410 to the electrodes of the sensor. The detection circuit determines 1420 a level of the AC signal present at the input of the detection circuit. If the detection circuit determines that a low level AC signal is present at the input (indicating a high level of attenuation to the AC signal), an interrupt is sent 1430 to the microcontroller. Once an interrupt is sent to the microcontroller, the microcontroller knows to start or initiate 1440 a stabilization sequence, i.e., applying a plurality of voltage pulses to the electrodes of the sensor, as described above. For example, the microcontroller may cause the digital-to-analog converter to apply three voltage pulses (of amplitude +0.535 volts) to the sensor, each of which is followed by a period of three voltage pulses (of amplitude 1.07 volts). This may be referred to as transmitting a stable sequence of voltages. The microcontroller may accomplish this by executing a software program in Read Only Memory (ROM) or random access memory. After the stabilization sequence has completed execution, the sensor may generate 1450 a sensor signal that is measured and transmitted to the microcontroller.

The detection circuit can determine 1432 that the high level AC signal continues to be present at the input of the detection circuit (e.g., the input of the comparator) even after the hydration time threshold has elapsed. For example, the hydration time threshold may be 10 minutes. After 10 minutes, the detection circuit may still detect the presence of a high level AC signal. At this point, the detection circuit may transmit 1434 the hydration assistance signal to the microcontroller. If the microcontroller receives a hydration assistance signal, the microcontroller may transmit 1436 a signal that causes the DAC to apply a voltage pulse or series of voltage pulses to assist in sensor hydration. In one embodiment, the microcontroller may transmit a signal that causes the DAC to apply a portion of a stabilization sequence or other voltage pulse to assist in hydrating the sensor. In this embodiment, the application of the voltage pulse may cause 1438 a low level AC signal (or a highly attenuated signal) to be detected at the detection circuit. At this point, the detection circuit may transmit an interrupt, as disclosed in step 1430, and the microcontroller may initiate a stabilization sequence.

Fig. 14B shows a second embodiment combining the hydration method with the stabilization method, wherein feedback is utilized during stabilization. The sensor is connected 1405 to the sensor electronics. An AC signal (or DC signal) is applied 1411 to the sensor. In one embodiment, an AC signal (or DC signal) is applied to an electrode of the sensor, such as a reference electrode. An impedance measurement device (or resistance measurement device) measures 1416 an impedance (or resistance) within a specified region of the sensor, such as an impedance between the reference electrode and the working electrode. The measured impedance (or resistance) may be compared 1421 to an impedance or resistance value to see if the impedance (or resistance) in the sensor is low enough, indicating that the sensor is hydrated. If the impedance (or resistance) is below an impedance (or resistance) value or other set criteria (which may be a threshold), an interrupt is transmitted 1431 to the microcontroller. After receiving the interrupt, the microcontroller transmits 1440 a signal to the DAC indicating that the digital-to-analog converter applied a stable sequence of voltages (or currents) to the sensor. After the stabilization sequence is applied to the sensor, a sensor signal is generated in the sensor (e.g., at the working electrode), which is measured by the sensor signal measurement device, transmitted by the sensor signal measurement device, and received 1450 by the microcontroller. Because the sensor is hydrated and a stable sequence of voltages has been applied to the sensor, the sensor signal accurately measures the physiological parameter (i.e., blood glucose).

Fig. 14C shows a third embodiment of the combination stabilization and hydration method. In this embodiment, the sensor is connected 1500 to sensor electronics. After the sensor is physically connected to the sensor electronics, an AC signal (or DC signal) is applied 1510 to an electrode (e.g., a reference electrode) of the sensor. At the same time, or at about the same time, the microcontroller transmits a signal that causes the DAC to apply 1520 a steady voltage sequence to the sensor. In an alternative embodiment, a steady sequence of currents may be applied to the sensor instead of a steady sequence of voltages. The detection circuit determines 1530 the level of the AC signal (or DC signal) present at the input of the detection circuit. If a low level AC signal (or DC signal) is present at the input of the detection circuit representing a highly attenuated AC signal (or DC signal), an interrupt is transmitted 1540 to the microcontroller. Because the microcontroller has initiated the stabilization sequence, the microcontroller receives an interrupt and sets 1550 the first indicator that the sensor is sufficiently hydrated. After the stabilization sequence is complete, the microcontroller sets 1555 a second indicator that indicates that the stabilization sequence is complete. The application of the stabilization sequence voltage causes the sensor (e.g., working electrode) to generate 1560 a sensor signal that is measured by the sensor signal measurement circuit and sent to the microcontroller. If the second indicator of completion of the stabilization sequence is set and the first indicator of completion of hydration is set, the microcontroller can utilize the 1570 sensor signal. If one or both indicators are not set, the microcontroller may not utilize the sensor signal because the sensor signal may not represent an accurate measurement of the physiological measurement of the subject.

Generally, the hydration and stabilization processes described above can be used as part of a larger Continuous Glucose Monitoring (CGM) method. The current state of the art of continuous glucose monitoring is largely ancillary, meaning that readings provided by CGM devices (including, for example, implantable or subcutaneous sensors) cannot be used to make clinical decisions without reference values. In turn, the reference value must typically be obtained from the fingertip using, for example, a BG meter. The reference value is required because the amount of information provided by the sensor/sensing assembly is limited. In particular, only the raw sensor value (i.e., sensor current or Isig) and the counter voltage (which is the voltage between the counter electrode and the reference electrode) (see, e.g., fig. 5) may be provided by the sensing component for processing. Thus, during analysis, if the raw sensor signal appears abnormal (e.g., if the signal is decreasing), the only way that sensor failure can be distinguished from physiological changes in the user/patient (i.e., changes in glucose level within the body) may be to acquire a reference glucose value through the fingertip. It is well known that the reference fingertip is also used to calibrate the sensor.

Embodiments of the invention described herein relate to advances and improvements in continuous glucose monitoring leading to a more autonomous system, and related devices and methods, in which the need for a reference fingertip can be minimized or eliminated, and thereby clinical decisions can be made based on information obtained only from one or more sensor signals with a high level of reliability. From a sensor design perspective, such autonomy may be achieved through electrode redundancy, sensor redundancy (including, for example, pseudo-orthogonal redundancy between two or more sensors), sensor diagnostics, and Isig and/or Sensor (SG) fusion, in accordance with embodiments of the present invention.

In the discussion herein, "redundancy" refers to the presence/use of two or more electrodes, whether contained on/within a single sensor or on/within two or more sensors, for purposes of the present invention. That is, redundancy may be achieved, for example, by using multiple working electrodes (within a single sensor, or between two or more sensors, whether or not identical) to generate multiple signals indicative of a patient's Blood Glucose (BG) level. In turn, the multiple signals can be used to generate a fused glucose value, as well as to assess the relative health of the (working) electrodes, the overall reliability of the sensor, and the frequency (if any) required to calibrate the reference value.

For example, it is well known that acquiring signals from multiple electrochemical sensors can provide improved performance in the form of simple redundancy, which can be achieved by multiple electrodes on the same probe (or "flexure"), or by using spatial separation and two separate probes. For example, a hospital glucose sensor sold by Medtronic, inc. contains two probes, each having two working electrodes, that produce four independent glucose signals.

In contrast to simple redundancy, orthogonal redundancy can be defined as two devices employing two different techniques to achieve the same goal, where the failure modes of the two devices are completely unique and disjoint. Thus, orthogonal redundancy can be created by combining, for example, optical sensing and electrochemical sensing techniques. It is clear that one advantage of orthogonal redundancy is that both types of sensors (e.g., optical sensors and electrochemical (or "echem") sensors) are subject to different types of interference, failure modes, and body reactions. On the other hand, the use of two disparate techniques introduces an additional design layer and computational complexity to the measurement and analysis of glucose levels in a patient.

Pseudo-orthogonal redundancy, on the other hand, can be achieved by utilizing the same technique, but with minor but significant variations, in order to produce complementary glucose measurements, while minimizing additional design and/or computational complexity. In this regard, as will be discussed in more detail below, in one embodiment of the invention, two or more electrochemical sensors may be employed, wherein the sensor(s) may be conventional peroxide-based sensors (of the type discussed in detail above), and the sensor(s) may measure glucose by calculating the difference in oxygen between the two working electrodes (typically located on the same sensor).

Sensor diagnostics involve the use of additional (diagnostic) information that can provide real-time knowledge of the sensor health. In this regard, Electrochemical Impedance Spectroscopy (EIS) has been found to provide such additional information in the form of sensor impedance and impedance related parameters at different frequencies. Furthermore, advantageously, it has further been found that for certain frequency ranges the impedance and/or the impedance related data is substantially independent of glucose. This glucose independence enables the use of a variety of EIS-based markers or indicators, which are used not only to generate stable, highly reliable sensor glucose values (via a fusion method as will be described in more detail below), but also to assess the condition, health, age, and efficiency of individual electrode(s) and integral sensor(s) substantially independently of the glucose-dependent Isig.

For example, analysis of glucose independent impedance data provides information about the efficiency of the sensor (how fast the sensor hydrates) and is ready for data acquisition using values such as 1kHz real impedance, 1kHz imaginary impedance, and Nyquist Slope (described in more detail below). Furthermore, the glucose-independent impedance data provides information about potential blockage(s) that may be present on the sensor membrane surface that may temporarily prevent glucose from entering the sensor and thus cause a signal dip (using, for example, 1kHz real impedance values). In addition, the glucose-independent impedance data uses values of phase angle and/or imaginary impedance at frequencies of, for example, 1kHz and higher to provide information about the loss of sensitivity of the sensor during extended wear, which may be due to local hypoxia at the insertion site.

In the case of (electrode) redundancy and EIS, a fusion algorithm may be used to obtain diagnostic information provided by the EIS for each redundant electrode and to independently assess the reliability of each electrode. Then, a weight can be added as a reliability measure for each individual signal, and a single fused signal can be calculated that can be used to generate the sensor glucose value seen by the patient/subject.

From the above, it can be seen that the combined use of redundancy, sensor diagnostics using EIS, and EIS-based fusion algorithms enables a more reliable CGM system. Redundancy is advantageous in at least two respects. First, redundancy eliminates the risk of a single point of failure by providing multiple signals. Second, providing multiple (working) electrodes where a single electrode may be sufficient allows the output of the redundant electrode to be used as a check on the main electrode, thereby reducing and possibly eliminating the need for frequent calibration. Furthermore, EIS diagnostics can automatically check the health of each electrode without the need for reference glucose values (fingertips), thereby reducing the number of reference values required.

EIS or AC impedance methods study the response of the system to the application of periodic small amplitude AC signals. This is schematically illustrated in fig. 15A, where E is the applied potential, I is the current, and the impedance (Z) is defined as Δ E/Δ I. However, although the impedance itself can be mathematically defined simply as Δ E/Δ I, to date, there has been no commercial success in applying EIS technology to continuous glucose monitoring. This is due in part to the fact that glucose sensors are very complex systems and to date, no mathematical model has been developed that can fully account for the complexity of the EIS output of a glucose sensor.

FIG. 15B shows a simplified circuit model for describing electrochemical impedance spectroscopy. In this illustration, IHP represents the inner Helmholtz plane, OHP represents the outer Helmholtz plane, CE represents the counter electrode, WE represents the working electrode, C_dRepresenting a double layer capacitance, R_pRepresents the polarization resistance, Z_wRepresenting the Valley impedance, R_sRepresenting the solution resistance. The last four components, i.e. the double layer capacitor (C)_d) Valurg impedance (Z)_w) Polarization resistance (R)_p) And solution resistance (R)_s) Each of which may play an important role in sensor performance and may be measured individually by applying a low or high frequency alternating operating potential. For example, the valberg impedance is closely related to the diffusion impedance (mainly low frequency impedance) of the electrochemical system and is therefore present in all diffusion limited electrochemical sensors. Thus, by associating one or more of these components with one or more components and/or layers of a glucose sensor, EIS techniques may be used as a sensor diagnostic tool.

As is well known, the impedance may be defined in terms of its amplitude and phase, where amplitude (| Z |) is the ratio of the voltage difference amplitude to the current amplitude, and phase (θ) is the phase shift of the current leading the voltage. When the circuit is driven by Direct Current (DC) only, the impedance is the same as the resistance, i.e. the resistance is a special case of a zero phase angle impedance. However, as a complex number, the impedance may also be represented by its real and imaginary parts. In this regard, the real and imaginary impedances may be derived from the impedance magnitude and phase using the following equations:

real impedance (ω) amplitude (ω) × cos (phase (ω)/180x π)

Imaginary impedance (ω) amplitude (ω) × sin (phase (ω)/180x π)

Where ω represents the input frequency measuring amplitude (ohms) and phase (degrees). The relationship between impedance on the one hand and current and voltage on the other hand, including how the former is calculated based on measurements of the latter, will be discussed more fully below in connection with sensor electronics, including Application Specific Integrated Circuits (ASICs) developed for use in embodiments of the invention described herein.

Continuing with the circuit model shown in FIG. 15B, the total system impedance can be simplified as:

wherein Z is_w(ω) is the Valley impedance, ω is the angular velocity, j is the imaginary component (instead of the conventional "i" to avoid confusion with current), C_d、R_pAnd R_sDouble layer capacitance, polarization resistance and solution resistance, respectively (as described previously).

The Valurberg impedance can be calculated as follows

Where D is the diffusivity, L is the sensor film thickness, C is the peroxide concentration, and m: 1/2 corresponds to a 45 Nyquist slope.

The nyquist curve is a graphical representation in which the real part of the impedance (real Z) is plotted over the entire frequency spectrum relative to its imaginary part (Img Z). Fig. 16A shows a general example of a nyquist curve, in which the X value is the real part of impedance and the Y value is the imaginary part of impedance. The phase angle is the angle between the impedance point (x, y) and the x-axis that defines a vector of magnitude | Z |.

The nyquist curve of fig. 16A is generated by applying an AC voltage plus a DC voltage (DC bias) between the working electrode and the counter electrode at a selected frequency (i.e., frequency sweep) from 0.1Hz to 1000 MHz. Starting from the right, the frequency increases from 0.1 Hz. For each frequency, real and imaginary impedances can be calculated and plotted. As shown, a typical nyquist curve for an electrochemical system may look like a semicircle connected to a straight line at an inflection point, where the semicircle and the straight line represent plotted impedances. In some embodiments, the impedance at the inflection point is particularly important because it is most easily identified in the nyquist curve and the intercept can be defined. Generally, the inflection point is near the X axis, and the X value of the inflection point is approximately the sum of the polarization resistance and the solution resistance (R)_p+R_s)。

Referring to fig. 16B, the nyquist curve may generally be described in terms of a lower frequency region 1610 and a higher frequency region 1620, where the labels "higher frequency" and "lower frequency" are used in a relative sense and are not intended to be limiting. Thus, for example, low frequency region 1610 may illustratively contain data points obtained for a frequency range between about 0.1Hz and about 100Hz (or higher), and high frequency region 1620 may illustratively contain data points obtained for a frequency range between about 1kHz (or lower) and about 8kHz (or higher). In the low frequency region 1610, the nyquist slope represents the gradient of a linear fit 1630 of low frequency data points in the nyquist curve. As shown in the figure, in the high frequency region 1620, the imaginary impedance value is minimum and can become negligible. Thus, intercept 1600 is essentially a real impedance value at higher frequencies (e.g., in this case, approximately in the range of 1kHz to 8 kHz). In fig. 16B, the intercept 1600 is at about 25 kohms.

FIGS. 16C and 16D show how the glucose sensor responds to a sinusoidal (i.e., alternating) operating potential in these figures, G L M is the glucose limiting membrane of the sensor, AP is the adhesion promoter, HSA is human serum albumin, and GOX isGlucose oxidase (layer), E_dcIs a DC potential, E_acIs AC potential, C'_{Peroxides and their use in the preparation of pharmaceutical preparations}Is the peroxide concentration during AC application. As shown in fig. 16C, if the sensor diffusion length as a function of AC potential frequency, molecular diffusivity, and film thickness is small compared to the film (GOX) length, the system gives a relatively linear response with a constant phase angle (i.e., infinity). Conversely, if the diffusion length is equal to the film (GOX) length, the system response will become finite, such that the nyquist curve is semicircular, as shown in fig. 16D. The latter is generally applicable to low frequency EIS where the non-faradaic process is negligible.

In performing an EIS analysis, AC voltages and DC biases of different frequencies may be applied, for example, between the working electrode and the reference electrode. In this regard, EIS is an improvement over previous methods that may apply an AC voltage limited to a simple DC current or a single frequency. Although, in general, EIS may be performed at frequencies in the μ Hz to MHz range, in embodiments of the invention described herein, a narrower frequency range (e.g., between about 0.1Hz to about 8 kHz) may be sufficient. Thus, in some embodiments, an AC potential having a programmable magnitude of up to at least 100mV (and preferably at about 50mV) may be applied over a frequency range between about 0.1Hz and about 8 kHz.

Within the above frequency range, the capacitive characteristics of the sensor are carefully examined using relatively high frequencies (i.e., frequencies typically between about 1kHz to about 8 kHz). Depending on the thickness and permeability of the membrane, a typical range of impedance at relatively high frequencies may be, for example, between about 500 and 25 kilo-ohms, and a typical range of phase may be, for example, between 0 and-40 degrees. On the other hand, the resistive properties of the sensor are carefully examined using relatively low frequencies (i.e., frequencies typically between about 0.1Hz and about 100 Hz). Here, depending on the electrode design and degree of metallization, a typical operating range for the output real impedance may be, for example, between about 50 kohms and 300 kohms, and a typical range for the phase may be between about-50 degrees to about-90 degrees. The illustrative ranges described above are shown, for example, in the baud curves of fig. 16E and 16F.

As previously mentioned, the phrases "higher frequency" and "lower frequency" are used relative to each other, not in an absolute sense, and these phrases, as well as the typical impedance and phase ranges mentioned above, are intended to be illustrative, not limiting. However, the basic principle remains unchanged: the capacitive and resistive behavior of the sensor can be scrutinized by impedance data across the analysis spectrum, where, in general, lower frequencies provide information about more resistive components (e.g., electrodes, etc.) while higher frequencies provide information about capacitive components (e.g., films). However, the actual frequency range in each case depends on the overall design, including, for example, the type or types of electrode or electrodes, the surface area of the electrode or electrodes, the thickness of the membrane, the permeability of the membrane, etc. See also figure 15B, which relates to the general correspondence between high frequency circuit components and the sensor membrane, and between low frequency circuit components and the faraday process (including, for example, electrodes).

EIS can be used in sensor systems where the sensor contains a single working electrode as well as sensor systems where the sensor comprises multiple (redundant) working electrodes. In one embodiment, the EIS provides valuable information about the age (or aging) of the sensor. In particular, the impedance has different magnitudes and phase angles at different frequencies. As shown in fig. 17, the sensor impedance (specifically, the sum of Rp and Rs) reflects the usage time of the sensor and the operating state of the sensor. Thus, as can be seen from the different curves in fig. 17, a new sensor typically has a higher impedance than a used sensor. In this way, by considering the X value of the sum of Rp and Rs, a threshold can be used to determine when the age of the sensor exceeds a specified operating life of the sensor. It should be noted that although for the illustrative examples shown in fig. 17-21 and discussed below, the real impedance value at the inflection point (i.e., Rp + Rs) is used to determine the aging, state, stability, and hydration of the sensor, alternative embodiments may use other EIS-based parameters in addition to, or instead of, real impedance, such as, for example, imaginary impedance, phase angle, nyquist slope, etc.

Fig. 17 shows an example of a nyquist curve over the life of the sensor. The point indicated by the arrow is the corresponding inflection point of each sweep across the spectrum. For example, Rs + Rp is higher than 8.5 kohms before initialization (at time t ═ 0); after initialization (at time t ═ 0.5 hours), the value of Rs + Rp drops below 8 kiloohms. The Rs + Rp continued to drop over the next six days, dropping below 6.5 kohms at the end of the specified sensor life. Based on these examples, a threshold may be set to specify when the value of Rs + Rp indicates the end of the specified operating life of the sensor. Thus, EIS techniques allow for plugging leaks that allow sensors to be reused beyond a specified operating time. In other words, if the patient attempts to reuse the sensor by disconnecting and then reconnecting the sensor again after the sensor has reached its designated operating time, the EIS will measure an abnormally low impedance, thereby enabling the system to reject the sensor and prompt the patient to install a new sensor.

In addition, EIS can enable detection of sensor failure by detecting when the impedance of the sensor drops below a low impedance threshold level, indicating that the sensor may be worn out and not operating properly. The system may then terminate the sensor before the specified operational life. As will be discussed in detail below, the sensor impedance may also be used to detect other sensor faults (modes). For example, when the sensor enters a low current state (i.e., sensor failure) for various reasons, the sensor impedance may also increase beyond a certain high impedance threshold. The system may also terminate the sensor before a specified sensor operating life if the impedance becomes abnormally high during sensor operation, for example, due to protein or polypeptide fouling, macrophage attachment, or any other factor.

FIG. 18 illustrates how EIS techniques can be applied during sensor settling and when detecting sensor age, in accordance with certain embodiments. The logic of FIG. 18 begins at 1800 after the previously described hydration process and sensor initialization process have been completed. In other words, the sensor has been considered to be sufficiently hydrated and a first initialization process has been applied to initialize the sensor. The initialization process may preferably be in the form of voltage pulses, as described in detail above. However, in alternative embodiments, different waveforms may be used for the initialization process. For example, a sine wave may be used instead of a pulse to accelerate wetting or conditioning of the sensor. Furthermore, some portions of the waveform may need to be greater than the normal operating voltage of the sensor, i.e., 0.535 volts.

At block 1810, the EIS procedure is applied and the impedance is compared to a first high threshold and a first low threshold. Examples of the first high threshold and the first low threshold are 7 kilo-ohms and 8.5 kilo-ohms, respectively, although these values may be set higher or lower as desired. If the impedance (e.g., Rp + Rs) is above the first high threshold, then at block 1820 the sensor undergoes additional initialization processes (e.g., applying one or more additional pulses). Ideally, the number of total initialization procedures used to initialize the sensor would be optimized to limit the impact on both the battery life of the sensor and the total time required to stabilize the sensor. Thus, by applying EIS, less initialization may be performed initially, and the number of initializations may be incrementally increased to give the right amount of initialization to prepare the sensor for use. Similarly, in an alternative embodiment, the EIS may be applied to the hydration process, as described in fig. 13-14, to minimize the number of initializations required to assist the hydration process.

On the other hand, if the impedance (e.g., Rp + Rs) is below the first low threshold, the sensor will be determined to be malfunctioning and will be immediately terminated at block 1860. A message will be sent to the user requesting that the sensor be replaced and the hydration process started again. If the impedance is within the high and low thresholds, the sensor will begin normal operation at block 1830. The logic then proceeds to block 1840, where additional EISs are performed to check the age of the sensor. When the logic first reaches block 1840, the microcontroller will execute an EIS to measure the age of the sensor to block leaks where the user can insert and remove the same sensor. In future iterations of the EIS program, the microprocessor will perform EIS at fixed intervals during the specified lifetime of the sensor when the logic returns to block 1840. In a preferred embodiment, the fixed interval is set every 2 hours, however, longer or shorter periods of time may be readily used.

At block 1850, the impedance is compared to a second set of high and low thresholds. Examples of such second high and low thresholds may be 5.5 kilo-ohms and 8.5 kilo-ohms, respectively, although these values may be set higher or lower as desired. As long as the impedance value remains within the second high and low thresholds, the logic proceeds to block 1830 where the sensor operates normally until a specified sensor life (e.g., 5 days) is reached. Of course, as described with respect to block 1840, EIS is performed at regularly scheduled intervals throughout a specified sensor lifetime. However, if after performing EIS, it is determined at block 1850 that the impedance has dropped below a second lower threshold or risen above a second higher threshold, the sensor is terminated at block 1860. In a further alternative embodiment, a secondary check may be performed on faulty sensor readings. For example, if the EIS indicates that the impedance is outside of the second high and low thresholds, then the logic may execute the second EIS to confirm that the second set of thresholds are indeed not met (and that the first EIS was executed correctly) before determining the end of the sensor at block 1860.

FIG. 19 builds on the above description and details a possible schedule for performing a diagnostic EIS procedure. Each diagnostic EIS procedure is optional and, as desired, no diagnostic EIS procedure or any combination of one or more diagnostic EIS procedures may be scheduled. The schedule of fig. 19 begins with sensor insertion at point 1900. After sensor insertion, the sensor undergoes a hydration period 1910. This hydration period is important because an insufficiently hydrated sensor may provide the user with inaccurate readings, as previously described. The first optional diagnostic EIS procedure at point 1920 is scheduled during the hydration period 1910 to ensure that the sensor is sufficiently hydrated. The first diagnostic EIS program 1920 measures the sensor impedance value to determine if the sensor has been sufficiently hydrated. If the first diagnostic EIS program 1920 determines that the impedance is within the set high and low thresholds, indicating sufficient hydration, the sensor controller will allow the sensor to power up at point 1930. Conversely, if the first diagnostic EIS program 1920 determines that the impedance is outside of the set high and low thresholds, indicating inadequate hydration, the sensor hydration period 1910 may be extended. After prolonged hydration, once a certain capacitance is reached between the electrodes of the sensor, meaning that the sensor is sufficiently hydrated, power may be applied at point 1930.

After the sensor is powered up at point 1930, but before sensor initialization begins at point 1950, a second optional diagnostic EIS routine 1940 is scheduled. After this scheduling, the second diagnostic EIS program 1940 may detect whether the sensor is being reused before the 1950 initialization begins. The test to determine whether the sensor is reused is described in detail in the description of fig. 18. However, unlike the previous description with reference to FIG. 18, where the burn-in test is performed after initialization is complete, the burn-in test is shown in FIG. 19 as being performed prior to initialization. It is important to recognize that the timeline of the EIS process depicted in fig. 19 may be rearranged and the order of some steps may be interchanged without affecting the overall teachings of the present application. As previously described, the second diagnostic EIS program 1940 detects a sensor that is being reused by determining the impedance value of the sensor and then comparing it to set high and low thresholds. If the impedance falls outside of the set threshold, indicating that the sensor is being reused, the sensor may be rejected and the user prompted to replace the new sensor. This prevents complications due to repeated use of old sensors. Conversely, if the impedance falls within a set threshold, sensor initialization 1950 may begin with a confidence that a new sensor is being used.

After initialization begins at point 1950, a third optional diagnostic EIS procedure 1960 is scheduled. The third diagnostic EIS program 1960 tests the impedance value of the sensor to determine if the sensor is fully initialized. The third diagnostic EIS procedure 1960 should be performed in the shortest time required for any sensor to fully initialize. When executed at this time, sensor life is maximized by limiting the time that a fully initialized sensor is not used, and over-initialization is avoided by confirming that the sensor is fully initialized before too much initialization occurs. It is important to prevent over-initialization, which can result in current being suppressed and thus inaccurate readings. However, insufficient initialization is also a problem, so if the third diagnostic EIS program 1960 indicates that sensor initialization is insufficient, an optional initialization may be performed at point 1970 to fully initialize the sensor. Insufficient initialization is disadvantageous because excessive current can be generated independent of the actual glucose concentration. The third diagnostic EIS procedure plays an important role in ensuring that the sensor is functioning properly when in use, due to the risk of under-initialization and over-initialization.

In addition, an optional periodic diagnostic EIS program 1980 may schedule a time after the sensor is fully initialized. The EIS program 1980 may be scheduled at any set time interval. As will be discussed in more detail below, EIS program 1980 may also be triggered by other sensor signals, such as an abnormal current or an abnormal counter electrode voltage. In addition, EIS program 1980 can be arranged as little as possible or as much as possible, as desired. In a preferred embodiment, the EIS procedure used during the hydration process, sensor life check, initialization process, or periodic diagnostic tests is the same procedure. In alternative embodiments, the EIS process may be shortened or lengthened (i.e., fewer or more frequency ranges examined) for various EIS processes as needed to focus on a particular impedance range. Periodic diagnostic EIS program 1980 monitors the impedance values to ensure that the sensor continues to operate at an optimal level.

If the sensor current drops due to contaminants, sensor aging, or a combination of contaminants and sensor aging, the sensor may not operate at an optimal level. Sensors that age beyond a certain length are no longer useful, but sensors that are impeded by contaminant species may be repaired. The contaminant species may reduce the surface area of the electrode or diffusion paths for analytes and reaction byproducts, thereby causing a decrease in sensor current. These contaminants are charged and gradually accumulate on the electrode or membrane surface under a certain voltage. Previously, contaminant species have destroyed the usefulness of the sensor. Now, if the periodic diagnostic EIS program 1980 detects an impedance value indicating the presence of a contaminant, remedial action can be taken. When remedial action is taken is described with reference to fig. 20. Therefore, the periodic diagnostic EIS procedure 1980 becomes very useful because it can trigger remedial actions of the sensor that can restore the sensor current to normal levels and extend the life of the sensor. Two possible embodiments of sensor remedial actions are described below in the description of fig. 21A and 21B.

In addition, any scheduled diagnostic EIS procedure 1980 may be paused or rescheduled when certain events are determined to be imminent. Such events may include any situation that requires a patient to check a sensor reading, including, for example, when a patient uses a test strip meter to measure his or her BG level in order to calibrate a sensor, when a patient is alerted to a calibration error and needs to measure his or her BG level using a test strip meter a second time, or when a hyperglycemic or hypoglycemic alarm has been issued but is not acknowledged.

FIG. 20 illustrates a method of combining a diagnostic EIS procedure with sensor remedial action. The box 2000 diagnostic routine may be any periodic diagnostic EIS routine 1980 as described in detail in fig. 19. The logic of the method begins when a diagnostic EIS procedure is performed to detect the impedance value of the sensor at block 2000. As described above, in certain embodiments, the EIS process applies a combination of DC bias and AC voltage of varying frequency, wherein the impedance detected by performing the EIS process is mapped onto the nyquist curve, and the inflection point in the nyquist curve approximates the sum of the polarization resistance and the solution resistance (i.e., the real impedance value). After the diagnostic EIS program detects an impedance value for the sensor at block 2000, the logic proceeds to block 2010.

At block 2010, the impedance value is compared to set high and low thresholds to determine if it is normal. If the impedance is within the set boundaries of the high and low thresholds at block 2010, normal sensor operation is resumed at block 2020 and the logic of FIG. 20 will end until such time as another diagnostic EIS procedure is scheduled. Conversely, if it is determined at block 2010 that the impedance is abnormal (i.e., outside the set boundary of the high and low thresholds), then remedial action at block 2030 is triggered. Examples of acceptable high and low thresholds over the life of the sensor are 5.5 kilo-ohms and 8.5 kilo-ohms, respectively, although these values may be set higher or lower as desired.

Block 2030 remedial action is performed to remove any contaminating substances that may have resulted in an abnormal impedance value. In a preferred embodiment, the remedial action is performed by applying a reverse current or a reverse voltage between the working electrode and the reference electrode. Details of the remedial action will be described in more detail with reference to fig. 21. After the remedial action is performed at block 2030, the impedance value is tested again at block 2040 by the diagnostic EIS procedure. Then, when the impedance value from the diagnostic EIS process of block 2040 is compared to a set high or low threshold, the success of the remedial action is determined at block 2050. As in block 2010, if the impedance is within a set threshold, then it is considered normal, and if the impedance is outside of the set threshold, then it is considered abnormal.

If at block 2050 it is determined that the impedance value of the sensor has returned to normal, then normal sensor operation will occur at block 2020. If the impedance is still not normal, indicating that sensor aging is the cause of the abnormal impedance, or that remedial action has failed to successfully remove the contaminant, the sensor is terminated at block 2060. In an alternative embodiment, instead of terminating the sensor immediately, the sensor may generate a sensor message that initially requests the user to wait and then perform further remedial action after a set period of time has elapsed. This alternative step may be coupled with separate logic to determine whether the impedance value is increasingly close to being within the boundary of the high and low thresholds after the initial remedial action is performed. For example, if no change is found in the sensor impedance value, the sensor may decide to terminate. However, if the sensor impedance value is closer to the preset boundary, but still outside the boundary after the initial remedial action, additional remedial actions may be performed. In yet another alternative embodiment, the sensor may generate a message requesting the user to calibrate the sensor by taking a fingertip gauge measurement to further confirm whether the sensor is indeed malfunctioning. All of the above embodiments are intended to prevent the user from using a faulty sensor that produces inaccurate readings.

FIG. 21A illustrates one embodiment of the aforementioned sensor remedial action. In this embodiment, clogging by contaminant species is eliminated by reversing the voltage applied to the sensor between the working electrode and the reference electrode. The reverse DC voltage carries charged contaminants away from the electrode or membrane surface, clearing the diffusion path. After clearing the path, the current of the sensor returns to a normal level and the sensor can give an accurate reading. Thus, the remedial action saves the user time and money associated with replacing a valid sensor.

Fig. 21B illustrates an alternative embodiment of the sensor remedial measures previously mentioned. In this embodiment, the reverse DC voltage applied between the working electrode and the reference electrode is coupled with an AC voltage. By adding an AC voltage, some tightly absorbed substances or species on the surface layer can be removed, since the AC voltage can extend its force further from the electrodes and penetrate all layers of the sensor. The AC voltage may have a variety of different waveforms. Some examples of waveforms that may be used include square waves, triangular waves, sine waves, or pulses. As with the previous embodiment, once the contaminant material is removed, the sensor can resume normal operation and the life and accuracy of the sensor is improved.

Although the above examples primarily illustrate the use of real impedance data in sensor diagnostics, embodiments of the invention described herein also relate to the use of other EIS-based and substantially analyte-independent parameters (other than real impedance) in sensor diagnostic procedures. For example, as previously described, analysis of the (substantially) glucose independent impedance data (e.g., values of real impedance at 1kHz and imaginary impedance at 1 kHz) and nyquist slope provides information about the efficiency of the sensor with respect to the speed at which the sensor hydrates and is ready for data acquisition. Furthermore, impedance data (such as e.g. a value of 1kHz real impedance) that is (substantially) independent of glucose provides information about one or more potential blockages that may be present on the sensor membrane surface that may temporarily prevent glucose from entering the sensor, resulting in a signal dip.

Furthermore, impedance data (substantially) independent of glucose (e.g. values of higher frequency phase angle and/or imaginary impedance at 1kHz and higher) provide information about the loss of sensitivity of the sensor during long-term wear, which may be due to local hypoxia at the insertion site. In this regard, the potential mechanism by which hypoxia causes a loss of sensitivity can be described as follows: when local oxygen is insufficient, the sensor output (i.e., Isig and SG) will depend on oxygen rather than glucose, and thus, the sensor will lose sensitivity to glucose. As will be discussed in more detail below, embodiments of the pseudo-orthogonally redundant sensor system of the present invention described herein make the sensor output substantially oxygen independent.

Other markers (including 0.1Hz real impedance, counter electrode voltage (V)_cntr) And spikes in Isig caused by EIS) can also be used to detect sensitivity loss due to hypoxia. Further, in a sensor system with redundant electrodes, the relative differences between the 1kHz real impedance, the 1kHz virtual impedance, and the 0.1Hz real impedance between two or more working electrodes can be used to detect a loss of sensitivity due to biofouling.

