MIMO sonar reverberation suppression method and device

Through a compression perception-based method, combined with sparse STAP algorithm and regularization parameter estimation, the problems of reverb suppression and low-speed object detection in complex underwater environments are solved, and high-precision reverb suppression and target signal detection are achieved.

CN118962653AInactive Publication Date: 2024-11-15INST OF ACOUSTICS CHINESE ACAD OF SCI

Patent Information

Application Number
CN202411130479.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-16
Publication Date
2024-11-15
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing reverb suppression methods are not effective in complex underwater environments, especially in low-speed target detection, especially in shallow water environments, which leads to unsuitable statistical STAP methods and cannot effectively suppress reverbs. Impacts the detection performance of low-speed targets.

Method used

Using a reverb suppression method based on compression perception, the FDA-MIMO sonar target signal model and the space-time reverb signal model are constructed using a uniform circular array. The space-time spectrum estimation algorithm of weighted compression perception is combined with sparse STAP algorithm and regularized parameters. The estimation of NP is difficult to convert the l0 norm optimization problem into the solution l1 norm optimization problem, and the sparse coefficient is optimized to estimate distance, angle and velocity with high accuracy.

Benefits of technology

High-precision reverberation suppression and low-speed target detection in complex underwater environments, improving the detection performance of target signals, enhancing the robustness to noise, and obtaining high-precision data without requiring a large amount of sample data .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118962653A_ABST
    Figure CN118962653A_ABST
Patent Text Reader

Abstract

The invention relates to the field of civil and military monitoring and reconnaissance, and provides an MIMO sonar reverberation suppression method and device, and the method is characterized in that the method comprises the steps: transmitting and receiving an echo signal through an MIMO sonar, carrying out the filtering and sparsification of a received signal, representing an amplitude matrix as a sparse coefficient a, and representing the received signal as s = Ga + c + n; calculating a weight vector W of weighted compressed sensing according to the reverberation covariance matrix; constructing an optimization problem of weighted compressed sensing, and converting an l0 norm optimization problem into an l1 norm optimization problem; solving an optimization problem of weighted compressed sensing, and performing high-precision estimation on the distance, the angle and the speed according to the solved a; the method has good reverberation suppression performance, and can obtain high-precision space-time spectrum estimation for a low-speed moving target.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of civilian and military surveillance and reconnaissance, and in particular relates to a reverberation suppression method and device for MIMO sonar. Background Art

[0002] Multiple-input multiple-output (MIMO) sonar systems can be used for civil and military surveillance and reconnaissance. MIMO sonars effectively utilize the degrees of freedom in the time domain and space domain by transmitting orthogonal waveforms and using the array structure of transmitting and receiving array elements. MIMO systems are centralized and distributed. The present invention focuses on the centralized multiple-input multiple-output (MIMO) sonar. Underwater moving target detection and parameter estimation are the most important tasks of MIMO sonar systems. Underwater targets are measured by distance, angle, speed and reflection coefficient. Compared with traditional array sonars, MIMO sonars have higher angle and distance resolution. Using multiple pulses with a specified pulse repetition frequency (PRF), the speed of a moving target can be estimated in the Doppler domain. Low PRF means that the time interval between each pulse is too long, and underwater targets usually produce motion offsets within the time interval. The offset of the moving target causes the offset of the echo envelope in the multi-pulse accumulation time, resulting in energy accumulation and reduced detection performance. Modern sonars usually use high PRFs. However, when there is range ambiguity, high PRFs will cause range ambiguity, and the range of the target cannot be determined. In recent years, frequency diversity array (FDA) has attracted extensive attention to reduce the impact of range ambiguity on target detection. FDA uses a small frequency increment in the array elements. Due to the existence of the frequency increment, the FDA-MIMO system does not need to further solve the range ambiguity problem.

