CN-122017727-A - Anti-fake peak frequency difference azimuth estimation method by using sparse Bayes learning

CN122017727ACN 122017727 ACN122017727 ACN 122017727ACN-122017727-A

Abstract

The disclosure provides an anti-fake peak-to-frequency difference azimuth estimation method by using sparse Bayesian learning, and relates to the technical field of underwater acoustic array signal processing. The method comprises the specific implementation modes of signal acquisition, selection of processing frequency points, construction of a dictionary matrix and a joint spectrum matrix, quantification of stability of signal intensity along with the change of a reference high-frequency point, optimization of the joint spectrum matrix according to the stability to eliminate false peaks, and obtaining of an azimuth estimation result. According to the technical scheme, the false peak can be effectively removed, the side lobe height in the azimuth spectrum can be reduced, and the accuracy of azimuth estimation of the target can be improved.

Inventors

LIU XIAOYAN
Niu Haiqiang
WANG HAIBIN

Assignees

中国科学院声学研究所

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (5)

1. An anti-spurious peak frequency difference azimuth estimation method utilizing sparse Bayes learning is characterized by comprising the following steps: acquiring an acoustic wave signal by an array element array to obtain a multi-snapshot array signal; l reference high-frequency points are selected from the high signal-to-noise ratio frequency band of the sound wave signal, and difference frequency values are set to obtain L frequency pairs, wherein L is a positive integer greater than 1; Constructing a dictionary matrix for representing complex amplitude response of each array element to the acoustic wave signals in each grid direction according to the array element position and the azimuth search parameter of the array element array, wherein the azimuth search parameter comprises a preset scanning angle domain and grid points, and the grid direction is a direction obtained by dispersing the preset scanning angle domain according to the preset grid points; According to the multi-snapshot array signals, the L frequency pairs and the dictionary matrix, carrying out frequency difference sparse signal reconstruction based on sparse Bayesian learning, and constructing a joint spectrum matrix for representing signal intensity of each reference high-frequency point acoustic wave signal in each grid direction; identifying and removing false signal intensity generated by frequency difference cross terms in the joint spectrum matrix by quantifying the stability of the signal intensity along with the change of the reference high-frequency points, so as to obtain an optimized joint spectrum matrix; and carrying out frequency domain averaging on the optimized combined spectrum matrix to obtain an azimuth spectrum, and further obtaining an azimuth estimation result.
2. The method of claim 1, wherein the constructing of the joint spectrum matrix and the optimizing of the joint spectrum matrix, wherein the constructing of the joint spectrum matrix comprises: constructing a differential feature matrix for each frequency pair according to the multi-snapshot array signals of the two corresponding frequency points; Initializing sparse Bayesian super-parameters of each reference high-frequency point, wherein the sparse Bayesian super-parameters comprise signal strength super-parameters and noise estimation super-parameters; Constructing a cross covariance matrix for each reference high-frequency point according to the dictionary matrix and the current sparse Bayes super-parameters; Performing a super-parameter iteration formula obtained by maximizing likelihood function derivation by using a fixed point iteration method by using a evidence maximization framework based on sparse Bayesian learning, and iteratively estimating signal strength super-parameters and noise estimation super-parameters to preset times; and constructing the joint spectrum matrix according to the signal intensity super-parameters iterated to the preset times.
3. The method according to claim 1 or 2, characterized by an optimization of a joint spectral matrix, wherein quantifying the stability of the signal strength as a function of the reference high frequency bin, comprises: setting a hard threshold value, setting 1 to the elements greater than or equal to the hard threshold value in the joint spectrum matrix, and keeping the original values of the other elements to obtain a part of binarized joint spectrum matrix; Randomly selecting at least one other grid column vector for the grid column vector in the partial binarization joint spectrum matrix, multiplying the grid column vectors element by element in sequence, and taking the average value of the result to obtain an updated partial binarization joint spectrum matrix; carrying out frequency domain summation on the updated partial binarization joint spectrum matrix to obtain a pseudo spectrum; And calculating a stability threshold based on the pseudo spectrum, and realizing the quantification of the stability.
4. A method according to claim 3, wherein the number of remaining randomly selected grid column vectors is equal to L.
5. The method of claim 4, wherein optimizing a joint spectrum matrix, wherein identifying and removing spurious signal strengths in the joint spectrum matrix due to frequency differential cross terms comprises: Removing grid column vectors corresponding to the false signal intensity from the current dictionary matrix according to the quantification of the stability to obtain an updated dictionary matrix; According to the updated dictionary matrix, performing one-time estimation on sparse Bayes super-parameters corresponding to real signal strength in the current joint spectrum matrix, and endowing minimum values with the signal strength super-parameters corresponding to false signal strength to obtain the updated joint spectrum matrix; Utilizing the updated joint spectrum matrix to perform the quantization of the stability, dictionary matrix updating and joint spectrum matrix updating again; When the grid column vector of the dictionary matrix is smaller than or equal to the preset vector number, the current joint spectrum matrix is the optimized joint spectrum matrix.