According to embodiments of the invention described herein, EIS-based sensor diagnostics require consideration and analysis of EIS data relating to one or more of at least the following three major factors (i.e., potential sensor failure modes): (1) starting a signal; (2) suddenly dropping signals; and (3) sensitivity loss. Notably, it has been found that most impedance-related parameters used in such diagnostic analyses and procedures can be studied at a frequency or range of frequencies where the parameter is substantially independent of the analyte, which allows the sensor diagnostic procedure to be performed independently of the analyte level in the patient. Thus, while EIS-based sensor diagnostics may be triggered by, for example, large analyte-dependent fluctuations in Isig, the impedance-related parameters used in such sensor diagnostic procedures are essentially independent of analyte levels themselves. As will be discussed in more detail below, it has also been found that in most cases, glucose can be seen to have an effect on the magnitude (or other characteristic) of the EIS-based parameter, which is typically small enough (e.g., at least an order of magnitude difference between the EIS-based measurement and the glucose effect on it) so that it can be filtered out of the measurement, e.g., by software in the IC.

By definition, "activation" refers to the integrity of the sensor signal within the first few hours (e.g., t-0-6 hours) after insertion. For example, in many current devices, the signal within the first 2 hours after insertion is considered unreliable and, therefore, the sensor glucose value is not visible to the patient/user. In the case of sensors that take a long time to hydrate, the sensor signal is low for several hours after insertion. By using EIS, additional impedance information can be obtained (by running the EIS program) immediately after the sensor is inserted. In this regard, the total impedance equation can be used to explain the principle behind low onset detection using a real impedance of 1 kHz. At relatively high frequencies (in this case, 1kHz and above), the imaginary impedance is very small (as evidenced by the in vivo data), so the total impedance is reduced to:

double layer capacitance (C) as sensor wetting is gradually completed_d) And (4) increasing. Therefore, the total impedance will be reduced because, as shown in the above equation, the total impedance and C_dIn inverse proportion. This is shown, for example, in the form of intercept 1600 on the real impedance axis shown in fig. 16B. Importantly, an imaginary impedance of 1kHz can also be used for the same purpose, since it also includes a capacitive component and is inversely proportional to the capacitive component.

Another marker of low onset detection is the nyquist slope, which depends only on the impedance of the relatively low frequency, which in turn corresponds to the varburg impedance component of the total impedance (see, e.g., fig. 15B). Fig. 22 shows the nyquist curve for a normally operating sensor, where arrow a indicates the time course from t-0 (i.e. the sensor wear time). Thus, immediately after sensor insertion (time t ═ 0), a relatively low frequency EIS is performed, which produces real and imaginary impedance data plotted with a first linear fit 2200 having a first (nyquist) slope. At a time interval after t-0, a second (lower) frequency scan is run, resulting in a second linear fit 2210 with a second (nyquist) slope greater than the first nyquist slope, and so on. As the sensor is further hydrated, the nyquist slope increases and the intercept decreases as reflected by lines 2200, 2210, etc. becoming steeper and moving toward the Y-axis. For low start-up detection, clinical data indicate that the nyquist slope typically increases significantly after sensor insertion and initialization, then settles to a certain level. One explanation for this is that as the sensor becomes increasingly wet, there is a significant change in the species diffusivity and concentration, which is reflected in the Valley impedance.

In fig. 23A, Isig 2230 of first working electrode WE1 was initially lower than expected (about 10nA) and it took some time to catch up with Isig 2240 of second working electrode WE 2. Thus, in this particular example, WE1 is designated as having a low start. EIS data reflects this low start in two ways. First, as shown in FIG. 23A, the real impedance of WE1 at 1kHz (2235) is much higher than the 1kHz real impedance 2245 of WE 2. Second, the nyquist slope of WE1 (fig. 23B) starts lower, has a larger intercept 2237, and requires more time to settle when compared to the nyquist slope of WE2 (fig. 23C). As will be discussed later, these two characteristics (real impedance at 1kHz and nyquist slope) can be used as diagnostic inputs in the fusion algorithm to decide which of the two electrodes can carry higher weight when calculating the fusion signal. In addition, one or both of these flags may be used in a diagnostic procedure to determine whether the sensor as a whole is acceptable or should be terminated and replaced.

By definition, a signal (or Isig) dip refers to a condition of low sensor signal, which is mostly temporary in nature, e.g. on the order of hours. Such a low signal may be caused by, for example, some form of biological blockage on the sensor surface or by pressure applied at the insertion site (e.g., while sleeping on its side). During this time, the sensor data is considered unreliable; however, the signal eventually recovers. In the EIS data, this type of dip (as opposed to a dip caused by blood glucose changes in the patient) is reflected in the 1kHz real impedance data, as shown in FIG. 24.

Specifically, in fig. 24, Isig 2250 of first working electrode WE1 and Isig2260 of second working electrode WE2 both started at the leftmost end (i.e., 6 pm) at 25 nA. Over time, both Isigs fluctuate, reflecting glucose fluctuations near the sensor. Around about the first 12 hours (i.e., up to about 6 am), both isigs are fairly stable, as are their respective 1kHz real impedances 2255, 2265. However, between about 12 and 18 hours (i.e., between 6 am and noon), Isig2260 of WE2 begins to dip sharply and continues to dip over the next several hours until about 9 pm. During this time Isig 2250 of WE1 also had some degree of dip but Isig 2250 was much more stable than Isig2260 of WE2 and had considerably less dip. The Isigs characteristics of WE1 and WE2 are also reflected in their respective 1kHz real impedance data. Therefore, as shown in fig. 24, during the above-described period of time, although the 1kHz real impedance of WE1(2255) remains fairly stable, the 1kHz real impedance of WE2(2265) increases significantly.

By definition, a loss of sensitivity is a condition where the sensor signal (Isig) becomes low and unresponsive for long periods of time, and is generally unrecoverable. Loss of sensitivity can occur for a variety of reasons. For example, electrode poisoning (electrodeionization) greatly reduces the effective surface area of the working electrode, thereby severely limiting the current amplitude. Loss of sensitivity may also occur due to under-or hypoxia at the insertion site. In addition, because some forms of extreme surface blockage (i.e., a more persistent form of signal drop caused by biological or other factors) restricts the passage of glucose and oxygen through the sensor membrane, a loss of sensitivity may occur, thereby reducing the number/frequency of chemical reactions that generate current in the electrodes and ultimately the sensor signal (Isig). Note that the various causes of sensitivity loss described above apply to short-term (7-10 day wear) sensors and long-term (6 month wear) sensors.

In EIS data, sensitivity loss is usually premised on an increase in the absolute value of the phase (| phase |) and the absolute value of the imaginary impedance (| imaginary impedance |) in a relatively high frequency range (e.g., 128Hz and above and 1kHz and above, respectively). FIG. 25A shows an example of a normally operating glucose sensor in which the sensor current 2500 is responsive to glucose (i.e., Isig 2500 tracks glucose fluctuations), but all relevant impedance outputs (such as, for example, a 1kHz real impedance 2510, a 1kHz imaginary impedance 2530, and a phase (2520) at a frequency of about 128Hz or above) remain stable because they are substantially glucose independent.

Specifically, the top plot in FIG. 25A shows that after the first few hours, the 1kHz real impedance 2510 remains fairly stable at about 5 kilo-ohms (the 1kHz virtual impedance 2530 remains fairly stable at about-400 ohms). In other words, at 1kHz, real impedance data 2510 and virtual impedance data 2530 are substantially glucose independent, so that they can be used as a signature or independent indicator of health, condition, and reliability of the particular sensor ultimately being analyzed. However, as previously mentioned, different impedance-related parameters may exhibit no glucose dependence at different frequency ranges, and in each case the range may depend on the overall sensor design, e.g. electrode type, surface area of the electrode, thickness of the membrane, permeability of the membrane, etc.

Thus, in the example of fig. 25B, the top graph again shows that the sensor current 2501 is responsive to glucose for 90% of the short tubeless electrode designs, and that the 1kHz real impedance 2511 remains fairly stable at about 7.5 kohms after the first few hours. The bottom plot in fig. 25B shows real impedance data for frequencies between 0.1Hz (2518) and 1kHz (2511). It can be seen that the real impedance data at 0.1Hz (2518) is very dependent on glucose. However, as indicated by reference numerals 2516, 2514 and 2512, as the frequency is increased from 0.1Hz to 1kHz (i.e. for impedance data measured at frequencies closer to 1 kHz), the real impedance becomes increasingly glucose independent.

Returning to fig. 25A, the middle graph shows that phase 2520 is substantially independent of glucose at relatively high frequencies. It should be noted, however, that for the sensor under analysis, the "relatively high frequency" associated with such a parameter (phase) refers to frequencies of 128Hz and above. In this regard, the figure shows that the phase of all frequencies between 128Hz and 8kHz is stable throughout the time period shown. On the other hand, as can be seen from the bottom graph of fig. 25C, while phase 2522 is stable at (and above) 128Hz, phase 2524 fluctuates at a frequency that is increasingly less than 128Hz (i.e., it becomes increasingly glucose dependent, and to a different degree). Note that the electrode design of the example of fig. 25C is the same as that used in fig. 25B, and the top view in the former is the same as that in the latter.

Figure 26 shows an example of sensitivity loss due to hypoxia at the insertion site. In this case, the insertion site becomes anoxic after day 4 (indicated by the black vertical line in fig. 26), resulting in a low and non-responsive sensor current 2600. The real impedance 2610 at 1kHz remains stable, indicating that there is no physical blockage on the sensor. However, the changes in the relatively high frequency phase 2622 and 1kHz virtual impedance 2632 are consistent with a loss in sensitivity, as indicated by the corresponding downward arrows, indicating that this type of loss is due to hypoxia at the insertion site. Specifically, fig. 26 shows that the phase at the higher frequency (2620) and the imaginary impedance (2630) at 1kHz become more negative before the sensor loses sensitivity (indicated by the black vertical lines), and it continues to decline as the loss of sensor sensitivity continues. Thus, as described above, at relatively high frequency ranges (e.g., 128Hz and above and 1kHz and above, respectively), this sensitivity loss is advanced or predicted by an increase in the absolute value of the phase (| phase |) and the absolute value of the imaginary impedance (| imaginary impedance |).

The above characteristics can be verified by in vitro tests, an example of which is shown in fig. 27. Figure 27 shows the results of in vitro testing of the sensor, in which hypoxia was simulated at different glucose concentrations. In the top panel, Isig fluctuates as the glucose concentration increases from 100mg/dl (2710) to 200mg/dl (2720), 300mg/dl (2730), and 400mg/dl (2740) and then again falls to 200mg/dl (2750). In the bottom graph, the phase is generally stable at relatively high frequencies, indicating that it is not dependent on glucose. However, at very low oxygen concentrations (e.g., at 0.1% O)₂Lower), the phase may fluctuate at relatively high frequencies, as indicated by the circled areas and arrows 2760, 2770. Note the amplitude and/or direction of the wave(i.e., positive or negative) depends on various factors. For example, the higher the ratio of glucose concentration to oxygen concentration, the greater the amplitude of the phase fluctuation. In addition, the particular sensor design and the time of use of the sensor (i.e., measured in terms of time after implantation) can affect such fluctuations. Thus, for example, the older the sensor, the more susceptible it is to interference.

Fig. 28A-28D illustrate another example of sensitivity loss due to oxygen starvation of redundant working electrodes WE1 and WE 2. As shown in fig. 28A, the 1kHz real impedance 2810 is stable even if the sensor current 2800 fluctuates and eventually becomes unresponsive. In addition, as previously described, the change in virtual impedance 2820 at 1kHz is consistent with a loss in sensor sensitivity. However, in addition, fig. 28B shows real impedance data and imaginary impedance data (2830 and 2840, respectively) at 0.105 Hz. The latter (more commonly referred to as "0.1 Hz data") indicates that while the virtual impedance at 0.1Hz appears to be fairly stable, the real impedance at 0.1Hz 2830 increases significantly as the sensor loses sensitivity. Furthermore, as shown in fig. 28C, sensitivity is lost due to hypoxia, V_cntr2850 orbit to 1.2 volts.

In short, these figures show the finding of sensitivity loss due to hypoxia with a lower imaginary impedance of 1kHz (i.e. the latter becomes more negative), a higher real impedance of 0.105Hz (i.e. the latter becomes more positive) and V_cntrThe tracks are related. In addition, the anoxic process and V_cntrThe rails are often associated with an increase in the capacitive component of the electrochemical circuit. Note that in some diagnostic procedures described later, a real impedance of 0.105Hz may not be used, as this relatively low frequency real impedance data may be dependent on the analyte.

Finally, in connection with the examples of FIGS. 28A-28D, it should be noted that impedance measurements at 1kHz and higher frequencies typically cause EIS-induced spikes in Isig. This is shown in fig. 28D, where the original Isig of WE2 is plotted against time. When the spike begins, the sharp increase in Isig is a non-faradaic process due to the double layer capacitance charging. Thus, in addition to the lower 1kHz imaginary impedance, the higher 0.105Hz real impedance and V as described above_cntrIn addition to the rails, sensitivity loss due to hypoxia may also be associated with higher EIS-induced spikes。

Fig. 29 shows another example of sensitivity loss. This case can be considered as an extreme case of the Isig dip described above in connection with fig. 24. Here, a lower sensor current 2910 was observed from the time of insertion, indicating that there was a problem with the insertion process leading to electrode clogging. The 1kHz real impedance 2920 is significantly higher than the same parameter values for a normally operating sensor shown in FIG. 25A, while the relatively higher frequency phase 2930 and the 1kHz imaginary impedance 2940 both move to more negative values. The movement of the relatively high frequency phase 2930 and the 1kHz imaginary impedance 2940 indicates that the loss of sensitivity may be due to oxygen depletion, which in turn may be caused by blockages on the sensor surface.

30A-30D show data for another redundant sensor in which the relative differences between 1kHz real impedance and 1kHz imaginary impedance and 0.1Hz real impedance between two or more working electrodes can be used to detect a loss of sensitivity due to biofouling. In this example, WE1 shows greater sensitivity loss than WE2, as evident from WE2 at 0.105Hz (3030) with a higher 1kHz real impedance 3010, a lower 1kHz imaginary impedance 3020, and a higher real impedance. However, also in this example, V_cntr3050 do not move along the track. Furthermore, as shown in fig. 30D, the height of the spikes in the raw Isig data does not vary much over time. This indicates that for the loss of sensitivity due to biofouling, V_cntrThe increase in rail and spike height is correlated. Furthermore, the fact that the height of the spike in the raw Isig data does not vary greatly over time indicates that the capacitive component of the circuit does not vary significantly over time, and therefore, the loss of sensitivity due to biofouling is related to the resistive component (i.e., diffusion) of the circuit.

Various of the above-described impedance-related parameters may be used, alone or in combination, as inputs to: (1) an EIS-based sensor diagnostic program; and/or (2) fusion algorithms for producing more reliable sensor glucose values. With respect to the former, fig. 31 shows how EIS-based data (i.e., impedance-related parameters or characteristics) are used in the diagnostic process to determine in real time whether a sensor is functioning properly, or should be replaced.

The diagnostic routine shown in the flowchart of fig. 31 is based on the periodic collection of EIS data (such as, for example, every hour, every half hour, every 10 minutes, or any other interval (including continuously)), which may be appropriate for the particular sensor being analyzed. At each such interval, the EIS may operate over the entire frequency spectrum (i.e., a "full scan"), or may operate over a selected frequency range, or even at a single frequency. Thus, for example, for an hourly data collection scheme, EIS may be performed at frequencies in the μ Hz to MHz range, and may also be run at smaller frequency ranges, such as, for example, between about 0.1Hz to about 8kHz, as described above. In various embodiments, EIS data collection can be performed alternately between full scan and smaller spectrum, or according to other schemes.

The time frequency of EIS implementation and data collection may be determined by various factors. For example, each embodiment of the EIS consumes a certain amount of power, which is typically provided by the battery of the sensor (i.e., the battery running the sensor electronics) (including the ASIC described later). Thus, battery capacity and remaining sensor life may be helpful in determining the number of EIS runs, as well as the sampling frequency width for each run. Furthermore, certain situations may require monitoring EIS parameters (e.g., real impedance at 1 kHz) at certain frequencies based on a first schedule (e.g., once every few seconds or minutes), while monitoring other parameters and/or the same parameters at other frequencies based on a second schedule (e.g., less frequently). In these cases, the diagnostic routines can be tailored to the specific sensors and requirements so that battery power can be maintained and unnecessary and/or redundant EIS data acquisition can be avoided.

Note that in some embodiments, the diagnostic procedure (e.g., the procedure shown in FIG. 31) requires a series of separate "tests" that are performed in order to perform real-time monitoring of the sensors. Multiple tests or flags (also referred to as "multiple flags") are implemented because each time the EIS is run (i.e., each time the EIS procedure is performed), data or characteristics relating to multiple impedance-based parameters may be collected, which may be used to detect sensor status or quality (including, for example, whether or not the sensor is malfunctioning or is malfunctioning). In performing sensor diagnostics, there may be times when one diagnostic test indicates a fault, while the other diagnostic(s) may indicate no fault. Thus, the availability of multiple impedance-related parameters and the implementation of multiple test procedures is advantageous, as the diversity of some tests may serve as a validity check for some other tests. Thus, real-time monitoring using a multi-marker procedure may involve a degree of built-in redundancy.

In view of the above, the logic of the diagnostic routine shown in FIG. 31 begins at 3100 after a sensor has been inserted/implanted and an EIS run has been performed to provide EIS data as input. At 3100, using the EIS data as input, it is first determined whether the sensor is still in place. Thus, if the | Z | slope is found to be constant across the frequency band (or range) of the test and/or the phase angle is about-90 °, it is determined that the sensor is no longer in place and an alert is sent to the patient/user indicating that the sensor has been unplugged, for example. The specific parameters (and their respective values) described herein for detecting sensor pull-out are based on the discovery that once the sensor is off the body and the membrane is no longer hydrated, the impedance spectral response appears like a capacitor.

If it is determined that the sensor is still in the proper position, the logic moves to step 3110 to determine if the sensor is properly initialized. As shown, the "initialization check" is performed by determining: (1) whether or not at 1kHz (Z)_n-Z₁)/Z₁|>30% of the total amount of Z₁Is the real impedance at the first measurement, and Z_nIs the impedance measured at the next interval, as described above; and (2) whether the phase angle change at 0.1Hz is greater than 10 deg. If the answer to either question is "yes," then the test is satisfactory, i.e., test 1 has not failed. Otherwise, test 1 is marked as failed.

At step 3120, test 2 interrogates two consecutive EIS runs at a phase angle of-45 ° (f)₂-f₁) Whether the frequency difference therebetween is greater than 10 Hz. Likewise, the answer of "NO" is labeledRecording as failure; otherwise, test 2 is satisfactory.

Test 3 in step 3130 is a hydration test. Here, the problem is the current impedance Z_nImpedance Z after initialization at 1kHz or not_pi. If so, then the test is satisfactory; otherwise, test 3 is marked as failed. Test 4 of step 3140 is also a hydration test, but this time at a lower frequency. Thus, the test query is during sensor operation after initialization, Z_nWhether less than 300 kilo-ohms at 0.1 Hz. Likewise, the answer "no" indicates that the sensor failed test 4.

In step 3150, test 5 asks whether the low frequency Nyquist slope increases overall from 0.1Hz to 1 Hz. As previously mentioned, the nyquist slope for a relatively low frequency should increase with time for a properly functioning sensor. Thus, if the answer to the question is "yes," this test is satisfactory; otherwise, the test is marked as failed.

Step 3160 is the last test of this diagnostic program embodiment. The question here is whether the real impedance is reduced overall. Also, as previously mentioned, in a normally operating sensor, the real impedance is expected to decrease over time. Thus, a "yes" here means that the sensor is functioning properly; otherwise, the sensor fails test 6.

Once all 6 tests have been conducted, a determination is made at 3170 as to whether the sensor is functioning properly, or is malfunctioning. In this embodiment, the sensor is determined to be functioning properly (3172) if it passes at least 3 of the 6 tests. In other words, to determine that a fault has occurred (3174), the sensor must fail in at least 4 of the 6 tests. In alternative embodiments, different rules may be used to evaluate normal operation versus sensor failure. Further, in some embodiments, each test may be weighted such that when overall sensor operation (normal versus fault) is determined, the assigned weight reflects, for example, the importance of the test, or the importance of a particular parameter queried for the test. For example, one test may be twice as weighted as another test, but only half as weighted as the third test, and so on.

In other alternative embodiments, a different number of tests and/or different sets of EIS-based parameters may be used for each test. Fig. 32A and 32B show an example of a diagnostic program for real-time monitoring containing 7 tests. Referring to FIG. 32A, the logic begins at 3200 after a sensor has been inserted/implanted and an EIS procedure has been performed to provide EIS data as input. At 3200, using the EIS data as input, it is first determined whether the sensor is still in place. Thus, if the | Z | slope is found to be constant across the frequency band (or range) of the test and/or the phase angle is about-90 °, it is determined that the sensor is no longer in place and an alert is sent to the patient/user indicating that the sensor has been unplugged, for example. On the other hand, if the sensor is determined to be in the proper position, the logic moves to the initiation of a diagnostic check (3202).

At 3205, test 1 is similar to test 1 of the diagnostic procedure discussed above in connection with FIG. 31, except that instant test 1 specifies that a later measurement Z is taken 2 hours after the first measurement_n. Also in this example, Z_n＝Z_{2 hours}. More specifically, test 1 compares the real impedance 2 hours after (sensor implantation and) initialization with the pre-initialization value. Similarly, the second part of test 1 asks whether the difference between the 2 hours post-initialization phase and the pre-initialization phase is greater than 10 at 0.1 Hz. As previously described, if the answer to either query is affirmative, then it is determined that the sensor is normally hydrated and initialized, and test 1 is satisfactory; otherwise, the sensor fails the test. It should be noted that even if the instant test interrogates impedance and phase changes 2 hours after initialization, the time interval between any two consecutive EIS runs may be shorter or longer depending on a number of factors, including, for example, sensor design, electrode redundancy, the degree to which the diagnostic routine includes redundant tests, battery power, etc.

Turning to 3210, the logic next performs a sensitivity loss check by asking if the percentage change in impedance magnitude at 1kHz after the 2 hour interval (n +2) and the percentage change in Isig is greater than 30%. If the answer to both questions is "yes," then it is determined that the sensor is losing sensitivity and, therefore, test 2 is determined to have failed. Note that although test 2 is described herein based on a preferred percentage difference of 30%, in other embodiments, the percentage difference in impedance magnitude at 1kHz and the percentage difference in Isig may be in the range of 10% -50% for purposes of this test.

Test 3 (at 3220) is similar to test 5 of the algorithm shown in fig. 31. Here, as before, the question is whether the low frequency nyquist slope increases overall from 0.1Hz to 1 Hz. If so, then the test passes; otherwise, the test fails. The test may also set a threshold or acceptable range of percentage change in the low frequency nyquist slope beyond which the sensor may be deemed to be malfunctioning, or at least may trigger further diagnostic tests, as shown at 3220. In embodiments of the present invention, such a threshold/acceptable range of percentage change in the low frequency nyquist slope may be in the range of about 2% to about 20%. In some preferred embodiments, the threshold may be about 5%.

The logic next moves to 3230, which is another low frequency test, this time involving phase and impedance magnitude. More specifically, the phase test asks whether the phase at 0.1Hz continuously increases with time. If so, the test fails. As with other tests that monitor parameter trends, the low frequency phase test of test 4 may also set a threshold or acceptable range for the percentage change in low frequency phase beyond which the sensor may be deemed to be malfunctioning, or at least of concern. In some preferred embodiments, this threshold/acceptable range of percentage change in low frequency phase may be in the range of about 5% to about 30%. In some preferred embodiments, the threshold may be about 10%.

As previously mentioned, test 4 also includes a low frequency impedance magnitude test, where the question is whether the impedance magnitude at 0.1Hz continuously increases with time. If so, the test fails. Note that if either the phase test or the impedance magnitude test fails, test 4 is considered to be a "failure". The low frequency impedance magnitude test of test 4 may also set a threshold or acceptable range for the percentage change in low frequency impedance magnitude beyond which the sensor may be deemed to be malfunctioning, or at least of concern. In some preferred embodiments, such a threshold/acceptable range of percentage change in low frequency impedance magnitude may be in the range of about 5% to about 30%. In some preferred embodiments, the threshold may be about 10%, with the impedance magnitude of a normal sensor typically ranging between about 100 kilo-ohms to about 200 kilo-ohms.

Test 5 (at 3240) is another sensitivity loss check that can be considered complementary to test 2. Here, if both the percent change in Isig and the percent change in impedance magnitude at 1kHz are greater than 30%, then it is determined that the sensor is recovering from the loss of sensitivity. In other words, it is determined that the sensor has previously experienced some loss of sensitivity, even if for some reason the loss of sensitivity was not detected by test 2. As with test 2, although test 5 is illustrated based on a preferred percentage difference of 30%, in other embodiments, to perform this test, the percentage difference in impedance magnitude at Isig and 1kHz may be in the range of 10% -50%.

Turning to 3250, test 6 provides a sensor function test that includes specific failure criteria determined based on observed data and a specific sensor design. Specifically, in one embodiment, it may be determined that the sensor has failed and is therefore unlikely to respond to glucose if at least two of the following three criteria are met: (1) isig is less than 10 nA; (2) the virtual impedance at 1kHz is less than-1500 ohms; and (3) a phase at 1kHz of less than-15 deg. Therefore, if any two of (1) - (3) are not satisfied, it is determined that test 6 has passed. Note that in other embodiments, the Isig fork of the test may fail if Isig is less than about 5nA to about 20 nA. Similarly, if the imaginary impedance at 1kHz is less than about-1000 ohms to about-2000 ohms, the second fork may fail. Finally, if the phase at 1kHz is less than about-10 ° to about-20 °, the phase fork may fail.

Finally, step 3260 provides another sensitivity check in which the parameters are evaluated at a lower frequency. Thus, test 7 asks whether the magnitude of the difference between the ratio of the imaginary impedance at 0.1Hz to Isig (n +2) on the one hand and the previous value of the ratio on the other hand is greater than 30% of the magnitude of the difference of the previous value of the ratio. If so, the test fails; if not, the test passes. Here, although test 7 is illustrated based on a preferred percentage difference of 30%, in other embodiments, the percentage difference may be in the range of 10% -50% for purposes of this test.

Once all 7 tests have been conducted, a determination is made at 3270 as to whether the sensor is functioning properly, or whether an alarm should be issued indicating that the sensor has failed (or may be failing). As shown, in this embodiment, if the sensor passes at least 4 of the 7 tests, it is determined that it is functioning properly (3272). In other words, to determine that a fault has occurred or is at least of concern (3274), the sensor must fail in at least 4 of the 7 tests. If the sensor is determined to be "problematic" (3274), an alert may be sent to, for example, the patient/user to obtain this effect. As previously described, in alternative embodiments, different rules may be used to evaluate normal operation versus sensor failure/concern. Further, in some embodiments, each test may be weighted such that when overall sensor operation (normal versus fault) is determined, the assigned weight reflects, for example, the importance of the test, or the importance of a particular parameter queried for the test.

As previously mentioned, in embodiments of the invention described herein, various of the above-described impedance-related parameters may be used, alone or in combination, as inputs to one or more fusion algorithms to produce more reliable sensor glucose values. In particular, it is well known that, unlike a single sensor (i.e., single working electrode) system, multiple sensing electrodes provide more reliable glucose readings, as multiple signals obtained from two or more working electrodes can be fused to provide a single sensor glucose value. This signal fusion uses the quantitative inputs provided by EIS to calculate the most reliable output sensor glucose values from the redundant working electrodes. It should be noted that although the discussion that follows may describe various fusion algorithms in terms of the first working electrode (WE1) and the second working electrode (WE2) as redundant electrodes, this is exemplary and not limiting, as the algorithms described herein and their underlying principles are applicable to and can be used in redundant sensor systems having more than 2 working electrodes. Furthermore, as previously described, the redundant electrodes may be contained on/within a single sensor, multiple (identical) sensors, or multiple non-identical sensors.

Fig. 33A and 33B show top level flow charts of two alternative methods, each of which includes a fusion algorithm. Specifically, fig. 33A is a flow chart relating to the current (Isig) based fusion algorithm, while fig. 33B is a flow chart for Sensor Glucose (SG) fusion. As can be seen from the graph, the main difference between these two methods is the calibration time. Thus, fig. 33A shows that for Isig fusion, calibration 3590 is performed after fusion 3540 is complete. That is, the redundant Isig from WE1 to WEn is fused into a single Isig 3589, which is then calibrated to produce a single sensor glucose value 3598. On the other hand, for SG fusion, calibration 3435 is done for each individual Isig from WE1 to WEn to produce a calibrated SG value (e.g., 3436, 3438) for each working electrode. Thus, the SG fusion algorithm provides independent calibration of each of the multiple isigs, which may be preferred in some embodiments of the invention described herein. Once calibrated, the multiple calibrated SG values are fused into a single SG value 3498.

It is important to note that each of the flow diagrams shown in fig. 33A and 33B contain spike filtering processes (3520, 3420). As discussed above with respect to sensitivity loss, impedance measurements at 1kHz and higher frequencies typically cause EIS-induced spikes in Isig. Thus, once the EIS procedure has been performed for each of the electrodes WE1 through WEn, for SG fusion and Isig fusion it is preferable to first filter Isig 3410, 3412, etc. and 3510, 3512, etc. to obtain the respective filtered Isig 3422, 3424, etc. and 3522, 3524, etc. The filtered Isig is either used for Isig fusion or first calibrated and then used for SG fusion, as described below. As will become clear in the discussion that follows, both fusion algorithms need to compute and assign weights based on various factors.

Fig. 34 shows details of a fusion algorithm 3440 for SG fusion. Basically, four factors need to be checked before determining the fusion weight. First, integrity check 3450 involves determining whether each of the following parameters is within a specified range for normal sensor operation (e.g., predetermined lower and upper threshold values): (i) isig; (ii)1kHz real impedance and imaginary impedance; (iii)0.105Hz real and imaginary impedances; and (iv) the nyquist slope. As shown, integrity check 3450 includes a boundary check 3452 and a noise check 3456, where for each check, the above parameters are used as input parameters. It should be noted that for the sake of brevity, the real and/or imaginary impedances at one or more frequencies are simply represented in fig. 33A-35 as "Imp" of impedances. In addition, both real and imaginary impedances can be calculated using impedance magnitude and phase (which is also shown as input in fig. 33A and 33B).

The output of each of the bounds check 3452 and the noise check 3458 is a respective Reliability Index (RI) for each redundant working electrode. Thus, the output from the bounds check includes, for example, RI _ bounds _ We₁(3543) And RI _ Border _ We₂(3454). Similarly, for noise checking, the output includes, for example, RI _ NOT _ We₁(3457) And RI _ noise _ We₂(3458). The boundary and noise reliability index for each working electrode is calculated based on a range consistent with normal sensor operation as described above. Thus, if any parameter is outside the specified range for a particular electrode, the reliability index for that particular electrode is reduced.

Note that the threshold or range of the above parameters may depend on various factors, including the particular sensor and/or electrode design. However, in a preferred embodiment, typical ranges for some of the above parameters may be, for example, as follows: the boundary threshold of 1kHz real impedance is [0.3e + 42 e +4 ]; the boundary threshold of the 1kHz imaginary impedance [ -2e +3, 0 ]; a boundary threshold of 0.105Hz real impedance ═ 2e + 47 e + 4; boundary threshold of 0.105Hz imaginary impedance [ -2e +5-0.25e +5 ]; the threshold of the nyquist slope is [ 25 ]. For example, the noise may be calculated using a second order center difference method, where the noise is considered to be beyond the noise boundary if the noise is above a certain percentage (e.g., 30%) of the median of each variable buffer.

Second, sensor dips can be detected using the sensor current (Isig) and a real impedance of 1 kHz. Therefore, as shown in fig. 34, Isig and "Imp" are used as inputs to the dip detection 3460. Here, the first step is to determine whether there is divergence between Isigs, and whether such divergence is reflected in the real impedance data at 1 kHz. This can be achieved by using a mapping 3465 between the Isig similarity index (RI _ similarity _ Isig12)3463 and the 1kHz real impedance similarity index (RI _ similarity _ imp12) 3464. This mapping is important because it helps to avoid false positives in cases where the dip is not true. If Isig divergence is true, the algorithm will select the sensor with the higher Isig.

According to one embodiment, the divergence/convergence of two signals (e.g., two Isig or two 1kHz real impedance data points) may be calculated as follows:

diff_va1＝abs(va1-(va1+va2)/2)；

diff_va2＝abs(va2-(va1+va2)/2)；

RI _ similarity ═ 1- (diff _ va1+ diff _ va 2)/(average value (abs (va1+ va2))/4)

Where va1 and va2 are two variables, and RI _ sim (similarity index) is an index of convergence or divergence of a measurement signal. In this embodiment, RI _ similarity must be limited to between 0 and 1. Therefore, if the RI _ similarity calculated as above is less than 0, it will be set to 0, and if greater than 1, it will be set to 1.

Mapping 3465 is performed by using ordinary linear regression (O L R). however, when O L R does not work well, robust median slope linear regression (RMS L R) may be used-for example, for the Isig similarity index and the 1kHz real impedance index, two mapping processes are required (i) to map the Isig similarity index to the 1kHz real impedance similarity index and (ii) to map the 1kHz real impedance similarity index to the Isig similarity index both mapping processes produce two residuals, res12 and res 21. each of the dip reliability indices 3467, 3468 may be calculated as follows:

RI _ dip ═ 1- (res12+ res21)/(RI _ similarity _ isig + RI _ similarity _1K _ real _ impedance)

The third factor is the sensitivity loss 3470, which can be detected using, for example, the 1kHz virtual impedance trend over the past 8 hours. If the trend of one sensor becomes negative, the algorithm will rely on the other sensor. If both sensors lose sensitivity, a simple average is taken. The trend may be calculated by smoothing the imaginary impedance at 1kHz, which tends to be noisy, using a strong low pass filter, and determining whether the correlation coefficient is negative or the slope is negative by using a correlation coefficient or linear regression with respect to time, e.g., over the past 8 hours. Each of the sensitivity loss reliability indices 3473, 3474 is then assigned a binary value of 1 or 0.

The total Reliability Index (RI) for each of we1, we2, …, wen is calculated as follows:

RI_we₁RI _ dip _ we₁× RI _ sensitivity _ loss _ we₁× RI _ Border _ we₁× RI _ NOISE _ WE₁

RI_we₂RI _ dip _ we₂× RI _ sensitivity _ loss _ we₂× RI _ Border _ we₂× RI _ NOISE _ WE₂

RI_we₃RI _ dip _ we₃× RI _ sensitivity _ loss _ we₃× RI _ Border _ we₃× RI _ NOISE _ WE₃

RI_we₄RI _ dip _ we₄× RI _ sensitivity _ loss _ we₄× RI _ Border _ we₄× RI _ NOISE _ WE₄

RI_we_nRI _ dip _ we_n× RI _ sensitivity _ loss _ we_n× RI _ Border _ we_n× RI _ NOISE _ WE_n

After calculating the reliability index corresponding to each working electrode, the weight of each electrode can be calculated as follows:

weight of_we₁＝RI_we₁/(RI_we₁+RI_we₂+RI_we₃+RI_we₄+…+RI_we_n)

Weight _ we₂＝RI_we₂/(RI_we₁+RI_we₂+RI_we₃+RI_we₄+…+RI_we_n)

Weight _ we₃＝RI_we₃/(RI_we₁+RI_we₂+RI_we₃+RI_we₄+…+RI_we_n)

Weight _ we₄＝RI_we₄/(RI_we₁+RI_we₂+RI_we₃+RI_we₄+…+RI_we_n)

Weight _ we_n＝RI_we_n/(RI_we₁+RI_we₂+RI_we₃+RI_we₄+…+RI_we_n)

Based on the above, the fused SG 3498 is calculated as follows:

SG is weight _ we₁×SG_we₁+ weight _ we₂×SG_we₂+ weight _ we₃×SG_we₃+

Weight _ we₄×SG_we₄+ … + weight _ we_n×SG_we_n

The last factor is related to artifacts in the final sensor reading, as may be caused by the instantaneous weight change of the sensor fusion. This can be avoided by applying a low pass filter 3480 to smooth the RI of each electrode or by applying a low pass filter to the final SG. When the former is used, the filtered reliability indices (e.g., RI _ We1 and RI _ We2 (3482, 3484)) are used to calculate the weight of each electrode and thus the fused SG 3498.