[0003] In shallow water sonar detection, the received signal contains reverberation caused by water surface reflection and other scatterers. Therefore, moving targets, especially low-speed targets, will be overwhelmed by the severely diffuse reverberation Doppler. Space-time adaptive processing (STAP) technology is considered to be an effective method to suppress Doppler spread reverberation in the joint angle-Doppler domain. In the past few decades, many statistical STAP methods have been developed. They mainly rely on independent and identically distributed (IID) quadratic range samples to estimate the statistical characteristics of reverberation. Inspired by the rapid development of compressed sensing (CS) theory, in order to suppress the severe heterogeneous characteristics of reverberation, direct data domain methods that do not require training data are proposed. Only test units are used instead of training data, thus avoiding non-stationarity in training data. However, the reverberation in shallow water environments is severely non-uniform, resulting in the reverberation spectrum changing with distance. The training samples are no longer independent and identically distributed in different distance units. Statistical STAP methods are no longer suitable for complex underwater environments.

[0004] The direct data domain method that does not require a large number of samples usually assumes that moving targets can be well separated from reverberation in the spatio-temporal spectrum. However, the severe broadening of the spatio-temporal spectrum of reverberation makes this assumption invalid. The low-speed targets and the extended reverberation spectrum are fused, resulting in poor reverberation suppression effect, which brings great challenges to the detection of low-speed targets. Therefore, the existing reverberation suppression methods cannot be directly applied to the detection of low-speed targets in complex underwater environments. Summary of the Invention

[0005] To solve the above problems, based on the FDA-MIMO sonar system, the present invention proposes a reverberation suppression method based on compressive sensing. The algorithm is based on a uniform circular array (UCA) and constructs an FDA-MIMO sonar target signal model and a spatio-temporal reverberation signal model. The spatio-temporal spectrum estimation algorithm based on weighted compressive sensing can estimate the range, angle, and velocity with high precision by using multi-dimensional information while maintaining the advantage of reverberation suppression. In the optimization process, a constant false alarm detector and an ML estimator are used to estimate the regularization parameter to ensure a certain robustness to noise parameters, including:

[0006] Transmit and receive echo signals using a MIMO sonar, filter and sparsify the received signals, represent the amplitude matrix as a sparse coefficient a, and represent the received signal as: s = Ga + c + n; where, is the result of vectorizing the amplitude matrix A column by column, represents the response vector at each point in the angle-Doppler grid, and represent reverberation and additive white Gaussian noise respectively. The grid on the angle-Doppler plane is H×K, where MNK < HK, that is, the number of targets is much smaller than the number of grids on the angle-Doppler plane, and the a is the sparse coefficient;

[0007] Use the sparse STAP algorithm to estimate the reverberation covariance matrix through the signals of range cells without targets, and calculate the weight vector W of weighted compressive sensing according to the reverberation covariance matrix, that is, W = R -1 G; the reverberation covariance matrix is used to describe the characteristics of the reverberation signal, and the weight vector W is a measurement matrix adjusted by the reverberation covariance matrix, which is used to filter reverberation when recovering the sparse coefficient a;

[0008] Solve the sparse coefficient a by optimizing the problem of the norm of weighted W, and transform the NP-hard l0 norm optimization problem into a solvable l1 norm optimization problem; the l1 norm optimization problem is

[0009] Solve the l1 norm optimization problem, and estimate the range, angle, and velocity according to the obtained a.

[0010] The filtering and sparse processing of the received signal comprises:

[0011] Perform matched filtering on the received signal;

[0012] The received signal is non-uniformly sampled and represented as a received data matrix Among them, A(i,R,θ,f d ) is the received signal amplitude of the target at position (R,θ) in the kth pulse, A vector for receiving a signal;

[0013] The received signal is represented by an H×K angle Doppler grid, and the sparse coefficient a is extracted. The received signal is represented as s=Ga+c+n, where a is the result of vectorizing the amplitude matrix A by column. represents the response vector at each point in the angle-Doppler grid, and represent reverberation and additive white Gaussian noise respectively, where H>MN.

