阅读说明：本技术 一种基于信号导频的射频指纹识别方法及系统 (Radio frequency fingerprint identification method and system based on signal pilot frequency ) 是由 朱丰超 曾盛 张颖 沈晓卫 金伟 张峰干 伍宗伟 袁丁 于 2021-01-12 设计创作，主要内容包括：本发明涉及一种一种基于信号导频的射频指纹识别方法及系统,所述识别方法包括：获取输入信号；对所述输入信号进行解调；采用滑动窗口的函数方法在解调后的输入信号中定位导频的起始位置；根据所述导频的起始位置确定信号导频部分,得到导频信号；计算所述导频信号的功率谱密度；对所述功率谱密度进行分类。本发明中的上述方法抗噪性能好,精度高。(The invention relates to a radio frequency fingerprint identification method and a system based on signal pilot frequency, wherein the identification method comprises the following steps: acquiring an input signal; demodulating the input signal; positioning the initial position of the pilot frequency in the demodulated input signal by adopting a sliding window function method; determining a signal pilot frequency part according to the initial position of the pilot frequency to obtain a pilot frequency signal; calculating a power spectral density of the pilot signal; classifying the power spectral density. The method has the advantages of good noise resistance and high precision.)

1. A method for radio frequency fingerprint identification based on signal pilot, the method comprising:

acquiring an input signal;

demodulating the input signal;

positioning the initial position of the pilot frequency in the demodulated input signal by adopting a sliding window function method;

determining a signal pilot frequency part according to the initial position of the pilot frequency to obtain a pilot frequency signal;

calculating a power spectral density of the pilot signal;

classifying the power spectral density.

2. The signal-pilot-based radio frequency fingerprinting method of claim 1, wherein demodulating said input signal specifically comprises:

down-converting the input signal;

low-pass filtering the input signal after the down-conversion;

performing analog-to-digital conversion on the input signal after the low-pass filtering;

sampling the input signal after analog-to-digital conversion;

performing automatic gain control on the sampled input signal;

performing matched filtering on the input signal after automatic gain control;

performing frequency compensation on the input signal after matched filtering;

performing time synchronization on the input signal after the frequency compensation;

and performing series-parallel conversion on the input signals after time synchronization, and then judging to obtain a demodulated Chip sequence.

3. The signal-pilot-based radio frequency fingerprinting method of claim 2, characterized in that the down-conversion, low-pass filtering and sampling of the input signal specifically use the following formulas:

r[s]＝r(sT_s)；

wherein, r [ s ]]For the sampled discrete digital signal sequence, with cos (omega)_ct) multiplication means a down-conversion operation,represents a convolution operation, h_LP(T) represents a low-pass filter function, s represents a sampling point, T_sRepresenting the sampling period.

4. The method of claim 2, wherein the automatic gain control of the sampled input signal specifically uses the following formula:

r_AGC[s]＝F_AGC(r[s])。

wherein, F_AGC(. to) is an AGC module.

5. The signal-pilot-based radio frequency fingerprint identification method according to claim 2, wherein the matched filtering of the input signal after the automatic gain control specifically adopts the following formula:

r_MF[s]＝H_MF(r_AGC[s])

wherein H_MF(. cndot.) is an MF module.

6. Method for signal-pilot based radio frequency fingerprinting according to claim 1, characterized in that the power spectral density of the pilot signal is calculated with the following formula:

wherein N is_FFTRepresenting the number of fast Fourier transform points, n representing the nth preamble sample point, k representing the kth power spectral density point, x_Preamble(n) denotes preamble data, and j is a complex unit.

7. The signal-pilot-based radio frequency fingerprinting method of claim 1, characterized in that the classification of the power spectral density uses in particular the following formula:

wherein, ω is^TRepresenting a projection matrix, p representing a sample, μ_iRepresents the mean of the class i samples.

8. A signal pilot based radio frequency fingerprinting system, characterized in that the system comprises:

the input signal acquisition module is used for acquiring an input signal;

a demodulation module for demodulating the input signal;

the initial position determining module of the pilot frequency is used for positioning the initial position of the pilot frequency in the demodulated input signal by adopting a function method of a sliding window;

a pilot signal determining module, configured to determine a signal pilot part according to an initial position of the pilot to obtain a pilot signal;

a power spectral density determination module for calculating a power spectral density of the pilot signal;

a classification module to classify the power spectral density.

Technical Field

The invention relates to the field of wireless communication, in particular to a radio frequency fingerprint identification method and a radio frequency fingerprint identification system based on signal pilot frequency.