Fig. 35 shows details of the fusion algorithm 3540 for Isig fusion. It can be seen that this algorithm is substantially similar to the algorithm for SG fusion shown in fig. 34, with two exceptions. First, as previously described, for Isig fusion, calibration constitutes the last step of the process, where a single fused Isig 3589 is calibrated to generate a single sensor glucose value 3598. See also fig. 33B. Second, although SG fusion uses SG values of a plurality of electrodes to calculate a final SG value 3498, a fused Isig value 3589 is calculated using filtered Isig (3522, 3524, etc.) of a plurality of electrodes.

In a closed loop study involving non-diabetic populations, the fusion algorithm described above was found to provide a considerable improvement in Mean Absolute Relative Difference (MARD), with the low start problem being most pronounced on day 1, and thus potentially having a significant impact on the accuracy and reliability of the sensor as well as the overall (i.e., over the 7-day lifetime of the sensor). The study evaluated data for an 88% distributed layout design with high current density (nominal) plating using three different methods (1) calculating a sensor glucose value (SG) by fusion using Ferrari algorithm 1.0 from Medtronic Minimed (as described above SG fusion algorithm); (2) calculate an SG (by ISIG fusion algorithm described above) by identifying better ISIG values using 1kHz EIS data; and (3) calculate an SG by using the higher ISIG value (i.e., without using EIS). The data details of this study are as follows:

(1) ferrari 1.0Alg based SG for 88% distributed layout with high current density (nominal) plating

(2) More optimal ISIG based SG using 1kHz EIS for 88% distributed layout with high current density (nominal) plating

(3) Higher ISIG based SG for 88% distributed layout with high current density (nominal) plating

From the above data, it was found that with the first method, the MARD (%) at day 1 was 19.52% and the total MARD was 12.28%. For the second method, the MARD at day 1 was 15.96%, and the total MARD was 11.83%. Finally, for the third method, the MARD at day 1 was 17.44%, and the total MARD was 12.26%. Therefore, calculation of SG using 1kHz EIS (i.e., the second method) based on better ISIG seems to provide the greatest advantage for this design with redundant electrodes. In particular, the reduction in day 1 MARD may be due to better low start tests, e.g. using EIS. Furthermore, in this study, the total percent MARD was more than 1% lower than the overall average MARD value of 13.5% for WE1 and WE 2. Note that in the above approach, the data transformation may be processed, for example, by a filtering approach to minimize the severity of the transformation, for example, by using a low pass filter 3480 as discussed above in connection with fig. 33A-35.

It is worth repeating that sensor diagnostics, including, for example, low start-up, sensitivity loss, and evaluation of slump events, depend on various factors, including sensor design, number of electrodes (i.e., redundancy), electrode distribution/configuration, and the like. In this way, the EIS-based parameters may be substantially independent of the actual frequency or frequency range of glucose and thus the independent signature or predictor for one or more of the above-described failure modes may also depend on the specific sensor design. For example, although it has been found that, as described above, the imaginary impedance at relatively high frequencies can be used to predict the loss of sensitivity (where the imaginary impedance is substantially independent of glucose), the level of glucose dependence, and thus the particular frequency range in which the imaginary impedance is used as a marker for the loss of sensitivity, can be shifted (higher or lower) depending on the actual sensor design.

More specifically, as sensor designs increasingly tend to use redundant working electrodes, the latter must be smaller in size to maintain the overall size of the sensor. In turn, the size of the electrodes can affect the frequency with which a particular diagnosis may be queried. In this regard, it is important to note that the fusion algorithms described herein and shown in FIGS. 33A-35 should be considered illustrative, rather than limiting, in that each algorithm may be modified as needed to use EIS-based parameters at a frequency that exhibits the least amount of glucose dependence based on the type of sensor under analysis.

Furthermore, experimental data indicate that human tissue architecture may also affect glucose dependence at different frequencies. For example, in children, a real impedance at 0.105Hz has been found to be a substantially glucose independent indicator of low onset detection. It is believed that this is a result of changes in the child's anatomy, such as the valburg impedance, which is primarily related to the resistive component. See also the subsequent discussion regarding interference detection.

Embodiments of the invention described herein also relate to using EIS in optimizing sensor calibration. By way of background, in the current method, the slope of the BG versus Isig curve that can be used to calibrate subsequent Isig values is calculated as follows:

where α is an exponential function of the time constant, β is a function of the change in blood glucose, and the offset is a constant.

However, there are cases where the above linear relationship does not hold, for example, during a period in which the sensor undergoes a transition. As shown in FIG. 37, it is clear that Isig-BG pair 1 is significantly different from Isig-BG pair 3 and 4 in terms of Isig and BG relationships. For these types of conditions, using a constant offset tends to produce inaccurate results.

To address this issue, one embodiment is directed to using EIS-based dynamic offsets, where EIS measurements are used to define a sensor state vector as follows:

v-real _ impedance _1K, imaginary _ impedance _1K, nyquist _ slope, nyquist _ R _ square }

Wherein all elements in the vector are essentially BG independent. Note that nyquist _ R _ square is the R-square of the linear regression used to calculate the nyquist slope, i.e., the square of the correlation coefficient between real and imaginary impedance at relatively low frequencies, a low R-square indicating sensor performance anomalies. For each Isig-BG pair, a state vector is assigned. If a significant difference in the state vector is detected (e.g., | V2-V3|, for the example shown in FIG. 37), then 3 and 4 are assigned different offset values when compared to 1 and 2. Therefore, by using this dynamic offset method, a linear relationship between Isig and BG can be maintained.

In a second embodiment, the calibration may be performed using an EIS-based segmentation method. Using the example of fig. 37 and vector V, it can be determined that the sensor states during 1 and 2 are significantly different from the sensor states during 3 and 4. Thus, the calibration buffer can be divided into two parts, as follows:

isig _ buffer 1 ═ Isig1, Isig 2; BG _ BUFFER 1 ═ BG1, BG 2%

Isig _ buffer 2 ═ Isig3, Isig 4; BG _ BUFFER 2 ═ BG3, BG 4%

Therefore, when the sensor is operating during periods 1 and 2, Isig _ buffer 1 and BG _ buffer 1 will be used for calibration. However, when the sensor is operating during periods 3 and 4, i.e. during the transition, calibration will be performed using Isig _ buffer 2 and BG _ buffer 2.

In yet another embodiment, an EIS-based dynamic slope method (where EIS is used to adjust the slope) may be used for calibration purposes. Fig. 38A shows an example of how this approach can be used to improve sensor accuracy. In this figure, data points 1-4 are discrete blood glucose values. As can be seen in FIG. 38A, there is a sensor dip 3810 between data points 1 and 3, which can be detected using the sensor state vector V described above. As shown at reference numeral 3820 in fig. 38A, during a dip, the slope can be adjusted upward to reduce the reading from being too low.

In a further embodiment, EIS diagnostics may be used to determine the timing of sensor calibration, which may be useful, for example, for low start events, sensitivity loss events, and other similar situations. It is well known that most current methods require periodic calibration based on a preset schedule, for example 4 times per day. However, with EIS diagnostics, calibration becomes event driven, so it can only be performed when necessary, and when most efficient. Here again, the state vector V may be used to determine when the sensor state has changed and request calibration if it has indeed changed.

More specifically, in one illustrative example, FIG. 38B shows a flow chart of EIS assisted sensor calibration involving low start detection. Using the nyquist slope, 1kHz real impedance, and boundary check 3850 (see, e.g., boundary check and associated thresholds previously described for EIS-based parameters in connection with the fusion algorithm of fig. 33A-35), a reliability index 3853 may be developed for startup such that when the 1kHz real impedance 3851 and the nyquist slope 3852 are below their respective upper limits, RI _ startup is 1, and the sensor is ready for calibration. In other words, the reliability index 3853 is "high" (3854), and the logic may be calibrated at 3860.

On the other hand, when the 1kHz real impedance and nyquist slope are above their corresponding upper limits (or thresholds), RI _ start is 0 (i.e., it is "low") and the sensor is not ready for calibration (3856) (i.e., there may be a low start-up problem). Here, the trend of 1kHz real impedance and Nyquist slope can be used to predict when both parameters are within range (3870). If it is estimated that this will only take a very short amount of time (e.g., less than an hour), the algorithm waits until the sensor is ready (i.e., until the EIS-based parameter described above enters range (3874)), at which time the algorithm calibrates. However, if the latency is relatively long (3876), then the sensor can now be calibrated and then the slope or offset can be gradually adjusted according to the 1kHz real impedance and nyquist slope trend (3880). Note that by performing the adjustment, severe false high readings or too low readings caused by low start-up can be avoided. As previously mentioned, the EIS-based parameters and related information used in the instant calibration algorithm are substantially independent of glucose.

It should be noted that while the above description in connection with fig. 38B shows a single working electrode and the calculation of a reliability index for the activation of the working electrode, this is exemplary and not limiting. Thus, in a redundant sensor comprising two or more working electrodes, for each of a plurality of (redundant) working electrodes, a boundary check may be performed and a start-up reliability index calculated. Then, based on the corresponding reliability indicator, at least one working electrode from which glucose measurements may continue to be obtained may be identified. In other words, in a sensor with a single working electrode, if the latter exhibits a low start, the actual use of the sensor (for measuring glucose) may have to be delayed until the end of the low start period. This period of time may typically be about one hour or more, which is clearly disadvantageous. In contrast, in redundant sensors, utilizing the methods described herein allows for adaptive or "smart" start-up, where electrodes that can perform data collection can be identified in a fairly short sequence (e.g., on the order of a few minutes). This in turn reduces the MARD, since a low start-up cost typically results in an increase of about 1/2% in MARD.

For existing calibration algorithms, the buffer size is always 4, i.e., 4 Isig-BG pairs, the weights are based on α and β, α is an exponential function of the time constant, and β is a function of the blood glucose variance, here, EIS can help determine when to refresh the buffer, how to adjust the buffer weights, and the appropriate buffer size.

In some embodiments, the EIS may also be used for interferent detection. In particular, it may be desirable to provide a drug infusion device comprising a combination sensor and drug infusion catheter, wherein the sensor is placed within the infusion catheter. In such systems, the physical location of the infusion catheter relative to the sensor may be of concern, primarily due to potential effects on (i.e., interference with) the sensor signal that may be caused by the drug being infused and/or its inactive components.

For example, diluents used with insulin contain m-cresol as a preservative. In vitro studies, it has been found that if insulin (and m-cresol) is injected near the sensor, m-cresol can negatively affect the glucose sensor. Therefore, a system combining a sensor and an infusion catheter in a single needle must be able to detect and adjust the effect of m-cresol on the sensor signal. Since m-cresol affects the sensor signal, it is desirable to have a method for detecting such interferents that does not rely on the sensor signal itself.

Experiments have shown that the effect of m-cresol on the sensor signal is temporary and therefore reversible. However, when insulin is infused too close to the sensor, the m-cresol tends to "poison" the electrode or electrodes such that the electrode or electrodes can no longer detect glucose until the insulin (and m-cresol) is absorbed into the patient's tissue. In this regard, it has been found that there is typically a time interval of about 40 minutes between the initiation of the insulin infusion and the sensor's ability to again obtain a measurement of glucose. Advantageously, however, it has been found that the 1kHz impedance amplitude increases over the same time period, which is completely independent of the glucose concentration.

Specifically, FIG. 39 shows Isig data and impedance data for an in vitro experiment in which the sensor was placed in a 100mg/d L glucose solution and the 1kHz impedance was measured every 10 minutes, as shown by the circled data points 3920, then m-cresol was added, bringing the solution to 0.35% m-cresol (3930). it can be seen that Isig 3940 initially increases sharply and then begins to decrease once m-cresol is added.

On the other hand, m-cresol has a significant effect on both the magnitude and phase of the impedance. FIG. 40A shows the phase baud curve, and FIG. 40B shows the impedance amplitude before and after the addition of m-cresol. It can be seen that the impedance magnitude 4010 increases by at least one order of magnitude across the entire frequency spectrum from its initialized value 4020 after the addition of m-cresol. At the same time, phase 4030 is completely changed compared to its initialized value 4040 on the nyquist curve of fig. 40C. Here, the pre-initialization curve 4050 and post-initialization curve 4060 of a properly functioning sensor appear as expected. However, curve 4070 became completely different upon addition of m-cresol.

The above experiment identified an important practical drawback of continuing to rely on Isig after the addition of m-cresol. Referring back to fig. 39, the patient/user monitoring the sensor signal may be mistakenly believing that his glucose level has just peaked and that a large dose should be administered. The user then administers a bolus when Isig has begun to drift downward. In other words, it looks normal to the patient/user at all. In practice, however, it does happen that the patient is only administered an unwanted dose of insulin that may risk the patient to a hypoglycemic event (depending on the patient's glucose level before administration of the bolus). This situation enhances the desirability of methods for detecting interferents that are as independent of glucose as possible.

FIG. 41 shows another experiment in which the sensor was initialized to a 100mg/d L glucose solution, after which the glucose was raised to 400mg/d L for one hour, then returned to 100mg/d L, then m-cresol was added, increasing the concentration to 0.35%, and the sensor was held in this solution for 20 minutes.

From the above experiments, it can be seen that EIS can be used to detect the presence of an interfering agent (in this case, m-cresol). In particular, since the interferents affect the sensor in a manner that increases the impedance magnitude across the entire frequency spectrum, the impedance magnitude may be used to detect the interference. Once the disturbance is detected, either the sensor operating voltage may change to a voltage at which the interferent is not measured, or the data reporting may be temporarily suspended and the sensor indicates to the patient/user that the sensor is unable to report data due to administration of the drug (until the measured impedance returns to the pre-infusion level). Note that since the effect of the interferent is due to the preservatives contained in the insulin, the impedance amplitude will exhibit the same behavior as described above regardless of whether the infused insulin acts rapidly or slowly.

Referring to FIG. 41, it can be seen that as the glucose concentration rises from 100mg/d L to 400mg/d L (by a factor of four), the 1kHz impedance amplitude increases from about 2000 ohms to about 2200 ohms, or by about 10% in other words, the effect of glucose on the impedance amplitude measurement appears to be an order of magnitude less than the measured impedance.

Embodiments of the invention described herein also relate to an analog front end integrated circuit (AFE IC), which is a custom Application Specific Integrated Circuit (ASIC) that provides the necessary analog electronics to provide the following functions: (i) supporting a plurality of potentiostats and connected to an oxygen or peroxide-based multi-terminal glucose sensor; (ii) interfacing with a microcontroller to form a micropower sensor system; and (iii) performing EIS diagnostics, fusion algorithms, and other EIS-based processes based on the measurement of the EIS-based parameters; and so on. More specifically, the ASIC incorporates diagnostic capabilities to measure real and imaginary impedance parameters of one or more sensors over a wide frequency range, as well as digital interface circuitry to enable bi-directional communication with a microprocessor chip. In addition, the ASIC contains a power control circuit capable of operating at very low standby and operating power, as well as a real-time clock and crystal oscillator, so that the power supply of the external microprocessor can be shut down.

Fig. 42A and 42B show block diagrams of the ASIC, and table 1 below provides a pad signal description (shown on the left side of fig. 42A and 42B) in which some signals are multiplexed onto a single pad.

The ASIC will now be described with reference to fig. 42A and 42B and table 1.

Power layer

The ASIC has a power plane powered by power pad VBAT (4210) and has an operational input in the range of 2.0 volts to 4.5 volts. The power plane has a regulator for reducing the voltage of certain circuits of the plane. This power supply is called VDDBU (4212) and has one output pad for testing and bypassing. The circuitry on the VBAT power supply includes an RC oscillator, a real time clock (RCosc)4214, battery protection circuitry, regulator control, power on reset circuitry (POR), and various inputs/outputs. The pads on the VBAT power plane are configured to draw less than 75nA of current at 40 ℃ and 3.50V volts VBAT.

The ASIC also has a VDD supply for powering the logic. The VDD supply voltage range may be programmed to be at least 1.6 volts to 2.4 volts. The circuitry of the VDD power plane contains most of the digital logic, a timer (32khz), and a real-time clock (32 khz). The VDD power plane includes level shifters that interface with other voltage planes as needed. The interface of the level shifter is adjusted so that if another power plane is not powered, the current increment of any powered power plane will not exceed 10 nA.

The ASIC contains an on-board regulator (with shutdown control) and an external VDD power option. The regulator input is a separate pad, REG _ VDD _ IN (4216), which has the same electrostatic discharge (ESD) protection as the other I/os on VBAT. The on-board regulator has an output pad, REG _ VDD _ OUT (4217). The ASIC also has a VDD input pad that is separate from the REG _ VDD _ OUT pad.

The ASIC includes an analog power plane, referred to as VDDA (4218), powered by a VDD on-board regulator or external power supply, typically filtered VDD. The VDDA power supply circuit is configured to operate within 0.1 volts VDD, so no level shifting between the VDDA power plane and the VDD power plane is required. The VDDA power supply provides power for the sensor analog circuitry, the analog measurement circuitry, and any other noise sensitive circuitry.

The ASIC contains a pad power supply VPAD for the designated digital interface signal. The operating voltage range of the pad power supply is at least 1.8V to 3.3V. These pads have one or more independent power supply pads and are powered by an external power supply. The pad also incorporates a level shifter to other on-board circuitry to allow a flexible pad power supply range independent of the VDD logic supply voltage. The ASIC can adjust the VPAD pad ring signal so that the other supply currents do not increase by more than 10nA when the VPAD power supply is not enabled.

Bias generator

The ASIC has a BIAS generator circuit BIAS _ GEN (4220) which is powered by the VBAT supply and which generates a BIAS current which is stable with the supply voltage of the system. The output current has the following specifications: (i) power supply sensitivity: supply voltage of <1.6V to 4.5V ± 2.5%; (ii) current accuracy: after fine adjustment, is < + -. 3 percent.

The BIAS GEN circuit generates switched and unswitched output currents to power circuits that require BIAS current for operation. At VBAT 2.5V to 4.5V, the operating current consumption of the BIAS GEN circuit is less than 0.3uA at 25 ℃ (without any BIAS output current included). Finally, the temperature coefficient of the bias current is typically between 4,000 ppm/deg.C and 6,000 ppm/deg.C.

Voltage reference

The ASICs described herein are configured with a low power voltage reference, which is powered by the VBAT power supply. The voltage reference has an enable input that can accept signals from logic powered by VBAT or VDDBU. The ASIC is designed so that when VBAT is powered on, the enable signal does not cause the current from any power supply of the signal interface to increase by more than 10 nA.

The voltage reference has the following specifications: (i) output voltage: 1.220 +/-3 mV after fine adjustment; (ii) power supply sensitivity: <1.6V to 4.5V input ± 6 mV; (iii) temperature sensitivity is from <0 ℃ to 60 ℃ +/-5 mV; and (iv) output voltage default accuracy (no adjustment): 1.220V. + -. 50 mV. In addition, at 4.5V, 40 ℃, the supply current will be less than 800 nA. In this embodiment, when the reference is disabled, the reference output will be forced to VSSA in order to keep the VDD voltage regulator from overshooting to a level that exceeds the logic's breakdown voltage.

32kHz oscillator

The ASIC contains a low power 32.768kHz crystal oscillator 4222, powered by a supply from the VDDA supply, which can adjust the capacitance of the crystal oscillator pad (XTA L I, XTA L O) with software, specifically, the frequency adjustment range is at least-50 ppm to +100ppm, and the step size is at most 2ppm throughout the adjustment range, here, the load capacitance of the crystal can be assumed to be 7pF, L s 6.9512kH, Cs 3.3952fF, Rs 70k, shunt capacitance of 1pF, and PC board parasitic capacitance on each crystal side to be 2 pF.

The ASIC has a VPAD stage output available at pad C L K _32kHZ, where the output can be disabled under software and logic control, a default output 32kHz oscillator, input pin OSC32K _ BYPASS (4224) can disable the 32kHz oscillator (no power consumption) and allow digital input to the XTA L I pad circuits related to this function are configured so that at low OSC32K _ BYPASS, no ASIC current in excess of 10nA is added to the oscillator current in either state of the OSC32K _ BYPASS signal.

If OSC32K _ BYPASS is true, the 32KHZ oscillator analog circuitry is placed into a low power state and the XTA L I pad is configured to accept a digital input at a level of 0 to VDDA. Note that the duty cycle of the 32kHz oscillator output is between 40% and 60%.

Timer

The ASIC contains a timer 4226, which is clocked by a 32kHz oscillator divided by 2, which is presettable and has two programmable timeouts, the timer has 24 programmable bits, a total time count of 17 minutes 4 seconds, the timer also has a programmable delay (see section "Microprocessor Wake-Up control Signal" below) to disable the clock to the C L K _32KHz pad and set the microprocessor (uP) interface signal on the VPAD plane to a predetermined state, which will allow the microprocessor to enter suspend mode without an external clock.

The timer also includes a programmable delay to wake UP the microprocessor by enabling the C L K _32KHz clock output and setting the UP _ WAKEUP high level transition of POR2(VDD POR) from a Power Low state to a Power OK state will enable the 32kHz oscillator, the C L K _32KHz clock output and set the UP _ WAKEUP high level.

Real Time Clock (RTC)

The ASIC also has a 48-bit readable/writable binary counter operated by an unencrypted, free-running 32kHz oscillator. Writing to real time clock 4228 requires writing the address with the key before the clock is written. The write access to the clock is configured to terminate between 1 millisecond and 20 milliseconds after the write operation to the key address.

The real time clock 4228 is configured to be reset by a power-on reset of POR1_ IN (vbat POR) or POR2_ IN (VDD _ POR) to half count (MSB ═ 1, with all other bits being 0). In embodiments of the invention, the real time clock has programmable interrupt capability and is designed to be robust to Single Event Upsets (SEU), which can be achieved by topology techniques or by adding capacitance to the appropriate node, if desired.

RC oscillator

The ASIC further includes an RC clock that is powered by the VBAT power source or the VBAT derived power source. The RC oscillator is running all the time except that the oscillator can be bypassed by writing register bits in the analog test mode (see "digital test" section) and applying a signal of 0 to VBAT level to GPIO _ VBAT. The RC oscillator is not trimmable and contains the following specifications: (i) the frequency is between 750Hz and 1500 Hz; (ii) the duty cycle is between 50% ± 10%; (iii) the current consumption at 25 ℃ is lower than 200 nA; (iv) the frequency variation of the 1V to 4.5V VBAT power supply is less than +/-2%, and the frequency variation of the 1.8V to 4.5V VBAT power supply is more than 1%; and (V) a frequency change of less than +2, -2% over a temperature range from 15 ℃ to 40 ℃ in the case of VBAT ═ 3.5V. The RC frequency can be measured using a 32kHz crystal oscillator or an external frequency source (see oscillator calibration circuit).

Real-time RC clock (based on RC oscillator)

The ASIC contains a 48-bit readable/writable binary ripple counter based on an RC oscillator. Writing to the RC real time clock requires writing the address with the key before the clock can write. The write access to the clock terminates between 1 millisecond and 20 milliseconds after the write operation to the key address, wherein the time of the protection window is configured to be generated by the RC clock.

The real-time RC clock is allowed relative time stamping if the crystal oscillator is off and is configured to reset to half count (MSB ═ 1, all other values are 0) on POR1_ in (battle). The real-time RC clock is designed to be robust to single event disturbances (SEU) by placement techniques or adding capacitance to the appropriate nodes when needed. On the falling edge of POR2_ IN, or if the ASIC enters a battery low state, the RT real time clock value may be captured IN a register that can be read through the SPI port. The registers and associated logic are located at the VBAT or VDDBU power plane.

Battery protection circuit

The ASIC includes battery protection circuitry 4230 which uses a comparator to monitor the battery voltage and is powered by power from the VBAT power plane. The battery protection circuit is configured to always operate with the VBAT power supply. The battery protection circuit can use an RC oscillator to time the signal and has an average current consumption of less than 30nA, including a total resistive external voltage divider of 3 mega ohms.

For a battery threshold of 2.90V, the battery protection circuit uses an external switched voltage divider with a ratio of 0.421. The ASIC also has an internal voltage divider with a ratio of 0.421 + -0.5%. The divider is connected between BATT _ DIV _ EN (4232) and VSSA (4234), and the divider output is one pin (called BATT _ DIV _ INT (4236)). To save pins in the packaged part, BATT _ DIV _ INT in the present embodiment is connected to BATT _ DIV inside the package. Also in this configuration, BATT _ DIV _ EN does not need to come out of the package, saving two package pins.

The battery protection circuit is configured to sample the voltage on the input pin BATT _ DIV (4238) at a rate of about 2 times per second, with the sampling time being generated by an RC oscillator. The ASIC can adjust the frequency divider of the RC oscillator to adjust the sampling time interval to 0.500 seconds ± 5 milliseconds if the remote controlled oscillator is operating within its operating tolerance. In a preferred embodiment, the ASIC has a test mode that enables more frequent sampling intervals during testing.

The comparator input is configured to accept an input from 0 to VBAT volts. For an input of 0 to VBAT volts, the input current BATT _ DIV at the comparator input is less than 10 nA. The comparator sample circuit outputs a positive pulse to the pad BATT _ DIV _ EN that can be used by external circuitry to enable the off-chip resistor divider only during sampling in order to save power. The voltage high logic level is the VBAT voltage and the low level is the VSS level.

The output resistance of the BATT _ DIV _ EN pad should be less than 2 kilo ohms when VBAT is 3.0V. This allows the voltage divider to be driven directly by the output. In the event that a programmable number of consecutive samples indicate that the battery is low, the comparator control circuitry triggers an interrupt to the interrupt output pad UP _ INT. The default number of samples is 4, but the number of consecutive samples can be programmed to be 4 to 120.

After the programmable number of consecutive samples indicate that the battery is low after the UP INT is generated, the comparator control circuitry is configured to generate a signal that will place the ASIC in a low power mode: the VDD regulator will be disabled and the low level signal will be set to pad VPAD _ EN. This will be referred to as a battery low state. Likewise, the number of consecutive samples can be programmed to be 4 to 120 samples, with a default of 4 samples.

The comparator has separate programmable thresholds for the falling and rising voltages on BATT _ DIV, which is implemented in digital logic that multiplexes two values into circuitry based on the state of the battery low state, thus, a falling threshold is applied if the battery low state is low and a rising threshold is applied if the battery low state is high.

The comparator threshold varies by less than +/-1% between 20 ℃ and 40 ℃. The default threshold for the falling voltage is 1.44V (nominal voltage divider VBAT threshold 3.41V) and the default threshold for the rising voltage is 1.53V (nominal voltage divider VBAT threshold 3.63V). After the ASIC is placed in the battery low state, the ASIC will initiate a microprocessor start sequence if the comparator detects 4 consecutive battery OK indications.

Power-on reset of battery power supply layer

If the input VBAT drops above 1.2 volts within a 50 microsecond period, or the VBAT voltage is below 1.6 ± 0.3 volts, a power-on-reset (POR) output is generated on the pad npro 1_ OUT (4240). The POR is stretched to a minimum pulse width of 5 milliseconds. The output of the POR circuit is configured to be active low and is connected to a pad npro 1_ OUT on the VBAT power plane.

The IC has an input pad for the battery power plane POR, npro 1_ IN (4242). The input pad has an RC filtering function so that a pulse shorter than 50 nanoseconds does not cause a logic reset. IN this embodiment, IN normal operation, npro 1_ OUT is externally connected to npro 1_ IN, thereby separating the analog circuitry for testing from the digital circuitry. nport 1_ IN will reset all logic on any power plane and initialize all registers to default values. Thus, the reset status register POR bit is set and all other reset status register bits are cleared. The POR reset circuitry is configured to consume no more than 0.1uA of current from the VBAT power supply for more than 5 seconds after power-up.

VDD Power-on reset (POR)

The ASIC also has a voltage comparator circuit that generates a VDD voltage plane reset signal at power-up or if VDD drops below a programmable threshold. This range can be programmed by several voltage thresholds. The default value is 1.8V-15% (1.53V). POR2 has a programmable rising voltage threshold that implements hysteresis. The rise threshold is also programmable and defaults to 1.60V ± 3%.

The POR signal is active low and has an output pad npar 2_ OUT on the VDD power plane (4244). The ASIC also has an active low POR open drain output npro 2_ OUT _ OD (4246) on the VBAT power plane. This can be used to apply POR to other system components.

The POR for the VDD supply logic comes from the input pad npro 2_ IN (4248). The npro 2_ IN pad is located at the VDD power plane and has an RC filtering function, so that a pulse shorter than 50 ns does not cause a logic reset. The npro 2_ OUT is configured to be externally connected to the npro 2_ IN input pad under normal use conditions, thereby separating analog circuitry from digital circuitry.

After VDD exceeds a programmable threshold, the resulting reset is extended to an active time of at least 700 milliseconds to ensure the crystal oscillator is stable. After power-on, the POR reset circuit system consumes VDD power supply for more than 5 seconds and VBAT power supply for more than 5 seconds, wherein the time of the VDD power supply for consuming no more than 0.1uA is not more than 0.1 uA. The registers storing POR thresholds are powered by the VDD power plane.

Sensor interface electronics

In embodiments of the invention described herein, the sensor circuit supports up to five sensor WORK electrodes (4310) in any combination of peroxide or oxygen sensors, although in further embodiments a greater number of such electrodes may be accommodated. When the peroxide sensor WORK electrode supplies current, the oxygen sensor WORK electrode sinks current. With the present embodiment, the sensor may be configured in a potentiostat configuration as shown in fig. 43.

The sensor electronics have programmable power control for each electrode interface circuit to minimize current consumption by turning off current to unused sensor electronics. The sensor electronics also include electronics to drive the COUNTER electrode 4320 using feedback from the RE (reference) electrode 4330. When not in use, the current to the circuitry may be programmed off to save power. The interface electronics include a multiplexer 4250 so that the COUNTER and RE electrodes can be connected to either of the (redundant) WORK electrodes.

The ASIC is configured to provide the following sensor interfaces: (i) RE: a reference electrode establishing a reference potential for a solution of an electronic device for setting a WORK voltage; (ii) WORK1-WORK 5: a working sensor electrode where the desired reduction/oxidation (redox) reaction occurs; and (iii) COUNTER: the output of this pad holds a known voltage on the RE electrode relative to the system VSS. In this embodiment, the ASIC is configured to be able to set the WORK voltages of up to 5 WORK electrodes, respectively, and has a resolution and accuracy greater than or equal to 5 mV.

In the oxygen mode, the one or more WORK voltages may be programmed to be at least 0 to 1.22V relative to the VSSA. In the peroxide mode, one or more WORK voltages can be programmed to be at least 0.6V to 2.054V relative to VSSA. If VDDA is less than 2.15V, the WORK voltage can be operated at VDDA-0.1V. The ASIC includes a current measurement circuit for measuring the WORK electrode current in the peroxide sensor mode. This may be achieved by, for example, a current-to-voltage converter or a current-to-frequency converter, which may have the following specifications: (i) current range: 0-300 nA; (ii) voltage output range: same as the WORK electrode in peroxide/oxygen mode; (iii) output offset voltage: maximum +/-5 mV; and (iv) uncalibrated resolution: . + -. 0.25 nA.

After applying the calibration factor to the gain and assuming an acquisition time of 10 seconds or less, the current measurement accuracy is:

5pA-1nA：±3％±20pA

1nA-10nA：±3％±20pA

10nA-300nA：±3％±0.2nA

for the current-to-frequency converter (ItoF) only, the frequency range may be between 0Hz and 50 kHz. In the peroxide mode, the current converter must operate within a specified voltage range relative to the WORK electrode VSS. Here, the leakage current of the 2.5V power supply is less than 2uA, and the WORK electrode current of each converter is less than 10nA (including digital-to-analog converter (DAC) current).

The current converter may be enabled or disabled by software control. When disabled, the WORK electrode will exhibit a very high impedance value, i.e., greater than 100 mega ohms. Likewise, for ItoF only, the output of the I-F converter will go to a 32-bit counter that the microprocessor and test logic can read, write, and clear. During the counter read, the timing of the counter is suspended to ensure an accurate read.

In embodiments of the invention described herein, the ASIC also includes a current measurement circuit that measures the WORK electrode current in the oxygen sensor mode. The circuit may be implemented as a current-to-voltage converter or a current-to-frequency converter, and the current converter may be configured to operate in an oxygen mode using programmable bits. As previously mentioned, in the oxygen mode, the current converter must operate within a specified voltage range of the WORK electrode relative to VSS. Here, too, the current range is 3.7pA to 300nA, the voltage output range is the same as that of the WORK electrode in the oxygen mode, the output offset voltage is at most + -5 mV, and the uncorrected resolution is 3.7pA + -2 pA.

After applying the calibration factor to the gain and assuming an acquisition time of 10 seconds or less, the current measurement accuracy is:

5pA-1nA：±3％±20pA

1nA-10nA：±3％±20pA

10nA-300nA：±3％±0.2nA

for current-to-frequency converters (ItoF) only, the frequency range can be between 0Hz and 50kHz, the leakage current of 2.5V power supplies is less than 2uA, and the WORK electrode current per converter is less than 10nA (including DAC current). The current converter may be enabled or disabled by software control. When disabled, the WORK electrode will exhibit a very high impedance value, i.e., greater than 100 mega ohms. Likewise, for ItoF only, the output of the I-F converter will go to a 32-bit counter that the microprocessor and test logic can read, write, and clear. During the counter read, the timing of the counter is suspended to ensure an accurate read.

In the embodiments of the invention described herein, the Reference Electrode (RE)4330 has an input bias current of less than 0.05nA at 40 ℃. The COUNTER electrode adjusts its output to maintain the desired voltage on the RE electrode. This is accomplished by an amplifier 4340 whose output to the COUNTER electrode 4320 attempts to minimize the difference between the actual RE electrode voltage and the target RE voltage (the latter being set by the DAC).

The RE set voltage can be programmed to be at least 0 volts to 1.80V and the common mode input range of the COUNTER amplifier comprises at least 0.20 to (VDD-0.20) V. Register bits may be used to select the common mode input range and provide programming for the COUNTER mode of operation, if desired. The resolution and accuracy of the WORK voltage is set to better than or equal to 5 mV. Note that in the normal mode, the COUNTER voltage attempts to maintain the RE voltage at the level of the programmed RE target value. However, in forced COUNTER mode, the COUNTER electrode voltage is forced to the programmed RE target voltage.

