Rapid reverberation suppression method based on orthogonal subspace model

Abstract

The invention provides a rapid reverberation suppression method based on an orthogonal subspace model, belongs to the field of underwater acoustic signal processing, and particularly belongs to the field of underwater active sonar detection. Aiming at the underwater target detection problem at a port and the like, reverberation and a target in multi-frame data are modeled into a low-rank matrix and a sparse matrix respectively according to the strong correlation among the multi-frame reverberation. And low-rank sparse decomposition is realized through robust orthogonal subspace learning, separation of reverberation and a target is completed, and the signal-to-mix ratio is improved. The effectiveness of the method provided by the invention is verified through a sea-necked sprain reverberation experiment and a Songhua river target detection experiment. Experimental results show that compared with an existing method, the detection performance of the method provided by the invention is competitive, and the calculation speed is greatly improved.

Description

Technical Field

[0001] The invention belongs to the field of underwater acoustic signal processing, more precisely to the category of underwater active sonar detection, and is mainly a fast reverberation suppression method based on an orthogonal subspace model. Background Art

[0002] Underwater early warning detection has always been a research focus in the field of hydroacoustics. As one of the main working modes of sonar, the performance of active sonar is often seriously disturbed by reverberation. Especially in ports and other places, the interference of reverberation is more intense, which seriously limits the range and detection performance of active sonar. Unlike the noise limit, the frequency band overlap between the reverberation and the transmitted signal is high and the waveform is similar. Many signal processing methods that perform well under noise often have unsatisfactory performance under the background of reverberation. Therefore, reverberation suppression has become an important research content in active sonar work. At present, the research on reverberation suppression can be mainly divided into the following categories: pre-whitening processing, subspace processing, array space processing, time-frequency domain processing, waveform design, etc. The above research can effectively suppress reverberation and improve the detection performance under the background of reverberation. However, most of the above research focuses on the reverberation characteristics in single-frame data, while ignoring the correlation between multi-frame reverberations. In the past decade, with the development of low-rank matrix recovery theory, a new idea for reverberation suppression of multi-frame data has been provided. In a stable hydroacoustic environment, the reverberation in multi-frame data obtained in a short time has a strong correlation, while the targets between multi-frame data are uncorrelated. Taking advantage of this difference, through the signal processing method of joint processing of multi-frame data, the reverberation and target can be separated and the signal-to-mix ratio can be improved through low-rank matrix recovery. Based on the above research ideas, Ge Fengxiang used the accelerated proximal gradient method (APG) to suppress the reverberation in the multi-frame data matrix. Liu Bing introduced the alternating direction multiplier method (ADMM) to achieve the separation of reverberation and target, and discussed in detail the influence of algorithm parameters on the reverberation suppression performance. Zhu Guangping improved the detection performance of active sonar in the reverberation background under the framework of variational Bayesian robust principal component analysis (VBRPCA), and the performance of the algorithm does not depend on a large number of parameter adjustments. The above research contents have performed very well in suppressing reverberation and greatly improved the detection performance of active sonar. However, the application of these methods in actual engineering is limited. The reason is not the lack of performance, but the problem faced is more the limitation of calculation speed. The above methods all involve a large amount of singular value decomposition or cyclic inference of multiple parameters, which brings a large amount of calculation. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art and to provide a fast reverberation suppression method based on an orthogonal subspace model. On the basis of the orthogonal subspace learning model, reverberation suppression is achieved, and the calculation speed is greatly improved without reducing the detection performance.

[0004] The object of the present invention is achieved through the following technical solution: A fast reverberation suppression method based on an orthogonal subspace model, comprising the following steps:

[0005] Step 1: Beamform the active sonar receiving signal to obtain a single frame of data Where N rr is the distance dimension, N θ is the angle dimension;

[0006] Step 2: Quantize the single frame data into columns and arrange multiple frames in sequence to form a multi-frame data matrix N is the number of frames of multi-frame data;

[0007] Step 3: Separate the reverberation and moving target energy in Y through a robust orthogonal subspace learning model;

[0008] Step 4: Perform inverse column quantization on the sparse matrix S output by the model, and output the detection result after reverberation suppression.