Background

With the rapid development of the internet of things technology, the times of all things interconnection have come, and the wireless communication technology has become an essential part in our lives. However, as wireless communication technologies are more open than wired communication technologies and thus more vulnerable and threatened, network security issues are increasingly becoming more prominent. The traditional way for preventing network attacks is usually based on layers above the physical layer in the OSI network model, and the security of the wireless network is ensured by establishing a secret key, but there is always a certain vulnerability in the way based on secret key authentication. Therefore, it is highly desirable to find a more secure way to secure wireless communications. In the past decade, the concept of radio frequency fingerprint identification has gained widespread attention at home and abroad once it is put forward. In the communication process, even if the same batch of communication equipment manufactured by the same manufacturer always generates slight differences of signals radiated outwards during operation due to inherent characteristics of the equipment, such as carrier frequency errors generated by a crystal oscillator, errors of a timer and the like, and the errors are combined to make the signals sent by the communication equipment always unique, just like human fingerprints, so the errors are also called radio frequency fingerprints, and the technology for reversely identifying the communication equipment according to the errors is also called radio frequency fingerprint identification technology. The concept of radio frequency fingerprint is widely concerned at home and abroad once being put forward, the transient process of signals is firstly researched and feature extraction at home and abroad, the transient process of the signals refers to the process from starting to stable work of equipment, the signals radiated in the process are called transient signals, and the transient signals contain a large amount of fingerprint feature information of the equipment because the transient signals do not contain useful information and are only influenced by the characteristics of various components of the equipment. In 2002, Shaw et al established a multi-fractal model to determine the point in time at which a transmitter transient started from a channel containing noise, which requires that the channel noise exhibit a different multi-fractal characteristic than the transmitter transient signal. In 2003, Hall et al proposed to detect signal transients using the characteristic that the phase slope of the signal becomes linear at the onset of the transient, with a success rate of 85-90% using bluetooth signals for verification. In 2005, Ureten et al proposed a Bayesian step change point detection method that uses a Bayesian ramp detector to determine the transient start point by estimating the time when the signal power starts to increase gradually, with a better detection effect on the step signal. In 2009 Klein et al analyzed the instantaneous amplitude response of the signal using frame-Bayesian step detection and Variance Trace (VT) and evaluated the performance of each method using the 802.11a orthogonal frequency division multiplexed signal under different signal-to-noise ratios. However, although the transient signal contains abundant radio frequency fingerprint information, since the duration of the transient signal is very short, usually in microseconds, the sampling rate of the receiver is very high by using the transient signal to extract fingerprint features, and most receivers in the market do not meet the requirement.

With the development of communication technology, almost all signals are added with pilot parts at present for the convenience of receiver design, and the stable pilot provides a steady-state signal for extracting steady-state characteristics. In 2008, Kennedy et al first proposed a radio frequency fingerprint study based on steady-state signals, extracted a signal pilot portion, performed a frequency spectrum analysis after performing a fast fourier transform, then extracted signal frequency domain features, and classified 8 identical Universal Software Radio Peripheral (USRP) transmitters using a K-neighbor classification algorithm, achieving 97% accuracy at 30dB signal-to-noise ratio, and 66% accuracy at 0dB signal-to-noise ratio. In 2008, Suski et al proposed in the literature that transient detection was performed by using a variance trace method based on amplitude and phase, after a transient starting point was estimated, a subsequent pilot signal was selected to draw a Power Spectral Density (PSD), and then a radio frequency fingerprint was extracted from the PSD to classify the device, correct estimation of the transient starting point greatly affected the classification result, and finally classification and identification were performed on three devices, and the accuracy was only 80% at 6 dB. In 2008, Brik et al designed a fingerprint identification system for identifying IEEE 802.11b devices using 5 features of frequency offset, I/Q offset, pilot correlation, amplitude error and phase error of signals as radio frequency fingerprints of the devices, and the classification effect is better over a plurality of frames than that using one frame. However, the existing radio frequency fingerprint identification technology based on signal pilot frequency can not accurately and efficiently extract the pilot frequency part of the signal, only one pilot frequency can be extracted for continuous signals at a time based on a window function and power rising method, the method based on transient starting point estimation is greatly influenced by channel noise, the starting position of the pilot frequency can not be accurately found, and the method based on human eye identification is low in efficiency. Aiming at the current situation, the invention designs a new pilot frequency extraction algorithm, and effectively solves the problems.

Disclosure of Invention

The invention aims to provide a radio frequency fingerprint identification method and a radio frequency fingerprint identification system based on signal pilot frequency, which can accurately extract the signal pilot frequency.

In order to achieve the purpose, the invention provides the following scheme:

a signal pilot based radio frequency fingerprinting method, the identifying method comprising:

acquiring an input signal;

demodulating the input signal;

positioning the initial position of the pilot frequency in the demodulated input signal by adopting a sliding window function method;

determining a signal pilot frequency part according to the initial position of the pilot frequency to obtain a pilot frequency signal;

calculating a power spectral density of the pilot signal;

classifying the power spectral density.