[0014] The weight vector W of weighted compressed sensing calculated according to the reverberation covariance matrix includes:

[0015] The sparse coefficient a is solved using the l0 norm optimization problem for the distance units on both sides of the tested unit:

[0016] min||a||0,st||s-Ga||2≤ε

[0017] Solve the l0 norm optimization problem to obtain the angle Doppler spectrum of the range unit on both sides of the tested unit;

[0018] The reverberation covariance matrix is ​​estimated using the signals of the distance units on both sides of the unit under test:

[0019] R=ss H =Gaa H G H

[0020] Calculate the weight vector W based on the reverberation covariance matrix:

[0021] W=R -1 G

[0022] The l0-norm optimization problem can be solved by using methods including but not limited to greedy algorithms and iterative threshold algorithms.

[0023] The transformation of the NP-hard l0-norm optimization problem into a solvable l1-norm optimization problem includes:

[0024] The sparsity of Wa is expressed using the Laplace distribution density function:

[0025]

[0026] Denotes s with respect to a, G, σ 2 The probability density function of is:

[0027]

[0028] Substitute the probability density function of W and the probability density function of s into the problem of maximizing a using the maximum a posteriori probability method:

[0029]

[0030] The l0 optimization problem can be transformed into an l1 optimization problem:

[0031]

[0032] The solution to the l1 norm optimization problem includes:

[0033] The regularization parameter is estimated in the angle Doppler spectrum using the CFAR detector and the ML estimator, where the regularization parameter is η=2σ 2 γ, where σ 2 is the variance of the additive noise, and γ is the scale coefficient of the Laplace density function about Wa, which is:

[0034]

[0035] The initialization of a is completed through the delay superposition algorithm and fast Fourier transform;

[0036] Solve the norm optimization problem using sparse recovery algorithm;

[0037] The dynamic threshold method is used to detect multiple target peaks. and Obtain the detected target signal angle and speed parameters by solving Estimate the target's position.

[0038] The sparse recovery algorithm for solving the norm optimization problem may be any one of the OMP algorithm, the FOCUSS algorithm, the adaptive moment estimation algorithm and the BFGS method.

[0039] In a second aspect, a sparse reverberation suppression device for frequency diversity MIMO sonar is disclosed, comprising:

[0040] A signal processing unit, used for filtering and sparse processing the received signal, expressing the amplitude matrix as a sparse coefficient a, and expressing the received signal as: s = Ga + c + n;

[0041] A weight vector calculation unit, used to estimate the reverberation space-time spectrum and covariance matrix, and calculate the weight vector of the weighted sparse algorithm;

[0042] The optimization problem conversion unit is used to solve the sparse coefficient a based on the optimization problem of weighted compressed sensing, and convert the NP-hard l0-norm optimization problem into a solvable l1-norm optimization problem;

[0043] The optimization problem solving unit is used to optimize the regularization parameter and initialize a, solve the l1 norm optimization problem, and calculate the distance, angle and speed information of the target signal model from the optimization results.

[0044] 1. Since the present invention estimates the reverberation covariance matrix through the reverberation signals in the distance units on both sides of the target signal, calculates the weight vector W, and completes the calculation of weighted sparse recovery, a better reverberation suppression effect and high-precision parameter data are obtained.

[0045] 2. Since the present invention uses a compressed sensing algorithm, a large amount of independent and identically distributed sample data is not required, and high-precision data can be obtained even when the training samples are no longer independent and identically distributed in different distance units.

[0046] 3. Using the constant false alarm detector and ML estimator, a regularization parameter estimation method is proposed to ensure the robustness of the algorithm to reverberation and noise. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A flow chart of a reverberation suppression method for MIMO sonar provided by the present invention;

[0048] Figure 2 A schematic diagram of a MIMO array system of a MIMO sonar reverberation suppression method provided by the present invention;

[0049] Figure 3 1 is a comparison of the space-time spectrum and profile results of a specific embodiment of the present invention. DETAILED DESCRIPTION

[0050] The best mode of implementation of the present invention is described below by way of example. It should be understood that the specific embodiments herein are used to explain the present invention in detail and should not be construed as limiting the present invention. It should be noted that various changes and modifications may be made under the premise of following the principles and core scope of the present invention, and these changes should all be deemed to be within the scope of protection of the present invention. In conjunction with the accompanying drawings, the specific implementation steps of the present invention are described in detail.

