Source number estimation method and system based on deep learning under color noise and array error conditions

Abstract

The invention relates to the technical field of signal processing, in particular to an information source number estimation method and system based on deep learning under color noise and array error conditions, and the method comprises the steps: deploying a sensor receiving array; the method comprises the following steps: receiving signals of K far-field narrow-band information sources by using a sensor receiving array, then acquiring each information source signal to obtain N snapshots, and establishing an array output signal model containing amplitude-phase errors, mutual coupling errors and array element position disturbance errors and having color noise; obtaining a covariance matrix by using the array output signal, and establishing a covariance matrix feature extraction relational expression; extracting time sequence information of an array output signal by using a gating unit to construct a pseudo covariance matrix; and extracting features from the covariance matrix and the pseudo-covariance matrix, splicing the features, and inputting the spliced features into a neural network to obtain the number of incident information sources. According to the method, the time sequence information in the array original signal is utilized, so that the influence of color noise and array errors on the information source estimation performance can be effectively inhibited.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and particularly relates to a method and system for estimating the number of signal sources based on deep learning under the conditions of colored noise and array error. Background Art

[0002] Estimation of the number of signal sources is a classical problem in signal processing and has important applications in multiple fields such as wireless communication, array signal processing, and radar. In addition, the number of signal sources is an important prior information for blind signal separation and angle-of-arrival estimation, and its accuracy has a great impact on the performance of algorithms.

[0003] Classical algorithms for estimating the number of signal sources are various algorithms based on information-theoretic criteria. These mainly include the AIC criterion, the MDL criterion, and the BIC criterion. Such methods obtain the sampling covariance matrix through signal sampling data, then perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues, and use the eigenvalues to establish the maximum likelihood function. Classical algorithms for estimating the number of signal sources can estimate the number of signal sources well in the case of Gaussian white noise, but the algorithms fail in the colored noise environment. The Gerschgorin disk algorithm is another type of algorithm. This algorithm does not need to know the specific eigenvalues, but judges the positions of the eigenvalues through the Gerschgorin disk theorem to estimate the number of signal sources. The above algorithms are traditional model-driven methods, which highly depend on the accuracy of the established models, thus greatly limiting their applications in complex actual environments.

[0004] In recent years, with the rapid development of deep learning, many data-driven methods for estimating the number of signal sources based on deep learning have been proposed. Most of the existing algorithms for estimating the number of signal sources are for Gaussian white noise, and the algorithm performance drops sharply in the case of colored noise, and most algorithms do not consider the influence of array error. In addition, most deep learning-based algorithms use the covariance matrix and its eigenvalues as network inputs and do not fully utilize the information contained in the original signal data, resulting in a reduction in the performance of estimating the number of signal sources. Summary of the Invention

[0005] The present invention aims to solve the influence of colored noise and array error, and proposes a method and system for estimating the number of signal sources based on deep learning under the conditions of colored noise and array error. Since the timing information in the original array signal is utilized, the influence of colored noise and array error on the performance of signal source estimation can be effectively suppressed, and the accuracy of estimating the number of signal sources can be improved.

[0006] To achieve the above object, the technical solutions adopted are as follows:

[0007] The present invention provides a method for estimating the number of signal sources based on deep learning under the conditions of colored noise and array error, including:

[0008] Step 1, deploy a sensor receiving array;

[0009] Step 2: Use the sensor receiving array to receive the signals of K far-field narrowband sources, then collect N snapshots for each source signal, and establish an array output signal model containing amplitude-phase error, mutual coupling error, array element position perturbation error, and colored noise.

[0010] Step 3: Obtain the covariance matrix using the array output signal, and establish a covariance matrix feature extraction relationship.

[0011] Step 4: Use the gating unit to extract the timing information of the array output signal to construct a pseudo-covariance matrix.

[0012] Step 5: Extract features from the covariance matrix and the pseudo-covariance matrix respectively, and splice the features and input them into the neural network to obtain the number of incident sources.