All electrode drive circuits are configured to be able to drive the electrodes to the electrode load and do not oscillate under any use conditions. FIG. 44 shows an equivalent ac inter-electrode circuit according to an embodiment having a potentiostat configuration as shown in FIG. 43. The equivalent circuit shown in FIG. 44 may be between any of the poles (i.e., WORK1-WORK5, COUNTER, and RE), and the values of the corresponding circuit components range as follows:

ru ═ 200-5k ohm

Cc＝[10-2000]pF

Rpo ═ 1-20 kilo ohms

Rf [ < 200-

Cf＝[2-30]uF

During initialization, the drive currents of the WORK and COUNTER electrodes need to provide higher currents than the normal potentiostat operation described previously. In this way, the programmable register bits can be used to program the electrode driver circuit to a higher power state if additional driving is required. It is important to achieve low power operation in the normal potentiostat mode, where the electrode current is typically less than 300 nA.

In a preferred embodiment, during initialization, the WORK1 to WORK5 electrodes can be programmed in steps of 0 to VDD volts equal to or less than 5mV, and the drive or sink current output capability of the electrodes is at least 20uA, from 0.20V to (VDD-0.20V). Also, during initialization, the ASIC is typically configured to be able to measure the current of a WORK electrode up to 20uA with an accuracy of ± 2% ± 40nA of the measured value. Furthermore, during initialization, as previously described, the RE set voltage is programmable, the COUNTER DRIVE CIRCUIT output must be able to generate or sink a minimum of 50uA of current at the COUNTER electrodes of 20V to (VDD-.20V), and the power supply currents (VDD and VDDA) of the initialization CIRCUIT are required to exceed any generated output current by less than 50 uA.

Current calibrator

In an embodiment of the invention, the ASIC has a current reference that can be directed to any WORK electrode for calibration purposes. In this regard, the calibrator includes programmable bits that cause the current output to sink current or supply current. Assuming that the tolerance of the external precision resistor is 0, the programmable current at least comprises 10nA, 100nA and 300nA, and the precision is better than 1% 1 nA. The calibrator was connected to the pad TP _ RES (4260) using a precision resistor of 1 megaohm as a reference resistor. Furthermore, a current reference may be directed to the COUNTER electrode or the RE electrode for initialization and/or sensor state purposes. A constant current may be applied to the COUNTER or RE electrodes and the electrode voltage may be measured with an ADC.

High-speed RC oscillator

Referring back to FIG. 42, the ASIC further includes a high-speed RC oscillator 4262 that provides power for an analog-to-digital converter (ADC)4264, ADC sequencer 4266, and other digital functions that require a higher speed clock than 32 kHz. The high-speed RC oscillator is phase locked to a 32kHz clock (32.768kHz) to provide a programmable output frequency of 524.3kHz to 1048 kHz. In addition, the duty ratio of the high-speed RC oscillator is 50% +/-10%, the phase jitter is less than 0.5% rms, the current is less than 10uA, and the frequency is stable in the whole VDD working range (the voltage range is 1.6 to 2.5V). The default value of the high speed RC oscillator is "off" (i.e., disabled), in which case the current consumption is less than 10 nA. However, the ASIC has programmable bits for enabling the high speed RC oscillator.

Analog-to-digital converter

The ASIC includes a 12-bit ADC (4264) having the characteristics of (i) being able to complete conversion in less than 1.5 milliseconds when operating at a clock frequency of 32kHz, (ii) being able to perform faster conversion when clocked from a high speed RC oscillator, (iii) having at least 10 bits of precision (12 bits + -4 counts), (iv) a reference voltage input of 1.220V, a sensitivity of less than 0.2 mV/deg.C at a temperature of 20 deg.C to 40 deg.C, (V) a full scale input range of 0 to 1.22V, 0 to 1.774V, 0 to 2.44V, and 0-VDDA, wherein the 1.774 and 2.44V ranges have programmable bits to reduce the conversion range to lower values to accommodate lower VDDA voltages, (vi) a power supply current consumption of less than 50uA, (vi) having a converter capable of operating with either a 32kHz or high speed RC clock, (vii) L being less than 1L SB, and (viii) issuing interrupts at the end of conversion.

As shown in fig. 42A and 42B, the ASIC has an analog multiplexer 4268 at the input of the ADC 4264, both of which may be controlled by software. In a preferred embodiment, at least the following signals are connected to the multiplexer:

(i) VDD-core Voltage and regulator output

(ii) VBAT-battery power supply

(iii) VDDA-analog power supply

(iv) Reference electrode of RE-sensor

(v) COUNTER-SENSOR COUNTER ELECTRODE

(vi) WORK1-WORK 5-working electrode of sensor

(vii) Temperature sensor

(viii) At least two external pin analog signal inputs

(ix) EIS integrator output

(x) ItoV current converter output

The ASIC is configured so that the ADC load of the inputs COUNTER, RE, WORK1-WORK5, temperature sensor, and any other input adversely affected by the load does not exceed + -0.01 nA. The multiplexer includes a voltage divider for any input having a voltage above the ADC input voltage range and a buffer amplifier for reducing the divided input resistance of the load sensitive input to below 1 nA. In turn, the common mode input range of the buffer amplifier is offset by less than 3mV between at least 0.8V to VDDA voltage and an input range of 0.8V to VDDA-0.1V.

In a preferred embodiment, the ASIC has a mode in which ADC measurements are made in a programmed sequence. Thus, the ASIC includes a programmable sequencer 4266 that monitors measurements of up to 8 input sources with the following programmable parameters for ADC measurements:

(i) ADC MUX input

(ii) ADC range

(iii) A delay time before measurement, wherein the delay can be programmed to 0 in steps of 0.488 milliseconds

62 ms

(iv) Number of measurements per input from 0 to 255

(v) Measuring the number of cycles: 0-255, where a measurement cycle refers to a sequence of up to 8 input measurements repeated multiple times (e.g., as an outer loop in a procedure).

(vi) The delay between cycles is measured, where the delay can be programmed to 0 to 62 milliseconds in steps of 0.488 milliseconds.

Sequencer 4266 is configured to start upon receipt of an automatic measurement start command, and the measured values may be stored in the ASIC for retrieval through the SPI interface. Note that the sequencer timebase may be programmed between the 32kHz clock and the high speed RC oscillator 4262.

Sensor diagnostics

As described in detail above, embodiments of the invention described herein are directed to the use of impedance and impedance related parameters in, for example, sensor diagnostics programs and Isig/SG fusion algorithms. To this end, in a preferred embodiment, the ASIC described herein has the ability to measure the impedance magnitude and phase angle of any WORK sensor electrode relative to the RE and COUNTER electrodes when in a potentiostat configuration. This is achieved by measuring the amplitude and phase of the current waveform, for example in response to a sine-like waveform superimposed on the WORK electrode voltage. See, for example, diagnostic circuitry 4255 in fig. 42B.

The ASIC has the capability to measure the resistive and capacitive components of any electrode to any electrode through, for example, the electrode multiplexer 4250. Note that such measurements may disturb the sensor balance and may require settling time or sensor initialization to record a stable electrode current. As previously described, although an ASIC may be used to make impedance measurements across a wide spectral range, for purposes of embodiments of the present invention, a relatively narrow frequency range may be used. In particular, the sine wave measurement capability of the ASIC may include a test frequency of about 0.10Hz to about 8192 Hz. In making such measurements, the minimum frequency resolution according to an embodiment of the present invention may be limited as shown in table 2 below:

TABLE 2

The sine wave amplitude can be programmed in 5mV steps to at least 10mVp-p to 50mVp-p, and can be programmed in 10mV steps to 60mVp-p to 100 mVp-p. In a preferred embodiment, the amplitude accuracy is better than ± 5% or ± 5mV, whichever is greater. In addition, the ASIC can measure the electrode impedance with the accuracy specified in table 3 below:

in one embodiment of the invention, the ASIC may measure the input waveform phase relative to a time base, which may be used in impedance calculations to improve accuracy. The ASIC may also have an on-chip resistor for calibrating the electrode impedance circuit described above. Conversely, the on-chip resistor may be calibrated by comparing it to a known 1 megaohm off-chip precision resistor.

Data samples of the waveform may also be used to determine impedance. Data may be transmitted to an external microprocessor for computation and processing through a Serial Peripheral Interface (SPI). The converted current data is sufficiently buffered, and 2000 times of ADC data conversion can be transmitted to an external device through the SPI interface without losing data. This assumes a maximum delay time of 8 milliseconds for service data transfer request interruption.

In embodiments of the invention, rather than (or in addition to) measuring the electrode impedance with a sine wave, the ASIC may measure the electrode current with a step input. Here, the ASIC can provide programmable amplitude steps of 10 to 200mV to the electrodes with a resolution better than 5mV and sample (measure) the resulting current waveform. The duration of the sampling may be programmed in steps of 0.25 seconds to at least 2 seconds, and the sampling interval for measuring the current may comprise at least 5 programmable binary weighted steps of about 0.5 to 8 milliseconds.

The resolution of the electrode voltage samples was less than 1mV and ranged up to ± 0.25 volts. Such measurements may be made for a suitable regulated voltage in order to reduce the dynamic range required for data conversion. Similarly, the resolution of the electrode current samples was less than 0.04uA, and ranged up to 20 uA. If the measurement polarity is programmable, the current measurement may be unipolar.

In an embodiment of the invention, the current measurement may use an I-V converter. Further, the ASIC may have an on-chip resistor for calibrating the current measurement. Conversely, the on-chip resistor may be calibrated by comparing it to a known 1 megaohm off-chip precision resistor. The accuracy of the current measured sample is better than +/-3% or +/-10 nA, and the larger one is taken as the standard. As described above, the converted current data is sufficiently buffered, and 2000 ADC data conversions can be transmitted to an external device through the SPI interface without data loss. This assumes a maximum delay time of 8 milliseconds for service data transfer request interruption.

Calibrating voltage

The ASIC contains a precision voltage reference for calibrating the ADC. The output voltage is 1.000V +/-3%, the production deviation is less than 1.5%, and the stability is better than +/-3 mV in the temperature range of 20-40 ℃. The precision calibration voltage may be calibrated by an on-chip ADC by comparing it to an external precision voltage during manufacturing. In manufacturing, the calibration factor may be stored in system non-volatile memory (not on this ASIC) to achieve higher accuracy.

The current consumption of the calibration voltage circuit is preferably less than 25 uA. Furthermore, the calibration voltage circuit can be powered down below 10nA when not in use to save battery power.

Temperature sensor

The ASIC has a temperature sensor with a sensitivity between-10 ℃ and 60 ℃ and between 9 and 11mV per degree celsius. The output voltage of the temperature sensor allows the analog-to-digital converter to measure a temperature dependent voltage in the input range of the 0 to 1.22V ADC. The current consumption of the temperature sensor is preferably less than 25uA, and the temperature sensor can be powered down to less than 10nA when not in use to conserve battery power.

VDD Voltage regulator

The ASIC has a VDD voltage regulator with the following characteristics:

(i) minimum input voltage range: 2.0V-4.5V.

(ii) Minimum output voltage: 1.6-2.5V + -5%, default value is 2.0V.

(iii) Dropping voltage: in I_load＝100uA、V_inWhen 2.0V，V_in-V_out<0.15V。

(iv) The output voltage is programmable and has an accuracy within 2% of the values shown in table 4 below:

TABLE 4

(v) The regulator can provide an output of 1mA at 2.5V when the input voltage is 2.8V.

(vi) If an external regulator is used, the regulator may also have open input and output pads.

In this non-operating mode, the current consumption of the regulator circuit is preferably less than 100 nA.

(vii) The output voltage variation from a 10uA load to a 1mA load is preferably less than 25 mV.

(viii) The source current consumption (excluding the output current under 1mA load) is less than 100 uA.

(ix) The source current consumption (excluding the output current under 0.1mA load) is less than 10 uA.

(x) The source current consumption (excluding the output current under a load of 10 uA) is less than 1 uA.

Universal comparator

The ASIC contains at least two comparators 4270, 4271 powered by VDDA. The comparator generates the threshold using 1.22V as a reference. The output of the comparator can be read by the processor and will generate a maskable interrupt on the rising or falling edge as determined by the configuration register.

The comparators have power control that reduces power when not in use, and the current supply of each comparator is less than 50 nA. The response time of the comparator is preferably less than 50 microseconds for a 20mV overdrive signal and the offset voltage is less than + -8 mV.

The comparator also has programmable hysteresis, where the hysteresis option includes a threshold of the rising input of 1.22V + V_hystThreshold value of falling input is 1.22-V_hystOr no hysteresis (V)_hyst25 ± 10 mV). The output of any comparator can be used for any GPIO on any power plane.(see GPIO section).

Sensor connection sensing circuitry on RE

An analog switched capacitor circuit monitors the impedance of the RE connection to determine if the sensor is connected. Specifically, a capacitor of about 20pF is driven by the inverter, switching at a frequency of 16Hz, with its output swing from VSS to VDD. The comparator will sense the voltage swing on the RE pad and if the swing is less than the threshold, the comparator output will indicate a connection. The comparison is performed on two transitions of the pulse. A swing of both transitions is required to be below a threshold to indicate a connection and a comparison indicating a high swing for either phase will indicate a disconnection. The connect/disconnect signal is debounced such that its state transition requires a steady indication of at least 1/2 seconds to a new state.

This circuit has six thresholds defined by the following resistances in parallel with a 20pF capacitor: 500 kilo-ohms, 1 mega-ohm, 2 mega-ohms, 4 mega-ohms, 8 mega-ohms and 16 mega-ohms. The parallel equivalent circuit is located between the RE pad and virtual ground, which may be any voltage between the power supply rails. The threshold accuracy is better than ± 30%.

The output of the sensor connection sensing circuitry can be programmably interrupt or processor enabled if the sensor is connected or disconnected. As long as npro 2_ IN is high and VDD and VDDA are present, the circuit will be active. The average current consumption of the circuit is less than 100 nA.

WAKEUP pad

The WAKEUP circuitry is powered by the VDD power supply and has an input range of 0V to VBAT. WAKEUP pad 4272 has a weak pull down of 80 + -40 nA. This current may come from the output of BIAS _ GEN 4220. At 0v input, the average current consumed by the circuit is less than 50 nA.

The rising input voltage threshold Vih of the WAKEUP input is 1.22 + -0.1V, and the falling input threshold is-25 mV + -12 mV of the rising threshold. In the preferred embodiment, the circuit associated with the WAKEUP input draws no more than 100nA of current (which does not include the input pull-down current) for any input having a value of-0.2 to the VBAT voltage. The WAKEUP pad is debounced for at least 1/2 seconds.

The output of the WAKEUP circuit can programmably generate an interrupt or processor boot if the WAKEUP pad changes state. (see event handler section). It is important to note that the WAKEUP pad circuitry is configured to assume a low current <1nA if the battery protection circuit indicates a low battery condition.

UART WAKEUP

The ASIC is configured to monitor nRX _ EXT pad 4274. If the nRX _ EXT level continues high (UART BREAK) for more than 1/2 seconds, a UART WAKEUP event will be generated. Due to the sampling, the UART WAKEUP event may be generated at a continuous high level as short as 1/4 seconds. The UART WAKEUP event may programmatically generate an interrupt, WAKEUP, and/or microprocessor reset (nRESET _ OD). (see event handler section).

In a preferred embodiment, the circuitry associated with the UART WAKEUP input draws no more than 100nA, and if the battery protection circuitry indicates a battery low state, the UART WAKEUP pad circuitry is configured to assume a low current <1 nA. The rising input voltage threshold Vih of the UART Wakeup input is 1.22 +/-0.1V. The falling input threshold is-25 mV. + -. 12mV of the rising threshold.

Microprocessor wake-up control signal

The ASIC can generate signals that help control power management of the microprocessor. Specifically, the ASIC may generate the following signals:

(i) nSHUTDN-nSHUTDN may control power up of an off-chip VDD regulator. The nSHUTDN pad is located on the VBAT power rail. nSHUTDN should be low if the battery protection circuitry indicates a battery low state, otherwise nSHUTDN should be high.

(ii) VPAD _ EN-VPAD _ EN can control power up of an external regulator that supplies power to VPAD. When the VPAD power supply is disabled, the internal signal corresponding to the external signal ensures that the input from the VPAD pad does not generate additional current due to the floating input. The VPADEN pad is an output on the VBAT power rail. If the battery protection signal indicates that the battery is low, the VPAD _ EN signal is low. The VPAD _ EN signal may be set to a low level by a software command that starts a timer; the terminal count of the timer forces VPAD _ EN low. If the battery protection signal indicates that the battery is in good condition, the following events may cause the VPAD _ EN signal to go high (see event handler for more details): npro 2_ IN transitions from low to high; SW/timer (programmable); WAKEUP conversion; low to high, and/or high to low (programmable); sensor connection conversion; low to high and/or high to low (programmable); UART interruption; and RTC time event (programmable).

(iii) UP _ WAKEUP-UP _ WAKEUP may be connected to the microprocessor wake-UP pad. It is intended to wake up the microprocessor from a sleep mode or similar power down mode. The UP _ WAKEUP pad is an output on the VPAD power rail. The UP _ WAKEUP signal may be programmed to be active low, active high, or pulsed. The UP _ WAKEUP signal may be set to a low level by a software command to start a timer; the terminal count of the timer forces UP WAKEUP low. If the battery protection signal indicates that the battery is in good condition, the UP WAKEUP signal may be high (see event handler for more details): npro 2_ IN transitions from low to high; SW/timer (programmable); WAKEUP conversion; low to high, and/or high to low (programmable); sensor connection conversion; low to high and/or high to low (programmable); UART interruption; and RTC time event (programmable). The WAKEUP signal may be delayed by a programmable amount. If WAKEUP is programmed as a pulse, the pulse width can be programmed.

(iv) The C L K _32KHZ-C L K _32KHZ pad may be connected to the microprocessor to power the low speed clock, the clock is programmable ON-off and may be programmably turned ON to wake up events, the C L K _32KHZ pad is an output ON the VPAD power rail, if the battery protection signal indicates that the battery is low, the C L K _32KHZ signal is low, the C L K _32KHZ output may be programmed off by a programmable bit, default "ON", the C L K _32KHZ signal may be disabled by a software command that starts a timer, the terminal count of the timer forces the C L K _32KHZ low, if the battery protection signal indicates that the battery is IN good, events that may cause the C L K _32KHZ signal to be enabled (see event handler for more details), the nPOR2 IN to transition from low to high, the programmable SW/timer (programmable high), the low to WA level transition, and/or high level transition to low to high (and high level of battery interrupt detection events), and the programmable UART detection circuit (via a programmable low to high level or high level sensor interrupt event).

(v) nRESET _ OD-nRESET _ OD can be connected to the microprocessor to cause the microprocessor to reset. nRESET _ OD can be programmed to wake up an event. The nRESET _ OD pad is the output on the VPAD power rail. The pad is an open drain (nfet output). If the battery protection signal indicates that the battery is low, the nRESET _ OD signal is low. The nRESET _ OD active time can be programmed to be 1 to 200 milliseconds. The default time is 200 milliseconds. The following events may result in the nRESET _ OD signal being set low (see event handler for more details): npro 2_ IN; SW/timer (programmable); WAKEUP conversion; low to high, and/or high to low (programmable); sensor connection conversion; low to high, and/or high to low (programmable); UART interruption; and RTC time event (programmable).

(vi) UP _ INT-UP _ INT may be coupled to a microprocessor to transmit interrupts. UP _ INT can be programmed as a wake event. The UP _ INT pad is the output on the VPAD power rail. If the battery protection signal indicates that the battery is low, the UP _ INT signal is low. The UP _ INT signal may be set to a high level by a software command to start a timer; the terminal count of the timer forces UP _ INT high. If the battery protection signal indicates that the battery is in good condition, the following events may result in the UP _ INT signal being asserted high (see event handler for more details): SW/timer (programmable); WAKEUP conversion; low to high, and/or high to low (programmable); sensor connection conversion; low to high and/or high to low (programmable); UART interruption; RTC time event (programmable); detecting that the battery is low through a battery protection circuit; and ASIC interrupts when unmasked.

POR2 event will reset a 2-bit counter whose bits map to GPIO1 and GPIO0 (MSB and L SB, respectively). the rising edge of the UART interrupt increments the counter by one, where the counter counts in modulo 4, and if incremented in state 11, the counter is zeroed.

Event processor/monitor

The ASIC incorporates an event handler to define responses to events, including changes in system state and input signals. Events include all interrupt sources (e.g., UART _ BRK, WAKE _ UP, sensor connections, etc.). The response of the event handler to the stimulus can be programmed by software through the SPI interface. However, some responses may be hardwired (not programmable).

Event handler operations include enabling/disabling VPAD _ EN, enabling/disabling C L K _32KHZ, setting nRESET _ OD, setting UP _ WAKEUP, and setting UP _ INT event monitor timers 1 through 5 may be individually programmed to 250 milliseconds through 16,384 seconds in 250 millisecond increments, timeout of event monitor timers 6 through 8 is hard coded, timeout of timer 6 and 7 is 1 minute, timeout of timer 8 is 5 minutes.

The ASIC also has a monitor function that can monitor the response of the microprocessor when it is triggered by an event. The event monitor is activated when the microprocessor fails to confirm the event-induced activity. Once activated, the event monitor performs a series of programmable actions, i.e., event monitor timers 1-5, followed by a series of hardwired actions, i.e., event monitor timers 6-8, to regain the response of the microprocessor. The series of actions includes interrupt, reset, wake up, set 32kHz clock, power down, and microprocessor power up.

If the ASIC fails to obtain confirmation from the microprocessor, the event monitor shuts down the microprocessor under conditions that allow the UART _ BRK to restart the microprocessor and activates an alarm, after activation, the alarm state generates a square wave on pad A L ARM in a programmable repetitive pattern at a frequency of about 1 kHz.

Digital-to-analog conversion (D/A)

In a preferred embodiment, the ASIC has two 8-bit D/a converters 4276, 4278 with the following characteristics:

(i) at loads less than 50pF, the D/A stabilizes in less than 1 millisecond.

(ii) The accuracy of D/a is at least 8 bits.

(iii) The output range can be programmed to 0 to 1.22V or 0 to VDDA.

(iv) The temperature sensitivity of the D/A voltage reference is less than 1 mV/DEG C.

(v) DN L is lower than 1L SB.

(vi) The current consumed by D/a is less than 2uA of the VDDA power supply.

(vii) Each digital to analog conversion has an output l to the pad.

(viii) The D/a output is high impedance. The load current must be less than 1 nA.

(ix) The D/a pad may be programmed to output a digital signal from the register. The output swing is from VSSA to VDDA.

Charger/data downloader interface

TX _ EXT _ OD 4280 is an open drain output with the input being the signal on the TX _ UP input pad. This would allow the TX EXT OD pad to be turned on in the UART idle state. The TX EXT OD pad has a comparator to monitor its voltage. If the voltage is higher than the comparator threshold voltage in the debounce period (1/4 seconds), the output nBAT _ CHRG _ EN (4281) will go low. The comparator and other related circuitry with this functionality are located on the VBAT and/or VDDBU layers.

The circuitry associated with this function must allow a low level on the TX _ EXT _ OD pad resulting from normal communication with the external device without disabling the setting nBAT _ CHRG _ EN. If POR1 is active, nBAT _ CHRG _ EN will be high (not set). The threshold voltage of the comparator is between 0.50V and 1.2V. The comparator will exhibit hysteresis; the falling threshold is about 25mV lower than the rising threshold.

The nRX _ EXT pad inverts the signal on this pad and outputs it to RX _ UP. In this way, the nRX _ EXT signal will be idle low. nRX _ EXT must accept input up to VBAT voltage. The nRX _ EXT threshold is 1.22V ± 3%. The output of the comparator will be provided to the microprocessor via the SPI bus for reading.

The nRX _ EXT pad also incorporates a means to programmably supply current, which will be 80 ± 30nA, with a maximum voltage VBAT. The ASIC layout has a mask programmable option to adjust the current from 30nA to 200nA in steps of less than 50nA with minimal change in the number of mask layers. Programmable bits may be used to prevent UART interrupt detection and force RX _ UP high. In normal operation, this bit is set high before the current source is enabled to nRX _ EXT and then low after the current source is disabled to ensure that no glitches (glitches) or UART interrupt events are generated on RX _ UP. Note that to implement a wet connector detector, when the current source into nRX _ EXT is active, the RX comparator output indicating a low input voltage will indicate leakage current. The ASIC contains a pull-down resistor of about 100k ohms on the nRX _ EXT pad. When the current source is active, the pull-down resistor will be turned off.

Sensor connecting switch

The ASIC should have a pad SEN _ CONN _ SW (4282) that can detect low resistance to VSS (4284). In the case where SEN _ CONN _ SW is 0V, SEN _ CONN _ SW supplies a current of 5 to 25uA, and the maximum open circuit voltage is 0.4V. The ASIC layout has a mask programmable option that can adjust the current from 1uA to 20uA in steps less than 5uA with minimal replacement of the mask layer. SEN _ CONN _ SW has associated circuitry that detects the resistance between SEN _ CONN _ SW and VSSA (4234), which has a threshold value between 2k ohms and 15k ohms. The average current consumption of the circuit is 50nA at most. Sampling must be used to achieve this low current.

Oscillator calibration circuit

The ASIC has a counter whose input can be directed to an internal or external clock source. One counter generates a programmable strobe interval for the other counter. The gating interval comprises a time of 1 to 15 seconds from the 32kHz oscillator. The clock that can be directed to either counter is the input of the 32kHz RC oscillator, the high speed RC oscillator and any one of the GPIO pads.

Oscillator bypass

The ASIC may replace the output of each oscillator with an external clock. The ASIC has a register that can only be written to when a particular TEST MODE is set. This register has a bit that enables an external input of the RC oscillator and can be shared with other analog test control signals. However, if TEST MODE is inactive, the register will not allow any oscillator bypass bit to be active.

The ASIC also has an input pad for an external clock to bypass the RC oscillator. The pad GPIO _ VBAT is positioned on a VBAT power layer. The ASIC further includes a BYPASS enable pad for the 32KHZ oscillator OSC32K BYPASS. At high, the 32KHZ oscillator output is provided through the OSC32KHZ _ IN pad. Note that typically, the OSC32KHZ _ IN pad is connected to the crystal.

The ASIC has an input for an external clock to bypass the HS _ RC _ OSC. The bypass is enabled by a programmable register bit. HS _ RC _ OSC may be provided programmably by GPIO on the VDD plane or GPIO on the VPAD plane.

SPI slave port

The SPI slave port includes an interface consisting of chip select input (SPI _ nCS)4289, clock input (SPI _ CK)4286, serial data input (SPI _ MOSI)4287, and serial data output (SPI _ MISO) 4288. The chip select input (SPI nCS) is an active low input that is set by the off-chip SPI master to initiate and define an SPI transaction. When SPI _ nCS is set low, the SPI slave port configures itself as an SPI slave and performs data transactions according to the clock input (SPI _ CK). When the SPI nCS is inactive, the SPI slave port will reset itself and remain in reset mode. Since the SPI interface supports data block transfers, the host should keep SPI nCS low until the transfer is complete.

The SPI clock input (SPI _ CK) will always be set by the SPI master. The SPI slave port latches the input data on the SPI _ MOSI input with the rising edge of SPI _ CK and drives the output data on the SPI _ MISO output with the falling edge of SPI _ CK. Data is transmitted from the SPI master to the SPI slave using the serial data input (SPI _ MOSI). All data bits are set after the SPI _ CK falling edge. The serial data output (SPI _ MISO) is used to transmit data from the SPI slave to the SPI master. All data bits are set after the SPI _ CK falling edge.

SPI _ nCS, SPI _ CK, and SPI _ MOSI are always driven by the SPI host unless the SPI host is powered down. If VPAD _ EN is low, then these inputs are regulated so that the leakage current associated with these inputs is less than 10nA, and the SPI circuitry remains in a reset or inactive state. When SPI _ nCS is active, SPI _ MISO is driven only by SPI slave port, otherwise SPI _ MISO is tri-state.

The chip select (SPI nCS) defines and constructs the data transport packets for the SPI data transaction. The data transmission packet is composed of three parts. There is a 4-bit command segment, then a 12-bit address segment, then any number of 8-bit data bytes. Command bit 3 is used as a direction bit. "1" represents a write operation and "0" represents a read operation. The combination of command bits 2, 1 and 0 has the following definition. Unused combinations are undefined.

(i) 0000: read data and an incremental address.

(ii) 0001: reading data with unchanged address

(iii) 0010: reading data, decrementing addresses

(iv) 1000: write data and incremental address

(v) 1001: writing data with unchanged address

(vi) 1010: writing data, decrementing addresses

(vii) x 011: test port addressing

The 12-bit address field defines the starting byte address. If SPI _ nCS remains active after the first data byte, then to indicate a multi-byte transfer, the address is incremented by one after each byte transfer. Bit <11> of the address (address <11:0>) represents the highest address bit. The address wraps around after reaching the boundary.

The data is in byte format and block transfer can be performed by extending SPI nCS to allow all bytes to be transferred in one data packet.

Microprocessor interrupts

The ASIC has an output UP _ INT of VPAD logic level for sending an interrupt to the host microprocessor. The microprocessor interrupt module consists of an interrupt state register, an interrupt mask register and a function of logically OR' ing all interrupt states into one microprocessor interrupt. Interrupts are implemented to support edge-sensitive and level-sensitive types. The polarity of the interrupt is programmable. The default interrupt polarity is TBD.

In a preferred embodiment, all interrupt sources on the AFE ASIC will be recorded in the interrupt status register. Writing a "1" to the corresponding interrupt status bit will clear the corresponding pending interrupt. All interrupt sources on the AFE ASIC may be masked by the interrupt mask register. Writing a "1" to the corresponding interrupt mask bit may mask the corresponding pending interrupt. Writing a "0" to the corresponding interrupt mask bit will disable the mask for the corresponding interrupt. The default state of the interrupt mask register is TBD.

General purpose input/output (GPIO)/parallel test port

In an embodiment, the ASIC may have eight GPIOs operating on VPAD level signals. The ASIC has a GPIO for operating on a VBAT level signal and a GPIO for operating on a VDD level signal. All GPIOs have at least the following characteristics:

(i) the register bits control the selection and direction of each GPIO.

(ii) The ASIC has means to configure the GPIO as an input that can be read through the SPI interface.

(iii) The ASIC has means to configure the GPIO as an interrupt generating input.

(iv) The ASIC may configure each GPIO to an output and be controlled by register bits that may be written to by the SPI interface.

(v) By the programmable manner, the ASIC can output signals applied to GPIO _ VBAT and GPIO _ VDD to GPIOs (on the VPAD power plane). (level shift function).

(vi) The ASIC has means to configure each GPIO as an input to the oscillator calibration circuit.

(vii) The ASIC has means to configure each general purpose comparator output to at least one GPIO on each power plane. The polarity of the comparator output can be programmed by a programmable bit.

(viii) GPIOs have microprocessor interrupt generation capability.

(ix) The GPIO may be programmed to turn on the drain output.

(x) The GPIO on the VPAD power plane may be configured to implement startup control of the microprocessor.

The parallel test ports share 8-bit GPIOs on the VPAD voltage plane. The test port will be used to observe the register contents and various internal signals. In normal mode, the output of the port is controlled by a port configuration register. Writing 8'hFF to both GPIO _ O1S _ REG and GPIO _ O2S _ REG registers will direct the testport data ON the GPIO output, while writing 8' h00 to the GPIO _ ON _ REG register will disable the testport data and enable the GPIO data ON the GPIO output.

The target register is addressed through the SPI slave port, and the register and pre-subgroup internal signals can be observed on the test port. The command bit of the SPI packet is set to 4' b0011 followed by a 12-bit destination register address. The parallel test port continues to display the contents of the addressed register until the next test port addressing command is received.

Simulation test port

The IC has a multiplexer feeding pad TP _ amamux (4290) which provides visibility to the internal analog circuit nodes for testing. The IC also has a multiplexer feeding pad TP _ RES (4260) which provides visibility to the internal analog circuit nodes for testing. In a typical application, the pad will also accommodate a precision 1meg resistor to perform various system calibrations.

Chip ID

The ASIC includes a 32-bit mask programmable ID. A microprocessor using the SPI interface will be able to read this ID. The ID will be placed in the analog electronics module, so no chip rerouting is required to change the ID. The design should be such that changing the ID only requires changing the metal once or changing the contact shield once.

Spare test output

The ASIC has 16 spare digital output signals that can be multiplexed to 8-bit GPIOs under commands sent over the SPI interface. These signals will be organized into two 8-bit bytes and will be connected to VSS when not in use.

Digital testing

The ASIC has a TEST mode controller that uses two input pins TEST _ CT L0 (4291) and TEST _ CT L1 (4292). the TEST controller generates signals from a combination of TEST control signals having the following functions (TEST _ CT L < 1:0 >):

(i)0 is the normal operating mode;

(ii)1 is a simulation test mode;

(iii)2 is a scanning mode;

(iv) and 3 is an analog test mode, and VDD _ EN is controlled by the input of GPIO _ VBAT.

In scan mode, test L T _ VBAT should be set high to condition the analog output of the digital logic.

Leakage test pin

The ASIC has a pin named L T _ VBAT that puts all analog blocks in inactive mode when high, so that only leakage current will be drawn from the power supply L T _ VBAT puts all digital outputs of the analog blocks in a stable high or low state to avoid impacting interface logic current consumption L T _ VBAT pad is on the VBAT plane and has a pull-down resistance between 10k ohms and 40k ohms.

Power supply requirement

In an embodiment of the invention, the ASIC includes a low power mode in which at least the microprocessor clock is off, the 32kHz real time clock is running, and the circuit is active to detect a change in the level of the WAKE _ UP pin or BREAK on the nRX _ EXT input. In this mode, the total current consumption of vbat (vddbu), VDD and VDDA is at most 4.0 uA. When the battery protection circuit detects that the battery is low (see description of battery protection circuit), the ASIC enters a mode where only the VBAT and VDDBU power planes are active. This is referred to as a low battery condition. The VBAT current in this mode is less than 0.3 uA.

In the case of programming the ASIC into a potentiostat configuration, either WORK electrode is at H₂O₂Active in (peroxide) mode with voltage set to 1.535V, COUNTER amplifier operated with VSET _ RE set to 1.00V, 20MEG load resistor connected between WORK and COUNTER, COUNTER and RE connected together, and assuming that WORK electrode current is measured once per minute, the average current consumption of all power supplies is less than 7 uA. The calibrated measurement current should be 26.75nA ± 3%. With a WORK electrode current of 25nA, enabling additional working electrodes would increase the combined leakage current by less than 2 uA.

In the case where the ASIC is programmed into a potentiostat configuration and a diagnostic function is enabled to measure the impedance of one of the WORK electrodes relative to the COUNTER electrode, the ASIC is configured to meet the following requirements:

(i) testing frequency: 0.1, 0.2, 0.3, 0.5Hz, 1.0, 2.0, 5.0, 10, 100, 1000 and 4000 Hz.

(ii) The frequency must not be measured for more than 50 seconds.

(iii) The total charge supplied to the application specific ASIC is less than 8 millicoulombs.

Environment(s)

In a preferred embodiment of the invention, the ASIC:

(i) operate in the commercial temperature range of 0 to 70 ℃ and meet all requirements.

(ii) Functionally between-20 ℃ and 80 ℃, but accuracy may be reduced.

(iii) It is expected that it will not operate until stored at a temperature of-30 to 80 c.

(iv) Operation is expected in the relative humidity range of 1% to 95%.

(v) Unless otherwise specified, ESD protection on all pins is greater than ± 2KV mannequin when packaged in a TBD package.

(vi) Configured to enable the WORK1-WORK5 pads, COUNTER pads, RE pads, TX _ EXT _ OD pads, and nRX _ EXT pads to withstand greater than + -4 KV manikins.