[0051] The process of the conventional STAP algorithm is:

[0052] The STAP (Space-Time Adaptive Processing) algorithm is an advanced sonar signal processing technology used to suppress reverberation and improve target detection capabilities. The STAP algorithm effectively separates and suppresses reverberation and enhances target signals by jointly processing data in the spatial and temporal dimensions. The following are the general steps for the STAP algorithm to suppress reverberation:

[0053] Get the echo signal from the sonar receiver, select the range unit that contains reverberation but not the target signal as the training unit. The data of these units are used to estimate the reverberation covariance matrix, and organize the data of the training units into a training data matrix;

[0054] The reverberation covariance matrix is ​​calculated using the training data, and a target-oriented vector is constructed, which represents the desired target direction and speed characteristics. The adaptive filter weights W are calculated using the reverberation covariance matrix and the target-oriented vector;

[0055] Apply the filter weights to the designed filter to obtain the filtered signal;

[0056] Detect the filtered signal to determine whether there is a target. Common detection methods include constant false alarm rate (CFAR) detection. If a target is detected, further determine the target's azimuth, distance, speed and other parameters;

[0057] According to the newly received data, the reverberation covariance matrix is ​​dynamically updated to adapt to the changes in the environment. In each processing cycle, the filter weights are recalculated based on the latest covariance matrix and steering vector.

[0058] The STAP algorithm jointly processes data in spatial and temporal dimensions, dynamically estimates and updates the reverberation covariance matrix, and adaptively calculates the filter weights to effectively suppress reverberation and enhance the target signal.

[0059] The present invention is similar to the STAP algorithm in terms of ideas, including data collection, preprocessing, selecting distance units that contain reverberation but not targets as training units to obtain the reverberation covariance matrix, and using the reverberation covariance matrix and the target-oriented vector to calculate the adaptive filtering weights. However, the present invention introduces the idea of ​​compressed sensing, converts the signal amplitude matrix into sparse coefficients, and does not design a STAP filter when calculating the STAP weight vector W. Instead, the weight vector W is introduced during sparse recovery, thereby suppressing reverberation in the process of reconstructing the signal and achieving better results.

[0060] The present invention only uses the distance units on both sides of the unit to be detected to estimate the covariance matrix, which has adaptive capabilities. Therefore, compressed sensing can obtain high-precision data using a small number of training samples that are not independent and identically distributed.

[0061] Figure 1A flow chart of a reverberation suppression method for MIMO sonar provided by the present invention is shown in the figure. The method comprises the following steps:

[0062] Step 1: Use MIMO sonar to transmit and receive echo signals, filter and sparse the received signals, represent the amplitude matrix as sparse coefficient a, and represent the received signal as: s = Ga + c + n;.

[0063] In a specific embodiment, step one includes the following sub-steps:

[0064] Step 101: Based on a uniform circular array (UCA), according to the narrowband assumption, the FDA-MIMO sonar receiving signal is represented.

[0065] like Figure 2 As shown, the present invention uses a centralized FDA-MIMO system of a regular uniform circular array (UCA) composed of M transmitting array elements and N receiving array elements, and the array elements are evenly spaced along a circle with a radius A. t and A r are the radii of the emitting circular array respectively.

[0066] In a specific embodiment, it is assumed that K pulses are transmitted in total, and the pulse repetition time (PRT) is T. The pulse repetition time is the time interval between two consecutive pulses when a pulse train is transmitted.

[0067] Let f m =f0+Δf m (m=0,1,......,M-1), f0 is the carrier frequency, Δf m is the frequency increment of the mth array element, Ψ m (t) is the transmission waveform. Then the transmission signal at the mth array element in the kth pulse can be expressed as:

[0068] s m,k (t) = exp(j2πf m t)Ψ m (t) (1)

[0069] During signal propagation, information can be transmitted from one transmitter to multiple receivers, and at the same time it can be transmitted from another transmitter to different receivers. Considering this bidirectional propagation situation, the center point of the uniform circular array is selected as the reference.