[0013] According to the method for estimating the number of sources based on deep learning under the conditions of colored noise and array errors of the present invention, further, in step 2, the array output signal model containing amplitude-phase error, mutual coupling error, array element position perturbation error, and colored noise has the following expression:

[0014]

[0015] where F is the amplitude-phase error matrix, B is the mutual coupling error matrix, is the array manifold matrix in the presence of array element position perturbation error, S represents the received signal, and W is the array additive noise, which represents Gaussian white noise or colored noise.

[0016] According to the method for estimating the number of sources based on deep learning under the conditions of colored noise and array errors of the present invention, further, the expression for obtaining the covariance matrix using the array output signal in step 3 is:

[0017]

[0018] According to the method for estimating the number of sources based on deep learning under the conditions of colored noise and array errors of the present invention, further, in step 3, a covariance matrix feature extraction relationship is established, and the features include eigenvalues, Gerschgorin disk radii, and weighted Gerschgorin disk radii:

[0019] First, perform singular value decomposition on the covariance matrix to obtain:

[0020]

[0021] where U is the eigenvector matrix, Σ = diag(λ1, λ2,..., λ L ) is the eigenvalue matrix, and the eigenvalues satisfy λ1 ≥ λ2 ≥ … ≥ λK ≥ λ K+1 ≥ … ≥ λ L , where L is the number of elements in the receiving array;

[0022] Then, using the covariance matrix of the last column to expand the dimension to obtain Construct a unitary matrix V based on the eigenvector matrix U;

[0023] Use V to perform a unitary transformation to obtain:

[0024]

[0025] where the superscript * represents the conjugate;

[0026] Further define the following transformation matrix:

[0027]

[0028] Transform the matrix S to obtain:

[0029]

[0030] Finally, the eigenvector containing three characteristics of eigenvalue, Gerschgorin disk radius, and weighted Gerschgorin disk radius is obtained as:

[0031] f feature = [λ T , (|ρ|) T , (|κ|) T T

[0032] where f feature represents the eigenvector, λ represents the eigenvalue vector, |ρ| represents the Gerschgorin disk radius vector, and |κ| represents the weighted Gerschgorin disk radius vector.

[0033] According to the method for estimating the number of signal sources based on deep learning under the conditions of colored noise and array error of the present invention, further, the process of constructing the pseudo-covariance matrix in step 4 is as follows:

[0034] First, for the complex signal it is necessary to convert it into a real part matrix, and combine the real part and the imaginary part of the array output signal to obtain the input matrix;

[0035] Then, input the normalized input matrix into the gated unit to extract the timing information in the array output signal; subsequently, pass the output of the gated unit through a dense layer to match the data size with the size of the covariance matrix, and then rearrange the output result according to the arrangement of the real part and the imaginary part again, and recombine the real part and the imaginary part to obtain the pseudo-covariance matrix. ​

[0036] According to the method for estimating the number of signal sources based on deep learning under the conditions of colored noise and array error of the present invention, further, step 5 extracts features from the covariance matrix and the pseudo-covariance matrix respectively, and inputs the concatenated features into the neural network, including:

[0037] Using the covariance matrix feature extraction relationship established in step 3, the eigenvectors of the pseudo-covariance matrix and the covariance matrix are calculated respectively as:

[0038]

[0039]

[0040] Combining and normalizing the above features to obtain

[0041] According to the method for estimating the number of signal sources based on deep learning under the conditions of colored noise and array error of the present invention, further, the neural network adopts a fully connected network, and inputs the multi-dimensional features into the fully connected network, and the last fully connected layer and the sigmoid activation function output the confidence of each neuron corresponding to the number of signal sources.

[0042] Further, the present invention also provides a system for estimating the number of signal sources based on deep learning under the conditions of colored noise and array error, which is used to implement the method for estimating the number of signal sources based on deep learning under the conditions of colored noise and array error as described above. The system includes:

[0043] A receiving array deployment module for deploying a sensor receiving array;

[0044] An array output signal modeling module for receiving the signals of K far-field narrowband signal sources by using the sensor receiving array, then collecting N snapshots of each signal source signal, and establishing an array output signal model containing amplitude-phase error, mutual coupling error, array element position perturbation error, and colored noise;

[0045] A covariance matrix construction module for obtaining a covariance matrix by using the array output signal and establishing a covariance matrix feature extraction relationship;

[0046] A pseudo-covariance matrix construction module for extracting the timing information of the array output signal by using a gated unit to construct a pseudo-covariance matrix;

[0047] A signal source number estimation module for extracting features from the covariance matrix and the pseudo-covariance matrix respectively, and inputting the concatenated features into a neural network to obtain the number of incident signal sources.