(vii) Is configured so that leakage current of the WORK1-WORK5 pad and the RE pad is less than 0.05nA at 40 ℃.

In an embodiment of the present invention, the ASIC may be fabricated in a 0.25 micron CMOS process, with the ASIC's backup data on the DVD disk 916-TBD.

As described in detail above, the ASIC provides the necessary analog electronics to provide the following functions: (i) supporting a plurality of potentiostats and interfacing with a multi-terminal glucose sensor based on oxygen or peroxide; (ii) interfacing with a microcontroller to form a micropower sensor system; and (iii) performing an EIS diagnosis based on the measurement of the EIS-based parameter. The measurement and calculation of EIS-based parameters will now be described in accordance with an embodiment of the present invention.

As previously mentioned, impedance at frequencies ranging from 0.1Hz to 8kHz can provide information about the state of the sensor electrodes. The AFE IC circuitry combines circuitry that generates the measurement forcing signal with circuitry that takes the measurement for calculating the impedance. Design considerations for this circuitry include current consumption, accuracy, measurement speed, required throughput, and on-time required to control the microprocessor.

In a preferred embodiment of the invention, the AFE IC technique for measuring electrode impedance is to superimpose a sine wave voltage on the DC voltage driving the electrodes and measure the phase and amplitude of the resultant AC current. To generate the sine wave, the AFE IC incorporates a digitally synthesized sine wave current. This digital technique is used because the frequency and phase can be precisely controlled by the crystal time base and frequencies up to 8kHz can be easily generated from DC. The sine wave current is applied across a resistor in series with the voltage source to add the AC component to the voltage of the electrode. This voltage is an AC forcing voltage. And then buffered by an amplifier driving the selected sensor electrode.

The current driving the electrodes contains a resultant AC current component from the forced sine wave and is converted to a voltage. The voltage is then processed by multiplying it by a square wave having a fixed phase relative to the composite sine wave. The multiplied voltage is then integrated. After a programmable number of integration intervals (which are integer multiples of 1/2 driving the sine wave period) have ended, the voltage is measured by the ADC. By calculations involving integrated voltage values, the real and imaginary parts of the impedance can be obtained.

The advantage of using an integrator for the impedance measurement is that the noise bandwidth of the measurement is significantly reduced compared to sampling the waveform only. Furthermore, the sampling time requirements are significantly reduced, thereby reducing the speed requirements of the ADC.

Fig. 45 shows the main blocks of the EIS circuitry in the AFE IC (denoted by reference numeral 4255 in fig. 42B). IDAC 4510 generates a stepped sine wave synchronized to the system clock. The high frequency of the system clock steps the IDAC through a lookup table containing digital codes. The code drives the IDAC, producing an output current that approximates a sine wave. The sine wave current is forced through a resistor to provide a DC offset VSET8(4520) to the AC component Vin _ AC. When the IDAC circuit is disabled, the DC output voltage is restored to VSET8 and therefore disturbance of the electrode balance is minimized. This voltage is then buffered by amplifier 4530, which passes through series resistor R_senseAnd a drive electrode. R_senseThe differential voltage across is proportional to the current. This voltage is provided to multiplier 4540 which multiplies the voltage by either +1 or-1. This is achieved by switches and differential amplifiers (instrumentation amplifiers). The system clock is divided to produce a phase clock 4550 which controls the multiplication function and may be set to 0, 90, 180 or 270 degrees relative to the sine wave.

The graphs in FIGS. 46A-46F and 47A-47F show a simulation of the signal of the circuit shown in FIG. 45 for a current with a 0 degree phase shift representing the actual resistance. For these example simulations, the analog input value was selected to give a current sense voltage equal to 0.150V. To obtain enough information to derive the impedance and phase, two integrations are required: the 0-degree phase multiplication is performed once (fig. 46A to 46F), and the 90-degree phase multiplication is performed once (fig. 47A to 47F).

Impedance calculation

The equation describing the integrator output is as follows. For simplicity, only 1/2 for the sine wave period is considered. As can be seen from the graphs of fig. 46A-46F and 47A-47F, the total integrator output will approximate the integrated value of 1/2 sine wave cycles times the number of integrated 1/2 cycles. Note that the integration time dependent multiplication switch performs the "gating" function of the integrator signal; this can be seen as setting a limit for the integration. The multiplication signal has a fixed phase with the generated sine wave. This may be set by software to 0, 90, 180 or 270 degrees. If the sine wave is in phase (0 degrees of motion) with respect to the multiplying square wave, the limits of the integration will be π (180) and 0 (0). If the sine wave is shifted by 90 degrees, the limits of integration can be considered to be 3/4 π (270) and 1/4 π (90).

The formula for multiplying a square wave in phase (0 deg.) with respect to the driving sine wave is shown below. This will produce a voltage proportional to the real component of the current. Note that Φ is the phase shift of the sine wave relative to the multiplied square wave; v_outIs the integrator output, A_amplIs the current sine wave amplitude. The period of the sine wave is also 1/f, and RC is the time constant of the integrator.

If phi is 0, thenThis corresponds to the real part of the current.

For a multiplicative square wave quadrature phase (90 °) with respect to the driving sine wave, an output proportional to the imaginary part of the current is produced:

if phi is 0, thenThis corresponds to the imaginary part of the current.

In a first exemplary plot shown in FIGS. 46A-46F, A_amplAt 0.150v, frequency 1kHz, Φ 0, RC of the integrator 20M ohm, 25pF, giving RC 0.5 ms. Substituting these numbers into the equation gives 0.09549v, which is advantageous compared to the integrator output of the curve in fig. 46. Note that the integrator output during integration is the voltage from the start of integration to the measurement.

For a square wave multiplication of 90 °, the result should be 0, since sin (0) is 0. The simulation results are close to this value.

To calculate the phase:

because of the fact thatIt follows that:wherein, V_out90Is the integrator output multiplied by a 90 phase shift，V_out0Is the 0 ° phase shift integrator output. V_out90Output sum V_out0The output must be integrated for the same 1/2 cycles or normalized by the number of cycles. It is important to note that in practical software (e.g., ASIC) implementations, only an integer number of cycles (360 °) are allowed, as the integer number of cycles compensates for any offset in the circuit before the multiplier.

The magnitude of the current may be based onOrOrAnd (4) determining. The current has a phase angle calculated as above.

The above analysis shows that the current amplitude and its phase relative to the multiplication signal can be determined. The forcing voltage is generated at a fixed phase (0, 90, 180 or 270 degrees) with respect to the multiplication signal-this is done digitally for precise control. But before the forced sine wave is applied to the electrodes, there is at least one amplifier in the path; this will introduce unnecessary phase shift and amplitude errors. This can be compensated by integrating the forced sine wave signal obtained electrically in the vicinity of the electrodes. Thus, the magnitude and any phase shift of the forcing voltage can be determined. Since the paths of the current and voltage waveforms will be handled by the same circuit, any analog circuit gain and phase errors will be cancelled out.

Because the variable of interest is impedance, it may not be necessary to actually calculate A_ampl. Because the current and voltage waveforms are integrated through the same path, there is a simple relationship between the ratio of current to voltage. The integrated current sense voltage is referred to as V_{I_out}The integral electrode voltage is referred to as V_{V_out}And the phase of the multiplication function is described with the additional subscript:

the impedance will be the voltage divided by the current. Therefore, the temperature of the molten metal is controlled,

the magnitudes of the voltage and current can also be obtained from the square root of the square of the 0 and 90 degree phase integrated voltages. Thus, the following may also be used:

for relatively high frequencies (e.g., frequencies above about 256 Hz), the integration of the waveform may be accomplished with a hardware integrator. High frequency requires four measurement cycles: (i) one for the in-phase sensor current; (ii) one for 90 degrees out of phase sensor current; (iii) one for the in-phase forcing voltage; and (iv) one for 90 degrees out of phase forcing voltage.

Two integrators may be used for relatively low frequencies (e.g., frequencies below about 256 Hz), and the integration value is made up of the digitally combined integrator results in the system microprocessor. Knowing how many integrations there are per cycle, the microprocessor can properly calculate the 0 degree component and the 90 degree component.

Synchronizing the integration with the forced AC waveform and splitting the integration into at least four parts at lower frequencies would eliminate the need for hardware multipliers since the combination of integrated parts in the microprocessor could implement the multiplication function. Thus, only one integration is needed to obtain real and imaginary current information. For lower frequencies, the amplifier phase error will become smaller, so below one frequency, for example between 1Hz and 50Hz, and preferably below about 1Hz, the forcing voltage phase will not need to be determined. Furthermore, for lower frequencies, the amplitude may be assumed to be constant, so that only one measurement cycle may be required to determine the impedance after settling.

As described above, although one hardware integrator is used for relatively high frequencies, two integrators may be used for relatively low frequencies. In this regard, the schematic diagram in fig. 45 shows EIS circuitry for relatively high EIS frequencies in the AFE IC. At these frequencies, the integrator does not saturate during integration over one period. In practice, multiple periods are integrated for the highest frequency, as this will provide a larger output signal, resulting in a larger signal-to-noise ratio.

For relatively low frequencies (e.g., frequencies below about 500 Hz), the integrator outputs may be saturated with common parameters. For these frequencies, therefore, two integrators are used that are alternately switched. That is, when the first integrator integrates, the second integrator is read by the ADC and then reset (zeroed out) so that it is ready to integrate at the end of the integration time of the first integrator. In this way, the signal can be integrated without gaps in the integration. This would add a second integrator and associated timing control to the EIS circuitry shown in fig. 45.

Stability period considerations

The above analysis is for a steady state condition, where the current waveform does not vary with period. Due to the initial state of the capacitor, this condition is not immediately met when a sine wave is applied to the resistance (capacitance (RC) network). The current phase starts at 0 degrees and proceeds to a steady state value. However, it is desirable that the measurement consume a minimum amount of time in order to reduce current consumption, and also allow sufficient time for the DC sensor measurement (Isig). Therefore, it is necessary to determine the number of cycles required to obtain a sufficiently accurate measurement.

The equation for a simple RC circuit (resistor and capacitor in series) is

The above discussion of solving I (t) gives:

wherein, V_c0Is the initial value of the capacitor voltage, V_mIs the amplitude of the driving sine wave and ω is the radian frequency (2 π f).

The first term comprises a term that defines a non-steady state condition. One way to accelerate system stabilization (setting) is to make the first term equal to 0, which can be achieved, for example, by setting

Or

Although this may not be necessary in practice, the initial phase of the forcing sine wave may be set to jump immediately from the DC steady-state point to V_cinit. The technique may be evaluated for a particular frequency and expected phase angle to look for possible time reductions.

The non-steady state term is multiplied by an exponential function of time. This will determine how quickly the steady state condition is reached. The RC value may be determined as a first order approximation based on the impedance calculation information. The following are given:

it follows that:

for a sensor at 100Hz and a phase angle of 5 degrees, this means a time constant of 18.2 milliseconds. For less than 1% stabilization, this means a stabilization time of about 85 milliseconds or 8.5 cycles. On the other hand, for a sensor at 0.10Hz and a phase angle of 65 degrees, this means a time constant of 0.75 seconds. For a build-up of less than 1%, this means a settling time of about 3.4 seconds.

Thus, in the embodiments of the invention detailed above, the ASIC contains (at least) 7 electrode pads, 5 of which are designated as WORK electrodes (i.e., sense or working electrodes, or WE), one of which is labeled COUNTER (i.e., COUNTER electrode, or CE) and one of which is labeled REFERENCE (i.e., REFERENCE electrode, or RE). COUNTER amplifier 4321 (see fig. 42B) may be programmably connected to COUNTER, REFERENCE, and/or any WORK distribution pad, as well as any combination thereof. As noted above, embodiments of the present invention can include, for example, more than five WEs. In this regard, embodiments of the present invention may also be directed to ASICs that interface with more than 5 working electrodes.

It is important to note that with the ASIC described herein, each of the five working, counter and reference electrodes described above is individually and independently addressable. Thus, any of the 5 working electrodes can be turned on and Isig (electrode current) measured, and any can be turned off. Further, any of the 5 working electrodes may be operably connected/coupled to EIS circuitry to measure EIS related parameters, such as impedance and phase. In other words, the EIS may be selectively operated on any one or more working electrodes. In addition, the respective voltage levels of each of the 5 working electrodes can be independently programmed in magnitude and sign relative to the reference electrode. This has many applications, for example, changing the voltage on one or more electrodes to make one or more electrodes less sensitive to interference.

In embodiments using two or more working electrodes as redundant electrodes, the EIS techniques described herein may be used, for example, to determine which of a plurality of redundant electrodes functions best (e.g., in terms of faster start-up, minimal or no dips, minimal or no sensitivity loss, etc.), so only the best working electrode or working electrodes may be addressed to obtain a glucose measurement. The latter, in turn, can greatly reduce (or even eliminate) the need for continuous calibration. At the same time, the other working electrode or electrodes (redundant) may: (i) shut down, which will aid in power management, since the "shut down" electrode may not run EIS; (ii) powering off; and/or (iii) periodically monitored by EIS to determine if it has recovered so that they can be brought back online. On the other hand, one or more non-optimal electrodes may trigger a calibration request. The ASIC is also capable of fabricating any electrode-including, for example, a failed electrode or an off-line working electrode (counter electrode). Thus, in embodiments of the present invention, the ASIC may have more than one counter electrode.

While the above generally addresses simple redundancy where the redundant electrodes have the same size, the same chemistry, the same design, etc., the diagnostic algorithms, fusion methods, and related ASICs described above can also be used in conjunction with spatially distributed, similarly sized, or different working electrodes as a way to assess the change in sensor implant integrity over implant time. Thus, in embodiments of the invention, sensors may be used that contain electrodes on the same flexure, which may be of different shapes, sizes and/or configurations, or contain the same or different chemicals for, for example, specific environments.

In yet other embodiments, the entire sensor design may contain different sizes of WEs. Such smaller WEs typically output a lower Isig (smaller geometric area) and may be used specifically for hypoglycemia detection/accuracy, while larger WEs (output a larger Isig) may be used specifically for both normal and hyperglycemia accuracy. In view of the size differences, different EIS thresholds and/or frequencies must be used for diagnosis between these electrodes. As described above, ASICs accommodate these requirements by enabling programmable, electrode-specific EIS standards.

As previously described, the ASIC contains a programmable sequencer 4266 that controls the start and stop of stimulation and coordinates the measurement of EIS-based parameters for frequencies above about 100 Hz. At the end of the sequence, the data is in a buffer memory and the microprocessor can quickly obtain (the value of) the desired parameter. This saves time and reduces system power requirements by requiring less microprocessor intervention.

For frequencies below about 100Hz, programmable sequencer 4266 coordinates the start and stop of EIS stimulation and buffers the data. At the end of the measurement period, or if the buffer becomes close to full, the ASIC may interrupt the microprocessor to indicate that it needs to collect available data. The depth of the buffer will determine how long the microprocessor can complete other tasks or sleep while collecting the EIS-based parameters. For example, in one preferred embodiment, the depth of the buffer is 64 measurements. Also, this saves energy because the microprocessor does not need to collect data one by one. It should also be noted that sequencer 4266 also has the ability to initiate stimulation at a phase other than 0, which has the potential for faster settling.

As described above, the ASIC may control the power provided to the microprocessor. Thus, for example, based on detection of sensor connection/disconnection using, for example, a mechanical switch or capacitive or resistive sensing, it may shut down the power supply completely and power up the microprocessor. In addition, the ASIC may control the wake-up of the microprocessor. For example, the microprocessor may place itself in a low power mode. Then, if sensor connection/disconnection detection is made, for example, by the ASIC, the ASIC may send a signal to the microprocessor, which wakes up the processor. This involves responding to the ASIC generated signal using techniques such as mechanical switches or capacitance-based sensing schemes. This allows the microprocessor to sleep for a long time, thereby significantly reducing power consumption.

It is important to reiterate that with an ASIC as described above, oxygen sensing and peroxide sensing can be performed simultaneously, since the five (or more) working electrodes are independent and independently addressable, and thus can be configured in any desired manner. In addition, the ASIC allows multiple flags to have multiple thresholds, so that the EIS resistance can be triggered by various factors, e.g., V_cntrLevel, capacitance change, signal noise, large changes in Isig, drift detection, etc. (each with its own threshold or thresholds). Further, for each such factor, the ASIC supports multiple levels of thresholds.

According to an embodiment of the present invention, an equivalent circuit model as shown in FIG. 48 can be used to simulate the measured EIS between the working and reference electrodes WE and RE, respectively. The circuit shown in FIG. 48 has a total of six (6) elements, which can be divided into three broad categories, (i) reaction-related elements; (ii) a component associated with the membrane; (iii) solution-related elements. In the latter category, Rsol is the solution resistance and corresponds to a property of the environment outside the sensor system (e.g., interstitial fluid in the body).

The reaction-related element includes R_p(i.e., polarization resistance (i.e., voltage bias and charge transfer resistance between the electrode and the electrolyte)) and Cdl (i.e., double layer capacitance at the electrode-electrolyte interface). Note that while in this model, the double layer capacitance is shown as a Constant Phase Element (CPE) due to the non-uniformity of the interface, it can also be modeled as a pure capacitance

Thus, the model contains two (2) reaction-related elements-R_pAnd Cdl, which is represented by a total of three (3) parameters: r_pCdl and α.

Membrane-related elements include Rmem (i.e., membrane resistance (or resistance due to a chemical layer)), and Cmem (i.e., membrane capacitance (or capacitance due to a chemical layer)). Although Cmem is shown as pure capacitance in fig. 48, it can also be modeled as CPE in special cases. As shown, W is a bounded Valley element and has two parameters: y is₀Which means due to glucose/H₂O₂Admittance of the valburg element by diffusion within the chemical layer; and λ, which represents the diffusion time constant of the walbauer element. Note that the walbauer may also be modeled in other ways (e.g., unbounded). The frequency dependent impedance of a bounded Valley element may be calculated as follows

Thus, this model contains three (3) membrane-associated elements — Rmem, Cmem, and W, which are represented by a total of four (4) parameters: rmem, Cmem, Y₀And λ.

Here, it is important to note that while a single electrode is depicted, this is by way of illustration only and not by way of limitation, as this model may be applied to sensors having a greater number of layers and more electrodes than the illustrative 3-layer single electrode configuration shown in FIG. 48.

In the discussion that follows, the equivalent circuit model of FIG. 48 will be used to explain some of the physical characteristics of the sensor behavior. However, it should be mentioned that other circuit configurations are possible depending on the way the glucose diffusion is simulated. In this regard, fig. 49A-49C show illustrations of additional circuit models, some of which contain a greater number of elements and/or parameters. However, for purposes of this discussion, it has been found that the circuit model of fig. 48 provides a best fit to empirical data, where the mass transport limit (i.e., the walburg component) is due to the diffusion of glucose through the membrane. Fig. 50A is a nyquist curve showing that the equivalent circuit simulation 5020 fits the empirical data 5010 very closely. FIG. 50B is an enlarged view of the high frequency portion of FIG. 50A, showing that the simulation also tracks the actual sensor data in this region very accurately.

Each of the above circuit elements and parameters affect the output of the EIS in various ways. Fig. 51 shows a nyquist curve in which Cdl increases in the direction of arrow a. It can be seen that as the value of Cdl increases, the length of the (lower frequency) nyquist curve decreases and its slope increases. Thus, the length of the nyquist curve decreases from curve 5031 to curve 5039, and each of curves 5033, 5035, and 5037 has a respective length that gradually decreases as Cdl increases from curve 5031 to curve 5039. In contrast, the slope of the nyquist curve increases from curve 5031 to curve 5039, and each of curves 5033, 5035, and 5037 has a corresponding slope that gradually increases as Cdl increases from curve 5031 to curve 5039. However, the high frequency region of the nyquist curve is generally unaffected.

FIG. 52 shows a Nyquist curve where α increases in the direction of arrow A. Here, as α increases, the slope of the Nyquist curve increases in the lower frequency region, in FIG. 53, as R increases_pIncreasing in the direction of arrow a, the length and slope of the low frequency nyquist curve increases. The higher the Rp, the more resistant to chemical reactions and hence the slower the rate of electron and ion exchange. Thus, phenomenologically, fig. 53 shows that the length and slope of the low frequency nyquist curve increase with decreasing electron-ion exchange rate (i.e., with increasing resistance to chemical reaction), which in turn means lower current (Isig) output. Also, there is little to no effect on the high frequency region of the Nyquist curve.

FIG. 54 illustrates the effect of a change in the Walberg admittance. As the varburg admittance increases in the direction of arrow a, both the length and the slope of the low frequency nyquist curve increase. Phenomenologically, this means that the length and slope of the low frequency nyquist curve increase with increasing reactant inflow. In fig. 55, the slope of the nyquist curve decreases as λ increases in the direction of arrow a.

In contrast to the above elements and parameters, the elements and parameters associated with the membrane generally affect the high frequency region of the nyquist curve. FIG. 56 shows the effect of film capacitance on the Nyquist curve. As can be seen in FIG. 56, the variation in Cmem affects the visibility of the high frequency region semicircle. Thus, as the membrane capacitance increases in the direction of arrow a, a gradual decrease in the semi-circle can be seen. Similarly, as shown in fig. 57, as the membrane resistance increases in the direction of arrow a, more high-frequency region semicircles become visible. Furthermore, as Rmem increases, the nyquist curve moves generally from left to right. The latter parallel shift phenomenon is also applicable to Rsol, as shown in fig. 58.

The discussion above in connection with the equivalent circuit model of FIG. 48 can be summarized as follows_imagMainly depending on Cdl, α, Rp and λ, low frequency length/Z_magnitudeMainly depending on Cdl, Rp and the varburg admittance. Second, Rmem andcmem controls the high frequency response. In particular, Rmem determines the high frequency semi-circle diameter and Cmem determines the turning frequency, with minimal overall impact on the nyquist curve. Finally, changes in Rmem and Rsol cause a parallel shift of the nyquist curve.

Figures 59A-59C, 60A-60C and 61A-61C show the results of in vitro experiments with the above described changes in circuit elements during sensor start-up and calibration, figures 59A, 60A and 61A are identical, as shown in figure 59A, experiments were typically performed using two redundant working electrodes 5050, 5060 for (7 to) 9 days, using a baseline glucose amount of 100mg/d L, although the glucose amount varied (5070) between zero and 400mg/d L at various points throughout the experiment, furthermore, the effect of (solution) temperature variations between 32 ℃ and 42 ℃ (5080) and 0.1mg/d L acetaminophen reaction (5085) was also investigated, finally, experiments included an oxygen stress test in which the supply of oxygen dissolved in solution varied (i.e. limited) between 0.1% and 5% (5075) for the purposes of these experiments, one complete EIS scan was run (i.e. from 0.1 to 8kHz) and data was recorded at approximately every 30 minutes (however, again using shorter or shorter time intervals).

In fig. 59C, the sum of Rsol and Rmem (again, estimated by the magnitude of the real impedance at the inflection point of the nyquist curve) shows an overall downward trend over time. This is primarily because the membrane takes time to hydrate and thus, over time, its resistance to charge decreases. A slight correlation can also be seen between the plot for Isig (fig. 59A) and the plot for Rsol + Rmem (fig. 59C).

FIG. 60B shows the EIS output of Cdl. Here, there is initially a relatively rapid drop in time of several hours due to the sensor activation/sensor charging process (5087). However, thereafter, Cdl remained fairly constant, exhibiting a strong correlation with Isig (fig. 60A). In view of the latter correlation, Cdl data as an EIS parameter may be less useful in applications requiring glucose independence. As shown in fig. 60C, the trend of Rp can be generally described as a mirror image of the curve of Cdl. As the hydration level of the membrane increases, the influx increases, which is reflected in the walbauer admittance diagram in fig. 61B. As shown in fig. 61C, λ is always kept substantially constant.

FIGS. 62-65 show the actual EIS response of various portions of the experiment described above. Specifically, the changes that occurred within the first 3 days (i.e., glucose changes, oxygen stress, and temperature changes), as shown in FIGS. 59A, 60A, and 61A, are boxed (5091), V in FIG. 62_cntrResponse 5093 is shown at the bottom of the figure and in FIG. 59B. Figure 63 shows that the slope and length of the nyquist curve were reduced by Isig calibration with increased glucose. In FIG. 64, oxygen (or V)_cntr) The response was shown on day 2, where V_cntrBecoming more negative as the oxygen content decreases. Here, the length of the nyquist curve becomes shorter, and the slope thereof decreases (5094), indicating that the imaginary impedance is greatly reduced. The curve length is mainly determined by Cdl and Rp, and is related to V_cntrAre closely related, whereas V_cntrAnd also to changes in glucose and oxygen. In fig. 65, the change in Isig from day 2 to day 3 is negligible. However, the nyquist curve shifts horizontally (from the curve at 37 ℃) for data acquired at 32 ℃ (5095) and 42 ℃ (5097). However, there is no significant effect on the nyquist curve length, slope or Isig.

Putting the EIS output and the characteristic information together: during sensor startup, the magnitude of Rmem + Rsol decreases over time, corresponding to a shift in the nyquist curve from right to left. During this time, Cdl decreases and Rp increases, with a corresponding increase in the nyquist slope. Finally, the Walberg admittance also increases. As previously mentioned, the foregoing is consistent with the hydration process, and the EIS curves and parameter values take approximately 1-2 days (e.g., 24-36 hours) to stabilize.

Embodiments of the present invention are also directed to real-time self-calibration, and more particularly to in vivo self-calibration of glucose sensors based on EIS data. Any calibration algorithm, including self-calibration algorithms, must address the sensitivity loss problem. As previously mentioned, two types of sensitivity loss may occur: (1) isig dips, a temporary loss of sensitivity, typically occurring in the first few days of sensor operation; (2) permanent sensitivity loss, which usually occurs at the end of the sensor life, sometimes associated with V_cntrThe presence of tracks being related。

It has been found that the loss of sensitivity can be manifested as an increase in Rsol or Rmem (or both), which can be observed as a parallel shift to the right in the nyquist curve, or, if Rmem changes, it can be manifested as the beginning of a semicircle being more pronounced at higher frequencies (causing an increase in high frequency imaginary impedance). In addition to or instead of Rsol and Rmem, Cmem may be increased. This can be observed as a change in the high frequency half circle. The sensitivity loss will be accompanied by a Cdl change (a longer tail in the low frequency part of the nyquist curve). The above features provide an apparatus for determining how to use different changes in EIS output to compensate for changes in sensitivity.

For a normally functioning glucose sensor, there is a linear relationship between Blood Glucose (BG) and the current output of the sensor (Isig). Therefore, the temperature of the molten metal is controlled,

BG＝CF×(Isig+c)

where "CF" is the calibration factor and "c" is the offset. This is shown in fig. 66, where the calibration curve is shown as line 6005, "c" is the baseline offset 6007 (in nA). However, when Rmem increases and/or Cmem decreases, c will be affected. Thus, line 6009 depicts the case where Rmem increases and Cmem decreases, which represents a change in the membrane properties, resulting in the shift "c" to 6011, i.e. the calibration curve moves downwards. Similarly, when the (non-glucose related) change in Cdl occurs and Rp increases, and the length of the (low frequency) nyquist curve is increased, then the slope will suffer, where the slope is 1/CF. Therefore, in fig. 66, the line 6013 has a different (smaller) slope than the line 6005. A combined change may also occur, as shown by line 6015, indicating a loss of sensitivity.

Nyquist curve (L)_nyquist) The length of the low frequency band of the nyquist curve can be estimated illustratively as the length between 128Hz and 0.105Hz (true) impedance for simplicity, which is highly correlated with glucose variations. The model fitting found that the only parameter that changes during glucose change is the double layer capacitance Cdl, especially the double layer admittance. Thus, in the equivalent circuit model of FIG. 48, the only Isig-dependent parameter and thus extended dependenceThe parameter for glucose is Cdl, all other parameters being essentially independent of Isig.

In view of the above, in one embodiment, changes in Rmem and Cmem can be tracked to achieve readjustment of the calibration factor (BG/Isig) and thereby achieve real-time self-calibration of the sensor without the need for continuous finger touch testing. This is possible in part because changes in Rmem and Cmem result in changes in the offset (c) of the calibration curve, but do not result in changes in the slope. In other words, such a change in the membrane-related parameter of the model typically indicates that the sensor is still functioning properly.

Fig. 67A graphically illustrates actual Blood Glucose (BG) data 6055 being recorded, overlaid by Isig output 6060 from the working electrode. Comparing data (6051) from a first time period (or time window) comprising about days 1-4 with data (6053) from a second time period comprising about days 6-9, fig. 67A shows that the sensor drifts generally downward during the second time period, indicating that a moderate loss of sensitivity may occur in the sensor. As shown in fig. 67B, during the second period of time, V_cntrThere is also an increase.

Referring to fig. 68 and 69, it can be seen that the loss of sensitivity is clearly illustrated by a fairly significant increase in the membrane resistance 6061 and a corresponding decrease in the walbau admittance 6063 during the second time period between days 6 and 9, thus, fig. 70 illustrates that the calibration curve 6073 of the second time period 6053 is parallel to, but shifted down from, the calibration curve 6071 of the first time period 6051, further, as discussed above in connection with fig. 57, as the membrane resistance (Rmem) increases, the total nyquist curve shifts from left to right, and more high frequency region semicircles become visible, for the data of fig. 67A-70, this phenomenon is illustrated in fig. 71, where the amplified high frequency region of the nyquist curve illustrates that the curve shifts from left to right compared to the data from the first time period 6051, and as the nyquist curve shifts from left to right (6080), further, the amplified low frequency region of the curve indicates that L becomes increasingly visible (6080)_nyquistThe frequency does not change significantly.

On the other hand, changes in Cdl and Rp often indicate that one or more electrodes may have been damaged, and thus recovery may no longer be possible. However, the changes in Cdl and Rp may also be tracked (e.g., as a diagnostic tool) to determine whether the drift or loss of sensitivity has actually reached a point where proper sensor operation is no longer recoverable or achievable based on the direction/trend of the changes in these parameters. In this regard, in embodiments of the present invention, a respective lower threshold and/or upper threshold or threshold range may be calculated for each of Cdl and Rp or for changes in slope, such that EIS output values of these parameters that fall outside the respective threshold (range) may trigger termination and/or replacement of the sensor, for example, due to an unrecoverable loss of sensitivity. In particular embodiments, a sensor design and/or patient-specific range or threshold may be calculated, where the range/threshold may be, for example, a change relative to Cdl, Rp, and/or slope.

Fig. 72A graphically illustrates actual Blood Glucose (BG) data 6155 being recorded overlaid by Isig output from two working electrodes WE 16160 and WE 26162. The graphs show data for a first time window (6170) from day 1, a second time window (6172) from day 3-5, a third time window (6174) from day 3, and a fourth time window (6176) from day 5-9. From day 3 on, FIG. 72B shows V at 1.2 volts_cntrMoving along the rail. However, from about day 5 (6180), the sensitivity decreases. Once V is_cntrThe shift, Cdl, increased significantly and Rp decreased accordingly, indicating a higher resistance of the overall electrochemical reaction As expected, the slope of the calibration curve also changed (decreased), and L_nyquistBecomes shorter (see fig. 73-75). Note that, in the embodiment of the present invention, V_cntrThe presence of the rail can be used to trigger the unrecoverable termination of the sensor.

FIGS. 76A-76B and 77-80 show the increase in membrane resistance, decrease in Cdl and V_cntrThe combined effect of the rails. In fig. 76A, actual Blood Glucose (BG) data 6210 is overlaid by Isig outputs from two working electrodes WE 16203 and WE 26205. It can be seen that WE1 generally tracks the actual BG data 6210, i.e., WE1 is functioning properly.On the other hand, Isig of WE2 seems to start from a lower point and continues to trend downward from the beginning to day 10, thus implying a gradual loss of sensitivity. This is consistent with Cdl of WE2(6215) being lower than that of WE1(6213), as shown in fig. 77, although Cdl of both working electrodes generally shows a downward trend.

Fig. 79 illustrates the combined effect on the calibration curve, where both the offset and slope of the linear fit of the sensitivity loss period (6235) are varied relative to the calibration curve 6231 for the normal operating time window. Further, the nyquist curve of fig. 80 shows that, in the lower frequency region, the length of the nyquist curve is longer when sensitivity loss occurs (6245) than when the sensor is operating normally (6241). Further, near the inflection point, the semicircle (6255) becomes more and more pronounced when there is a sensitivity loss. Importantly, the nyquist curve of fig. 80 moves horizontally from left to right over time when there is a loss of sensitivity. In an embodiment of the invention, the latter movement may be used as a measure for compensation or self-correction in the sensor.

Thus, as discussed herein, as an EIS characteristic, an increase in membrane resistance (Rmem) and/or a local increase in Rsol may cause a temporary dip. The increase in Rmem is in turn reflected in an increase in the high frequency virtual impedance. This increase may be represented by the slope (S) at high frequencies_nyquist) Characterized, for simplicity, the slope may be exemplarily estimated as a slope between 8kHz and 128 Hz. In addition, V_cntrIncreasing Cdl and decreasing Rp along the track, thereby decreasing length and slope; this may be accompanied by a gradual decrease in Cdl and increase in Rp associated with loss of sensitivity. In general, a decrease in Cdl, plus an increase in Rp (length increase) and Rmem, may be sufficient to result in a loss of sensitivity.

In accordance with an embodiment of the present invention, a sensor self-calibration algorithm based on detection of sensitivity changes and/or losses is shown in FIG. 81 at blocks 6305 and 6315, the baseline Nyquist curve length is set (L)_nyquist) And a baseline higher frequency slope to reflect the EIS state at the beginning of sensor life. As mentioned above, the Nyquist curve length is related to Cdl, a higher frequency of NyquistThe stedt slope is related to the sheet resistance. The process then proceeds by monitoring the Nyquist curve length (6335) and higher frequency slope (6345) and V_cntrThe value (6325) continues. When V is_cntrIn translation, baseline L_nyquistIs adjusted or reset 6355 because V_cntrCdl is significantly altered along the track. Thus, there is a feedback loop 6358 to accommodate real-time changes in the monitored EIS parameters.

As block 6375 shows, as the length of the nyquist curve is monitored, a significant increase in that length will indicate a decrease in sensitivity. In particular embodiments, a sensor design and/or patient-specific range or threshold may be calculated, wherein the range/threshold may be, for example, a variation with respect to the length of the nyquist curve. Similarly, a more negative high frequency slope S_nyquistCorresponding to the increased occurrence of the high frequency semi-circle and will indicate a possible dip 6365 monitoring L_nyquistAnd S_nyquistAnd based on the duration and trend of the sensitivity decline, determining whether a total (i.e., severe) loss of sensitivity has occurred such that one or more particular Sensor Glucose (SG) values should be discarded (6385). In block 6395, the calibration factor may be adjusted based on the monitored parameter to provide a "calibration-free" CGM sensor. Note that in the context of the present invention, the term "calibration-free" does not mean that a particular sensor does not require calibration at all. Instead, this means that the sensor can self-calibrate in real time (e.g., based on the output data of the EIS) without requiring additional finger touches or meter data. In this sense, self-calibration may also be referred to as "smart" calibration, since calibration is not performed based on a predetermined schedule, but is performed in real-time as needed.