[0070] Assume τ m,n,k is the time delay of the target echo from the mth transmitting element to the nth receiving element in the kth pulse, that is:

[0071]

[0072] Where c is the propagation velocity, V is the radial velocity, θ n =2πn / N(n=0,1,...,N-1), θ m =2πm / M(m=0,1,...,M-1).

[0073] For a far-field point target with coordinates (R, θ), the kth echo received by the nth array element can be written as:

[0074]

[0075] According to the narrowband assumption, assuming that the delay spread is much smaller than the signal period, the complex envelope delay of the signal can be considered to be approximately 2(R+VkT) / c, that is, Ψ m (t-τ m,n,k )=Ψ m (t-τ k ).

[0076] Substituting the assumption into the echo received signal expression, the received signal can be written as:

[0077]

[0078] Where a(R,θ) is the reflection coefficient, λ0=c / f0 is the wavelength, and f d =2V / λ0 is the Doppler frequency.

[0079] Since Δf m The induced Doppler frequency and array response can be ignored because its frequency increment is insignificant compared to the carrier frequency.

[0080] Step 102: Perform matched filtering on the received signal.

[0081] In order to improve the signal-to-noise ratio, matched filtering is also used in this method. For each receiving array element and pulse, the received signal is processed by M filters that match the transmitted waveform to generate M waveforms.

[0082] If the transmission signals between different transmitting array elements are correlated, then the signals of other transmitting array elements will remain after the matched filtering of different receiving array elements. That is, after the matched filtering, the output of the mth transmitting array element, the nth receiving array element, and the kth pulse is:

[0083]

[0084] Among them, psf m (t) = F -1 [|Ψ m (f)| 2 ], psf(·) is the point distribution function, Ψ m (f) is the waveform Ψ mThe spectrum of (t).

[0085] Step 103: Sample the received signal and construct a received data matrix.

[0086] Assume that the sampling rate of the signal is T s , at t = iT s (i=0,1,...,N r ) is sampled at, where N r is the number of range cells, ignoring the cross-correlation and range migration issues of the signals, the received data matrix of the target in the i-th range cell is rearranged as:

[0087]

[0088] Among them, A(i,R,θ,f d ) is the received signal amplitude of the target at the position (R,θ) in the kth pulse, specifically:

[0089]

[0090] in, represents the launch steering vector, Represents the received steering vector, g d (f d ) is the Doppler steering vector. is the guidance vector of the target at position (R,θ) in the kth pulse.

[0091] in, is the Kronecker product.

[0092] Specifically, unlike the traditional MIMO system, the transmission steering vector of the FDA-MIMO system is related to the range and angle and can be written as:

[0093] g t (R,θ)=g t (R)⊙g t (θ) (8)

[0094] Where ⊙ represents the Hadamard product, g t (R) = [exp(-j4πΔf m R / c)] T , (m=0,1,...,M-1) is the launch distance guidance vector, g t (θ) = {exp[j2πAcos(θ-θ m ) / λ0]} T , (m=0,1,...,M-1) is the transmission angle steering vector. The transmission steering vector describes the phase and amplitude distribution of the antenna array transmitting the beam in a specific direction.

[0095] g r (θ) = {exp[j2πAcos(θ-θ n ) / λ0]} T , (n=0,1,...,N-1) is the receiving angle steering vector, and the receiving steering vector describes the phase and amplitude distribution of the signal received by the antenna array in a specific direction.

[0096] g d (f d )=[exp(-j2πf d kT)] T , (k=1,2,...,K) is the Doppler steering vector, which is used to describe the frequency offset of the signal caused by the Doppler effect.