[0048] Adopting the above technical solutions, the beneficial effects obtained are:

[0049] The present invention first deploys a receiving array, uses the receiving array to receive the signal data of multiple far-field narrowband signal sources, and establishes an array output signal model under the conditions of colored noise and array error; then, derives the relational expression for calculating the eigenvector from the covariance matrix; next, uses a gating unit to extract the timing information in the array output signal, and rearranges the output elements to obtain a pseudo-covariance matrix. The eigenvalues, Gerschgorin disk radii, and weighted Gerschgorin disk radii are respectively extracted from the pseudo-covariance matrix and the covariance matrix, the respective eigenvectors are constructed, and the eigenvectors are concatenated and then input into the subsequent fully connected network. The present invention makes full use of the original signal data and the features in the covariance matrix, and can effectively improve the performance of estimating the number of signal sources under the conditions of colored noise and array error. Description of the Drawings

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings of the embodiments of the present invention will be briefly introduced below. Among them, the drawings are only used to show some embodiments of the present invention, rather than limiting all embodiments of the present invention thereto.

[0051] Figure 1 It is a network framework diagram of a method for estimating the number of signal sources based on deep learning under the conditions of colored noise and array error according to an embodiment of the present invention;

[0052] Figure 2 It is a curve of the estimation accuracy of the number of signal sources of a multi-input network and a single-input network varying with the signal-to-noise ratio according to an embodiment of the present invention;

[0053] Figure 3 It is a curve of the estimation accuracy of the number of signal sources varying with the signal-to-noise ratio according to an embodiment of the present invention;

[0054] Figure 4 It is a curve of the estimation accuracy of the number of signal sources varying with the non-whitening level of the noise according to an embodiment of the present invention;

[0055] Figure 5 It is a curve of the estimation accuracy of the number of signal sources varying with the number of snapshots according to an embodiment of the present invention;

[0056] Figure 6 It is a curve of the estimation accuracy of the number of signal sources varying with the angular separation of the signal source incidence according to an embodiment of the present invention;

[0057] Figure 7 It is a curve of the estimation accuracy of the number of signal sources varying with the number of signal sources according to an embodiment of the present invention;

[0058] Figure 8 It is a curve of the estimation accuracy of the number of signal sources varying with the array error according to an embodiment of the present invention. Detailed Embodiments

[0059] In the following, the exemplary solutions of the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings of the specific embodiments of the present invention. Unless otherwise defined, the technical terms or scientific terms used in the present invention should have the ordinary meaning understood by those with ordinary skills in the art.

[0060] This embodiment discloses a method for estimating the number of signal sources based on deep learning - MFNet under the conditions of colored noise and array errors, as Figure 1 shown, which includes the following steps:

[0061] Step S101, deploy a sensor receiving array.

[0062] Place a uniform linear array with L array elements, and the element spacing is half a wavelength. This array has amplitude-phase errors, mutual coupling errors, and element position perturbation errors, and the environmental noise is colored noise or white noise.

[0063] Step S102, use the sensor receiving array to receive the signals of K far-field narrowband signal sources, and then collect N snapshots for each signal source. Due to the non-ideality of the receiving array, an array output signal model containing amplitude-phase errors, mutual coupling errors, element position perturbation errors, and colored noise is established.

[0064] For K narrowband signals, their arrival directions are θ k (k = 1, 2,..., K), where -90° ≤ θ k ≤ 90°. Without considering array errors, after receiving N snapshots, the array output signal can be expressed as:

[0065] X = AS + W (1)

[0066] where, is the array output, is the array manifold matrix; represents the received signal, s(t) = [s1(t), s2(t),..., s K (t)] T . is the array additive noise, which represents unknown Gaussian white noise or colored noise. The generation of colored noise is as follows:

[0067]

[0068] where, is the colored noise of the mth element at time t, w l (t) is the white noise of the lth element of the array at time t, w l (t) has a zero-mean Gaussian distribution. ζ m,l is the control coefficient, and its calculation method is:

[0069]

[0070] Among them, γ represents the non-whitening degree of the generated colored noise.

