阅读说明：本技术 基于特征值对角加载和构造二阶统计量的信源数估计方法 (Information source number estimation method based on characteristic value to angular loading and construction of second-order statistics ) 是由 王川川 曾勇虎 董晓博 汪连栋 王华兵 朱宁 于 2019-11-21 设计创作，主要内容包括：一种基于特征值对角加载和构造二阶统计量的信源数估计方法,针对天线阵列观测信号,计算其样本协方差矩阵,并对样本协方差矩阵进行特征值分解得到特征值序列,然后对特征值序列进行对角加载,再针对加载后的相邻特征值计算其差分值,并计算差分值的统计方差,然后再构造差分值的二阶统计量,根据二阶统计量的分布规律即可得到信源数的估计值。本发明提出的信源数估计方法具有较为广泛的适用性,既适用于经典渐近体系下的信源数估计,也适用于一般渐近体系下的信源数估计；既适用于高斯白噪声环境下的信源数估计,也适用于色噪声环境下的信源数估计；为电磁环境中辐射源的数量估计、信号波达方向估计等要求信源数作为条件的技术提供重要支撑。(A signal source number estimation method based on eigenvalue angular loading and second-order statistic construction includes calculating a sample covariance matrix of antenna array observation signals, decomposing eigenvalues of the sample covariance matrix to obtain an eigenvalue sequence, carrying out diagonal loading on the eigenvalue sequence, calculating difference values of loaded adjacent eigenvalues, calculating statistical variance of the difference values, constructing second-order statistic of the difference values, and obtaining estimated values of signal sources according to distribution rules of the second-order statistic. The information source number estimation method provided by the invention has wider applicability, is suitable for information source number estimation under a classical asymptotic system, and is also suitable for information source number estimation under a general asymptotic system; the method is suitable for estimating the number of information sources in a Gaussian white noise environment and in a color noise environment; the method provides important support for the technology requiring the number of information sources as conditions in electromagnetic environment such as the estimation of the number of radiation sources and the estimation of the arrival direction of signals.)

1. A method for estimating the number of information sources based on the characteristic value to angular loading and construct second-order statistics is characterized in that a sample covariance matrix is calculated for an antenna array observation signal, the characteristic value of the sample covariance matrix is decomposed to obtain a characteristic value sequence, then the characteristic value sequence is diagonally loaded, then the difference value of adjacent characteristic values after angular loading is calculated, the statistical variance of the difference value is calculated, then the second-order statistics of the difference value is constructed, and the estimated value of the number of the information sources can be obtained according to the distribution rule of the second-order statistics.

2. The method for estimating the source number based on the characteristic value pair angular loading and the construction of the second order statistic as claimed in claim 1, wherein the method for estimating the source number specifically comprises the following steps:

step 1: setting M array elements in the array, wherein M observation signals obtained by measuring at t moment are X (t), and X (t) is [ X [)₁(t),X₂(t),…,X_M(t)]^TWhen the sampling time t is 1,2, …, N and N are signal sampling numbers, calculating a sample covariance matrix of the observed signal

Step 2: the covariance matrix r (t) is subjected to eigenvalue decomposition,wherein the characteristic value lambda_iAnd a feature vector u_iAlso called sample eigenvalues and sample eigenvectors; expressing the sequence of characteristic values asIt is a numerical sequence arranged in descending order;

and step 3: for a sequence of characteristic valuesAnd (3) solving the diagonal loading quantity, wherein the calculation formula is represented as:

and 4, step 4: according to the sample covariance matrix R (t) and the eigenvalue diagonal load calculation formula, calculating a new sample covariance matrix after diagonal load, which is expressed as:

in the formula I_MIs an M-dimensional identity matrix;

and 5: for diagonally loaded sample covariance matrixDecomposing the characteristic value to obtain a new characteristic value sequence

Step 6: for a sequence of characteristic valuesCalculating a difference value of adjacent feature values, expressed as:

the sequence of differential values is represented as

And 7: calculating a sequence of differential valuesIs expressed as:

and 8: constructing a sequence of differential valuesThe second order statistic sosid of (a), expressed as:

and step 9: and the information source number estimation is realized, and the expression is as follows:

in the formula (I), the compound is shown in the specification,the number of the obtained information sources is estimated.