[0097] In the case of moving targets, the Doppler effect will cause frequency shifts. Targets in different range units will have different Doppler frequency shifts due to different relative speeds. Compensating the transmission range steering vector can partially correct these frequency shifts and reduce the errors caused by the Doppler effect. Therefore, g is performed on each range unit. t (R) compensation, that is, substitute the distance data of the unit to be tested into g t (R) = [exp(-j4πΔf m R / c)] T After the launch, the steering vector becomes g t (θ), in the kth pulse, the guidance vector of the point target at position (R,θ) becomes:

[0098]

[0099] Step 104: Establish a measurement matrix G and construct a multi-pulse FDA-MIMO signal model.

[0100] The received data cube set s(i) is stored as NMK×1 vectors, dividing the entire angle-Doppler plane into an H×K grid. The angle-Doppler grid is a two-dimensional matrix constructed during the sonar signal processing, where:

[0101] One dimension represents the azimuth (angle), which is usually obtained using the antenna array pattern. In the present invention, there are MN angles, and H>MN grids are designed, where Δθ represents the angular resolution after gridding. The discrete angle set is represented by θ={θ(0),...,θ(h),...,θ(H-1)}.

[0102] The other dimension represents the Doppler frequency, which reflects the radial velocity of the target. There are K of them in the present invention. The discrete Doppler set is represented by f d ={f d (0),...,fd (k),...,f d (K-1)}.

[0103] (θ,f d ) The target signal model at the position is rewritten in vector form as:

[0104] s=Ga(10)

[0105] Among them, a is the result of vectorizing the amplitude matrix A by column, that is, matrix Corresponding to the space-time spectrum, the target is represented by a non-zero element.

[0106] in, Indicates the received signal of the target.

[0107] in, Represents the response vector at each point in the angular-Doppler grid. The entire data matrix can be expressed as: (steering vector matrix)

[0108] G={g[θ(0),f d (0)],g[θ(1),f d (0)],...,

[0109] g[θ(H-1),f d (0)],g[θ(0),f d (1)],

[0110] g[θ(1),f d (1)],...,g[θ(H-1),f d (1)],

[0111] ...,g[θ(H-1),f d (K-1)]} (11)

[0113] In active sonar detection, reverberation is the main cause of degradation, especially in shallow water environments. In addition to the target signal, there are also a lot of reverberation signals and background noise in the received signal. The reverberation is strongly correlated with the target echo signal. The reverberation signal can be regarded as the superposition of echoes from multiple independent scatterers in an area with uniform angle distribution. In addition, background noise also seriously affects the detection performance of weak targets. Considering the cross-correlation sidelobes and noise, the following multi-pulse FDA-MIMO signal model is established:

[0114] s=Ga+c+n (12)

[0115] in, and They represent reverberation and additive white Gaussian noise respectively.

[0116] Since the product of the number of transmitting and receiving array elements is less than the number of angular grids, i.e., MN < H, and the number of measurements is less than the number of unknown coefficients, i.e., MNK < HK, recovering a from s is an ill-posed problem. Compressed sensing theory provides a way to recover an unknown signal from limited measurements with high probability. For target detection, since the number of targets is much less than the number of grids on the angular Doppler plane, the coefficient vector a represents strong sparsity.

[0117] Step 2: Calculate the weight vector W of the sparse STAP processor according to the reverberation covariance matrix.

[0118] There is a coupling problem with the measurement matrix G in the angular Doppler domain, specifically the aliasing effect between the angle and the Doppler frequency. The change in the Doppler frequency of one target may affect its angle estimation, and vice versa. Due to the coupling effect, it is difficult for the sonar system to accurately resolve the angle and velocity of the target simultaneously, resulting in a decrease in detection and positioning accuracy. Therefore, the present invention performs joint angle-Doppler spectrum estimation for each range cell.

[0119] Since there are no targets on both sides of the unit to be measured, the high-resolution reverberation spatio-temporal spectrum can be estimated through the range cells on both sides of the unit to be tested by the sparse STAP algorithm. An l0 optimization problem is constructed and solved to obtain a:

[0120] min||a||0, s.t.||s - Ga||2 ≤ ε (13)

[0121] Among them, any method that can solve the l0 optimization problem can be used, including but not limited to the matching pursuit algorithm, greedy algorithm, l1 relaxation method, non-convex optimization method, etc. This step is a conventional method, so no more explanation is given.