Optionally, demodulating the input signal specifically includes:

down-converting the input signal;

low-pass filtering the input signal after the down-conversion;

performing analog-to-digital conversion on the input signal after the low-pass filtering;

sampling the input signal after analog-to-digital conversion;

performing automatic gain control on the sampled input signal;

performing matched filtering on the input signal after automatic gain control;

performing frequency compensation on the input signal after matched filtering;

performing time synchronization on the input signal after the frequency compensation;

and performing series-parallel conversion on the input signals after time synchronization, and then judging to obtain a demodulated Chip sequence.

Optionally, the down-conversion, the low-pass filtering, and the sampling of the input signal adopt the following formulas:

r[s]＝r(sT_s)；

wherein, r [ s ]]For the sampled discrete digital signal sequence, with cos (omega)_ct) multiplication means a down-conversion operation,represents a convolution operation, h_LP(T) represents a low-pass filter function, s represents a sampling point, T_sRepresenting the sampling period.

Optionally, the following formula is specifically adopted for performing automatic gain control on the sampled input signal:

r_AGC[s]＝F_AGC(r[s])。

wherein, F_AGC(. to) is an AGC module.

Optionally, the following formula is specifically adopted for performing matched filtering on the input signal after automatic gain control:

r_MF[s]＝H_MF(r_AGC[s])

wherein H_MF(. cndot.) is an MF module.

Optionally, the following formula is specifically adopted to calculate the power spectral density of the pilot signal:

wherein N is_FFTRepresenting the number of fast Fourier transform points, n representing the nth preamble sample point, k representing the kth power spectral density point, x_Preamble(n) denotes preamble data, and j is a complex unit.

Optionally, the following formula is specifically adopted for classifying the power spectrum:

wherein, ω is^TRepresenting a projection matrix, p representing a sample, μ_iRepresents the mean of the class i samples.

The present invention additionally provides a signal pilot based radio frequency fingerprint identification system, the system comprising:

the input signal acquisition module is used for acquiring an input signal;

a demodulation module for demodulating the input signal;

the initial position determining module of the pilot frequency is used for positioning the initial position of the pilot frequency in the demodulated input signal by adopting a function method of a sliding window;

a pilot signal determining module, configured to determine a signal pilot part according to an initial position of the pilot to obtain a pilot signal;

a power spectral density determination module for calculating a power spectral density of the pilot signal;

a classification module to classify the power spectral density.

According to the specific embodiment provided by the invention, the invention discloses the following technical effects:

the invention provides a pilot frequency extraction algorithm of an OQPSK signal, solves the problem of large workload of the traditional manual pilot frequency extraction method, and solves the problems that the method for performing transient detection based on signal amplitude and phase by using a variance trace method and then extracting the pilot frequency is sensitive to noise and has high requirements on the sampling rate of receiving equipment. According to the invention, OQPSK signal pilot frequency data sets under different transmission distances are established by utilizing a new pilot frequency extraction algorithm, and the established pilot frequency data sets are classified by utilizing an LDA supervised learning algorithm, and the classification result shows that the pilot frequency extraction algorithm designed in the invention has good noise resistance, and the extracted pilot frequency signals can be used for classifying and identifying different radio frequency devices and have good classification effect.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.

FIG. 1 is a flow chart of a method for radio frequency fingerprint identification based on signal pilot according to an embodiment of the present invention;

FIG. 2 is a diagram illustrating a modulation process according to an embodiment of the present invention;

fig. 3 is a block diagram of an OQPSK signal modulation circuit according to an embodiment of the present invention;

FIG. 4 is a diagram of a physical layer frame structure according to an embodiment of the present invention;

FIG. 5 is a block diagram of a wireless communication flow according to an embodiment of the present invention;

FIG. 6 is a block diagram of a fine frequency compensation algorithm according to an embodiment of the present invention;

FIG. 7 is a schematic view of a sliding process according to an embodiment of the present invention;

FIG. 8 is a schematic diagram of a positioning pilot start position algorithm framework according to an embodiment of the present invention;

FIG. 9 is a diagram illustrating a pilot detection result according to an embodiment of the present invention;

FIG. 10 is a diagram illustrating the influence of pilot length on classification accuracy according to an embodiment of the present invention;

FIG. 11 is a diagram illustrating a second exemplary embodiment of the present invention in which the influence of pilot length on classification accuracy is shown;

FIG. 12 is a diagram illustrating the effect of data set size on classification accuracy according to an embodiment of the present invention;

fig. 13 is a schematic structural diagram of a signal pilot-based radio frequency fingerprint identification system according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The invention aims to provide a radio frequency fingerprint identification method and a radio frequency fingerprint identification system based on signal pilot frequency, which can accurately extract the signal pilot frequency.

In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.