Technical Field

The invention belongs to the technical field of radar and communication reconnaissance, and further relates to an information source number estimation method based on characteristic value angular loading and second-order statistic construction in the technical field of radar and communication reconnaissance signal processing.

Background

In many signal processing studies and applications, the observed signal can be modeled as a mixture of a limited number of source signals and noise. Determining the number of signal sources is a necessary prerequisite for signal parameter estimation algorithms, and thus has attracted extensive attention and research in many fields, such as biomedicine, wireless communication, geophysical, radar, sonar, finance and the like.

The information source number estimation method is essentially based on the statistical analysis theory of the observed data and the moment function thereof, such as a hypothesis test method and an information theory rule method commonly used in the information source number estimation, and mainly utilizes the statistical distribution of the observed data and the statistics of the sample characteristic value. At present, the source number estimation method is mainly based on the classical asymptotic system established, namely, the classical statistical signal analysis theory that the dimension of an observed data matrix is fixed and the number of samples tends to be infinite, and is suitable for small-scale array signals of which the number of samples is far greater than the number of array elements.

However, in large-scale sensor arrays such as phased array radar and Multiple Input Multiple Output (MIMO) systems, due to the limitation of data storage space and the requirement of signal processing real-time, in practice, observed data often hardly meets the condition that the number of signal samples is far greater than the number of array elements, and usually belongs to high-dimensional limited sampled data or even small sampled data, that is, the number of signal samples and the number of array elements are in the same order of magnitude or even the number of signal samples is smaller than the number of array elements. The proportional relation between the signal sampling number and the array element number of the large-scale array observation data often does not meet the requirement of the classical statistical signal analysis theory, so that the appearance of the large-scale array brings a new challenge to the classical information source number estimation technology.

At the present stage, in the information source number estimation method under the classical asymptotic system, hypothesis testing methods include spherical testing, characteristic value testing and the like, observation statistics for hypothesis testing are mainly constructed by using a statistical distribution rule of sample characteristic values, and a decision threshold is set. The Information theory Criterion method includes Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Minimum Description Length (MDL), expected Description Length (PDL), etc., and usually assumes the observed data as gaussian distribution, establishes the Criterion of Information source number estimation according to the likelihood function of the observed data joint probability distribution, and the expression of the Information source number estimation is the function of the sample characteristic value. These methods are suitable for use in white gaussian noise environments. Under a classical asymptotic system, methods suitable for source number estimation under a color noise environment mainly include a Gerr circle method and an information theory criterion method based on diagonal loading, but both methods are not suitable for a large-scale array.

The source number estimation under the general asymptotic system is mainly an estimation method based on a random matrix theory, and comprises an RMT-AIC method, a BN-AIC method, a BIC-variant method, an LS-MDL method and an estimation method based on a spike model which are suitable for array elements with the number less than the number of signal samples, an estimation method based on a spherical test and an estimation method based on a modified Rao score test, wherein the estimation method is suitable for array elements with the number more than, less than or equal to the number of signal samples. The methods are not only suitable for source number estimation under a general asymptotic system, but also suitable for source number estimation under a classical asymptotic system, but the methods are only suitable for white noise environment, and source number estimation fails under a color noise environment.

The comprehensive analysis of domestic and foreign documents shows that an information source number estimation method which is suitable for both a classical asymptotic system and a general asymptotic system and is in a white noise or color noise environment is not available at present. Considering that the proportional relation between the number of antenna arrays and the number of signal samples in the actual array receiving signal environment and the fact that the noise of the aliasing of the observed signals is white noise or color noise are unknown, in order to improve the reliability of signal source number estimation, a signal source number estimation method which is suitable for a classical asymptotic system and a general asymptotic system and is suitable for both the white noise environment and the color noise environment needs to be developed.

Disclosure of Invention

Aiming at the actual situation that the proportional relation of the antenna array element number and the signal sampling number and the fact that the noise of the aliasing of the observed signal is white noise or color noise are unknown when the array antenna is applied to receive signals in the actual environment, the invention provides a signal source number estimation method based on the characteristic value to angle loading and construct second-order statistics, the relation of the antenna array element number and the signal sampling number does not need to be judged in advance (the application condition of the invention needs to be satisfied, the relation of the antenna array element number M, the signal source number K and the signal sampling number N is M-K is more than or equal to 2, K is less than N, and M is less than or equal to N), and whether the aliasing noise of the observed signal is white noise or not, and blind estimation can be carried out on the number of narrow-band signal sources such as radar, communication and the like under the complex electromagnetic environment.