[0071] In the actual process, the reception of the array is not ideal, so there are various errors. The main errors existing in the array are: amplitude error, phase error, mutual coupling error, and array element position error. After considering the four errors, the array output signal model is:

[0072]

[0073] Among them, is the amplitude-phase error matrix, F1 represents the amplitude error, φ represents the phase error; the mutual coupling error matrix is:

[0074]

[0075] Array manifold matrix in the presence of array element position perturbation error d represents the array element spacing, Δd represents the array element position perturbation, and λ represents the wavelength.

[0076] Step S103: Obtain the covariance matrix using the array output signal with a finite number of snapshots, and establish a covariance matrix eigenvalue extraction relationship.

[0077] The covariance matrix estimation of the array output signal is:

[0078]

[0079] Next, it is necessary to derive the relationships for calculating the eigenvalues, Gerschgorin disk radii, and weighted Gerschgorin disk radii from the covariance matrix.

[0080] First, perform singular value decomposition on the covariance matrix to obtain:

[0081]

[0082] Among them, is the eigenvector matrix, is the eigenvalue matrix, where the eigenvalues satisfy λ1≥λ2≥…≥λ K ≥λ K+1 ≥…≥λ L . u l and λ l are the corresponding eigenvectors and eigenvalues. Expand the covariance matrix to obtain:

[0083]

[0084] Among them, r represents The last column. Construct a unitary matrix based on the eigenvector matrix U:

[0085]

[0086] Use V to Perform a unitary transformation to obtain:

[0087]

[0088] where the superscript * represents the conjugate.

[0089] Further define the following transformation matrix:

[0090]

[0091] Transform the matrix S to obtain:

[0092]

[0093] Finally, the eigenvector containing three characteristics of eigenvalues, Gerschgorin disk radii, and weighted Gerschgorin disk radii can be obtained as:

[0094] f feature = [λ T , (|ρ|) T , (|κ|) T T (13)

[0095] where f feature represents the eigenvector, λ T = [λ1, λ2,..., λ L T represents the eigenvalue vector, (|ρ|) T = [|ρ1|, |ρ2|,..., |ρ L |] T represents the Gerschgorin disk radius vector, (|κ|) T = [|κ1|, |κ2|,..., |κ L |] T represents the weighted Gerschgorin disk radius vector.

[0096] Step S104: Use a gated unit to extract the timing information of the array output signal to construct a pseudo-covariance matrix.

[0097] First, for the complex signal It is necessary to convert it into a real part matrix. Combine the real part and the imaginary part of the array output signal to obtain the input matrix as:

[0098]

[0099] Then, the normalized input matrix is input into the gating unit to extract the timing information in the array output signal. Subsequently, the output of the gating unit passes through a dense layer to ensure that the data size matches the size of the covariance matrix. Then, the output result is rearranged according to the arrangement of real and imaginary parts again, and the real and imaginary parts are recombined to obtain the pseudo-covariance matrix.

[0100] Step S105: Extract features from the covariance matrix and the pseudo-covariance matrix respectively, splice the features and input them into the neural network to obtain the number of incident signal sources. Convert the problem of signal source number estimation into a classification problem, divide the data set into a training set, a validation set and a test set, and use the training set and the validation set to obtain a trained neural network.

[0101] Using the covariance matrix feature extraction relational expression established in step S103, calculate the eigenvectors of the pseudo-covariance matrix and the covariance matrix respectively as:

[0102]

[0103] Combine and normalize the above features to obtain:

[0104]

[0105] The above multi-dimensional features are input into a fully connected network, which includes five fully connected layers, and each layer contains a fully connected layer and an activation function. The last fully connected layer and the sigmoid activation function output the confidence of each neuron corresponding to the number of signal sources.

[0106] Step S106: Input the test set data into the trained neural network, calculate the probability distribution at each number of signal sources, and select the maximum probability as the signal source number estimation result.