[0122] The reverberation covariance matrix is a matrix that describes the reverberation effect. Reverberation will cause the correlation between signals. Therefore, the reverberation covariance matrix can represent this correlation. In the present invention, the reverberation covariance matrix is expressed as:

[0123] R = ss H = Gaa H G H (14)

[0124] The weight vector of the STAP processor can be calculated by the following formula:

[0125] W = R -1 G (15)

[0126] In sparse reconstruction, the measurement matrix G is used to sample from the sparse signal and project the high-dimensional signal into a low-dimensional space. The reverberation covariance matrix R describes the effect of reverberation on the signal and represents the correlation between signal components. By dividing the measurement matrix by the reverberation covariance matrix (i.e., multiplying it by its inverse matrix), the "weight" of each measured signal is adjusted in a certain sense, so that in a reverberant environment, these measured signals can more accurately reflect the original signal. This adjustment allows the sparse reconstruction algorithm to better handle the effects of reverberation and optimize the stability and accuracy of the reconstruction process.

[0127] Step 3: Use the weighted W norm optimization problem to solve the sparse coefficient a, and transform the NP-hard l0 norm optimization problem into a solvable l1 norm optimization problem.

[0128] For FDA-MIMO sonar, in addition to reverberation, there are other interference factors that greatly weaken target detection. In FDA-MIMO systems, completely orthogonal waveforms do not exist in practice. Weak targets are covered by the correlated sidelobe levels of the signal. Environmental noise has become one of the main obstacles to target detection. Noise suppression is another important issue that affects the performance of underwater sonar systems. Therefore, in order to suppress reverberation, noise and waveform correlation, a weighted sparse algorithm is proposed to improve the target detection performance, especially for slow-moving targets.

[0129] In the range unit of the assumed moving target, the joint estimation of the angular and velocity parameters of the moving target can be obtained by l0 optimization:

[0130] min||Wa||0,st||s-Ga||2≤ε (16)

[0131] In a specific embodiment, the weight vector W is considered to be a known priori vector and can be estimated using equation 15. Solving 16l0 has been proven to be an NP-hard problem and needs to be relaxed to the optimization under l1. In the present invention, the introduced weight vector W is equivalent to the measurement matrix adjusted by the reverberation covariance matrix, so it can effectively weaken the reverberation during sparse reconstruction.

[0132] In the Bayesian framework, we can assume that the prior distribution of the parameters is a Laplace distribution. The probability density function of the Laplace distribution is peaked and has a peak near zero. This indicates that most of the samples of the Laplace distribution are concentrated near zero and have a long tail distribution. The samples of the Laplace distribution are more likely to be zero or close to zero, which helps to produce sparse solutions.

[0133] By transforming the sparsity problem into a problem of maximizing the likelihood under the Laplace distribution, we can naturally introduce the L1 norm to achieve sparse representation in the optimization problem.

[0134] Therefore, assuming that the coefficient vector obeys the Laplace distribution, according to Bayesian compressed sensing, the sparsity of the target Wa is expressed by the Laplace distribution density function as follows:

[0135]

[0136] Where γ is the scale factor of the Laplace density function with respect to Wa. In the sparse representation problem, if the coefficients are assumed to follow the Laplace distribution, the scale parameter controls the strength of the sparsity penalty term. A smaller value of γ imposes a stronger penalty on non-zero coefficients, thereby obtaining a sparser solution.