The mathematical model of the information source number estimation method provided by the invention is as follows:

suppose K far-field signals are from direction θ₁,…,θ_KIncident on an array consisting of M sensors, and an array observation signal at the time t is X (t) and is expressed as

Wherein X (t) ═ X₁(t),X₂(t),…,X_M(t)]^T(the superscript T denotes transposition) into the array observation signal vector, a (θ)_k) Is an array direction vector, A (theta) ═ a (theta)₁),a(θ₂),…,a(θ_K)]Is a matrix of directional vectors, theta ═ theta₁,…θ_K]^TIs the parameter vector of the incoming wave angle of the signal, s (t) ═ s₁(t),s₂(t),…,s_K(t)]^TFor the incident signal vector, w (t) ═ w₁(t),w₂(t),…,w_M(t)]^TFor an additive noise vector, the sampling time t is 1,2, …, and N is the number of signal samples. The basic assumption of the array observation signal model shown in formula (1) is as follows:

(1) incident signals are mutually independent narrow-band steady signals, and satisfy the mean value E { s (t) } ═ 0 and a covariance matrixWhereinIs the power of the kth signal;

(2) the superimposed noise in the array observation signal vector is additive noise which is independent from the signal;

(3) the number of incident signals is smaller than the number of array elements and the number of samples at the same time, namely K is less than min (M, N);

(4) the signals are propagated in an ideal space, and the array sensor has omnidirectional consistency.

In practice, the sample data received by the antenna array contains noise, and may not be ideal white gaussian noise, but rather complex spatial color noise. In a complex spatial color noise environment, the noise eigenvalue portion of the covariance matrix of the antenna array received data becomes very divergent, and does not oscillate around the noise power as the noise eigenvalue portion in white gaussian noise. The result caused by color noise can disable various algorithms for estimating the number of information sources by using hypothesis testing and an information theory criterion, and the information source number estimation method based on the Gerr theorem and the information source number estimation method based on the characteristic value diagonal loading and the information theory criterion can only be applied to a classical asymptotic system, namely the relationship between the number M of antenna array elements and the number N of signal samples is as follows: m is fixed and M/N is less than 1, under the general asymptotic system, namely the relation between the antenna array element number M and the signal sampling number N is as follows: m and N tend to infinity at the same rate, M, N → ∞ and M/N → c ∈ (0, ∞), the above method is generally defeated for source number estimation, whether the noise is white Gaussian noise or color noise.

The existing source number estimation method based on the random matrix theory cannot solve the source number estimation problem under the condition of aliasing color noise of an observation signal under a general asymptotic system.

By analyzing the eigenvalue of the covariance matrix of the observation signals of the antenna array, the noise eigenvalue is found to be very divergent in a color noise environment. The invention is inspired by the existing idea of loading characteristic values to angles, and provides an information source number estimation method by combining the analysis of the distribution characteristics of the characteristic values, and the technical idea is as follows: and calculating a sample covariance matrix of the antenna array observation signal, decomposing the eigenvalue of the sample covariance matrix to obtain an eigenvalue sequence, carrying out diagonal loading on the eigenvalue sequence, calculating a difference value of the adjacent eigenvalue subjected to the diagonal loading, calculating the statistical variance of the difference value, constructing second-order statistic of the difference value on the basis, and obtaining an estimated value of the information source number according to the distribution rule of the second-order statistic.

On the basis of the mathematical model and the technical thought, the information source number estimation method provided by the invention specifically comprises the following steps:

step 1: setting M array elements in the array, wherein M observation signals obtained by measuring at t moment are X (t), and X (t) is [ X [)₁(t),X₂(t),…,X_M(t)]^T(the superscript T denotes transposition), the sampling time T is 1,2, …, N is the number of signal samples, and the covariance matrix of the observed signal is calculated