[0107] In the test stage, use the array output signal obtained in step S102, the pseudo-covariance matrix obtained in step S104, then calculate the fused eigenvector, and input the eigenvector into the fully connected network for classification prediction. Select the category with the highest confidence in the network output categories as the signal source number estimation result.

[0108] Corresponding to the above method, this embodiment also discloses a signal source number estimation system based on deep learning under the conditions of colored noise and array error, including:

[0109] A receiving array deployment module for deploying a sensor receiving array.

[0110] An array output signal modeling module is used to receive the signals of K far-field narrowband signal sources by using a sensor array, then collect N snapshots for each signal source, and establish an array output signal model containing amplitude-phase errors, mutual coupling errors, array element position perturbation errors, and colored noise.

[0111] A covariance matrix construction module is used to obtain a covariance matrix by using the array output signal and establish a covariance matrix feature extraction relationship.

[0112] A pseudo-covariance matrix construction module is used to extract the timing information of the array output signal by using a gating unit to construct a pseudo-covariance matrix.

[0113] A signal source number estimation module is used to extract features from the covariance matrix and the pseudo-covariance matrix respectively, splice the features and input them into a neural network to obtain the number of incident signal sources.

[0114] Figure 2 The following shows the comparison results of the method of the present invention with the networks that only input the array output signal and only input the covariance matrix. It can be seen that as the signal-to-noise ratio increases, the accuracy of the method of the present invention continuously improves and tends to be stable. Moreover, the accuracy of the method of the present invention is the highest under different signal-to-noise ratios.

[0115] Figures 3 to 8 The comparison results of the signal source number estimation accuracy between the method of the present invention and other methods are given in turn under different signal-to-noise ratios, different noise non-whitening levels, different snapshot numbers, different incident angle intervals, different signal source numbers, and different array error conditions. It can be seen from the results that the method of the present invention maintains the highest accuracy under different conditions.

[0116] The experimental results prove that the proposed method has better performance compared with the methods that only input the array output signal and only input the covariance matrix. In addition, compared with the traditional methods and the methods that only input the array output signal and only input the covariance matrix, the method of the present invention has stronger robustness to colored noise and array errors and can adapt to various scenarios.

[0117] Unless otherwise specifically stated, the components, steps, numerical expressions, and values described in these embodiments do not limit the scope of the present invention.

[0118] Each embodiment in this specification is described in a progressive manner. The key points of each embodiment are the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and the relevant parts can be referred to the description of the method part.

[0119] The units and method steps of the examples described in combination with the embodiments disclosed in this document can be implemented by electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those of ordinary skill in the art can use different methods to implement the described functions for each specific application, but such implementation is not considered to exceed the scope of the present invention.

[0120] Those of ordinary skill in the art can understand that all or part of the steps in the above methods can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a magnetic disk, or an optical disc, etc. Optionally, all or part of the steps of the above embodiments can also be implemented using one or more integrated circuits. Correspondingly, each module / unit in the above embodiments can be implemented in the form of hardware or in the form of a software functional module. The present invention is not limited to any specific form of the combination of hardware and software.

[0121] Finally, it should be noted that the above-described embodiments are only specific implementation manners of the present invention, used to illustrate the technical solutions of the present invention, rather than limiting them. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that any person skilled in the art within the technical scope disclosed by the present invention can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the protection scope of the claims.

Claims

1. A method for estimating the number of signal sources based on deep learning under the conditions of color noise and array error, characterized in that: Include: Step 1: deploy the sensor receiving array; Step 2: Receive the signals of K far-field narrowband sources using the sensor receiving array, then collect N snapshots of each source signal, and establish an array output signal model containing amplitude and phase errors, mutual coupling errors, array element position disturbance errors, and colored noise; Step 3: Obtain a covariance matrix using the array output signal, and establish a covariance matrix feature extraction relation; Step 4: Use the gating unit to extract the timing information of the array output signal to construct a pseudo covariance matrix; Step 5: Extract features from the covariance matrix and the pseudo-covariance matrix respectively, and concatenate the features and input them into the neural network to obtain the number of incident signal sources.