Description

Anti-fake peak frequency difference azimuth estimation method by using sparse Bayes learning Technical Field The disclosure relates to the technical field of underwater acoustic array signal processing, in particular to an anti-fake peak-to-frequency difference azimuth estimation method by using sparse Bayes learning. Background Direction of arrival (Direction Of Arrival, DOA) estimation refers to the process of finding the direction of sound source from the output spatial spectrum of a receiving sensor array, and is an important research content of array signal processing. Conventional beamforming (Conventional Beamforming, CBF) is a widely used array data processing method, the basic principle of which is a delay-sum mode, and has the advantages of small calculation amount and good robustness. However, for sparse arrays, when the array element spacing is greater than half the wavelength of the received signal, grating lobes can appear in the CBF method due to spatial aliasing, which seriously affects the azimuth estimation performance of the target. In addition, due to the fact that ocean environments are complex and changeable, acoustic velocity profile mismatch and array mismatch phenomena possibly exist, high-frequency processing is more sensitive to mismatch, and the problem of performance degradation easily occurs. Frequency differential processing (Frequency Difference, FD) is a method of suppressing grating lobes by reducing the operating frequency by combining information between multiple frequency points. The FD method multiplies the array receiving data of a group of high-frequency signals by element conjugation, simulates the frequency differenceThe performance and Frequency of Frequency differential beamforming (Frequency-DIFFERENCE BEAMFORMING, FDB) are as followsThe result of the CBF method is equivalent, and a new solution is provided for the failure scene of the traditional method. In recent years, the FD concept is also expanded to be applied to matching field processing, and is successfully applied to shallow sea and deep sea sound source positioning, and the FD method can effectively solve the problem of spatial aliasing in array signal processing and has good robustness in uncertain environments. However, the working frequency after the frequency difference processing is reduced, the direction-finding beam width is increased, the azimuth resolution capability of adjacent targets is reduced, meanwhile, the self-integration processing can amplify noise influence, so that beam sidelobes are increased, the detection performance is reduced under the condition of low signal-to-noise ratio, in addition, under the condition of multiple targets, the signals have additional cross terms after self-integration, so that the azimuth estimation result has false peaks, and the weak source detection in the strong source is greatly interfered. Disclosure of Invention The present disclosure provides an anti-spurious peak frequency difference azimuth estimation method using sparse Bayes learning, comprising: and acquiring acoustic signals by the array element array to obtain multi-snapshot array signals. L reference high-frequency points are selected from the high signal-to-noise ratio frequency band of the sound wave signal, and difference frequency values are set to obtain L frequency pairs, wherein L is a positive integer greater than 1. And constructing a dictionary matrix for representing complex amplitude response of each array element to the acoustic wave signals in each grid direction according to the array element position and the azimuth search parameter of each reference high-frequency point, wherein the azimuth search parameter comprises a preset scanning angle domain and grid points, and the grid direction is a direction obtained by dispersing the preset scanning angle domain according to the preset grid points. And carrying out frequency difference sparse signal reconstruction based on sparse Bayesian learning according to the multi-snapshot array signals, the L frequency pairs and the dictionary matrix, and constructing a joint spectrum matrix for representing the signal intensity of each reference high-frequency point acoustic wave signal in each grid direction. And (3) suppressing false signal intensity generated due to frequency difference cross terms in the joint spectrum matrix by quantifying the stability of the signal intensity along with the change of the reference high-frequency point, thereby obtaining the optimized joint spectrum matrix. And carrying out frequency domain averaging on the optimized joint spectrum matrix to obtain an azimuth spectrum, and further obtaining an azimuth estimation result. The matters described in this section are not intended to identify key or critical features of the embodiments of the disclosure, nor are they to be construed as limiting the scope of the disclosure. Other features of the present disclosure will be described in detail in the followin