Step 2: the covariance matrix r (t) is subjected to eigenvalue decomposition,wherein the characteristic value lambda_iAnd characteristic vector u_iAlso called sample eigenvalues and sample eigenvectors; expressing the sequence of characteristic values asArranged in descending orderA sequence of values;

and step 3: for a sequence of characteristic valuesAnd (3) solving the diagonal loading quantity, wherein the calculation formula is represented as:

and 4, step 4: according to the sample covariance matrix R (t) and the eigenvalue diagonal load calculation formula, calculating a new sample covariance matrix after diagonal load, which is expressed as:

in the formula I_MIs an M-dimensional identity matrix;

and 5: for diagonally loaded sample covariance matrixDecomposing the characteristic value to obtain a new characteristic value sequence

Step 6: for a sequence of characteristic valuesCalculating a difference value of adjacent feature values, expressed as:

the sequence of differential values is represented as

And 7: calculating a sequence of differential valuesIs expressed as:

and 8: constructing a sequence of differential valuesThe second order statistic sosid of (a), expressed as:

and step 9: and the information source number estimation is realized, and the expression is as follows:

in the formula (I), the compound is shown in the specification,the number of the obtained information sources is estimated.

The information source number estimation method based on the characteristic value angular loading and the second-order statistic construction can estimate the number of narrow-band signal sources such as radar signals, communication signals and the like in a complex electromagnetic environment, has wide applicability, and can obtain the following beneficial effects:

firstly, in terms of the relation between the array number and the signal sampling number, the method does not need to preset or assume the relation between the array antenna array element number and the signal sampling number, and is suitable for estimating the information source number under a classical asymptotic system and the information source number under a general asymptotic system;

secondly, from the noise characteristics, the method can be suitable for estimating the number of information sources in a Gaussian white noise environment and in a color noise environment, particularly for estimating the number of information sources in the case of observing the mixed color noise of signals under a general asymptotic system, and an effective solution is provided for the problem of the lack of the number of information sources estimation technology under the condition;

thirdly, the method can provide important support for technologies requiring the number of the information sources as conditions, such as the number estimation of the radiation sources in the electromagnetic environment, the estimation of the direction of arrival of signals and the like.

Drawings

Fig. 1(a) to 1(d) are source number estimation results of the sdi method, the information theory rule method and the bell circle method under the condition of gaussian white noise.

Fig. 2(a) to 2(d) are source number estimation results of the sdi method and the information theory rule method and the geur circle method under the color noise condition.

Fig. 3(a) to 3(c) are the comparison between the estimation results of the sdi method and the source number estimation method based on the stochastic matrix theory under the condition of gaussian noise mixture.

Fig. 4(a) to 4(d) are the comparison between the sdi method and the source number estimation method based on the random matrix theory, which are provided by the present invention, and the estimation results under the color noise condition.

Detailed Description

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the invention without limiting the invention.

The method comprises the following specific implementation steps:

step 1: setting an antenna array to have M array elements, wherein M observation signals obtained by one measurement are X (t), and X (t) is [ X [)₁(t),X₂(t),…,X_M(t)]^T(the superscript T denotes transposition), the sampling time T is 1,2, …, N is the number of signal samples, and the covariance matrix of the observed signal is calculated

Step 2: the covariance matrix r (t) is subjected to eigenvalue decomposition,wherein the characteristic value lambda_iAnd characteristic vector u_iAlso called sample eigenvalues and sample eigenvectors; expressing the sequence of characteristic values asIt is a numerical sequence arranged in descending order;

and step 3: for a sequence of characteristic valuesAnd (3) solving the diagonal loading quantity, wherein the calculation formula is represented as:

and 4, step 4: according to the sample covariance matrix R (t) and the eigenvalue diagonal load calculation formula, calculating a new sample covariance matrix after diagonal load, which is expressed as:

in the formula I_MIs an M-dimensional identity matrix;

and 5: for diagonally loaded sample covariance matrixDecomposing the characteristic value to obtain a new characteristic value sequence

Step 6: for a sequence of characteristic valuesCalculating a difference value of adjacent feature values, expressed as:

the sequence of differential values is represented as

And 7: calculating a sequence of differential valuesIs expressed as:

and 8: constructing a sequence of differential valuesThe second order statistic sosid of (a), expressed as:

and step 9: and the information source number estimation is realized, and the expression is as follows:

in the formula (I), the compound is shown in the specification,the number of the obtained information sources is estimated.

The invention is further described below in conjunction with experimental test charts.