[0009] Furthermore, the robust orthogonal subspace learning model represents the low-rank matrix L of the reverberation through the orthogonal subspace, expressed as L=Oα, and the low-rank sparse decomposition method is expressed as:

[0010]

[0011] in, ‖·‖1 is the L1 norm of the matrix, λ is the trade-off parameter, I is the identity matrix, S is the sparse matrix of the target, k is a constant proportional to the rank of the matrix Y.

[0012] Furthermore, the mathematical model of the low-rank sparse decomposition is solved by the Lagrange multiplier method and converted into:

[0013]

[0014] Among them, μ is the regularization parameter and C is the Lagrange multiplier.

[0015] Furthermore, the alternating direction method is used to solve S:

[0016]

[0017] Where μ is the regularization parameter, C is the Lagrange multiplier, and the contraction operator

[0018] Furthermore, the block coordinate descent method is used to solve O and α:

[0019]

[0020] The contraction operator

[0021] The beneficial effects of the present invention are as follows: the present invention proposes a fast reverberation suppression method based on an orthogonal subspace model, firstly, the single frame data obtained after active sonar beamforming is column-vectorized, and the multi-frame data is arranged in sequence through joint processing of multi-frame data to obtain a multi-frame data matrix, wherein the horizontal axis is the number of frames, and the vertical axis is the product of the angle dimension and the distance dimension of the single frame data. According to the correlation of multi-frame reverberation, it can be considered that the reverberation between each column of the multi-frame data matrix is ​​strongly correlated. The moving target is represented as several data points with strong energy in the multi-frame data matrix, showing sparsity. Secondly, the robust orthogonal subspace learning algorithm (ROSL) is applied to represent the low rank of the low rank matrix through the row-1 norm of the matrix composed of the orthogonal subspace, so as to greatly reduce the dimension of the problem to be solved. The solution is completed by combining the Lagrange multiplier method with the block coordinate descent method to obtain a sparse matrix composed of the target. The obtained sparse matrix is ​​inversely column-vectorized to obtain the detection result after reverberation suppression. The method of the present invention can effectively separate the reverberation and target between multi-frame beam domain data and improve the signal-to-mix ratio. Experimental data from Vladivostok and Songhua River show that compared with existing methods, this method can significantly improve the calculation speed without reducing the reverberation suppression performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art or ordinary technicians, other drawings can be obtained based on these drawings without paying creative work.

[0023] Figure 1 This is a schematic diagram of the detection results before processing when the signal-to-mixture ratio is -5dB in the Vladivostok reverberation experiment.

[0024] Figure 2 It is a schematic diagram of the result after being processed by the method of the present invention when the signal-to-mixture ratio is -5dB in the reverberation experiment in Vladivostok.

[0025] Figure 3 The performance comparison chart of different methods under different signal-to-mixture ratios is shown in the figure.

[0026] Figure 4 The figure below is a comparison chart of the calculation speed of several detection methods.

[0027] Figure 5 This is the result of real target detection carried out in the Songhua River. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the protection scope of the present invention.

[0029] The present invention proposes a fast reverberation suppression method based on an orthogonal subspace model, comprising the following steps:

[0030] Step 1: Beamform the active sonar receiving signal to obtain a single frame of data Where N rr is the distance dimension, N θ is the angle dimension;

[0031] Step 2: Quantize the single frame data of active sonar into columns and arrange the multi-frame data in sequence to form a multi-frame data matrix Where m = N rr ×N θ , N rr is the distance dimension of a single frame of data, N θ is the angular dimension of a single frame of data, n=N, where N is the number of frames.

[0032] Step 3: Separate the reverberation and moving target energy in Y through a robust orthogonal subspace learning model;

[0033] In the orthogonal subspace, the low-rank matrix L composed of reverberation is expressed as L = Oα, and the low-rank sparse matrix

[0034] Decomposition problem into

[0035]

[0036] in ‖·‖1 is the L1 norm of the matrix, λ is the trade-off parameter, I is the identity matrix, S is the sparse matrix of the target, k is a constant proportional to the rank of the matrix Y.

[0037] Step 4: Perform inverse column quantization on the sparse matrix S output by the model, and output the detection result after reverberation suppression.