In embodiments of the invention, the algorithm for adjusting the Calibration Factor (CF) and/or the offset may be based on the membrane resistance, which in turn may be estimated by the sum of Rmem and Rsol. Since the sheet resistance represents a physical characteristic of the sensor, it is generally not possible to estimate the sheet resistance from EIS data at a single frequency. In other words, it has been observed that no single frequency can consistently represent the membrane resistance, as the frequency shifts according to the sensor state. Thus, for example, fig. 82 shows that when there is a certain degree of sensitivity loss, there is horizontal movement in the nyquist curve, and therefore movement occurs in the inflection point where the value of Rmem + Rsol is estimated. In this case, the movement of the real component of the impedance is actually quite large. However, if only the high frequency (e.g., at 8kHz) real impedance is monitored, there is little or no offset, as shown by the circled area in fig. 82.

Therefore, there is a need to track the membrane resistance in a physically meaningful way. Ideally, this can be achieved by model fitting, where Rmem and Rsol are derived from the model fitting, Rm is calculated as follows: rm ═ Rmem + Rsol. However, in practice, this approach is not only computationally expensive, as it may require an unpredictable long time, but in some cases is prone to misconvergence. Thus, a heuristic metric may be developed to approximate or estimate the value of Rm + Rsol. In one such metric, Rmem + Rsol is approximated as a real impedance intercept value at a fairly stable imaginary impedance value. Thus, as shown in fig. 83, for example, the overall stable region of the imaginary impedance (on the Y-axis) may be identified as about 2000 Ω. Taking this as a reference value, and along a direction parallel to the X-axis, a value proportional to Rm can be approximated as the real impedance value where the reference line intersects the nyquist curve. Interpolation between frequencies may be performed to estimate Δ Rm ∈ Δ (Rmem + Rsol).

After estimating the value of Rm as described above, the relationship between Rm and the Calibration Factor (CF) and/or Isig may be explored. Specifically, fig. 84 shows the relationship between estimated Rm and CF, where the former is proportional to the latter. The data points for purposes of fig. 84 are derived for steady state sensor operation. FIG. 85 shows a graph of normalized Isig versus 1/Rm, where Isig has been normalized by the BG range (of Isig). As can be seen, Isig can be adjusted according to the change of Rm. Specifically, an increase in 1/Rm (i.e., a decrease in membrane resistance) will result in a proportional increase in Isig, since there is a linear relationship between Isig and 1/Rm.

Thus, in one embodiment, the algorithm for adjusting the calibration factor will need to monitor the change in membrane resistance based on a reference calibration factor and then modify the calibration factor proportionally based on the correlation between Rm and CF.

In other words:

in another embodiment, the calibration factor adjustment algorithm may need to modify Isig based on a 1/Rm scale change and independent of CF calculations. Thus, for the purposes of such an algorithm, the adjusted Isig may be derived as follows

Experiments have shown that the most drastic CF changes occur in the first 8 hours of sensor life. Specifically, in one set of in vitro experiments, Isig was plotted as a function of time while keeping various glucose levels constant over the lifetime of the sensor. The EIS was run every 3 minutes for the first 2 hours, with all model parameters being evaluated and tracked. As previously mentioned, Rmem and Rsol cannot (independently) be robustly estimated given a finite spectrum of EIS. However, Rm + Rsol may be estimated.

FIG. 86 shows Isig's curves over time for different glucose levels, including 400mg/d L (6410), 200mg/d L (6420), 100mg/d L (6430), 60mg/d L (6440), and 0mg/d L (6450). all parameters typically change significantly upon startup FIG. 87 shows an example where Cdl is plotted as a function of time, curve 6415 corresponds to 400mg/d L glucose, curve 6425 corresponds to 200mg/d L glucose, curve 6435 corresponds to 100mg/d L glucose, curve 6445 corresponds to 60mg/d L glucose, and curve 6455 corresponds to 0mg/d L glucose.

In particular, fig. 88 shows the same graph as in fig. 86 except for an indication of the presence of a peak or second inflection point that occurs at about T1 hour, particularly at low glucose levels, e.g., 100mg/d L or less.

Fig. 90 shows the relationship between the calibration factor and Rm for the in vivo data over the first 8 hours of sensor operation. Here, the EIS was run approximately every 30 minutes at start-up with a period of time interposed between the two. It can be seen that Rm + Rsol correlates to the Calibration Factor (CF) in the first 8 hours of sensor operation. For purposes of the graph in fig. 90, the baseline offset is assumed to be 3 nA.

As described above in connection with FIGS. 83-85, in one embodiment, the algorithm for adjusting the calibration factor at startup may comprise: selecting a reference value (CF) for the calibration factor_reference) (ii) a Estimating CF ═ CF_referenceMembrane resistance (R)_reference) A value of (d); monitoring a change in membrane resistance (Rm ═ Rmem + Rsol); and adjusting the calibration factor according to the relationship shown in figure 90 based on the magnitude of the change. Thus, it is possible to provide

CF(t)＝CF_reference-m(R_reference-R_m(t))

Where m is the associated gradient in figure 90. Note that for the purposes of the above algorithm, CF takes into account the differences between the sensors_referenceThe value of (b) is sensor specific.

In another embodiment, the adjustment may be made by using a limited range of R over which the adjustment occurs_mTo modify the calibration factor adjustment algorithm. Once R is present_mLess than 7000 omega, which may help with small differences, as noise may cause this. When R is_mVery large, finite R_mRange may also be helpful because very slow sensor hydration/stabilization is possibleThis may occur. In yet another embodiment, the range of allowable CF may be limited, for example, by setting the lower limit of CF to 4.5.

Fig. 91A is a graph showing in vivo results for MARD over all effective BGs over about the first 8 hours of sensor life. A single (first) calibration with the first BG was performed 1 hour, 1.5 hours, or2 hours after start-up. It can be seen that the MARD at 1 hour calibration is much higher than the MARD value at 2 hours calibration (22.23 versus 19.34) without any adjustment of the calibration factor. However, by adjusting or modifying the adjustment as described above, the difference between the respective MARD values becomes smaller. Thus, for example, by adjustment, the MARD at 1 hour calibration is 16.98 compared to 15.42 at 2 hours for calibration. Furthermore, the MARD adjusted over 1 hour of calibration was much less than the MARD without 2 hours of calibration (16.98 vs 19.34). Thus, according to embodiments of the present invention, calibration factor adjustment (and modified adjustment) may be used to extend the usable life of the sensor — for example, in this example, by starting the sensor one hour in advance while maintaining or improving MARD. The graph in fig. 91B provides the median ARD number for all available blood glucose in about the first 8 hours.

Fig. 92A-92C, 93A-93C, and 94A-94C illustrate examples where the above calibration factor adjustment algorithm works better than some current, non-EIS based methods. In one such method, commonly referred to as "day one compensation" (or FDC), a first calibration factor is measured. If the measured calibration factor falls outside the predetermined range, a constant linear decay function is applied to return the calibration factor to within the normal range at the expected time determined by the decay rate. As can be seen from fig. 92A-94C, the calibration factor adjustment algorithm (referred to as "compensation" in the figure) 6701, 6711, 6721 of the present invention produces results that are closer to the actual Blood Glucose (BG) measurements 6707, 6717, 6727 than the results obtained by the FDC methods 6703, 6713, 6723.

Given the complexity of estimating EIS-related parameter values, some current methods (including FDC) may not be computationally complex than the EIS calibration factor adjustment algorithm described herein. However, the two methods can also be implemented in a complementary manner. In particular, there may be cases where FDC is enhanced by an immediate calibration factor adjustment algorithm. For example, the latter may be used to define the rate of change of the FDC, or to determine the extent of FDC application (i.e., other than using CF alone), or to reverse the direction of FDC in special cases.

In other embodiments, the offset may be adjusted instead of the calibration factor. In addition, or alternatively, to R_mAnd the applicable range of CF imposes limitations. In particular embodiments, absolute values may be used rather than relative values. Furthermore, the relationship between calcium factors and cell membranes can be expressed as multiplications, rather than additions.

Therefore, the temperature of the molten metal is controlled,

in embodiments using EIS-based dynamic biasing, the measured total current may be defined as the sum of the faraday current and the non-faradaic current, where the former depends on glucose and the latter does not. Therefore, in a mathematical sense,

i_total＝i_Faradaic+i_non-F_aradaic

ideally, the non-faradaic current should be zero, with a fixed operating potential, such that

Wherein A is the surface area, andis a gradient of peroxide.

However, when the double layer capacitance is changing, the non-faradaic current is not negligible. Specifically, the non-faradaic current may be calculated as follows

Where q is the charge, V is the voltage and C is the (double layer) capacitance. From the above, it can be seen that when both voltage (V) and capacitance (C) are constant, both time derivatives on the right side of the equation are equal to zero, so i_non-Faradaic0. In this ideal case, the focus can be shifted to diffusion and reaction.

When V and C are both functions of time (e.g., at sensor initialization),

on the other hand, when V is a constant, and C is a function of time,

this occurs, for example, on day 1 of sensor operation. Fig. 95 shows an example of a typical (initial) decay of the double layer capacitance during day 1, in this case occurring in the first 6 hours after sensor insertion. As shown, curve 6805 shows raw Cdl data based on EIS data obtained over a half-hour interval, curve 6810 shows a spline fit of raw Cdl data over a 5 minute time interval, curve 6815 shows a smoothed curve over a 5 minute time interval, and curve 6820 shows a polynomial fit of smoothed Cdl data over a 5 minute time interval.

Note that Cdl decay is not exponential. Therefore, the attenuation cannot be modeled with an exponential function. In contrast, a polynomial fit of order 6 (6820) has been found to provide reasonable simulations. Thus, for the above case where V is a constant and C is a function of time, if the polynomial coefficients are known, i can be calculated_non-Faradaic. In particular, the present invention relates to a method for producing,

C＝P(1)t⁶+P(2)t⁵+P(3)t⁴+P(4)t³+P(5)t²+P(6)t¹+P (7)

where P is the polynomial coefficient array and t is time. The non-faradaic current can be calculated as follows:

finally, since i_total＝i_Faradaic+i_non-FaradaicThe non-faradaic component of the current may be removed by rearrangement, so that

i_Faradaic＝i_total-i_non-Faradaic

Graph 96 shows Isig based on total current as a function of time (6840), and Isig based on capacitance decay minus non-faradaic current (6850). The non-faradaic component of the current can be as high as 10-15 nA. As can be seen in the figure, removing the non-faradaic current helps remove most of the low start Isig data at the beginning of the sensor life.

It has been found that the above method can be used to reduce the MARD and adjust the calibration factor at the beginning of the sensor life. With respect to the latter, fig. 97A shows the calibration factor prior to removing the non-faradaic current of first working electrode (WE1)6860 and second working electrode (WE2) 6870. On the other hand, fig. 97B shows the calibration factors for WE1(6862) and WE2(6872) after the non-faradaic current is removed. Comparing the calibration factor for WE1 in fig. 97A (6860) with the calibration factor for WE1 in fig. 97B (6862), it can be seen that the calibration factor (6862) is closer to the expected range after removal of the non-faraday component.

Furthermore, a reduction in MARD can be seen in the example shown in FIGS. 98A and 98B, where the sensor glucose values are plotted over time. As shown in fig. 98A, calibration at low start-up resulted in the sensor reading at WE1(6880) being significantly too high and the MARD being 11.23% before the non-faradaic current was removed. After removal of the non-faradaic current, the MARD value of WE1 reached 10.53%. Note that for illustrative purposes of FIGS. 97A-98B, the relational expressions are used in the preprocessing To calculate and remove the non-faradaic current, where p is the polynomial coefficient (array) used to fit the double layer capacitance curve.

In real time, the separation of the faraday current and the non-faradaic current can be used to automatically determine when to perform the first calibration. Fig. 99 shows the decay of the double layer capacitance over time. In particular, during a constant time interval Δ T, the double layer capacitance experiences a change from a first valueTo a second value C_T(7010) A change in (c). The non-faradaic current may be calculated, for example, using a first order time difference method as follows

Other methods may also be used to calculate the derivativeSuch as, for example, second order exact Finite Value Method (FVM), Savitzky-Golay, and the like.

Next, the percentage of the total current (i.e., Isig) that consists of the non-faradaic current may simply be in accordance with the ratio i_non-Faradaicand/Isig calculation. Once the ratio reaches a lower threshold, a determination can be made in real time as to whether the sensor is ready for calibration. Thus, in one embodiment, the threshold may be between 5% and 10%.

In another embodiment, the above algorithm can be used to calculate the offset value in real time, i.e., an EIS-based dynamic offset algorithm. In view of

And considering that the sensor current Isig is the total current, containing both Faraday and non-Faraday components

i_total＝i_Faradaic+i_non-Faradaic

The Farada component is calculated as follows

i_Faradaic＝i_total-i_non-Faradaic

Thus, in one embodiment, the current i is not pulled normally_non-FaradaicCan be considered as an additional offset to Isig. In fact, when the double layer capacitance decreases (e.g. on the first day of sensor life), i_non-FaradaicIs negative and decreases with time. Thus, according to this embodiment of the invention, at the very beginning of the sensor life, a larger offset (i.e., the common offset calculated with the current method plus i) will be added to Isig_non-Faradaic) And allowed to decay according to a polynomial curve of order 5. That is, another offset i_non-FaradaicFollowing a polynomial of order 5, its coefficients must be determined. Depending on the severity of the change in double layer capacitance, the algorithm according to this embodiment may be applied for the first few hours of sensor life, e.g. the first 6-12 hours.

The polynomial fit may be calculated in various ways. For example, in an embodiment of the present invention, the coefficient P may be predetermined based on existing data. The dynamic offset discussed above is then applied, but only if the first calibration factor is above the normal range (e.g., about 7). Experiments have shown that this method works best in general when real-time double layer capacitance measurements are less reliable than expected.

In an alternative embodiment, an online (in-line) fitting algorithm is used. Specifically, an inline double-layer capacitive buffer is created at time T. P is then calculated using a polynomial fit over time T based on the buffer. Finally, the non-faradaic current (dynamic offset) at time T + Δ T is calculated using P at time T. Note that this algorithm requires that the double layer capacitance measurements be more frequent (every 30 minutes) than its current level, and that the measurements are reliable (i.e., free of artifacts). For example, EIS measurements may be taken every 5 minutes or every 10 minutes during the first 2-3 hours of sensor life.

In developing real-time self-calibrating sensors, the ultimate goal is to minimize or eliminate the reliance on BG meters altogether. However, this requires an understanding of the relationship between EIS related parameters and Isig, Calibration Factor (CF), and offset, among others. For example, in vivo experiments have shown that Isig is correlated with each of the Cdl and Warburg admittances, and thus each of the latter may be Isig dependent (at least to some extent). Furthermore, it has been found that in terms of factory calibration of the sensor, Isig and Rm (Rmem + Rsol) are the most important parameters (i.e. contributing factors) of the calibration factor, while Warburg admittance, Cdl and Vcntr are the most important parameters of the offset.

In vitro studies, indices extracted from EIS (e.g., Rmem) often show strong correlation with calibration factors. However, in vivo studies, the same correlation may be weak. This is due in part to the patient-specific or (sensor) insertion site-specific characteristics that mask the aspects of the sensor that allow self-calibration or factory calibration using EIS. In this regard, in certain embodiments, redundant sensors may be used to provide a reference point that may be used to estimate patient-specific opposition. This in turn will allow for more robust factory calibration and help identify the source of one or more sensor failure modes internal or external to the sensor.

Generally, EIS is a function of the electric field formed between the sensor electrodes. The electric field may extend beyond the sensor membrane and may detect properties of the (patient) body at the sensor insertion site. Thus, the EIS may be correlated with the properties of the sensor only if the environment in which the sensor is inserted/positioned is consistent throughout all tests (i.e., if the tissue composition is always the same in vivo (or if the buffer is always the same in vitro)). In other words, it can be assumed that a change in the sensor directly results in a change in the EIS, which can be correlated to, for example, a calibration factor.

However, it is well known that the in vivo environment is highly variable, as patient-specific tissue properties depend on the composition of the insertion site. For example, the electrical conductivity of the tissue surrounding the sensor depends on the amount of surrounding fat. It is well known that the conductivity of fat is much lower than that of pure interstitial fluid (ISF), and the ratio of local fat to ISF can vary significantly. The composition of the insertion site depends on the insertion site, insertion depth, patient-specific body composition, and the like. Thus, even if the sensors are identical, the Rmem variation observed from EIS studies is greater, since the reference environment is rarely the same. That is, the conductivity of the insertion site affects the Rmem of the sensor/system. Thus, Rmem may not be uniformly used as a reliable calibration tool.

As previously mentioned, EIS can also be used as a diagnostic tool. Thus, in embodiments of the present invention, an EIS may be used for overall fault analysis. For example, EIS can be used to detect severe sensitivity loss, which in turn helps determine whether and when to block sensor data, determine optimal calibration times, and determine whether and when to terminate sensors. In this regard, it is worth repeating that in continuous glucose monitoring and analysis, two main types of severe loss of sensitivity are generally considered: (1) temporary loss of sensitivity (i.e., Isig drop), which typically occurs early in the life of the sensor and is generally considered a result of external sensor clogging; and (2) permanent sensitivity loss, which typically occurs at the end of the sensor life and never recovers, thus requiring sensor termination.

In vivo and in vitro data show that the altered EIS parameter may be any one or more of Rmem, Rsol, and Cmem during loss of sensitivity and Isig decline. The latter variation is in turn manifested as a parallel shift of the high frequency region of the nyquist curve, and/or an increase in the high frequency half circle. Generally, the more severe the loss of sensitivity, the more pronounced these symptoms are. Graph 100 shows the higher frequency regions of the nyquist curve for data at 2.6 days (7050), 3.5 days (7055), 6 days (7060), and 6.5 days (7065). It can be seen that during the loss of sensitivity (7070), there may be a horizontal movement from left to right (i.e., Rmem + Rsol movement), indicating an increase in membrane resistance. Furthermore, the 6-day curve, especially the 6.5-day curve (7065), clearly shows the appearance of higher frequency semicircles during the sensitivity loss (7075), which is indicative of a change in membrane capacitance. Depending on the environment and the severity of the loss of sensitivity, one or both of the above two manifestations may appear on the nyquist curve.

In particular with respect to the detection of Isig dips, some current methods detect Isig dips using only Isig, for example by monitoring the rate at which Isig may decline, or the degree/absence of change in the increment of Isig over time, as opposed to a permanent loss of sensitivity, indicating that the sensor may not respond to glucose. However, this may not be very reliable, as there are cases where Isig remains within the normal BG range, even if there is an actual dip. In this case, the loss of sensitivity (i.e. Isig dips) is indistinguishable from hypoglycemia. Thus, in one embodiment, EIS can be used to supplement the information obtained from Isig, thereby increasing the specificity and sensitivity of the detection method.

Permanent sensitivity loss is generally possible with V_cntrThe translation is relevant. Here, some current sensor termination methods rely on V alone_cntrTrack data, such as when V_cntrThe sensor may be terminated one day of translation. However, according to embodiments of the present invention, one method of determining when to terminate a sensor due to a loss of sensitivity requires using EIS data to confirm that at V_cntrWhether and when sensitivity loss occurs after translation. Specifically, once V is observed_cntrOn the other hand, the parallel shift in the high frequency region of the nyquist curve can be used to determine whether a permanent loss of sensitivity actually occurs. In this regard, in some cases, V_cntrIt may move along the track within 5 days after the beginning of the sensor life, but the EIS data shows little movement of the nyquist curve. In this case, the sensor will normally end in 5-6 days. However, since EIS data indicate that, in fact, there is no permanent loss of sensitivity, the sensor is not terminated, thereby saving (i.e., using) the remaining useful life of the sensor.

As previously described, the detection of the loss of sensitivity may be based on a change in one or more EIS parameters. Thus, for example, a change in membrane resistance (Rm ═ Rmem + Rsol) may be manifested in the medium frequency (about 1kHz) real impedance region. For the membrane capacitance (Cmem), the variation may be manifested in the virtual impedance at higher frequencies (about 8kHz) due to the increase in the semi-circle. Double layer capacitance (Cdl) is positive with average IsigThus, it can be approximated as a low frequency Nyquist slope L_nyquistLength of (d). Because of V_cntrIn relation to oxygen levels, normal sensor behavior usually requires V_cntrDecreases with decreasing Isig. Thus, V_cntrAn increase in (i.e., more negative), combined with a decrease in Isig, may also indicate a loss of sensitivity. In addition, the average Isig level, rate of change, or variability of low or physiologically unlikely signals can be monitored.

However, the EIS parameters must first be determined. As described previously in connection with calibration factor adjustment and related disclosure, the most robust method of estimating EIS parameters is to perform model fitting, where the parameters in the model equations are varied until the error between the measured EIS and the model output is minimized. There are many ways to make this estimation. However, for real-time applications, model fitting may not be optimal due to computational load, variability in estimation times, and poor convergence. Generally, feasibility will depend on hardware.

When the above-described complete model fitting is not possible, in one embodiment, one method of real-time application is through the use of heuristics. The goal is to approximate the true parameter values (or corresponding metrics proportional to the trend displayed by each parameter) with a simple heuristic applied to the measured EIS. In this regard, the following is an embodiment of estimating the variation of each parameter.

Double layer capacitor (Cdl)

In general, a rough estimate of Cdl can be obtained from any statistical data that measures low frequency Nyquist slope length (e.g., frequencies below about 128 Hz)_nyquist(Cartesian distance between EIS at 128Hz and 0.1Hz in the Nyquist plot). Other frequency ranges may also be used. In another embodiment, Cdl can be estimated by using the magnitude of the low frequency impedance (e.g., at 0.1 Hz).

Membrane resistance (Rmem) and solution resistance (Rsol)

As described above, on the Nyquist curve, Rmem + Rsol corresponds to one of the low and high frequency semicirclesAnd an inflection point therebetween. Thus, in one embodiment, Rmem + Rsol may be estimated by detecting the directional change in the nyquist slope (e.g., by using derivatives and/or differences) to locate the inflection point. Alternatively, the relative change in Rmem + Rsol can be estimated by measuring the shift in the nyquist slope. To this end, a reference point in the imaginary axis may be selected (see fig. 83), and interpolation may be used to determine a corresponding point on the real axis. This interpolation can be used to track the change in Rmem + Rsol over time. For a given sensor configuration, the selected reference voltage should not be unduly affected by large variations in the low frequency part of the Nyquist slope (e.g., because V_cntrRail) of the value range. Typical values may be between 1k Ω and 3k Ω. In another embodiment, it is possible to use the real components (e.g., 1kHz, 8kHz) of a single high frequency EIS. In some sensor configurations, this may simulate Rmem in most cases, but it should be noted that a single frequency may not accurately represent Rmem in all cases.

Film capacitor (Cmem)

The increase in Cmem appears as a more pronounced (or pronounced) higher frequency half-circle. Thus, changes in Cmem can be detected by estimating the presence of this semicircle. Thus, in one embodiment, Cmem can be estimated by tracking the high frequency imaginary component of the impedance. In this regard, a more negative value corresponds to an increase in the semi-circle.

Alternatively, Cmem can be estimated by tracking the highest point in a semicircle within the frequency range (e.g., 1kHz-8 kHz). The frequency range may also be determined by identifying the frequency at which the inflection point occurs and obtaining the maximum imaginary impedance for all frequencies above the identified frequency. In this regard, the more negative the value, the more semi-circles occur.

In a third embodiment, Cmem can be estimated by measuring the cartesian distance between two higher frequency points in the nyquist curve (such as, for example, 8kHz and 1 kHz). This is the high frequency slope (S) previously defined in this application_nyquist). Here, a larger absolute value corresponds to an increasing semicircle, and a negative slope (y-axis is negative imaginary impedance, x-axis is positive real impedance) corresponds to no semicircle. Note that in the above method, there may be one of the semicirclesThese detected changes may also be attributable to the presence of changes in Rmem. However, the overlap was considered acceptable as a change in either of the two indicated a loss of sensitivity.

EIS independent metrics

In this context, it is noted that sensitivity loss is essentially detected according to several non-EIS criteria before EIS measurements are available. These metrics are often not reliable enough in themselves to achieve perfect sensitivity and specificity in detection. However, these metrics can be combined with EIS-related metrics to provide supporting evidence for the presence of sensitivity loss. Some of these metrics include: (1) the amount of time (in nA) that Isig is below a certain threshold, i.e. a "low Isig" period; (2) a first or second derivative of Isig resulting in a "low Isig" state, indicating whether a change in Isig is physiologically possible or caused by a loss of sensitivity; and (3) variability/variance of Isig during the "low Isig" period, which may indicate whether the sensor is reactive or inactive to glucose.

Sensitivity loss detection algorithm

Embodiments of the invention herein also relate to algorithms for detecting loss of sensitivity. The algorithm typically accesses a vector of parameters estimated from EIS measurements (e.g., as described above) and non-EIS-related metrics. Thus, for example, the vector may contain Rmen and/or horizontal axis movement (of the nyquist curve), changes in Cmen, and changes in Cdl. Similarly, the vector may contain data regarding the period of time Isig is in a "low" state, the variability of Isig, the rate of change of Isig. The vector of this parameter can be tracked over time, where the purpose of the algorithm is to collect reliable evidence of sensitivity loss. In this context, "robust evidence" may be defined, for example, by a voting system, combined weighted metrics, clustering, and/or machine learning.

In particular, the voting system may need to monitor one or more of the EIS parameters. For example, in one embodiment, this involves determining when more than a predetermined or calculated number of elements in the parameter vector exceed an absolute threshold. In an alternative embodiment, the threshold may be a relative (%) threshold. Similarly, vector elements may be monitored to determine when a particular combination of parameters in a vector exceeds an absolute or relative threshold. In another embodiment, when any element in a subset of elements in a vector exceeds an absolute or relative threshold, a check of the remainder of the parameters may be triggered to determine if sufficient evidence of sensitivity loss can be obtained. This is useful when at least one parameter of the subset of parameters is a necessary (but possibly insufficient) condition to reliably detect a loss of sensitivity.

The combined weighting metric requires weighting the elements in the vector according to, for example, how far it crosses a predetermined threshold. A loss of sensitivity may then be detected (i.e., determined to occur) when the aggregate weighting metric exceeds an absolute or relative threshold.

Machine learning can be used as a more complex "black box" classifier. For example, an Artificial Neural Network (ANN), a Support Vector Machine (SVM), or a genetic algorithm may be trained using parameter vectors extracted from real in vivo experiments to detect sensitivity loss. The trained network can then be applied in real time in a very time efficient manner.

Fig. 101A and 101B show two illustrative examples of flow charts for sensitivity detection using synthetic logic. As shown, in both approaches, one or more metrics 1-N may be monitored. In the method of FIG. 101A, each metric is tracked to determine if and when it exceeds a threshold, and is described above. The outputs from the multiple threshold determination steps are then aggregated by synthesis logic, and a decision is made regarding the loss of sensitivity based on the synthesized outputs. In fig. 101B, the values of the monitored metrics 1-N are first processed by combinatorial logic, and then the total output of the latter is compared to one or more thresholds to determine whether a loss of sensitivity has occurred.

Additional embodiments also relate to using EIS in intelligent diagnostic algorithms. Thus, in one embodiment, the EIS data may be used to determine whether a sensor is new or whether it is being reused (in addition to previously proposed methods related to patient reuse of sensors). In relation to the latter, it is important to know whether the sensor is new or whether it is reused, as this information helps to determine which type of initialization sequence (if any) should be used. In addition, the information allows for preventing off-tag use of the sensor and preventing sensor damage caused by multiple re-initializations (i.e., the sensor "thinks" it is a new sensor whenever it is disconnected and then reconnected, and thus attempts to re-initialize upon reconnection). The information also facilitates post-processing of the collected sensor data.

With respect to reuse and/or reconnection of sensors, it has been found that prior to initialization, the low frequency nyquist slope for a new sensor is different from (i.e., lower than) the low frequency nyquist slope for a sensor that has been disconnected and then reconnected again. In particular, in vitro experiments show that the nyquist slope of a re-used sensor is higher compared to a newly inserted sensor. Thus, the nyquist slope may be used as a marker to distinguish between new sensors and used (or reused) sensors. In one embodiment, based on the nyquist slope, a threshold may be used to determine whether a particular sensor is being used again. In one embodiment, the threshold may be 3 nyquist slope. Fig. 102 shows a low frequency nyquist plot for a reference slope of 3(8030), and plots for a new sensor (pre-initialization) 8010, a new sensor (post-initialization) 8015, a reconnected sensor (pre-initialization) 8020, and a reconnected sensor (post-initialization) 8020. As described above, the slope of the new sensor (before initialization) 8010 is below a reference value or threshold (8030), while the slope of the reconnected sensor (before initialization) 8020 is above the threshold (8030).

In another embodiment, the EIS data can be used to determine the type of sensor used. Here, it has been found that if the sensor designs are significantly different, on average, the corresponding EIS outputs should also be significantly different. Different sensors are configured with different model parameters. Thus, the identification of these parameters may be used at any point during the life of the sensor to determine the type of sensor currently inserted. For example, parameters may be estimated based on the methods described above in conjunction with overall fault/sensitivity loss analysis. The identification may be based on common methods of separating values, e.g. setting thresholds for certain parameters (single or multiple), machine learning (ANN, SVM) or a combination of both methods.

This information can be used, for example, to change algorithm parameters and initialization sequences. Thus, at the beginning of the sensor life, it can be used to let the single processing unit (GST, GSR) set the optimal parameters of the calibration algorithm. Offline (non-real-time), identification of sensor types may be used to assist in the analysis/evaluation of in-situ sensor performance.

It has also been found that the length of the low frequency nyquist slope can be used to distinguish between different sensor types. Fig. 103A-103C show nyquist curves for three different sensors (i.e., different sensor configurations) identified as Enlite (8050), Enlite 2 (i.e., "Enlite enhanced") (8060), and Enlite 3(8070), all manufactured by Medtronic Minimed (north ridge, ca). It can be seen that for each stage, including before, after and after initialization (see fig. 103A-103C, respectively), the low frequency nyquist slope length of the Enlite sensor (8050) is shortest, followed by Enlite 2(8060) and Enlite 3(8070), which are longest. The latter is also shown in the graph 104, where the nyquist (slope) length (calculated as the cartesian distance between EISs at 0.105Hz and 1 Hz) is plotted against time.

As previously described, the initialization sequence may be based on the detected sensor type (based on EIS or otherwise) and/or whether a new or old sensor is detected as inserted (based on EIS). however, in addition, EIS based diagnostics may also be used to determine a minimum hydration state prior to initialization (e.g., by tracking Valburg impedance) or to determine when to terminate initialization (e.g., by tracking reaction related parameters such as Rp, Cdl, α, etc.) in order to appropriately minimize sensor initialization time.

In this regard, the EIS may alternatively or additionally be measured during high pulses and compared to the EIS for the best initialization state to determine when the sensor is sufficiently initialized.

As mentioned before, sensor calibration, especially real-time sensor calibration, is the core of a robust Continuous Glucose Monitoring (CGM) system. In this regard, the calibration algorithm is typically designed such that once BG is received by the fingertip, the new BG value is used to generate an error message or update a calibration factor, which in turn is used to calculate the sensor glucose. However, in some previous algorithms, there may be a 10-20 minute delay between the time of the fingertip entry and the time of notifying the user that the finger stick is accepted or that a new fingertip is needed for calibration. This is cumbersome because the user does not know if he/she will need his/her BG meter again within a few minutes.

Furthermore, in some cases, the presence of older BG values in the calibration buffer results in a perceived system delay, either because the most recent BG value carries less than 100% weight, or the calculated SG is inaccurate (because the older BG values no longer represent the current state of the system). In addition, the wrong BG value is sometimes entered but is not captured by the system, which may result in large errors until the next calibration.

In view of the foregoing, embodiments of the present invention seek to address potential shortcomings in existing approaches, particularly with respect to sensor performance for closed loop systems, for example, to make the system more predictable, calibration errors may only be notified when a fingertip (BG value) is received by the transmitter (i.e., input), rather than after, for example, 10-15 minutes.

On the other hand, when the expected error is low, the system may use a more stringent calibration error limit, such as 40% or 40mg/d L. this reduces the likelihood that the wrong BG value may be used for calibration, while also allowing the status of the calibration attempt to be issued immediately (i.e., accepted for calibration, or calibration error).

With respect to the above-described problems related to BG values and BG buffers, embodiments of the present invention aim to reduce latency and latency perception by matching newer BG values with higher weights assigned in previous algorithms and by ensuring that early calibration updates occur more frequently. Furthermore, in the presence of a confirmed sensitivity change (as confirmed, for example, by the smart calibration logic mentioned earlier and discussed below, and by the most recent calibration BG/Isig ratio), the calibration buffer may be partially cleared. Finally, while existing algorithms may have employed expected Calibration Factor (CF) weights (constants), embodiments of the present invention provide variable CF values based on sensor usage time.

In short, a variable calibration error threshold may be provided based on an expectation of error during a calibration attempt and issuing one or more calibration error messages without waiting for additional sensor data, a shorter calibration delay (e.g., 5-10 minutes), updating the expected calibration factor values according to sensor lifetime, and clearing the correction buffer appropriately.

Thus, in one aspect, if a high CF threshold is triggered after the first calibration, the system requires that the next calibration be performed within 3 hours. However, if the high CF threshold is triggered after the second or subsequent calibration, the system requires that the next calibration be performed within 6 hours. The above procedure can be carried out for 12 hours starting from the sensor connection.

In another aspect, the expected calibration factor used to calculate the calibration factor during calibration increases over time, thereby reducing the likelihood of readings that are too low. By way of background, existing approaches may use a fixed expected calibration factor throughout the life of the sensor, regardless of possible shifts in sensor sensitivity. In this approach, the expected calibration factors may be weighted in calculating the final calibration factor and used to reduce noise.

However, in an embodiment of the present invention, the expected CF is calculated as a function of time, expressed in terms of the age of the sensor. In particular, the present invention relates to a method for producing,

where sensor usage time is expressed in days. In a further embodiment, the expected calibration factor may be calculated as a function of the existing CF and impedance, such that any change in sensitivity may be reflected in the expected CF. Further, in aspects of the invention, the expected CF may be calculated on every Isig packet, rather than just at the BG entry, in order to gradually adjust the calibration factor between calibrations.

In conjunction with calibration buffer and calibration error calculation, certain embodiments provide for modification of calibration buffer weights and/or purging of calibration buffers. Specifically, when the impedance measurements (e.g., by EIS) indicate that the calibration factor may have changed, and a calibration attempt indicates that a change may have occurred, the change in Calibration Ratio (CR) is checked by comparing the CR of the current BG to the latest CR in the calibration buffer. Here, such a change can be verified by, for example, a 1kHz impedance value as detailed above in connection with the relevant EIS procedure. Further, weights may be added in the calibration buffer calculation based on the reliability index, the direction in which the calibration factor is expected to change, and/or the rate of change of the calibration. In the latter case, for example, if the calibration is at a high rate of change, a lower weight may be assigned, or the CF may be updated only temporarily.

In embodiments of the invention, the selection of the filtered Isig (fisig) value for the calibration buffer may be initiated on the second Isig packet after the BG entry. Specifically, the latest of the past three (3) values of fIsig that does not result in calibration errors may be selected. Then, once calibration is accepted, the calibration process will continue without issuing calibration errors. Such calibration errors may be caused by, for example, invalid Isig values, calibration ratio range checks, percentage error checks, and the like.

In other embodiments, the value of fIsig may be interpolated to derive a resolution of one minute. Alternatively, the fIsig value may be selected from the most recent values based on the rate of change of the values (and taking into account the delay). In yet another alternative embodiment, the fIsig value may be selected based on the CR value closest to the predicted CR value. Conversely, the predicted CR value is closest to the current value of the calibration factor unless the calibration factor or EIS data indicates that the CF should change.