Taking an IEEE802.15.4 protocol-based wireless communication system as an example, the protocol specifies the media access control layer and physical layer architecture of wireless personal area networks (LR-WPANs), which lays the foundation for other high-level standards, and is often used for data communication devices for low-rate, low-power, and low-complexity short-range radio frequency transmission, such as ZigBee, WirelessHart, and MiWi. The physical layer of the IEEE802.15.4 protocol adopts an Orthogonal Quadrature Phase Shift Keying (OQPSK) modulation method, and a data signal needs to be mapped before modulation, and the complete modulation process is shown in fig. 2.

Suppose that a binary bit sequence to be sent at a sending end is recorded as d^1×N. In practical communication systems, signals are subjected to various influences such as noise interference and multipath interference during transmission, and in order to enhance the anti-interference, anti-multipath interference capability and concealment of the signals and to easily implement Code Division Multiple Access (CDMA), ieee802.15.4The protocol stipulates that a signal needs to be subjected to direct sequence spread spectrum once before transmission, a spread spectrum sequence with high code rate is directly utilized to spread a signal spectrum at a transmitting end, then the same spread spectrum code sequence is used for despreading at a receiving end, a spread spectrum signal is restored into an original signal, and the spread spectrum modulation process of the transmitting end is as follows:

firstly, d is^1×NMapping to a symbol S according to every 4 bits_i(Symbol), all symbols constituting a Symbol matrix S^4×m：

Wherein S is^4×mA matrix of symbols is represented by a matrix of symbols,each column of which represents a Symbol, and each Symbol is then mapped into a 32-bit Chip_iObtaining Chip sequence of sending end

Wherein, p ═ 32 × m, G (·) represents Symbol-To-Chip mapping relationship, and the Symbol-To-Chip mapping relationship is shown in table 1 when the transmission center frequencies are 2450MHz and 2380 MHz.

TABLE 1

And then, performing serial-to-parallel conversion on the mapped Chip sequence to obtain two paths of baseband digital signals, namely an I path and a Q path, and transmitting the baseband digital signals through a radio frequency front end after passing through an OQPSK modulation circuit, wherein the OQPSK modulation process is shown in FIG. 3.

Hypothesis Chip sequenceThe complex signal obtained after serial-to-parallel conversion is recorded as

Whereinx_I(k) And x_Q(k) Respectively representing in-phase components and quadrature components, and after the in-phase components and the quadrature components are subjected to a shaping filter and analog-to-digital conversion, the time domain expression of the in-phase components and the quadrature components is as follows:

wherein, a and h (t) respectively represent the amplitude correction factor of the modulation signal and the impulse response of the shaping filter, and then the I-path and Q-path signals are up-converted by the mixer, ideally, the I/Q-path mixer carriers should be orthogonal, and the OQPSK signal obtained by up-converting the mixer and adding should be:

however, due to the inherent tolerance of the hardware, the amplitude correction factors of the I-path and Q-path modulation signals have a certain correction error, which results in I/Q-path amplitude mismatch, and assuming that the mismatch coefficient is denoted as α, the same mixer also introduces a certain phase error Δ θ into the signals, so that the obtained OQPSK signal should be:

before OQPSK signals are transmitted, proper output power needs to be obtained through a power amplifier of a radio frequency front end, when the power amplifier works in a linear region, radio frequency fingerprint characteristics are not introduced, however, due to the low power efficiency characteristic of the power amplifier, more energy is consumed when the power amplifier works in the linear region, most of the power amplifiers work near a saturation region for the purpose of saving energy, however, the amplification benefit of the power amplifier is inversely proportional to the linearity degree of the power amplifier, so that a nonlinear effect is introduced when the power amplifier works near the saturation region, serious nonlinear distortion is generated on signals, and the signals passing through the power amplifier are marked as

x_PA(t)＝H_PA(x_Real(t)) (8)

Wherein H_PAThe nonlinear function of the power amplifier is shown by research, and the nonlinear characteristic introduced by the power amplifier is more obvious in the frequency domain, so that different transmitters can be distinguished by using the power spectral density as a radio frequency fingerprint characteristic.

The physical layer Frame structure in the signal transmission process specified by the IEEE802.15.4 communication protocol is shown in fig. 4, where each Frame is composed of a physical layer synchronization Frame header SHR, a physical layer Frame header PHR, and a physical layer load PHY payload, where the SHR includes a pilot Preamble and a start Frame delimiter SFD (start Frame delimiter), the pilot is 32 all-zero bits, the SFD has a fixed content [ 11100101 ], the first 7 bits of the PHR define the length of the Frame, the 8 th bit is a reserved bit, and the PHY payload is a useful load carried by the Frame.