2. The method for estimating the number of signal sources based on deep learning under the conditions of color noise and array error according to claim 1, characterized in that: In step 2, an array output signal model containing amplitude and phase errors, mutual coupling errors, array element position disturbance errors, and colored noise is established. The expression is: Where F is the amplitude phase error matrix, B is the mutual coupling error matrix, is the array manifold matrix when there is an array element position disturbance error, S represents the received signal, and W is the array additive noise, which represents Gaussian white noise or colored noise.

3. The method for estimating the number of signal sources based on deep learning under the conditions of color noise and array error according to claim 2, characterized in that: In step 3, the expression of the covariance matrix obtained by using the array output signal is:

4. The method for estimating the number of signal sources based on deep learning under the conditions of color noise and array error according to claim 3, characterized in that: In step 3, the covariance matrix feature extraction relationship is established, which includes eigenvalues, Gais disk radius and weighted Gais disk radius: First, the covariance matrix Perform singular value decomposition to obtain: Among them, U is the eigenvector matrix, Σ=diag(λ1,λ2,...,λ L ) is the eigenvalue matrix, where the eigenvalues ​​satisfy λ1≥λ2≥…≥λ K ≥λ K+1 ≥…≥λ L , L is the number of receiving array elements; Then, using the covariance matrix The last column will be Expand the dimension Construct a unitary matrix V based on the eigenvector matrix U; Using V Performing a unitary transformation yields: Wherein, the superscript * indicates conjugation; Further define the following transformation matrix: Transform the matrix S to get: Finally, the feature vector containing three features, eigenvalue, Gaussian disk radius and weighted Gaussian disk radius, is obtained as follows: f feature =[λ T ,(|p|) T ,(|κ|) T ] T Among them, f feature represents the eigenvector, λ represents the eigenvalue vector, |ρ| represents the Gay disk radius vector, and |κ| represents the weighted Gay disk radius vector.

5. The method for estimating the number of signal sources based on deep learning under the conditions of color noise and array error according to claim 2, characterized in that: Step 4 The process of constructing the pseudo covariance matrix is: First, for complex signals It needs to be converted into a real matrix, and the real and imaginary parts of the array output signal are combined to obtain the input matrix; The normalized input matrix is ​​then input into the gating unit to extract the timing information in the array output signal; the output of the gating unit is then passed through a dense layer to match the data size with the size of the covariance matrix, and the output result is rearranged according to the arrangement of the real and imaginary parts, and the real and imaginary parts are recombined to obtain a pseudo-covariance matrix.

6. The method for estimating the number of signal sources based on deep learning under the conditions of color noise and array error according to claim 4, characterized in that: Step 5 extracts features from the covariance matrix and pseudo-covariance matrix respectively, and concatenates the features and inputs them into the neural network, including: The covariance matrix feature extraction relationship established in step 3 is used to calculate the eigenvectors of the pseudo-covariance matrix and the covariance matrix respectively: The above features are combined and normalized to obtain 7. The method for estimating the number of signal sources based on deep learning under the conditions of color noise and array error according to claim 6, characterized in that: The neural network adopts a fully connected network to transform multi-dimensional features The input is fed into the fully connected network, and the last fully connected layer and sigmoid activation function output the confidence of each neuron in the number of corresponding sources.

8. A source number estimation system based on deep learning under the conditions of color noise and array error, characterized in that: Used to implement the method for estimating the number of signal sources based on deep learning under the conditions of color noise and array error as described in any one of claims 1 to 7, the system comprises: A receiving array deployment module, used for deploying a sensor receiving array; The array output signal modeling module is used to use the sensor receiving array to receive the signals of K far-field narrowband sources, and then collect N snapshots of each source signal to establish an array output signal model containing amplitude and phase errors, mutual coupling errors, array element position disturbance errors, and colored noise; A covariance matrix building module is used to obtain a covariance matrix using an array output signal and to establish a covariance matrix feature extraction relation; A pseudo covariance matrix construction module is used to construct a pseudo covariance matrix by extracting timing information of array output signals using a gating unit; The information source number estimation module is used to extract features from the covariance matrix and the pseudo-covariance matrix respectively, and concatenate the features and input them into the neural network to obtain the number of incident information sources.

9. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images