As previously described in connection with fig. 24 and 34, for example, a value of 1kHz real impedance provides information about one or more potential blockages that may exist on the sensor membrane surface that may temporarily prevent glucose from entering the sensor, resulting in a signal dip. More broadly, a 1kHz real impedance measurement can be used to detect a sensor event that is typically sudden and can indicate that the sensor is no longer fully inserted. In this regard, fig. 105 illustrates a flow diagram of a method of blanking sensor data or terminating a sensor, according to one embodiment.

This method begins at block 9005, where the real impedance value of 1kHz is filtered using, for example, a moving average filter, and based thereon, it is determined whether the EIS-derived value is stable (9010). If it is determined that the EIS derived value is not stable, the method proceeds to block 9015, where a further determination is made based on the magnitude of the 1kHz impedance. Specifically, if both the 1kHz real impedance filtered value and the unfiltered value are less than 7,000 Ω, the EIS is set to stable (9020). On the other hand, if neither the filtered nor unfiltered value of the 1kHz real impedance is less than 7,000 Ω, the EIS is set to unstable (9025). It should be noted that the above-mentioned 7,000 Ω threshold may prevent data blanking or sensor termination of an unstable sensor.

When the EIS is stable, the algorithm proceeds to block 9030. Here, if the 1kHz real impedance is less than 12,000 Ω (9030), and also less than 10,000 Ω (9040), the algorithm determines that the sensor is within the normal operating range, and thus allows the sensor data to continue to be displayed (9045). On the other hand, if the real impedance value of 1kHz is greater than 10,000 Ω (i.e., when the real impedance value of 1kHz is between 10k Ω and 12k Ω), the logic determines whether the real impedance value of 1kHz has been high (i.e., greater than 10k Ω) for the last 3 hours (9050). If it is determined that the real impedance value of 1kHz has been high for the last 3 hours, the sensor is terminated at 9060 because it is assumed that the sensor has been unplugged, rendering the sensor data invalid. Otherwise, the sensor will not terminate because the sensor signal may simply drift, which may be a recoverable phenomenon, as previously described. However, the sensor data is blanked (9055) giving the sensor an opportunity to recover.

Note that in further embodiments, in addition to the thresholds described above, logic may also account for sudden increases in impedance by, for example, comparing the impedance derivative to historical derivatives when determining whether data should be blanked or whether the sensor should be terminated. Furthermore, the algorithm may incorporate noise-based blanking or termination depending on the duration of the high noise-low sensor signal combination. In this regard, existing methods involve terminating the sensor after three (3) consecutive 2 hour windows of high noise and low sensor signal. However, to prevent unreliable data from being displayed to the user, embodiments herein employ noise-based blanking, where the algorithm stops calculating the SG value after 2 consecutive 2 hour windows involving high noise and low signal (i.e., at the beginning of the third consecutive window). In a further aspect, the algorithm may allow further calculation and display of the calculated SG value after one hour of blanking (rather than two hours after which the sensor signal appears to have recovered). This is an improvement over methods that can blank data that is otherwise reliable for a long time.

While a real impedance of 1kHz may be used to detect sudden sensor failures, imaginary impedance measurements at higher frequencies (e.g., 8kHz) may be used to detect more gradual changes in which the sensor sensitivity has significantly deviated from its typical sensitivity. In this regard, it has been found that a large shift in the virtual impedance of 8kHz generally indicates that the glucose sensitivity of the sensor has moved significantly, or is no longer stable.

FIG. 106 shows a flow diagram of a sensor termination method according to an embodiment of the invention. As shown in fig. 106, the algorithm takes a 1.5 day (self-sensor start-up) baseline, as doing so provides more robust logic and ensures that the logic is concerned with long-term sensitivity variations. Thus, if the sensor is not working for at least 1.5 days (9002), no action is taken and the algorithm "waits" (9012) (i.e., it periodically loops back to step 9002). Once the condition in block 9002 is satisfied, a determination is made as to whether the reference imaginary impedance value is set (9022). If the reference value is not preset, the algorithm will set by assigning a minimum 8kHz virtual impedance value since sensor initialization as the reference value (9032), which is clipped to be in the range of-1,000 Ω to 800 Ω. In the case of setting the reference value, the variation value is calculated as an absolute value of a difference between the reference value and the current value of the 8kHz imaginary impedance (9052). In box 9062, the algorithm determines whether the change value of two consecutive measurements is greater than 1,200 Ω and the calibration ratio is greater than 14. If at least one of the subsequent queries is answered in the negative, the sensor is allowed to continue operating and display the SG value (9072). However, if the change value of two consecutive measurements is greater than 1,200 Ω and the calibration ratio is greater than 14, the sensor terminates at block 9082.

Embodiments of the present invention are also directed to reliability assessment of sensor glucose values, as well as estimation of sensor data error direction, in order to provide an indicator to the user and to the automatic insulin delivery system (including the system in a closed loop system) of the reliability of the system when displaying the SG to the user. Depending on the reliability of the sensor data, such an automated system can then assign corresponding weights to the SG and determine how aggressively therapy should be provided to the user. In addition, the direction of the error may also be used to inform the user and/or the insulin delivery system that the SG is a "false low" or "false high" value. The foregoing may be accomplished by, for example, detecting a dip in sensor data on the first day (EIS dip detection), detecting sensor hysteresis, and low frequency (e.g., 10Hz) impedance changes.

In particular, according to one embodiment, it has been found that a Calibration Factor (CF) above about 9mg/d L/nA may indicate low sensor reliability and therefore a predictive value for higher errors.

FIG. 107 shows a flow chart of a method of signal dip detection according to an embodiment of the invention, where an increase in the unfiltered real 1kHz impedance can be used in combination with a low Isig value to identify the onset of a dip. As shown, at block 9102, the logic determines whether the sensor data is currently blanked due to a signal drop. If the data has not been blanked, the logic determines if less than 4 hours have elapsed since the sensor was started (9104). If more than 4 hours have elapsed since the sensor was started, the logic determines whether more than 12 hours have elapsed since the sensor was started (9106), in which case there will be no dip detection or blanking of data (9108). Thus, in this regard, the method is directed to identifying a transient drop during the first 12 hours of sensor data.

Returning to block 9106, if less than 12 hours have elapsed since the sensor started, then an inquiry is made as to the most recent EIS, Isig, and SG values. specifically, in block 9110, if the two most recent real impedance values (at 1kHz) are increasing (Isig <18nA and SG <80mg/d L), then the algorithm determines that the beginning of a dip has been detected and notifies the system to stop displaying SG values (9112). on the other hand, if all of the aforementioned conditions are not met, then no dip detection or data blanking is performed (9108).

When it is determined at block 9104 that less than 4 hours have elapsed since sensor activation, then a sensor dip event may still be encountered. Specifically, if the two most recent EIS (i.e., 1kHz impedance) values are increasing, and Isig <25nA, the algorithm determines that the beginning of a dip has been detected and notifies the system to stop displaying SG values (9114, 9116). However, if the two most recent 1kHz impedance values have not increased, or Isig is not less than 25nA, then no dip detection or data blanking is performed (9108), as before.

Returning to block 9102, if it is determined that the data is currently being blanked due to a dip, it may still be possible to display the data. That is, if Isig is greater than about 1.2 times Isig at the beginning of the slump event (9118), then it is determined that Isig has recovered (i.e., the slump event is over and the data display will recover (9122)). On the other hand, if Isig is not greater than about 1.2 times Isig at the beginning of the slump event (9118), it is determined that Isig has not recovered (i.e., the slump event has not ended, and the system will continue to blank sensor data (9120)).

In general, the direction of error in the SG (low reading or high reading) can be determined by considering one or more factors related to low reading and/or high reading. Thus, it has been found that too low a reading in the sensor may occur when: (1) v_cntrIs an extreme value (e.g., V)_cntr<-1.0V); (2) CF is higher (e.g., CF)>9) (ii) a (3) Lower frequency impedance (e.g., at 10Hz) higher (e.g., real 10Hz impedance)>10.2k Ω); (4) FDC in low CF mode; (5) sensor lag indicates too low a reading; (6) lower frequency impedance (e.g., at 10Hz) increases (e.g., 10Hz impedance increases over 700 Ω); and/or (7) the EIS detects a dip. On the other hand, when the following occurs, a false high of reading occurs: (1) lower frequency impedance (e.g., 10Hz) reduction (e.g., lower frequency impedance)<-200 Ω); (2) sensor lag indicates a false high reading; and/or (3) FDC when CF is in limit mode.

For example, a false high reading near the hypoglycemic range (i.e., <70mg/d L) may result in excessive insulin administration to the patient, in this regard, several indicators of error direction have been determined, which may be used as test criteria, including (1) a low sensitivity indicator, (2) sensor hysteresis, (3) FDC mode, and (4) loss/gain of sensitivity after calibration.

Two such low sensitivity indicators are high (low frequency) real impedance (e.g.,>10k Ω) and high V_cntr(e.g., V)_cntr<-1.0V), both generally indicating a loss of sensitivity. FIG. 108A shows V_cntr9130 is an example that gradually increases (i.e., becomes more negative) over time. At about 115 hours, V is shown as line 9135_cntrcrosses-1.0V, as shown by line 9137, and continues to increase (i.e., Vcntr<-1.0V) to about-1.2V. As shown, Isig trend 9132 generally follows the Vcntr trend by about 115 hours ago. However, once V_cntrBeyond the threshold (i.e., to the right of line 9135), Isig leaves V_cntrAnd continues to descend. Meanwhile, as shown in fig. 108B, glucose 9134 also has a generally decreasing trend, with calibration error 9136 indicated at about 130 hours and about 165 hours.

As previously mentioned, (EIS) sensor dips also indicate a temporary loss of sensitivity. Also, a high calibration factor indicates that the sensor is attempting to compensate for the reduced sensitivity. In one example shown in fig. 109A and 109B, the calibration factor 9140 steadily increases over time. At about 120 hours (9145), the calibration factor 9140 crosses the threshold 9 (9147). As shown in fig. 109B, once the calibration factor crosses the threshold, the glucose value 9142 shows a more frequent deviation from the BG value and several errors 9144 occur between about 135 hours and about 170 hours.

As previously mentioned, sensor lag is another indicator of the direction of error. Thus, in one embodiment of the invention, the error caused by sensor lag is compensated by approximating the glucose value. Specifically, in one embodiment, the error from sensor hysteresis may be approximated by defining:

where sg (t) is the sensor glucose function and "h" is the sensor lag. The error can then be calculated as follows

Or

When the Calibration Factor (CF) is not within the expected range, a First Day Calibration (FDC) is performed. CF is set to the value indicated by the calibration and then ramped up or down to the desired range as shown, for example, in fig. 110A and 110B. During this time, high but often predictable errors typically occur, resulting in potential false high or low readings. As can be seen from fig. 110A and 110B, when CF rises or falls, it changes with a substantially constant slope and then stabilizes (in this case, at 4.5 or 5.5).

Finally, the calibrated sensitivity change (i.e., the loss/gain of the calibrated sensitivity) is also an indicator of the error/direction of error. Under normal circumstances, the calibration factor is typically held constant, except for the first day calibration discussed above, until a new calibration is performed. Thus, a calibrated sensitivity shift will result in a low reading and a false high reading, which will be reflected by a low frequency (e.g., 10Hz) real impedance value.

In particular, it has been found that a drop in low frequency real impedance results in a reading that is falsely high, and the direction of error is indicated by the real impedance curve. Conversely, a lower frequency real impedance increase will result in a reading that is too low and the direction of error is also represented by the real impedance curve. However, current directionality tests may not easily decipher the peaks and valleys of the glucose curve. Thus, in one embodiment, the sharpness of such peaks and valleys may be reduced by filtering (such as, for example, by deconvolving with low pass filtering).

As previously described in connection with fig. 81, for example, sensitivity changes and/or losses may be used to inform appropriate sensor calibration. In this regard, in another aspect of the present invention, changes in sensor sensitivity may be predicted based on previous calibration factors or based on impedance to enable "smart calibration" which helps to address issues of continually generating and/or displaying inaccurate glucose data when, for example, the sensor sensitivity has changed.

It is well known that in some existing Continuous Glucose Monitoring Systems (CGMS), the fingertip needs to be calibrated every twelve hours. Calibration allows the CGMS to update the functionality for converting the measured sensor current to a displayed glucose concentration value. In such systems, a calibration interval of 12 hours is chosen as a balance between reducing the user burden (performing too many fingertips) and using an interval sufficient to adjust for sensor sensitivity variations before inaccuracies can cause too great a problem. However, while this interval may generally be appropriate, if the sensor sensitivity has changed, if high accuracy (supporting closed loop insulin delivery) is desired, the 12 hours may be too long to wait.

Accordingly, embodiments of the present invention address the foregoing problems by using previous calibration factors (see discussion of FDC below) or impedances (see discussion of "intelligent calibration" based on EIS below) to predict whether sensitivity has changed. Various embodiments also use time constraints to maintain user predictability and include steps (in related methods) to ensure that detection is robust to variations between sensors.

Fig. 111 shows a flow chart of an embodiment according to the First Day Calibration (FDC). Beginning in block 9150, if FDC is not turned on after successful calibration, there is simply no intelligent calibration request (9151). However, if the FDC is on, it is determined at block 9153 whether this is the first calibration, and if not, a smart calibration request is made and the timer is set to 6 hours (i.e., additional calibrations are requested within 6 hours (9155)). On the other hand, if this is the first calibration, block 9157 determines whether the calibration ratio is less than 4 or greater than 7. If the condition in block 9157 is not met, the logic proceeds to block 9155 where a smart calibration request is made, as described above, and the timer is set to 6 hours. However, if the criteria in block 9157 are not met, then a smart calibration request is made and the timer is set to 3 hours (i.e., additional calibration is requested within 3 hours (9159)). Therefore, in order to improve the accuracy of the sensor requiring calibration adjustments, additional (intelligent) calibration is required, which in turn limits the time when the adjustments are incorrect.

Compared to the FDC mode, the EIS-based smart calibration mode may provide additional calibration when impedance changes occur. Thus, in one embodiment shown in fig. 112, an allowable range associated with the impedance value (as defined below) is set within one hour after calibration, and after calibration, if the impedance is out of range, additional calibration is requested. Thus, if not within one hour after calibration, it is determined whether the filtered 1kHz imaginary impedance value is out of range (9160, 9162). If the impedance value is not out of range, no change is made (9164). However, if the filtered 1kHz imaginary impedance value is out of range, the calibration timer is updated, requiring calibration to be performed 6 hours from the last calibration (9168). Note that while higher frequency imaginary impedance tends to better identify changes in glucose sensitivity, towards the higher end of the spectrum, the measurements are typically more noisy and therefore may require filtering.

Returning to block 9160, if it is determined that less than one hour has passed since self-calibration, the range of impedance values may be updated (9166). Specifically, in one embodiment, the impedance range calculation is performed on the last EIS measurement 1 hour after calibration. In a preferred embodiment, this range is defined as

Range 3 median 3 × (| x)_i-x_j|)

Where j is the current measurement and i is the last 2 hours. In addition, the range may be limited to a value between 50 Ω and 100 Ω. Note that the range defined above allows a 3-fold median value. The latter has been found to be more robust than the 2 standard deviation approach used in some previous algorithms, which allows noise and outliers to cause inconsistencies.

Embodiments of the present invention for Continuous Glucose Monitoring (CGM) are also directed to sensor calibration using a kalman filter, regardless of the actual design of the subject sensor(s). As previously mentioned, sensor calibration typically involves determining a Calibration Factor (CF) based on a reference Blood Glucose (BG), a correlated Isig, and an offset value. BG and Isig in turn may contain noise, and the offset may be sensor (design) specific, so that the calibration factor is also sensor specific. However, it has been found that by using an unscented kalman filter, a basic calibration method can be developed that is not specific to the sensor, as long as the sensor is linear. Thus, individual sensors can be calibrated using a single calibration method (and associated system) without the need to recalculate calibration factors and/or offset values for each particular sensor, and without the need to design a (separate) filtering mechanism to compensate for noise. In this way, both the calibration factor and the offset can be changed over time without changing the code base on which the calibration algorithm is based.

In this respect, it is well known that each time a new glucose sensor is developed, the method/algorithm for calibration needs to be re-evaluated and re-generated. As part of this re-evaluation, assumptions and constants must be redefined for each new sensor design. Furthermore, the mathematics in the calibration methodology are often heuristically (and manually) reviewed. However, as described in detail below, the use of an unscented kalman filter provides a calibration method in which the only assumption is that the sensor is linear (although other sensors, including non-linear sensors, may also be accommodated by modified versions of the sensor). This in turn provides a significant advantage, since the method of the invention can be applied to any new linear sensor, thereby significantly reducing the development time of the new sensor.

In prior methods, when the relationship between Isig and BG is generally assumed to be linear, the calibration factor (single working electrode WE) can be calculated as follows

CF BG/(Isig + offset)

Given that there is typically noise in both the reference BG and Isig, a degree of filtering may be applied so that several BGs may be averaged over time, and/or a complex function of BG levels is used, providing a more robust calibration. The sensor glucose value (SG) can be calculated as follows

SG ═ CF × (Isig + offset)

More specifically, as already noted, the periodic sensor measurements (SG) may be represented by the following relationship

SG＝CF(Isig+offset)+_s

Where "Isig" represents the physical output of the sensor (current in nA) and "CF" represents the calibration factor that relates glucose level to the measured output. The calibration factor is not accurate and will change with time; thus, it is estimated and compensated in real time. The sensor bias is represented by "offset", which is a time-varying variable, and the random sensor error is represented by_sAnd (4) showing. The latter is completely random and therefore cannot be estimated.

Blood Glucose (BG) levels are measured with the fingertip, for example, by a meter. The difference between the general BG measurement and SG is the random error: (_B) That is to say that,

SG＝BG+_B

there is also a first order lag between the sensor glucose measurement (SG) and the physical output (Isig). Therefore, the temperature of the molten metal is controlled,

where τ is a time constant that defines the dynamic relationship between SG and Isig. In the above relationship, τ is not yet precisely known and may vary depending on the patient, sensor location, time, and/or other variables. Assuming that the time constant is constant (e.g., 1/6 h-10 minutes), a dynamic variable may be established, which may be considered an uncertain parameter, and then estimated and compensated for using a kalman filter.

In general, a kalman filter is an optimal estimator that uses a series of measurements containing noise and produces a statistically optimal estimate of an unknown variable. The kalman filter is recursive so that new measurements can be processed on arrival to update the estimate. While kalman filters typically require linearization or discretization of the underlying equations describing the state of the system being evaluated, unscented kalman filters directly address any such nonlinearities in the measurement equations.

Nonlinear dynamic process model

The three variables that can be used for the above estimation are Sensor Glucose (SG), Calibration Factor (CF) and offset. The measurement is Blood Glucose (BG), which, as mentioned above, is related to the sensor current (Isig). Based on the above variables, the following states may be defined:

x₁＝SG

x₂＝CF

x₃is as offset

U＝Isig

Using the previous equations relating BG, SG, CF and first order lag, the following results can be obtained:

where α -0.995, τ 1/6 h-10 minutes, and u (T) -isig, as has been previously indicated in this specification and description, the sensor response at the beginning of the sensor life (e.g., the first day) is typically different from the sensor response for the remainder of the sensor life_dIs defined as the first day.

Using the above state variable definitions, the SG measurement (i.e., the estimate of BG using the fingertip) becomes:

z(t)＝x₂(t)(u(t)+x₃(t))+v₁

where z is BG, u (t) is a BG measurementThe first subsequent measurement of Isig. Sensor glucose is an estimate of blood glucose, i.e.Because BG measurements are provided in a sampled form, no discretization is required to implement the discrete-time measurements in the above equation.

To apply the unscented Kalman Filter to continuous glucose monitoring, the aboveAnd z (t) the equation must be expressed in a non-linear format, namely:

where u is the input, w is the state noise, z is the measurement vector, and v is the measurement noise. Note that although it is assumed that v and w are both uncorrelated zero-mean white gaussian noise sequences, they can be modified based on statistical information obtained from the data. Unlike kalman filters and extended kalman filters, unscented kalman filters do not require linearization or discretization of equations. Instead, it uses a truly non-linear model and approximates the distribution of state random variables. Thus, while the goal remains to calculate the calibration factor, the complexity of the latter calculation is included in the underlying model and method described herein. In other words, in the context of glucose sensor calibration and operation, calibration is performed by the unscented kalman filter framework. In this regard, as described above, the (unscented) kalman filter involves robustness to noise in calibration by assuming that there is a noise distribution in BG (i.e., measurement noise v) and Isig (i.e., state noise w) and implicitly compensating for such noise in the algorithm. Thus, the unscented kalman filter enables real-time calibration of the estimated calibration factor and offset, taking into account the variation over time.

Initial conditions and covariance matrix

For the above framework, the state vector initialization and covariance are given as follows:

as shown below, the diagonal element Q of the process noise covariance matrix is the variance that represents the uncertainty in the knowledge of each state accumulated between measurements.

These values should be based on observations of unpredictable variations of these processes when scaling during the measurement time t. The measurement error variance R is equal to the square of 3% of the BG measurements. Therefore, the temperature of the molten metal is controlled,

R＝0.03×z(t)

with the above structure and method, BG measurements are run through an unscented Kalman filter and calibration factors are estimated. The calibration factor is used in turn to convert Isig to SG, as previously described.

FIG. 113 shows a block diagram of a prior art calibration process for a single working electrode. Using Isig (weisig) from the working electrode, a pre-processing step 9210 is first performed, which may for example include filtering, averaging and/or weighting several Isig values of a single WE to generate a single optimized Isig value. The latter then calibrates 9220 using the offset and calibration BG 9230 (e.g., a fingertip meter measurement) to calculate a calibration factor CF, which in turn is used to calculate the sensor glucose value SG. Post-processing 9240 is then performed on the SG to generate a more robust and reliable sensor glucose value SG.

FIG. 114 shows a block diagram of a calibration of a single working electrode sensor using a Kalman filter. As previously described, Isig (we Isig) from the working electrode is an input to the preprocessing step 9212, where multiple Isig values may be filtered, averaged, and/or weighted, for example, to generate a single optimized Isig value. Then, in step 9222, the calibration BG 9232 is used to calculate CF and SG. However, step 9222 is now performed using an unscented kalman filter, such that the calculation of the actual calibration factor and resulting sensor glucose value is performed by the kalman filter using the methods and relationships described above. In step 9242, the calculated SG is post-processed to generate a more robust and reliable sensor glucose value SG. In an alternative embodiment shown in fig. 115, a kalman filter may be used to perform the preprocessing function (9217) in addition to calibration and SG calculations.

Multi-electrode system and fusion

In another embodiment, a multi-electrode system may be calibrated using a Kalman filter. Specifically, as shown in fig. 116, a system with N working electrodes can have a respective Isig from each preconditioning electrode 9214, 9216, 9218, as described above. The processed Isig from each working electrode can then be calibrated using an unscented kalman filter and calibration BG 9234 and the corresponding SG calculated, as shown in blocks 9224, 9226, 9228. Then, in block 9244, the respective SG from each of the N working electrodes can be fused and post-processed to produce a final fused SG.

Note that while in the above description, the kalman filter is only applied to the calibration step, in alternative embodiments, the kalman filter may be used for one or more of the pre-processing steps 9214, 9216, 9218, calibration and SG calculation steps 9224, 9226, 9228, and/or SG fusion and/or post-processing steps 9244. Further, as shown in fig. 117, a single kalman filter may be used to calibrate all working electrodes together, for example, by including all electrodes in the same kalman filter state space equation. Further, the fusing step may be performed using one of the generalized Melman's equations and/or fusion algorithms previously discussed in this specification in connection with the fusion of multiple Isigs or multiple SG values (including, for example, weighting of individual Isig and/or SG values). Accordingly, an unscented kalman filter, for example in conjunction with EIS data, can be used to optimize SG (or Isig) fusion in a multi-electrode system.

It is also important to note that as part of the fusion method, the post-processing steps described above may contain predictive components, whereby the physiological delay between blood glucose and interstitial glucose may be accounted for. Here, the past values of the sensor glucose SG are used to predict the (future) value of SG, the prediction amount applied at each time step depending on the noise level in the system. FIG. 118 is a table comparing results of a current fusion algorithm ("4D algorithm") on the one hand and an unscented Kalman filter applied to various sensor data sets on the other hand. As shown in fig. 118, the application of the kalman filter provides a significant improvement in the Mean Absolute Relative Difference (MARD) at each instant, while allowing a single kalman filter model to be applied to all datasets, even though there are significant design differences between the sensors for which the datasets were collected. Thus, for example, while applying the 4D algorithm to the Australia dataset resulted in a fused MARD value of 9.72, using the unscented kalman filter on the dataset provided a MARD of 9.66.

As previously discussed in connection with fig. 33-35 and 116, a fusion algorithm may be used to produce more reliable sensor glucose values. In particular, the fusion algorithm fuses the individual sensor glucose values to provide a single optimal glucose value to the user. In turn, optimal performance may be defined by the accuracy, duration, and rate of data availability and minimization of fault conditions that may burden the user. As previously noted, it should be noted that while the following discussion may describe aspects of the fusion algorithm in terms of the first working electrode (WE1) and the second working electrode (WE2) as redundant electrodes, this is exemplary and not limiting, as the algorithms described herein and their underlying principles are applicable to and may be used in redundant sensor systems having more than 2 working electrodes. Furthermore, such redundancy may be simple, pseudo-orthogonal, or orthogonal.

In one embodiment of the invention, the SG fusion algorithm is driven by multiple inputs, such as Electrochemical Impedance Spectroscopy (EIS), noise, and calibration. These inputs instruct the algorithm how to combine the individual electrode sensor glucose values to provide the final fused sensor glucose value, as well as the logic to control calibration, data display, and user prompting. Specifically, the fusion algorithm calculates a weight for each individual sensor glucose value (i.e., the glucose value from each of the working electrodes). The sum of the weights must be 1. In other words, the fused glucose value is a weighted average of the individual sensor glucose values, defined by the relationship:

wherein, at a given time, FG is a fused glucose, SG_kIs the sensor glucose value, FW, of the kth working electrode_kIs the final fusion weight assigned to the kth SG value of the system with N working electrodes.

The weights (as will be discussed further below) are derived by a series of transformations of the fused input, including noise, EIS-based sensor membrane resistance (Rmem), and calibration factor (calibration factor, or CF). As previously described, noise and Rmem are endogenous inputs that are driven by the sensor without any explicit input by the user. In this regard, fusion algorithms generally favor electrodes with lower noise and lower membrane resistance. On the other hand, the calibration factor is the ratio between the calibrated blood glucose value and the raw sensor current value (Isig), and is therefore derived from the user input. Here, the fusion algorithm will prefer electrodes with calibration factors within the range defined as optimal. The fusion algorithm weighs more on the preferred electrodes in the final fusion glucose calculation, according to the "preferred electrodes" defined by the noise, Rmem and the calibration factor. As shown in fig. 119, each type of input calculates a set of values that assign weights in a hierarchical manner, and the weights of each type are combined to calculate the final original fusion weight.

The fusion input is transformed through a series of functions to produce a set of weights. The ratiocore function computes the original fusion weights on the electrode set for a given input (e.g., noise) and, in one embodiment, can be expressed as:

this function or equation is applied to the input, where lower values indicate better performance (e.g., noise and membrane resistance), and therefore a larger fusion weight will be obtained. Thus, for example, noise from all electrodes at a given time is passed to the ratiochore function, which assigns a score (also referred to as a weight or ratio) to each electrode that is inversely proportional to its amount of noise relative to the sum of the noise on all electrodes. Thus, in the above equation, for a value of N>1 working electrode system, working electrode k at a given time (r)_k) Is represented as the working electrode k (ratio)_k) As a function of noise.

In particular, the first argument in the ratioScore function above normalizes the value in parentheses for r on all working electrodes_kThe sum is 1. The second argument in parentheses is the ratio of the noise of the kth working electrode to the sum of the noise values of all working electrodes (sigma operator). This ratio is then subtracted from 1 so that the electrode with low noise receives a high value.

As described above, the above equation is applied to the case where the lower the input value is, the better the performance is. For inputs with larger values indicating better performance, a simple formula calculates the original fusion weights. In particular, the following ratiocore is used to simply normalize a given metric by the sum of all working electrodes:

in the above equation, for a signal having N>1 working electrode system, input on working electrode k is_kIt is given.

The original fusion weight score (or ratio) calculated using one of the two equations above is then passed to the ratioGain function, which emphasizes or deemphasizes the relative score based on the predefined parameters. While the raw ratiocore value provides appropriate weights in terms of ranking, it does not necessarily assign weights in an optimal way. Thus, an equation is defined that expands or thins the weight ratio distribution based on the "gain factor" parameter. Thus, in one embodiment of the invention, the increased ratio weight g is defined as follows:

where r is the original fusion weight ratio and m is the "gain factor" parameter a for a system with N >1 working electrodes. The output g may then saturate to the range [0,1], such that if the output is greater than 1, the output is set to 1, and if the output is less than zero, the output is set to 0. In this regard, a saturation function that may be used in connection with embodiments of the present invention may be defined as:

note that sigmoidal or other smoothing functions may also achieve similar results as described above in embodiments of the present invention.

Finally, the values are processed through the makeSumOne function to ensure a sum of 1, and normalized if necessary. Thus, dividing a single value by the sum of all values yields a relative ratio, and the makeSumOne function is defined as follows:

by way of illustration, the above algorithm for noise weight and Rmem weight can be displayed separately as follows:

as can be seen from the above figure, computing a set of noise weights from all the individual noise weights follows the same general algorithm as computing a set of Rmem weights from all the individual Rmem inputs.

In an embodiment of the invention, the Carl factor weights are calculated in a similar manner, but with additional steps, including the calFactorTransform function, as follows:

the calibration factor values for all electrodes in a given time are first passed to the calFactorTransform function. Specifically, the calibration factor is converted to a score by normalizing the following function of the log-normal curve:

where x is the original (input) calibration factor, f (x) is the conversion (output) calibration factor, and the parameters σ and μ describe the width and peak of the lognormal curve, respectively.

Next, the results are saturated to the range [0.001, clip]Where all transform scores greater than the parameter clip will be assigned equal scores. Here, a higher score will result in a greater weight, and therefore, using the second of the two ratio score functions mentioned above (i.e.,. As shown, the remainder of the algorithm follows the noise and Rmem process described above.

Returning to FIG. 119, the flow chart of FIG. 119 shows how each set of weights is combined to calculate the final original blending weights. In particular, the original fusion weights are calculated by weighting and averaging the noise (9302) and calibration factor (9304) weights by the noise balance parameter (9308). The combined noise and calibration factor weights are then weighted by a Rmem balance (rmembance) variable (9310) and averaged with Rmem weights (9306). For the foregoing purposes, the parametric noise balance (9308) is predefined as the balance between the specified noise (9302) and the calibration factor (9304) weight. In a preferred embodiment of the present invention, the noise balance may be a constant with a value of 0.524.

In addition, the variable Rmem balance (9310) is determined as follows (see also the discussion below in connection with fig. 120): from the time of sensor start, after a predetermined duration, Rmem balance is set to zero. In other words, after a predetermined time from the sensor, the raw fusion weight (9318) receives zero contribution from Rmem. On the other hand, prior to the predefined time, i.e. from the time of sensor activation to the predefined duration, the calculation of Rmem balance (9310) is as follows and described:

first, the minimum and maximum Rmem _ weights for all electrodes are selected. Then, the minimum value is subtracted from the maximum value, added to 1, and the total divided by 2; this operation approximates the variance of the weights. This value is then passed to the TukeyWindow function (described below), the output of which is finally subtracted from 1. The purpose of these steps is to compute the Rmem balance (9310) such that when the difference between the Rmem values is large, the Rmem weights more emphasize the fusion weights.

TukeyPlus defines a flat-top cone cosine (Tukey) window, where the parameter r defines the taper ratio over the interval [0,1 ]. The nominal tukeyWindow function is as follows. Modifications can be implemented to increase the taper rate by introducing an additional "frequency" parameter before the 2 pi argument or by exponentiating the entire piecewise function:

in view of the above, a detailed description of the SG fusion algorithm according to the embodiment of the present invention will now be provided. The graph 120 shows the overall profile of the fusion algorithm, which takes as input (9350) the individual sensor glucose values (SG) that have been calculated for the individual sensors (i.e., the individual working electrodes). Restated, by way of illustration and not limitation, fig. 120 describes the fusion process with reference to two working electrodes, each producing a respective SG (i.e., SG1 and SG 2). However, the algorithm can be applied to many working electrodes.

At block 9352, it is determined whether any SG is invalid. If both SGs are determined to be invalid (9354), the entire fusion is set to "invalid" (9356). However, if only one SG is invalid (9358), the other (valid) SG is set as the fused SG (9360, 9362). On the other hand, if all SGs are valid, the next step in process 9370 determines if "FUSION _ START _ TIME _ SWITCH" has been reached. As previously explained in connection with fig. 119, in an embodiment of the present invention, this is a predefined duration since sensor activation, after which Rmem balance is set to zero. In a preferred embodiment, the predefined duration of time (after sensor connection) for the fusion algorithm to switch from Rmem logic to calibration factor and noise logic is about 25 hours.

Thus, if the current TIME is after "FUSION _ START _ TIME _ SWITCH," then Rmem-based FUSION is disabled so that Rmem does not contribute to the final FUSION weight (9380). On the other hand, if the current TIME is before "FUSION _ START _ TIME _ SWITCH," then Rmem-based FUSION is enabled (9372), such that Rmem FUSION weights are computed as described above, and the relative contribution of the Rmem FUSION weights to the final FUSION weights is computed based on the magnitude of the Rmem difference (9374).

Regardless of whether Rmem-based fusion is disabled (9380) or enabled (9372, 9374), the algorithm next provides for the calculation of calibration factors and noise fusion weights in block 9376. The Combined Calibration Factor and Noise (CCFN) and Rmem fusion weights are then combined, the final fusion weights are calculated, and the values are smoothed (9377). Finally, SG _ Fusion is calculated (for dual working electrode systems) as ri _1 SG1+ ri _2 SG2, as indicated by block 9378, where ri _1 and ri _2 are variables used to calculate the Fusion weight.

In conjunction with the fusion algorithm described herein, the behavior of each constituent working electrode (which may then be replicated prior to fusion) may be described as follows in conjunction with a preferred embodiment:

the first stage of filtration: convert 1 minute value to 5 minute value

For each individual Working Electrode (WE), this algorithm creates a five minute Isig using the last 8 minutes of sensor current data. This is referred to as first stage filtering. The algorithm uses information from the system to identify the time period during which the sensor data is affected by the diagnostic module. The algorithm then modifies the raw sensor signal (1 minute sensor current) by replacing the data packets that detected the total noise and/or diagnostic disturbance.

The algorithm calculates (1) the discard and (2) the five minute Isig by applying a simple 7-order FIR filter on the one minute data using the following coefficients [ 0.0660; 0.2095; 0.0847; 0.1398; 0.1398; 0.0847; 0.2095; 0.0660] the discard flag will be true or false depending on the variability of the 1 minute sensor current measurement in the last 8 measurements (8 minutes). when less than 4 measurements are taken after the sensor is connected, the discard flag will be false on the other hand, if the 4 or more measurements in the buffer fail to satisfy the condition that (a) the 1 minute sensor current is less than 1nA, (b) the 1 minute sensor current is greater than 200nA, (c) the 1 minute sensor current is less than the average count 2 with a precision of two bits, (d) the 1 minute sensor current is greater than the average count × 2. here, "average count" is the average of the middle 4 values, if there are 8 measurements in the history, otherwise it will be the average of the existing FIR measurement in the preferred embodiment, and only if the buffer has 5 or more false trigger events.