Thus the signal x sent out via the radio frequency front end_PA(t) can be expressed as:

x_PA(t)＝[x_Preamble；x_SFD；x_PHR；x_Payload] (9)

wherein x_PreambleThe method is a signal pilot frequency which needs to be extracted, because the content of the signal pilot frequency is fixed, the radio frequency signal sent by each device has the same pilot frequency part, so that the method can be used for extracting and identifying the radio frequency fingerprint, however, according to the research and discovery of the inventor, the existing pilot frequency extracting method can not completely and accurately find and extract the pilot frequency of the signal, only one pilot frequency can be extracted once for continuous signals based on a window function and power rising method, the method based on transient starting point estimation is greatly influenced by channel noise, the starting position of the pilot frequency can not be accurately found, and the method based on human eye identification has low efficiency. Aiming at the current situation, the invention designs a brand-new pilot frequency extraction algorithm, and effectively solves the problems.

Fig. 1 is a flowchart of a radio frequency fingerprint identification method based on signal pilot according to an embodiment of the present invention, and as shown in fig. 1, the method includes:

step 101: an input signal is acquired.

Fig. 5 shows a wireless communication flow diagram, where a signal received by a receiving end is denoted as r (t):

r(t)＝x_PA(t)+n(t) (10)

where n (t) is channel superimposed noise.

Step 102: demodulating the input signal.

In the first step, the received signal r (t) needs to be demodulated, and the demodulation process includes down conversion, low pass filtering, analog-to-digital conversion, sampling, automatic gain control, matched filtering, frequency compensation and time synchronization. Firstly, a frequency mixer is used for carrying out down-conversion on a signal to a baseband signal, then out-of-band noise is filtered by a low-pass filter, and Nyquist sampling is carried out on the signal in an analog-to-digital converter to obtain a digital signal sequence r [ s ]:

wherein the content of the first and second substances,represents a convolution operation, h_LP(t) is a low-pass filter of the receiving end, because the signal will receive noise interference after propagating in the air, so the sampled signal is subjected to Automatic Gain Control (AGC) at the receiving end, so as to improve the signal-to-noise ratio:

r_AGC[s]＝F_AGC(r[s]) (12)

wherein, F_AGC(. to) is AGC module, after automatic gain control, the signal is processed with Matched Filter (MF) to get r_MF[s]The purpose of matched filtering is also to improve the signal-to-noise ratio:

r_MF[s]＝H_MF(r_AGC[s]) (13)

wherein H_MFThe signal is an MF module, frequency offset is generated in the transmission process of the signal, errors are introduced when the signal with the frequency offset is decoded, the error rate is increased, and therefore frequency compensation needs to be performed on the signal. First a Fast Fourier Transform (FFT) based method is used [11 ]]And performing coarse frequency compensation, namely squaring the OQPSK signal to obtain two spectral peaks, and then averaging and halving the frequencies of the two spectral peaks to obtain a coarse compensation frequency.

Where m is the modulation order, R_symIs the symbol rate and N is the number of samples. Obtaining a signal r after coarse frequency compensation_CC[s]. The fine frequency compensation is based on the OQPSK carrier recovery algorithm described in the above, and the algorithm block diagram is shown in fig. 6.

Wherein λ is_nIs the output of a Direct Digital Synthesizer (DDS), which is a voltage-controlled oscillator for processing time discrete sequences and is the core component of a discrete time phase-locked loop (PLL), in this documentIn the block diagram, the DDS acts as an integrating filter. In order to modify the input signal x_nThe algorithm first determines the phase error e_nThe magnitude of the phase error depends on the modulation scheme, and for the OQPSK modulation scheme,

in order to ensure the stability of the system, the phase error is passed through a biquad loop filter to obtain:

ψ_n＝g_Ie_n+ψ_n-1 (17)

wherein psi_nIs the output of the n-th sampling point of the loop filter, g_IIs the integrator gain coefficient:

wherein the content of the first and second substances,d＝1+2ζθ+θ²，B_nis the normalized loop bandwidth, ζ is the damping factor, phase recovery gain K_oEqual to the number of samples per symbol, the modulation determines the gain K of the phase error detector_PFor OQPSK modulation scheme, K_P2, and then the output of the loop filter ψ_nPassing to the DDS, yields:

λ_n＝(g_Pe_n-1+ψ_n-1)+λ_n-1 (19)

wherein the content of the first and second substances,finally, the maximum frequency lock delay is expressed as:

after fine frequency compensation, obtaining a signalr_FC[s]However, fine frequency compensation is introducedIs known, resulting in the true constellation possibly being rotated by 0,n, orRadian, therefore, the phase compensation needs to be considered for the above four cases in the subsequent pilot detection process. Finally utilize [11]The OQPSK time recovery algorithm described in (1) performs time synchronization, performs series-parallel conversion on the signals obtained after time synchronization, and then performs decision to obtain a demodulated Chip sequence, which is recorded asChip sequence at receiving endAnd sending end Chip sequenceHave the same length.

Step 103: and positioning the initial position of the pilot frequency in the demodulated input signal by adopting a function method of a sliding window.

As mentioned above, the second step is obtained after demodulationAnd (4) positioning a pilot starting position.