1. Setting experimental conditions:

the experimental verification of the invention is carried out under the simulation condition of a DELL9020MT personal computer, Intel (R) core (TM) i7-4770CPU @3.40GHz and 64-bit Windows operating system, and MATLAB R2010a is adopted as simulation software. In order to fully verify the effectiveness of the method (called SOSDI method), the technical scheme of the invention is compared with the technical scheme recorded in the existing literature, and four groups of experimental tests are carried out.

Experiment one: the SOSDI method provided by the invention is compared with information theory criterion methods (BIC method, AIC method, MDL method, KIC method) and Gell circle method (MGDE method) disclosed in the monograph document array signal processing and MATLAB realization under the Gaussian white noise environment. The experimental conditions were set as follows:

1)s₁for BPSK signal, the sampling frequency is 120MHz, the sub-pulse width is 3 × 10^-7s, carrier frequency is 10 MHz;

2)s₂is a CW signal, the sampling frequency is 120MHz, the sub-pulse width is 1.5 × 10^-5s, carrier frequency is 10 MHz;

3)s₃the carrier frequency is 10MHz and the pulse repetition frequency is 0.1MHz for LFM signals;

4)s₄is FSK signal, has a sampling frequency of 120MHz and a sub-pulse width of 10^-7s, the carrier frequency changes between two frequency points of 25MHz and 50MHz along with the binary baseband signal;

5)s₅for MPSK signal, the sampling frequency is 120MHz, the sub-pulse width is 4 × 10^-7s, carrier frequency 50 MHz.

If the source number K is set to 4 in the simulation, the source signal is divided by s₁、s₂、s₃、s₄And (4) forming. Setting different array antenna array element numbers M, generating a mixing matrix A by a random function randn, wherein the sampling frequency is 120MHz, the number of signal sampling points is N, the mixed signal is superimposed with Gaussian white noise, the variation range of the signal-to-noise ratio is-20 dB to 30dB, the step length is 2dB, carrying out 1000 times of Monte Carlo simulation on each signal-to-noise ratio, and the experimental result is shown in figure 1.

Experiment two: the SOSDI method provided by the invention is compared with information theory criterion methods (BIC method, AIC method, MDL method, KIC method) and Gerr method (MGDE method) based on characteristic value to angle loading under the color noise environment. The experimental conditions were set as:

1)s₁for BPSK signal, the sampling frequency is 120MHz, the sub-pulse width is 3 × 10^-7s, carrier frequency is 10 MHz;

2)s₂is a CW signal with a sampling frequency of 120MHzPulse width of 1.5 × 10^-5s, carrier frequency is 10 MHz;

3)s₃the carrier frequency is 10MHz and the pulse repetition frequency is 0.1MHz for LFM signals;

4)s₄is FSK signal, has a sampling frequency of 120MHz and a sub-pulse width of 10^-7s, the carrier frequency changes between two frequency points of 25MHz and 50MHz along with the binary baseband signal;

5)s₅for MPSK signal, the sampling frequency is 120MHz, the sub-pulse width is 4 × 10^-7s, carrier frequency 50 MHz.

If the source number K is set to 4 in the simulation, the source signal is divided by s₁、s₂、s₃、s₄And (4) forming. Setting different array antenna array element numbers M, generating a mixing matrix A by a random function randn, sampling frequency of 120MHz, the number of signal sampling points of N, observing signal superposition spatial color noise, wherein the elements of a covariance matrix are given by the following formula: wherein sigma_nIs an adjustable parameter, is used for setting the signal-to-noise ratio of the mixed signal, the change range of the signal-to-noise ratio is-20 dB to 30dB, the step length is 2dB, and 1000 times of Monte Carlo simulation is carried out on each signal-to-noise ratio, and the experimental result is shown in figure 2.

Experiment three: the invention provides an SOSDI method and an application of a large-dimension random matrix theory in array signal parameter estimation of a academic paper, and discloses a source number estimation method (BIC-variant, EEE, LS-MDL, RMT-AIC, BN-AIC and ST-GBIC) based on a random matrix theory, wherein the source number estimation performance is realized under the condition of mixing Gaussian noise. In the experiment, 8 mutually independent signals with K equal to space are respectively incident on a uniform linear array consisting of M array elements with half-wavelength spacing from an arrival angle DOA equal to-55 degrees, -40 degrees, -25 degrees, -10 degrees, 5 degrees, 20 degrees, 35 degrees and 50 degrees, and a direction vector in an array receiving signal isThe incident signal is generated by mutually independent gaussian random sequences through an AR (1) model with a parameter of 0.9. Define SNR as 10log₁₀(p_s/σ²) Wherein the noise power is set to σ²1. The range of the signal-to-noise ratio is 0-30 dB, the step length is 4dB, the probability that each algorithm correctly estimates the number of the information sources in 1000 independent repeated simulation experiments is counted, and the experimental result is shown in figure 3.