Identification of invalid data packets

For every 5 minutes of data packets, the signal will be checked to verify if the data packet is valid. A packet will be considered invalid if either of the following criteria is met: (a) the 5-minute Isig value is higher than MAX _ ISIG or lower than MIN _ ISIG; (b) v_cntrHigher than 0 volts or lower than-1.3 volts; (c) the packet is marked as an artifact; (d) when the 1 minute data is converted to 5 minutes Isig, the packet is marked as discarded; (e)1kHz real impedance is out of range; and (f) high noise (see noise check section discussed below). In a preferred embodiment of the present invention, the thresholds MAX _ ISIG and MIN _ ISIG for identifying invalid Isigs are 200nA and 6nA, respectively.

Artifact detection

In every 5 minute packet, artifact detection may be performed to identify size reduction in Isig, thereby preventing data from being used for SG calculations. For larger drops in Isig, the event can be classified as a "large artifact", for which all subsequent packets are marked as discarded and will be considered part of the artifact event until a termination condition is met. A small drop that can be classified as a "small artifact" allows only that single packet to be marked as discarded; the artifact detection algorithm can only mark the following packet as discarded if it is detected as a large artifact. If the packet is marked "init" (i.e., initialization, data refers to data during sensor warm-up), the artifact detection variable is set to a default value and no artifact is detected.

For each 5 minute packet that is not an initialization packet, two variables nA _ diff_iAnd pct _ diff_iThe definition is as follows:

nA_diff_i＝isig_i–isig_i-1

pct_diff_i＝100×(nA_diff_i/isig_i-1)

wherein isig is_iValues for ith Isig (in nA), Isig_i-1Representing the last Isig. If the previous packet was not a small artifact and not a large artifact state, then at pct _ diff_i<-25 and nA _ diff_i<The current packet can be marked as discarded at time-4.

Identifying onset of large artifacts

If the previous packet was not a large artifact, the current packet will be marked as dropped and considered to be the beginning of a large artifact if any of the following 3 conditions are true:

pct_diff_i<-40 and nA _ diff_i<-5

pct_diff_i+pct_diff_i-1<-50 and nA _ diff_i+nA_diff_i-1<-13

pct_diff_i+pct_diff_i-1+pct_diff_i-2<-60 and nA _ diff_i+nA_diff_i-1+nA_diff_i-2<-18

After large artifacts are detected

For each packet in the large artifact state, a packet with the detected artifact is included, which is marked for discard. Once the artifact is detected, the status of the artifact is determined on each packet. In this respect, the valid states are: (1) descending; (2) lowest point stability; (3) and (4) rising. The large artifact state may be exited if any of the following 4 conditions are met: (1) after being in the raised state, Isig is high and stable; (2) the former state is ascending, Isig is stable, and the system has several data packets in the ascending state; (3) the system has been in the artifact state for a long period of time, the maximum length being defined when the artifact is detected; (4) the connection is broken.

Small drop detection

Each packet updates the drop structure, which indicates whether the current packet is in a drop state and has associated variables so that the filter can consider the drop. The overall logic is as follows: the detected fall condition is any one of three general situations: (1) and (3) rapid descending: isig decreases rapidly, while previous packets show a more stable signal; (2) direction change: moderately fast decreasing Isig and increasing Isig with lower noise of previous packets; (3) moderate reduction: isig drops at a modest level and previous packets show very low noise. Once any of these events is detected, the measured Isig drop value will be added back to the original Isig and define the Isig threshold for exiting the roll-off state prior to filtering. If the state duration is too long or Isig increases sufficiently, the logic exits the fall state.

Noise estimation

The noise _ level is also used to identify roll-off and sensor-over conditions (see "noise check" section.) this process requires two most recent noise _ level/. specifically, the noise _ level is calculated based on the absolute values of the seven (7) most recent second order derivatives of the Isig (Isig _ acc) value, scaled by a 9 × calibration factor, and clipped between 0 and 10. in a preferred embodiment, the default noise _ level may be set to 7.5 if no current or previous second order derivative calculation is performed.five (5) most recent unfiltered Isig change rate values are used to calculate the variable frequency _ equivalent as follows:

frequency _ equivalent (abs (mean (roc))) calibration factor

Where "roc" is the rate of change (in nA/min.) after the above calculation, the frequency _ equivalent value is clipped to 0.2 to 4mg/d L/min if three or more isig _ acc values are not valid, or the calculated noise level exceeds 7, the frequency _ equivalent is set to a default value of 0.9.

Rate of change (ROC) estimation

When performing an on-the-fly calibration error check, the first and second derivatives of Isig are used to estimate noise, identify dips in the signal, compensate for delays, and reduce false errors. Filtered and unfiltered rates of change are calculated. In the former case, the Savitzky-Golay smooth rate of change is calculated using the 5 most recent Isig values, and any invalid Isig is replaced with the most recent valid Isig. Thus:

weight ═ 0.2; 0.1; 0; -0.1; -0.2 ]; the percentages are identical to coeff/Norm; [ 2; 1; 0; -1; -2]/10roc _ savitisig-sum (ravigig. times/time since last packet); percentage unit

nA/min

The unfiltered Isig rate of change (variable roc _ rawisig) is calculated by subtracting the previous Isig from the current Isig and then dividing by the time difference (5 minutes). The second derivative of the unfiltered Isig (acc _ rowsig) is calculated by subtracting (the first derivative) from the current packet the roc _ rowsig value calculated from the previous packet and dividing by the time difference as follows:

acc_rawisig＝(roc_rawisig(1)-roc_rawisig(2))/5

isig filtering

The filtered Isig values used to determine fIsig (i.e., for calibration and calculation of SG) will now be described. The filter parameter "q" is adjusted based on the noise _ level and frequency _ equivalent, so at low noise or high change rate, fIsig will approach the unfiltered value. When the Isig data is invalid, the filter output remains unchanged from the previous output. The filter will be reset at SENSOR _ warm _ TIME, which is defined as the TIME after the SENSOR is connected when SG starts to show to the user. In the preferred embodiment, the SENSOR _ WARMUP _ TIME is about one hour.

If fIsig is generated to an unexpected value, particularly above 202.5nA or below 3.5nA, a change sensor alarm is issued. If the resulting fIsig is greater than or equal to 3.5nA and less than MIN _ ISIG, it will be clipped to MIN _ ISIG. As described previously, in the preferred embodiment of the present invention, MIN _ ISIG may be set to 6 nA. However, if fIsig is generated that is less than or equal to 202.5nA and greater than MAX _ ISIG, it will be clipped to MAX _ ISIG. As previously described, in a preferred embodiment of the present invention, MAX _ ISIG may be set to 200 nA.

Isig delay compensation

Using a kalman filter, the predicted Isig is used as a measurement input. The prediction value, in turn, is calculated based on the Isig rate of change and clipped to prevent adding too much prediction. The amount of added prediction is adjusted by the presence of invalid data and noise (from noise level) calculations.

Kalman _ state computation

The q value (used in the equations that follow) is calculated using the noise _ level and frequency _ equivalent described in the noise estimation section. If the system signal is in a fall condition, roc will not be added to Isig. Instead, dips are added and the calculated kalman state q is modified to provide more filtering. The following calculations are used to determine the values to be stored for the kalman _ state.x and the kalman _ state.p. The current Isig value contains a delay compensation added to five minutes Isig.

Kalman _ state p ═ kalman _ state p + kalman _ state q

Kalman _ state k is kalman _ state p/(kalman _ state p + kalman _ state r)

Kalman _ state. x ═ kalman _ state. x + kalman _ state. k · (cur _ isig-kalman _ state. x)

Kalman _ state (1-kalman _ state. k) kalman _ state. p

EIS event

Each time an EIS event is triggered, measurements are made at the following frequencies (Hz) and this sequence is repeated according to WE: [0.105, 0.172, 0.25, 0.4, 0.667, 1, 1.6, 2.5, 4, 6.3, 10, 16, 25, 40, 64, 128, 256, 512, 1024, 2048, 4096, 8192 ]. If one of the EIS measurements is marked as saturated or obsolete, then the entire set of measurements for each WE will not be used.

Blood Glucose (BG) input

As previously described, the Calibration Ratio (CR) for calibration error checking may be calculated as follows:

CR is BG/(design + offset)

If no new or old sensor commands are received, or the most recent packet is marked as "init," then reject BG. to reject the BG input if no packets exist before the BG input (e.g., after a new sensor command), then reject this input.

On-the-fly calibration error checking

If the basic check does not reject BG, the latest fIsig of the two WE values will be used to check for calibration errors, in the preferred embodiment, this is the only place that would have been sending out calibration errors, if there were calibration errors on both WEs, then a new successful BG input is required to continue to display SG values, and the BG that caused the calibration error will not be used for calibration.

When the BG entry does not cause a calibration error, the single WE calibration error counter will be set to 0 and BG will be used to update the calibration factor. If the algorithm identifies a BG as causing a single WE calibration error, but the BG is still waiting for a final calibration, it will reject the BG and continue with the previous accepted BG on the WE. If a new BG passes the calibration error check, it will replace any currently pending final calibrated BG value. If the algorithm identifies that BG causes a calibration error, not due to invalid Isig, and the above does not apply: (1) if the calibration error counter is 1 and less than 5 minutes have elapsed since the transmitter identified the previous calibration error, BG does not increment the calibration error counter, thereby preventing a change sensor alarm from being generated by the same BG and design that previously caused the calibration error; and (2) otherwise, incrementing a calibration error counter. If the counter is 0, a new BG error is required to continue displaying the SG. Once the calibration error counter for a single WE reaches 2, WE terminates because SG can no longer be calculated.

Embodiments of the present invention include a dynamic maximum CR limit. Specifically, at sensor start-up, MAX _ CR may be set to 16 and linearly decrease to 12 over 4 days as time progresses. If V_cntrThe value is continuously higher for longer periods of time, and MAX _ CR may be further gradually decreased to 10. As previously mentioned, high V_cntrThe values are generally associated with high noise levels in Isig and sensitivity loss.

Working electrode calibration

As described herein, a single working electrode will request/require calibration at fixed intervals, or be determined in real time by intelligent calibration. In this regard, in embodiments of the present invention, a first successful calibration may expire after 6 hours, and subsequent calibrations will expire after 12 hours. Smart calibration based on EIS or day one calibration logic may shorten the validity period as described in the day one calibration and EIS sections.

In a preferred embodiment, the algorithm will continue to calculate SG for an additional amount of TIME after the expiration of the standard calibration (EXTRA _ TIME) and after the expiration of the EIS SMART calibration (EXTRA _ TIME _ SMART). Thus, if the calibration factor expires but within either EXTRA _ TIME or EXTRA _ TIME _ SMART, the working electrode state is set to 1, and if the calibration factor expires and after either EXTRA _ TIME or EXTRA _ TIME _ SMART, the working electrode state is set to 2. These SGs are stored in separate SG buffers that do not affect the display of the SGs. In an embodiment of the present invention, EXTRA _ TIME is set to 12 hours and EXTRA _ TIME _ SMART is set to 6 hours.

Single WE SG calculation

The calibration factor used to calculate SG is based on the most recent calibration calculation, and if in adjustment mode, the value is updated by the first day calibration logic or Isig dip calibration logic. The calibration factor used to calculate SG must be less than MAX _ CR and greater than MIN _ CR. If the calibration factor is outside this range, the system will invalidate the calibration factor and set the working electrode state equal to 2. Likewise, the filtered Isig used to calculate SG must be less than MAX _ Isig and greater than MIN _ Isig. If the filtered Isig is not within this range, the system will invalidate Isig and set the working electrode state equal to 2. If the calibration factor has expired or is invalid, or the current data packet is invalid, the working electrode status is set to 2.

BG to Isig pairing

After no BG entries causing calibration errors, the following steps are performed to update the calibration factor. If the current packet is invalid or when a new BG would cause a calibration error, then the calibration factor is not updated at this time. If the current data packet is valid and BG does not cause calibration errors, a temporary update of the calibration buffer is performed by adding BG and the currently paired sensor information to the calibration buffer and temporarily deleting the oldest paired information. The calibration factor is then calculated according to the following "calibration factor calculation" section. If calibration has also been performed before, the calculated calibration factor values have to be weighted according to the previous calibration factors. In a preferred embodiment, the weights are assigned as follows: the new value is weighted 70% and the old value is weighted 30%. Note that for packets that occur 5 to 10 minutes after BG was successfully entered, the calibration factor is updated by selecting the most recent value of fiisig, which is closest to the previous calibration factor and does not result in violation of the calibration error criteria.

Calibration buffer update

In an embodiment of the invention, the calibration buffer contains BG values and the following pairing information: each BG value in the buffer is associated with a pair Isig value, an expected value of high frequency virtual impedance, and an expected value of range impedance. There are typically 4 locations in the calibration buffer, with location 4 being the oldest entry. If the system is in Isig slump mode and the CR is less than the latest CR in the calibration buffer, the calibration buffer is updated by replacing the most recent entry (position 1) in the calibration buffer with a pending entry (instead of deleting the oldest entry). However, if the latter does not apply, the calibration buffer is updated by moving the previous entry (removing the oldest entry at position 4) and placing the new pending BG in position 1.

Calibration factor calculation

If no calibration error occurs, the calibration factor may be updated according to the following relationship, where Isig is the paired Isig value and n is the number of valid entries in the calibration buffer:

additionally, in the preferred embodiment, the α weight for each BG entry in the calibration buffer is fixed such that the most recent BG entry (i.e.: position 1) is weighted 0.80, position 2 is weighted 0.13, position 3 is weighted 0.05, and position 4 is weighted 0.02. in the preferred embodiment, the β weight for each BG entry is calculated using the following equation, where i represents the position in the calibration buffer:

β(i)＝2.655×(BG(i)-0.8041)–0.01812

if the system is not in FDC mode and the EIS does not detect a change in sensitivity, the calculated calibration factor is weighted by the expected cf value. The expected cf values are weighted 20% and the calculated calibration factors are weighted 80%. Calculation of the expected calibration factor as follows:

expected _ cf _ value 0.109 × t +4.731

Where t is the number of days since the sensor began. If the system is in Isig Dip calibration mode and the calculated calibration factor is less than 75% of CR, the calibration factor is set to 75% of CR. This ensures that the BG and SG values are reasonably close after Isig dip calibration.

Single WE SG calculation

The sensor glucose value is calculated according to the following relation

SG-calibration factor (design + offset) × + expected SG change

Where the expected SG change value is a predicted value of 5 minutes, calculated from the filtered Isig, and adjusted for signal noise and glucose concentration, if the expected SG change is greater than 6mg/d L or less than-6 mg/d L, the clipping will be 6mg/d L or-6 mg/d L, respectively.

First day calibration mode

For entering FDC mode, if the first successfully entered BG entry indicates a calibration rate outside the normal range of 4.5 to 5.5mg/d L/nA, but within a calibration error threshold, the FDC mode for the WE will be turned on.

When the "first day calibration" mode is active, the calibration factor for WE will be adjusted on every 5 minute packet according to:

cf adjustment (p1 × origCF + p2) × 5/60

Calibration factor + cf adjustment

Where P1-0.1721 hr-1 and P2-0.8432 mg/d L/nA/hr, no "first day calibration" adjustment is made to the current packet if one of (1) cf adjustment is negative and SG is already low (below 75mg/d L), or (2) the adjusted correction factor has reached the target range (4.5 to 5.5mg/d L/nA).

If the sensor has been activated for 12 hours, or the CR of a new calibration entry is within a stable range (4.5 to 5.5mg/d L/nA), the FDC mode for each WE will be stopped and no other adjustments to the sensor will be allowed.

Isig dip calibration mode

Embodiments of the present invention use Isig dip calibration logic in response to certain calibrations suspected of being performed on Isig with low glucose concentrations. This logic brings the calibration factor back closer to the previous value. If WE is not in FDC mode, then Isig Dip calibration mode is turned on, at calibration, the calibration indicates that Isig is low and the previous calibration was successful. This is verified by comparing the following thresholds:

CR >1.4 × previous calibration factor (referred to as origCF)

Previous calibration factor <6mg/d L/nA

Mean value of recently effective Isigs <20nA

The fIsig value used to calculate the calibration factor on Isig Dip will then be used in the adjustment logic described below and will be referred to as trigger Isig. Further, a previous calibration factor is used to determine whether the Isig dip calibration mode should be exited. The previous calibration factor is referred to as origCF.

In an embodiment of the present invention, recovery is detected when the current fIsig value is greater than 1.4 × trigger Isig.

Withdraw from Isig drop

The algorithm will stop tuning and exit the Isig Dip calibration mode if either of the following conditions is met, with the calibration factor up to date (possibly adjusted):

calibration factor < origCF × 1.2.2

Calibration factor <5.5

The self-test Isig Dip has passed for more than one day.

New BG showed CR <1.25 × origCF at calibration.

EIS Intelligent calibration

In each EIS measurement, a 1kHz imaginary impedance was filtered using a 5-point moving average filter. If less than one hour from the last calibration, the expected 1kHz virtual impedance value of the previous calibration is set to the current filtered value and the allowable range for the 1kHz virtual impedance value is set according to the most recent EIS measurement value. If one hour has elapsed since the last calibration and the current filtered impedance value is outside the two WE's allowed range, the calibration expiration time will be reduced to a maximum of six hours from the last calibration. If calibration is in progress when a sensitivity change is detected, then if CR differs by > 15% from the most recent calibration factor in the calibration buffer, only the new and previous BG's remain in the calibration buffer, with the expected CF value not being used to calculate CF.

Working electrode state

Each individual working electrode is assigned a state that determines how information from the electrode is used for subsequent processing. The status is determined by various error checking, diagnostic and calibration states. The following table summarizes the states:

noise(s)

If two consecutive windows are noisy (according to the above calculation), the Isig data will be considered invalid (state 2) until the end of the two hour window (at which point the working electrode may terminate or the logic no longer marks the data invalid). If high noise occurs (calculated as above) for three consecutive two hour windows, the working electrode state is irreversibly set to 2 and is considered to have terminated.

EIS-working electrode termination based on virtual impedance of 8kHz

In each EIS measurement, the 8kHz imaginary impedance is filtered using a 5-point moving average filter. The filtration value was monitored for 36 hours from the sensor connection. After 36 hours, the minimum 8kHz filtered imaginary impedance value was set as the reference value, excluding the value obtained during preheating. In a preferred embodiment of the invention, the latter reference value is clipped to a range of-1,000 Ω to 800 Ω. After the reference value is set, the absolute difference between the filtered 8kHz imaginary impedance value and the reference value is calculated for each EIS measurement. If the difference between two consecutive data packets is greater than 1,200 Ω, the working electrode state is irreversibly set to 2, and terminates.

EIS-WE termination and error based on 1kHz real impedance

A 5-point moving average filter was used in each EIS measurement to filter 1kHz real impedance. The filtered real impedance values are monitored until the filtered and unfiltered values are below 7,000 Ω. If the unfiltered 1kHz real impedance value is greater than 10,000 Ω, an error will be triggered and the state is set to 2. If this condition persists for 3 hours, the working electrode will terminate. If the filtered 1kHz real impedance is greater than 12,000 Ω, the state will be set to 2 and the working electrode will terminate.

Fusion

As described above in connection with fig. 120, in a preferred embodiment of the present invention, the fusion algorithm proceeds as follows: if both WE SGs are invalid or in state 2, the fused SG is set to invalid. If only one WE SG is invalid or in state 2, the fused SG is identical to the other valid WE SG. The fusion algorithm contains a weight calculation for both modes, and a logical description of how to transition between the two modes.

RMEM fusion model

Rmem fusion uses the difference in Rmem across each working electrode to determine fusion weights. Generally, a lower working electrode for Rmem will have a greater fusion weight. In this regard, the Rmem of the EIS measurement for each working electrode is calculated and stored prior to the last successful calibration.

Combined Calibration Factor and Noise (CCFN) fusion mode

The combined calibration factor and noise fusion mode use these two metrics to determine fusion weights. Calibration factor fusion utilizes a calibration factor on each working electrode to determine fusion weights. The calibration factor on each working electrode is converted by a look-up table or function to give CF greater weight within a preset range. Therefore, to calculate the "calibration factor weight" (cf weight 1), the "calibration factor" is converted according to the above so that the extreme values are weighted to zero, the optimal values are weighted to one, and the intermediate values are weighted to zero to one. As mentioned before, the transformation function is a normalized lognormal curve, which is determined by a parameter (Fusion) μ (describing the calibration factor transformation lognormal curve peak) and a parameter (Fusion) σ (describing the calibration factor transformation lognormal curve width). In a preferred embodiment, μmay have a value of 1.643 and σ may have a value of 0.13.

The output of the lognormal transform is saturated to [0.001, FUSION _ C L IP ], where the lower saturation limit is to prevent the downstream from being divided by zero error, the upper saturation limit is to equalize all fractions above the parameter FUSION _ C L IP.

Noise-based fusion

Noise fusion utilizes the noise difference on each working electrode to determine fusion weights. Generally, a less noisy working electrode will receive greater weighting. The filtered noise from each working electrode is calculated by a moving average filter of length FUSION _ NOISEWOW based on the absolute value of the variable containing the second derivative of the original Isig (acc _ rowsigs) from each working electrode. In the preferred embodiment, FUSION _ NOISEWOW is set to 36 hours. Note that before the number of fuse _ noise windows of a packet is available (e.g., during warm-up), the moving average filter length is equal to the number of available packets.

Next, to avoid being divided by zero, the filtered noise value of each WE is saturated, so that if the filtered noise (filteredNoise) <0.001, the filtered noise is 0.001. Each WE is then assigned a noise weighting metric by using the saturated filtered noise values of the other WEs normalized by the total noise. In this way, the lower noise WE receive greater weighting, as described in detail above. Finally, the calibration factor and noise metric are combined as described above in connection with FIG. 119.

Fusion mode conversion

Different fusion modes may be applied to the sensor depending on the state of the sensor. The Rmem fusion mode is usually most appropriate early in sensor wear. The calibration factor and noise fusion are most appropriate in the late stage of wear. To transition between these FUSION modes, in the preferred embodiment of the present invention, after FUSION _ START _ TIME _ SWITCH, the FUSION weight is determined entirely by CCFN. This timed switching logic replaces the Rmem affinity switch.

Rmem similarity transformation

The logic for switching the fusion mode depends on the similarity between the WE Rmem values. The large difference in Rmem means that the final fusion value will be determined by Rmem-based fusion. As the difference in Rmem values approaches zero, the Rmem fusion weight approaches 0.5. In this regard, it is appropriate that the combined calibration factor and noise fusion (CCFN) have a greater impact on the final fusion weight. For example, as shown in fig. 119, a fusion weight value is calculated.

Fusion weighted smoothing

After calculation, the symmetric weighted moving average is applied to the fusion weight. This avoids sharp transitions in the event that they occur because one of the working electrodes becomes unreliable. Sharp transitions are allowed at calibration. For this purpose, the coefficients of the filter are: [12344321]/20.

Fused SG computation and display

When fusion is enabled, the fused SG value is the final weighted sum of the plurality of working electrodes SG. Thus, for a system with 2 working electrodes:

filtered Ri _2(t) ═ 1-filtered Ri _1(t)

Fuse _ sg (t) (filtered Ri _1(t) × cur _ sg (1) + filtered Ri _2(t) × cur _ sg (2))

Where filtered Ri _1(t) is the filtered fusion weight of WE1, and the fused SG value is rounded to 0 bits after the decimal point. Note that in the preferred embodiment, the displayed fused SG must be in the range of [40, 400 ]. If the calculated fused SG is below 40mg/dl, the display will show "< 40 mg/dl" and if the calculated fused SG is above 400mg/dl, the display will show "> 400 mg/dl".

Fusion Rate of Change (ROC) calculation

The SG rate of change may be calculated every 5 minute packet. Here, roc1 and roc2 are first calculated using the three most recent fused SG values as follows, where fusion _ SG (1) is the most recent fused SG value:

roc1 ═ fusion _ sg (1) -fusion _ sg (2))/5

roc2 ═ fusion _ sg (2) -fusion _ sg (3))/5

If the direction (sign) of roc1 is different from that of roc2, or any of the latest 3 SGs are masked to display the SG, then the SG rate of change is set to zero mg/d L/min otherwise the fused _ SG rate of change is a value of roc1 or roc2 that is closer to zero.

Calibrating BG requirements and coordination

A single WE can trigger a calibration BG request. However, in one embodiment of the invention, the user is prompted to calibrate a BG request only if all of the running WE have an unresolved calibration request. An exception to the above is the first calibration request, which will occur at or after SENSOR _ warm _ TIME, as previously described. Here, when any running WE has an unresolved calibration request, the user will be prompted for a BG request for the first calibration.

The calibration may be displayed to the user as "recommended" or "forced". The "calibration recommendation" logic is triggered according to the calibration schedule (i.e., in the preferred embodiment, 2 calibrations plus smart calibrations per day). As described above, the EXTRA _ TIME is allowed to fail before calibration becomes mandatory and SG calculations cease. When calibration is due to SMART calibration, this TIME is set to EXTRA _ TIME _ SMART. The data may continue to be displayed for 6-12 hours based on when the smart calibration was triggered relative to the last successful calibration. The state of the SG is recorded so that the display device can determine whether or how to display the SG during the "recommended calibration" state. The following table is a graphical representation of the logic:

WE1 calibration status	WE2 calibration status	Fusing calibration states
			Is free of	Is free of	Is free of
Is free of	Recommending	Is free of
			Is free of	Force the	Is free of
Recommendation	Recommendation	Recommendation
			Recommendation	Force the	Recommendation
Force the	Force the	Force the

Note that the states in the table above are summarized for brevity. Thus, a complete logical table can be generated by switching WE1 and WE 2. In addition, the user is only exposed to the "fused calibration" state.

As previously mentioned, embodiments of the present invention are directed to pseudo-orthogonally redundant glucose sensors, systems, and related methods, including implementing such pseudo-orthogonally redundant sensors and/or systems within the context of the methods and algorithms (e.g., EIS, calibration, fusion, diagnostics, etc.) that have been discussed in this specification and related figures. Pseudo-orthogonal redundancy is achieved by utilizing similar techniques in two or more (working) electrodes, but with subtle but important changes in order to generate complementary glucose measurements, while minimizing additional design and/or computational complexity. Thus, illustratively, in one embodiment of the invention, two or more electrochemical (echem) sensors may be used, wherein the sensor(s) may be conventional peroxide-based sensors and the sensor(s) may be an "oxygen sensor" for measuring glucose by calculating the difference in oxygen between two working electrodes (which may or may not be located on the same flexure). In this way, the ASIC described above can be used to simultaneously measure glucose and oxygen to obtain two different but complementary glucose measurements. In embodiments of the present invention, a fusion algorithm (e.g., those already described herein) may be used to calculate a single fused sensor glucose value.

Current peroxide-based echem glucose sensors measure hydrogen peroxide (or simply "peroxide", H) according to the following relationship₂O₂)：

In the above reaction, glucose oxidase (GOx) catalyzes the production of H proportional to the amount of glucose and oxygen present at the site₂O₂And gluconic acid (C)₆H₁₂O₇) Structurally, as shown in FIG. 121A, the peroxide-based glucose sensor stack includes electrodes covered with a layer of glucose oxidase (GOx), or similar catalyst, and a glucose-limiting membrane (G L M). The sensor system then converts peroxide to an electrical current by applying an electrical potential across the electrode surface when used as a peroxide-based glucose sensor, the sensor stack of FIG. 121A is turned onOften operating at positive potentials greater than about +400mV (e.g., in one embodiment, in the range of +400mV to +535 mV). However, the potential may be reversed (e.g., by an ASIC) to negative, in which case the same sensor stack will operate as an "oxygen sensor" at a negative potential that may be below about-400 mV (e.g., in the range of-400 mV to-535 mV in one embodiment), with the oxygen in the sensor decreasing in proportion to the amount of oxygen present. In contrast, the oxygen sensor stack shown in fig. 121B does not include a GOx layer. Thus, when a negative potential is applied (e.g., a potential below about-400 mV), the measurement will be proportional to the oxygen entering the sensor stack, while when a positive potential is applied (e.g., a potential greater than about +400 mV), there should be no (or very little) measurement output associated with glucose or oxygen.

A chemical stack of a conventional (i.e., peroxide-based) glucose sensor and an oxygen sensor can be combined to produce an oxygen-based glucose sensor system. Specifically, in a preferred embodiment of the present invention, each of the sensor stacks shown in FIGS. 121A and 121B functionally operates as an "oxygen sensor" (i.e., at a negative potential). However, although the oxygen in the glucose sensor stack of fig. 121A decreases in proportion to the amount of incoming oxygen minus the glucose present (because of the presence of the GOx layer), in the oxygen sensor stack of fig. 121B, oxygen is directly detected. Glucose is then estimated as the difference between the respective outputs of the two sensors. Note that for this mode where both chemical stacks are used as "oxygen sensors," both sensors are operated at a negative potential (e.g., less than about-400 mV).

It is well known that measuring oxygen is important because echem sensors typically rely on oxygen, and oxygen deficiency can result in a significant loss of sensitivity. More specifically, the potential mechanism by which hypoxia causes a loss of sensitivity can be described as follows: when local oxygen is insufficient, the sensor output (i.e., Isig and SG) will depend on oxygen rather than glucose, and thus, the sensor will lose sensitivity to glucose.

Thus, one of the significant advantages of the pseudo-orthogonal redundant sensor system of the present invention described herein is providing a sensor output that is substantially independent of oxygen (i.e., resistant to low oxygen faults). That is, although the sensor may be an "oxygen sensor" as described above, the output is substantially independent of oxygen, as the dependence on oxygen is removed by subtraction (i.e., by calculating glucose as the difference between the respective outputs of the two sensors, or by calculating the difference in oxygen between the two working electrodes). It is noted here that such sensors are not completely oxygen independent, as one or more sensors will not operate at zero oxygen. However, the one or more sensors will be more resistant to oxygen under low oxygen conditions.

In an embodiment of the present invention, the pseudo-orthogonal redundant sensor system described herein will measure both background oxygen (via the oxygen sensor stack on the first flexure) and consumed oxygen (via the peroxide-based sensor stack on the same flexure or a second flexure). Fig. 122 shows details of an illustrative example of such first and second flexures, where each flexure contains (i.e., carries) one or more electrode and/or sensor stacks thereon and is configured for implantation or subcutaneous placement within a patient. Note that although two flexures are shown in fig. 122, embodiments of the present invention may operate with all electrodes contained on a single flexure capable of powering multiple electrodes at a time.

Returning to fig. 122, one flexure, identified as flexure 2(9400), may include a first Working Electrode (WE)₁)9410, and a Reference Electrode (RE) 9420. Three Counter Electrodes (CE)9430 are also shown, but embodiments of the invention can operate without any counter electrodes. Working electrode WE₁And GOx layer 9412 (and G L M, not shown) so that when the flexure is implanted or subcutaneously placed in a patient, it detects peroxide according to the foregoing relationship₁As electrodes for conventional peroxide-based sensors (stacks).

In fig. 122, a second flexure, identified as flexure 1(9450), may contain additional working electrodes. Specifically, fig. 122 shows 3 such working electrodes, namely (WE)₂)9460、(WE₃)9470 and (WE)₄)9480, more than 3 working electrodes may be included. In the working electrode shown, WE₂Further comprising a GOx layer 9462 to provide glucose regulated oxygen detection, and WE₄Detecting background oxygen. In this example, WE₃Is a background sensor, i.e., an electrode without GOx, which operates at a positive potential and detects any substance in the sensor other than glucose because there is no conversion of glucose to peroxide. This can be used for diagnostic purposes. In practice, this further electrode may be designed to sense any analyte, and may supplement a pseudo-orthogonal redundant sensor (e.g. it may be a further glucose sensor operating at a positive potential, it may be used to detect background interferents, or it may be designed to sense other analytes, such as lactate). Any number or combination of sensors may be added to this sensing framework. Thus, in general, the two flexures provide two different glucose measurements (i.e., one via the WE)₁And the other by WE₂And WE₄Combinations of (ii). In addition, WE₄Can also be used independently as an oxygen measurement, which can be used for diagnostic purposes. Thus, with 3 working electrodes, the system produced two pseudo-orthogonal methods of detecting glucose plus oxygen diagnostics.

The following example provides further details of the pseudo-orthogonally redundant glucose sensor system of the present invention, combining two glucose sensing methods. Note that the examples discussed below are illustrative, and other/additional arrangements may also be used to utilize the pseudo-orthogonally redundant glucose sensors and systems of the present invention described herein. Furthermore, embodiments of the invention may diagnose one or more electrodes (e.g., by EIS as described above) and employ one or more of the fusion algorithms described above to generate a single fused sensor glucose value.

In a first example, an embodiment of the invention may comprise a total of four working electrodes. Here, one or both working electrodes can always work as a conventional (peroxide-based) glucose sensor, wherein one or more working electrodes are accompanied by a GOx layer and operate at a positive potential. When two such working electrodes are used, they may also provide simple redundancy and diagnostic cross-checking between the two redundant electrodes. Then, the two additional (i.e., third and fourth) working electrodes together always operate as an oxygen-based differential glucose sensor. Thus, the third working electrode can be accompanied by a GOx layer (e.g., as shown in fig. 121A), and the fourth working electrode can operate without a GOx layer (e.g., as shown in fig. 121B). The oxygen-based differential glucose sensor may be used for diagnostics of overall sensor wear to help determine, for example, whether the peroxide-based sensor is functioning properly, and which sensing modality is to be used. In an embodiment of the present invention, a peroxide-based sensor may be determined to be functioning properly if the difference between the output of the peroxide-based sensor (Isig or calculated sensor glucose) and the output of the oxygen-based differential glucose sensor exceeds a threshold. In embodiments of the present invention, an example of such a threshold may be, for example, if the normalized Isig difference is greater than 30-50%, or if the calculated SG difference varies by more than 20-30% over a defined period of time.

In a second example, an embodiment of the present invention may include a total of three working electrodes. Here, as in the above-described embodiments, one or both working electrodes may operate as a conventional (peroxide-based) glucose sensor, with each of the one or more working electrodes being accompanied by a GOx layer and operating at a positive potential. The third working electrode can then be operated as a (permanent) "oxygen sensor" in which the electrode is not operated with the GOx layer and is operated at a negative potential (e.g., -400mV or lower) for diagnostic purposes. In this manner, when the oxygen measurement falls below a calculated threshold, e.g., by 50%, the potential of one of the conventional glucose sensors may be reversed (e.g., to-400 mV or less) to convert the latter to a second "oxygen sensor," thus enabling differential oxygen-based glucose sensing between the two "oxygen sensors. As previously described, alternative embodiments may include designs/configurations in which different numbers of flexures, different numbers of total working electrodes, different numbers of working electrodes on each flexure, and/or different numbers of working, reference, and counter electrodes, or combinations thereof, may be employed.

While the above description relates to particular embodiments of the present invention, it will be understood that many modifications may be made without departing from the spirit thereof. Additional steps and changes in the order of the algorithms may be performed while still performing the key teachings of the present invention. Therefore, it is intended that the appended claims cover such modifications as fall within the true scope and spirit of the invention. The presently disclosed embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description. All changes which come within the meaning and range of equivalency of the claims are to be embraced within their scope.