The ieee802.15.4 protocol specifies that the physical layer data frame includes a pilot, SFD, PHR and a signal Payload, and thusShould have the following format:

wherein the pilot frequency and the SFD have fixed contents, the pilot frequency is 4 × 8 ═ 32 all-zero bits, and according to the mapping relationship in table 1, the pilot frequency of the 32 all-zero bits is mapped to 8 chips with number 0 for transmission, so C_rx,Preamble＝8*Chip₀Therefore, we only need to complete the Chip sequence after demodulationIn, find out continuous 8 chips₀The pilot frequency can be found out by adopting a sliding window function method, the window length is one Chip length, namely 32bits, one bit is slid every time, and when a first Chip is found out₀Then, the next sliding is performed for the whole Chip length, if 8 chips appear consecutively₀If the segment of signal is corresponding to the pilot sequence in the original signal, to prevent the false detection, we use the same mapping relation to detect the SFD sequence after the pilot is ended, the SFD has fixed content [ 11100101 ]]Corresponding to Chip₇And Chip₁₀Therefore, if the consecutive 8 numbers are detected as Chip₀And the following Chip₇And Chip₁₀Then the pilot sequence of the detected signal is verified, the pilot sequence being from the first Chip₀Starting with the first bit, followed by a succession of 32 x 8 to 256 bits, the sliding process is given in fig. 7, and the algorithm framework is given in fig. 8.

At this time, the pilot sequence extracted by the user is not the pilot in the original signal but the demodulated pilot, the radio frequency fingerprint information of the device is no longer contained in the pilot sequence at this time, and the final task is to locate the pilot part in the original signal according to the demodulated pilot sequence, namely x in the formula (9)_PreambleAnd assuming that the signal is influenced by the channel in the wireless transmission process to be negligible, considering x_PreambleCan be represented by rs]The pilot part in (1) is equivalently represented.

Step 104: and determining a signal pilot frequency part according to the initial position of the pilot frequency to obtain a pilot frequency signal.

The third step being found on the basis of a sliding window as described aboveC_rx,PreambleCorresponding to the received signal rs]Pilot part of (1), rs]After automatic gain control, matched filtering, coarse frequency compensation and fine frequency compensation, only changing the amplitude of a frame, not changing the length of the frame, after time synchronization, firstly carrying out I/Q alignment on signals, namely recovering an OQPSK signal into a corresponding QPSK signal, then carrying out interpolation on the signals to recover into continuous signals, and finally resampling at an optimal sampling point, wherein the length of the signals is changed, which brings difficulty for positioning a pilot frequency in an original signal, in order to solve the problem, a method of a window function is adopted to carry out time synchronization on the original signals, the length of the window function is the length of one frame, namely, each time, carrying out pilot frequency detection on one frame, if the pilot frequency cannot be detected, the starting point of the window is not the starting point of one frame, and then sliding a sample point of the window in the original signal until the pilot frequency is successfully detected in the frame at present, the next time, the length of one frame is directly slid, so that the operation times and complexity of the algorithm are greatly reduced, and meanwhile, the problem that the position of a pilot frequency in an original signal cannot be located is solved, at this time, the pilot frequency in the original signal is 128 × 4 ═ 512 sampling points which are continuous from the starting point of a current frame window, finally, in the process of extracting the pilot frequency, 1635 frames are transmitted every time, 1635 pilot frequencies can be extracted, the extraction rate can reach 100%, and a pilot frequency detection result is given in fig. 9 (a light color part is the original signal, and a dark color part is the pilot frequency part).

Step 105: calculating a power spectral density of the pilot signal.

As mentioned previously, the fourth step is to calculate the power spectral density of the pilot signal. The detected pilot signal is a signal in a time domain, the difference of the pilot signals among different transmitters in the time domain is not obvious, and related researches show that the nonlinear characteristic introduced by a power amplifier at the front end of the radio frequency of the transmitter is very obvious in the performance of a frequency domain, so that the invention utilizes the power spectral density of the pilot signal as the radio frequency fingerprint characteristic to distinguish different transmitters, and for the signal pilot frequency x which is successfully extracted by us_PreambleThe power spectral density calculation method comprises the following steps:

wherein N is_FFTIs the length of the Fast Fourier Transform (FFT) transform point.

Step 106: classifying the power spectral density.

As mentioned above, in the fifth step, the power spectral density p (k) is classified by Linear Discriminant Analysis (LDA).

The linear discriminant analysis method is a classic feature dimension reduction algorithm, is widely applied to the classification and identification process of machine learning, and achieves good classification effect in the fields of target detection, face identification and the like. Like a Support Vector Machine (SVM), LDA is also a supervised learning algorithm, and its input is sample data with labels, and LDA reduces data dimensions by projecting the sample data onto a low-dimensional hyperplane, and the projected data should meet the requirements of relatively concentrated projection points of the same kind of data and relatively dispersed class centers of different kinds of data as much as possible, that is, the intra-class variance is minimum and the inter-class variance is maximum.