Experiment four: the SOSDI method provided by the invention is compared with a source number estimation method (BIC-variant, EEE, LS-MDL, RMT-AIC, BN-AIC and ST-GBIC) based on a random matrix theory disclosed in a academic paper application of a large-dimension random matrix theory in array signal parameter estimation under a color noise environment. The experimental conditions were set as follows:

1)s₁for BPSK signal, the sampling frequency is 120MHz, the sub-pulse width is 3 × 10^-7s, carrier frequency is 10 MHz;

2)s₂is a CW signal, the sampling frequency is 120MHz, the sub-pulse width is 1.5 × 10^-5s, carrier frequency is 10 MHz;

3)s₃the carrier frequency is 10MHz and the pulse repetition frequency is 0.1MHz for LFM signals;

4)s₄is FSK signal, has a sampling frequency of 120MHz and a sub-pulse width of 10^-7s, the carrier frequency changes between two frequency points of 25MHz and 50MHz along with the binary baseband signal;

5)s₅for MPSK signal, the sampling frequency is 120MHz, the sub-pulse width is 4 × 10^-7s, carrier frequency 50 MHz.

If the source number K is set to 4 in the simulation, the source signal is divided by s₁、s₂、s₃、s₄And (4) forming. Setting different array antenna array element numbers M, generating a mixing matrix A by a random function randn, sampling frequency of 120MHz, the number of signal sampling points of N, observing signal superposition spatial color noise, wherein the elements of a covariance matrix are given by the following formula: wherein sigma_nIs an adjustable parameter, is used for setting the signal-to-noise ratio of an observation signal, the change range of the signal-to-noise ratio is-20 dB to 30dB, the step length is 2dB, 1000 times of Monte Carlo simulation is carried out on each signal-to-noise ratio, and the experimental result is shown in figure 4.

2. And (3) simulation result analysis:

fig. 1 is a comparison between the sdi method and the information theory criterion method (BIC, AIC, MDL, KIC) and the gehr circle Method (MGDE) in the white gaussian noise environment. As can be seen from fig. 1(a), at this time, M/N is less than 1, the relationship between the antenna array element number and the sample number meets the requirement of the classical asymptotic system, under the condition of gaussian white noise, when the signal-to-noise ratio is more than 10dB, the SOSDI method, the MDL method and the BIC method can accurately realize the source number estimation with a probability of 100%, but the cover circle method needs to realize the source number estimation with a probability of 100% above about 22dB, and the estimation accuracy of other information theory criterion methods under the condition of 30dB still does not reach 100%; in the case of figure 1(b) of the drawings,the relation between the array element number and the sample number of the antenna array approximately meets the requirement of a classical asymptotic system, under the condition of Gaussian white noise, when the signal-to-noise ratio is more than 0dB, the SOSDI method and the information theory criterion method can accurately realize the source number estimation with the probability of 100%, but the Gerr method needs to be more than about 27dB to realize the source number estimation with the probability of 100%; in FIGS. 1(c) and 1(d),the relation between the array element number of the antenna array and the sample number meets the requirement of a general asymptotic system, under the condition of Gaussian white noise, the SOSDI method provided by the invention can accurately realize the source number estimation with the probability of 100% when the signal to noise ratio is higher than-2 dB and higher than-8 dB respectively, and other information theory criterion methods and the Gerr method fail to estimate;