[0038] 1. Solve the convex optimization problem described by formula (1) by Lagrange multiplier method:

[0039]

[0040] Among them, μ is the regularization parameter and C is the Lagrange multiplier.

[0041] 2. Obtain the updated expression of S by the alternating direction method:

[0042]

[0043] Where μ is the regularization parameter, C is the Lagrange multiplier, and the contraction operator

[0044] 3. Obtain the updated expressions of O and α by block coordinate descent

[0045]

[0046] The amplitude contraction operator

[0047] 4. Iterate until convergence or the maximum number of iterations is reached, and then output the sparse matrix S consisting of the target.

[0048] 5. Perform inverse column quantization on the matrix and output the detection results of each frame.

[0049] like Figure 1-2 The processing results when the signal-to-mix ratio is -5dB in the Vladivostok reverberation experiment are shown. Figure 1 is the detection result before processing, Figure 2 The results after processing based on the orthogonal subspace model method. In the experiment, the target energy is added to the actual collected reverberation data, so that the signal-to-mix ratio of the target echo and the position of the target are known, which is convenient for quantitative evaluation of the performance of the method. The energy and distance of the beam diagram are normalized, and the target is circled in the beam diagram. From the results in the figure, it can be seen that this method can effectively suppress reverberation and improve the signal-to-mix ratio.

[0050] like Figure 3 The performance comparison diagram of different methods under different signal-to-mix ratios is shown in FIG. As can be seen from the figure, compared with the existing reverberation suppression methods of alternating direction multiplier method (ADMM), accelerated proximal gradient method (APG) and variational Bayesian robust principal component analysis method (VBRPCA), when the signal-to-mix ratio is high, the performance of this method is close to ADMM and VBRPCA and better than APG, while when the signal-to-mix ratio is low, the performance of this method is optimal. This means that the reverberation suppression ability of this method is not inferior to the existing multi-frame joint processing method.

[0051] like Figure 4 The calculation speed comparison diagram of several detection methods shown in FIG. From the result diagram, it can be seen that the calculation speed of this method has been significantly improved compared with the existing methods. By taking the average of multiple calculations, it can be found that the calculation speed of this method has been improved by 91% compared with other methods.

[0052] like Figure 5 The figure shows the actual target detection results carried out in the Songhua River. Figure 5(a) is the result before treatment, Figure 5 (b) is the result after being processed by this method. The target selected in the experiment is a small ROV, which moves in an irregular curve. From the result graph, it can be seen that under real target detection, this method is effective and can suppress reverberation well.

[0053] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A fast reverberation suppression method based on an orthogonal subspace model, characterized in that: The steps include: Step 1: Beamform the active sonar receiving signal to obtain a single frame of data Where N r is the distance dimension, N θ is the angle dimension; Step 2: Quantize the single frame data into columns and arrange multiple frames in sequence to form a multi-frame data matrix N is the number of frames of multi-frame data; Step 3: Separate the reverberation and moving target energy in Y through a robust orthogonal subspace learning model; Step 4: Perform inverse column quantization on the sparse matrix S output by the model, and output the detection result after reverberation suppression.

2. The fast reverberation suppression method based on the orthogonal subspace model according to claim 1, characterized in that: The robust orthogonal subspace learning model represents the low-rank matrix L of the reverberation through the orthogonal subspace, expressed as L=Oα, and the low-rank sparse decomposition method is expressed as: in, ‖·‖1 is the L1 norm of the matrix, λ is the trade-off parameter, I is the identity matrix, S is the sparse matrix of the target, k is a constant proportional to the rank of the matrix Y.

3. The fast reverberation suppression method based on the orthogonal subspace model according to claim 2, characterized in that: The mathematical model of the low-rank sparse decomposition is solved by the Lagrange multiplier method and converted into: Among them, μ is the regularization parameter and C is the Lagrange multiplier.

4. The fast reverberation suppression method based on the orthogonal subspace model according to claim 3 is characterized in that: The alternating direction method is used to solve S: Where μ is the regularization parameter, C is the Lagrange multiplier, and the contraction operator 5. The fast reverberation suppression method based on the orthogonal subspace model according to claim 4 is characterized in that: Using block coordinate descent method to solve O and α, we can get: The contraction operator

Images