For our pilot signal power spectral density datasetWhere m denotes the sample class, i.e. number of transmitting devices, and the set of samples for each device is p_iIndicating that the corresponding label is y_iAs shown, the mean of the i-th class samples is recorded as μ_iThe covariance matrix is denoted as psi_iAnd the projection of the sample center on the low-dimensional hyperplane is recorded as omega^Tμ_iAnd the covariance matrix of the projection of the sample on the low-dimensional hyperplane is recorded as omega^Tψ_iOmega, the requirement that the similar sample points are gathered as much as possible and the heterogeneous sample points are dispersed as much as possible is metAs small as possible in the form of a capsule,as large as possible, so the objective function of LDA can be written as

Order toExpressed as a within-class scatter matrix (within-class scatter matrix),

S_b＝(μ_i-μ_j)(μ_i-μ_j)^Texpressed as a between-class scatter matrix (between-class scatter matrix).

Therefore, the formula (4) can be abbreviated as

When we find the optimal projection matrix omega to make the objective function J have the minimum value, we can multiply the projection matrix omega with our sample p to obtain the projected sample point omega^Tp, then comparing it with the projected sample centers ω of all classes^Tμ_iThe smallest one is taken as the classification result:

in order to verify the feasibility of the invention, we carried out experimental platform construction:

the device for pilot extraction comprises 11 ADALM-PLUTO software defined radios, 10 of which are transmitting devices and 1 of which is a receiving device, with several coaxial lines. The ADALM-PLUTO is an independent self-contained portable RF learning module, the RF frequency range is from 325MHz to 3.8GHz, the adjustable channel bandwidth is from 200KHz to 20MHz, the modulation precision is less than-40 dB, a 12-bit ADC (receiving end) and a DAC (transmitting end) are integrated, a transmitter and a receiver are included, and half-duplex or full-duplex is supported. We transmit 10s samples at a time each device as we acquire the data, each sample containing 16350 frames, so each device contains 16350 pilot samples, and we take 16000 pilot samples to construct the data set for convenience of computation.

Analysis of Experimental results

We discuss signal data under five different transmission distances, which are respectively under the coaxial line connection conditions of 0.1m, 1m, 5m and 10m, and each distance condition collects 16000 pilot data for each device, so that the pilot data set under each distance condition contains 16000 × 10 ═ 160000 pilot data of 10 devices.

Before classifying and identifying 10 different devices to be identified, we first performed a pre-experiment to discuss the effectiveness of classifying devices using pilot signals. Under the condition of a transmission distance of 10m, 120 pilot signals are selected for each device to serve as a data set, the first 80% of the pilot signals serve as a training set, the second 20% of the pilot signals serve as a test set, classification is carried out by utilizing an LDA supervised learning algorithm, then 120 data segments with the length of 512 sample points are randomly intercepted from the collected signals to serve as the data set of a comparison experiment, the data sets are divided into the training set and the test set according to the same proportion and method, classification is carried out by utilizing the LDA supervised learning algorithm, and finally the classification result is shown in a table 2.

TABLE 2 Pre-experiment Classification results

It can be seen from the classification result that the classification and identification effect by using fixed pilot frequency data is better than that by using random data, and it is noted that in our experiment, the data frame sent when each device sends data is the same, so even if the classification precision by using random data segment can reach 0.9542, in the actual communication process, the signal payload is usually a random signal with larger difference, and when the device identity information in the payload is introduced, the classification result can be more easily deceived by illegal devices.

Effect of Pilot Length on Classification accuracy

The pilot frequency extraction algorithm proposed by Kenndy and Suski et al cannot accurately extract the complete signal pilot frequency part, and the extracted pilot frequency signal also contains a part of the effective load of the signal, so that the influence of the effective load of the signal on the pilot frequency signal is increased when the length of the pilot frequency signal extracted by the algorithm is shortened, and the final equipment identification rate is influenced, and fig. 10 shows the comparison of the influence of the incomplete pilot frequency extracted by other methods and the influence of the length of the complete pilot frequency shortened pilot frequency extracted by the method of the present invention on the classification precision. The pilot length in the original signal is 512 sampling points, the influence of different pilot signal lengths of 512, 256, 128, 64, 32, 16, 8 and 4 bits on the classification precision is discussed, the size of the data set selects 1% of the total pilot data set, and the ratio of the training set to the test set is 8: 2.