FIG. 2 shows the SOSDI method and the information theory criterion method (BIC) based on the feature value to the angle loading proposed by the present inventionAIC, MDL, KIC), the geuer circle Method (MGDE) in a color noise environment. In fig. 2(a), at this time, M/N is less than 1, the relationship between the antenna array element number and the sample number meets the requirement of the classical asymptotic system, and under the condition of color noise, when the signal-to-noise ratio is more than 10dB, the signal source number estimation can be accurately realized by the SOSDI method and various methods with 100% probability, and comparatively speaking, the estimation effect of the bell circle method is still poor, which proves that the diagonal loading method provided in the scheme of the present invention can also improve the signal source estimation effect of the information theory criterion method under the condition of color noise; in the case of figure 2(b) of the drawings,the relation between the array element number of the antenna array and the sample number approximately meets the requirement of a classical asymptotic system, under the condition of color noise, when the signal-to-noise ratio is more than 17dB, the SOSDI method can accurately realize the source number estimation with the probability of 100%, but the Gerr circle method needs to be more than about 27dB to realize the source number estimation with the probability of 100%, and the estimation of other information theory criteria methods fails; in figures 2(c) and 2(d),the relation between the array element number of the antenna array and the sample number meets the requirement of a general asymptotic system, under the condition of white color noise, when the signal-to-noise ratio is more than 16dB and more than 17dB, the information source number estimation can be accurately realized with 100 percent probability, and other methods fail to estimate;

FIG. 3 shows the source number estimation comparison between the SOSDI method of the present invention and the source number estimation method based on the stochastic matrix theory (BIC-variant, EEE, LS-MDL, RMT-AIC, BN-AIC, ST-GBIC) under the mixed Gaussian noise condition. As can be seen from FIG. 3(a), at this time, M/N & lt 1, the relationship between the antenna array element number and the sample number meets the requirement of the classical asymptotic system, and when the signal-to-noise ratio is above 17dB, the SOSDI method can accurately realize the source number estimation with the probability of 100%, which is slightly inferior to other estimation methods; in the case of figure 3(b),the relation between the array element number of the antenna array and the sample number satisfies a general asymptotic bodyThe method is required to be capable of accurately realizing source number estimation with 100% probability when the signal-to-noise ratio is more than 12dB under the conditions of Gaussian white noise and mixed Gaussian noise, and is slightly superior to BN-AIC, EEE and RMT-AIC methods and slightly inferior to LS-MDL, BIC-variant and ST-GBIC methods; in the case of figure 3(c),the relation between the array element number of the antenna array and the sample number meets the requirement of a general asymptotic system, the technical scheme of the invention (SOSDI method) can accurately realize source number estimation with 100% probability when the signal-to-noise ratio is more than 19dB, the BIC-variant method fails in estimation, and the estimation effect of the invention is superior to that of the RMT-AIC method and is slightly inferior to that of other methods;

FIG. 4 is a comparison between the SOSDI method proposed by the present invention and the source number estimation method based on the stochastic matrix theory (BIC-variant, EEE, LS-MDL, RMT-AIC, BN-AIC, ST-GBIC) in the color noise environment. It can be seen from fig. 4(a) that, at this time, M/N is less than 1, the relationship between the array element number of the antenna array and the sample number meets the requirement of the classical asymptotic system, and under the condition of color noise, compared with the multiple source number estimation methods based on the random matrix theory, the SOSDI method can accurately realize the source number estimation with a probability of 100% when the signal-to-noise ratio is about 12dB or more, which is slightly inferior to the CRST-GBIC method and superior to other methods; in the case of figure 4(b),the relation between the array element number of the antenna array and the sample number approximately meets the requirement of a classical asymptotic system, under the condition of color noise, the SOSDI method can realize the source number estimation with the probability of more than 90% when the signal-to-noise ratio is more than 14dB, and the estimation of other methods fails; in figures 4(c) and 4(d),the relation between the array element number of the antenna array and the sample number belongs to a classical asymptotic system, under the condition of color noise, the SOSDI method can realize the source number estimation with the probability of 100% when the signal-to-noise ratio is respectively above 15dB and 14dB, and the estimation of other methods fails.

The comprehensive experiment results show that the information source number estimation method provided by the invention is not required to preset or assume the relationship between the array antenna array element number and the signal sampling number, and is suitable for information source number estimation under a classical asymptotic system and information source number estimation under a general asymptotic system; the method can be used for estimating the number of information sources in a Gaussian white noise environment and in a color noise environment, and particularly provides an effective solution for estimating the number of information sources in an observation signal mixed with superimposed color noise under a general asymptotic system. In conclusion, the invention can carry out blind estimation on the number of the information sources in the electromagnetic environment with more lack of prior information.