according to the classification result, when the pilot frequency length is set to 128 bits, the LDA classification can still reach the classification accuracy rate close to 100% under various distance conditions, the classification effect of continuously shortening the pilot frequency length, whether the pilot frequency is complete pilot frequency or incomplete pilot frequency, is reduced because the device fingerprint information contained in the pilot frequency data for classification is reduced along with the shortening of the pilot frequency length, when the pilot frequency length is shortened to a certain degree, the condition of wrong classification is caused because the fingerprint information is too little, so the classification accuracy is reduced, and after the device fingerprint information contained in the pilot frequency data is reduced, the influence of the channel fingerprint is increased, so after the pilot frequency length is shortened to below 128 bits, the coaxial line connection condition is less influenced by the channel change, so the classification accuracy reduction speed and degree are better than other wireless transmission conditions, as can be seen from fig. 10, when the pilot length is shortened to below 128 and above 64, the classification performance starts to change sharply, so we further discuss the change of the classification performance in the interval of pilot length from 128 to 64 bits, and the result is shown in fig. 11. According to the classification result, the influence of shortening the pilot frequency length on the classification effect of the incomplete pilot frequency is larger, when the pilot frequency length is shortened to 124 bits, the classification accuracy of the complete pilot frequency still can reach 96.00% under the transmission distance of 10m, which is improved by 6.56% compared with the incomplete pilot frequency, when the pilot frequency length is shortened to 96 bits, the classification accuracy of the complete pilot frequency still can reach 84.33% under the transmission distance of 10m, and the signal load is usually different random signals in the actual communication process, which is improved by 17.3% compared with the incomplete pilot frequency, so if the effective load of the signal is mixed in the pilot frequency signal, the influence of shortening the pilot frequency length on the classification accuracy is larger. Under the condition of ensuring the classification accuracy, the complexity of the classification algorithm can be reduced by shortening the pilot length, when the size of the data set is 100% of the whole data set, 5058.3s is needed for classification by using the complete pilot, and only 2304.5s is needed for classification by using 124-bit pilot, compared with 54.44% of the operation speed of the algorithm, and table 3 shows the influence result of shortening the pilot length on the operation time of the algorithm under the complete data set.

Shortening the Pilot Length versus Algorithm runtime at Table 310m Transmission Range

Effect of dataset size on Classification accuracy

When 10 devices are classified by utilizing LDA supervised learning algorithm, the influence of different data set sizes on the final classification precision is discussed, the classification method of the training set and the test set is the same as the previous method, and the classification result is shown in FIG. 12.

According to the final classification result, when the pilot frequency data set is classified and identified by utilizing the LDA supervised learning algorithm, the requirement on the size of the sample data set is not very high, when the size of the data set accounts for less than 10% of the whole pilot frequency data set, the classification precision can basically reach 100%, when the size of the data set reaches 40% of the size of the whole data set, the data set is increased, the classification precision of coaxial line connection and 0.1m transmission conditions is still kept at 100%, but the precision starts to decline under the three transmission conditions of 1m, 5m and 10m, aiming at the situation, the reason is considered to be that the channel characteristics are changed because the time interval between the training set and the testing set is prolonged along with the increase of the data set, and in the wireless transmission process, the channel fingerprints are introduced into the signals due to the influences of multipath effects and channel noises, so that the LDA classifier remembers the sum of the channel fingerprints and the device fingerprints, when the channel changes, the channel fingerprint characteristics contained in the pilot data of the test set are changed from the channel fingerprint characteristics in the training set, and therefore, the classification accuracy is reduced.

The pilot frequency extraction algorithm proposed by Kenndy et al classifies the pilot frequency data sent by 8 devices of the same model, and can achieve 97% of classification accuracy under the condition that the signal-to-noise ratio is 30dB and reduce the accuracy to 66% under the condition of 0 dB. The pilot frequency extraction algorithm proposed by Suski et al classifies three devices, and when the signal-to-noise ratio is greater than 6dB, the classification accuracy is only 80%, and the performance is general at low signal-to-noise ratio. The pilot frequency extraction algorithm provided by the invention can reduce the length of the pilot frequency data for classification to 124 bits, greatly reduces the calculation amount of the classification algorithm, classifies 10 Pluto devices of the same model under the condition of coaxial line connection, has the classification precision as high as 99.74%, can reach 96.00% under the condition of 10m wireless transmission distance, improves the precision by 10.94% compared with the classification of incomplete pilot frequency signals extracted by other pilot frequency extraction algorithms, and has larger promotion when the transmission data is different.

Fig. 13 is a schematic structural diagram of a radio frequency fingerprint identification system based on signal pilot according to an embodiment of the present invention, and the system shown in fig. 13 includes:

an input signal acquiring module 201, configured to acquire an input signal.

A demodulation module 202, configured to demodulate the input signal.

A pilot start position determining module 203, configured to locate the start position of the pilot in the demodulated input signal by using a sliding window function method.

A pilot signal determining module 204, configured to determine a pilot portion of the signal according to the starting position of the pilot, so as to obtain a pilot signal.

A power spectral density determining module 205 for calculating the power spectral density of the pilot signal.

A classification module 206 for classifying the power spectral density.

The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.

The principle and the implementation mode of the invention are explained by applying a specific example, and the description of the embodiment is only used for helping to understand the method and the core idea of the invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.