[0137] The noise is assumed to be zero-mean Gaussian noise, with n and unknown variance σ 2 The Gaussian probability density function is:

[0138]

[0139] The received signal gives a, G and σ 2 , the Gaussian probability density function about n can be transformed into the Gaussian probability density function about s by the above formula:

[0140]

[0141] In order to estimate a from the received signal s, the following maximization problem adopts the maximum a posteriori probability method (MAP). According to Bayes' theorem, the posterior probability can be expressed as:

[0142] P(a|s)=[f(s|a,G,σ 2 )·f(Wa;γ) (20)

[0143] By maximizing the posterior probability, the MAP estimate can be obtained:

[0144]

[0145] Taking the logarithm, the problem is equivalent to:

[0146]

[0147] Substituting 17 and 19 into 21, the MAP estimator is:

[0148]

[0149] Multiply 2σ in the formula 2 , extract the negative sign, and finally, the maximum value problem is equivalent to the l1 norm optimization problem:

[0150]

[0151] Where η = 2σ2 γ represents the regularization parameter, which controls the trade-off between the estimation error term and the sparsity constraint term. 2 is the variance of the additive noise.

[0152] Step 4: Solve the optimization problem of weighted compressed sensing, reconstruct the angle-Doppler spectrum, and make high-precision estimates of the distance, angle, and speed based on the solved a.

[0153] Step 401: Determine a regularization parameter η;

[0154] If η is set too large, weak targets and noise will be eliminated at the same time when reconstructing the angle-Doppler spectrum, leaving only strong targets. If η is set too small, some noise elements will remain in the spectrum. Therefore, the automatic selection of regularization parameters and the estimation of η from the angle-Doppler spectrum are key issues that need to be solved.

[0155] In the estimation of η, the target and reverberation are first separated from the noise in the initial space-time spectrum.

[0156] Specifically, the separation of target and reverberation from noise in the angle Doppler spectrum corresponds to the problem of detecting target in a noisy background, in which the CFAR detector (Constant False Alarm Detection) is straightforward. The working principle of the CFAR detector is that the signal with an amplitude greater than the CFAR threshold is determined as the target and reverberation components, and the rest are considered as noise components.

[0157] The noise should follow a zero-mean Gaussian distribution, and the noise variance estimate σ 2 is determined by the ML estimate. Second, the angular Doppler spectrum values ​​around zero Doppler are set to zero to roughly separate the target and reverberation. The target is defined as

[0158] In a specific embodiment, the CFAR device calculates a threshold value based on the signal strength of the target, reverberation and noise. The threshold value may be 0.5. Signals with amplitudes greater than the CFAR threshold value are determined to be target and reverberation components, and the rest are considered to be noise components.

[0159] γ is the scale coefficient of the Laplace density function about Wa, expressed as:

[0160]

[0161] In a specific embodiment, a CFAR detector is first used to separate the target and reverberation and noise parts in the angle Doppler spectrum, and the noise variance σ is estimated using an ML estimator. 2 According to the formula, we get the scale coefficient γ of the Laplace density function about Wa, and finally use η=2σ 2 γ determines the regularization parameter.

[0162] Step 402: Another optimization problem is the initialization of a. A good initialization can reduce the number of iterations and improve the accuracy of the algorithm.

[0163] In this method, the initial value a0 is used to estimate the target range after the matched filter, the DAS beamforming is used to determine the angle, and finally the target velocity is estimated in the slow time domain.

[0164] Step 403: Obtain a solution to the optimization problem by combining a regularization parameter with any algorithm such as the OMP algorithm, the FOCUSS algorithm, the adaptive moment estimation (ADaM) and the BFGS method, and restore the coefficient vector a.

[0165] Step 404: Convert the vector a into a matrix form A. The non-zero coefficients in the amplitude matrix A correspond to the coordinates in the angle Doppler spectrum. Indicates the angle, Represents the Doppler shift frequency.

[0166] Based on the relationship between the angle Doppler spectrum and the detection area coordinates, and The angle and speed parameters are obtained.

[0167] The simulation uses the dynamic threshold method to detect multiple target peaks. By finding the peaks on the spectrum, the angle and Doppler shift of the target can be estimated.

[0168] After coordinate transformation, the target position is estimated as:

[0169]

[0170] in, is the estimated range in the range angle domain. [x0,y0] is the UCA center position.

[0171] In a specific embodiment, the effectiveness of the weighted sparse enhancement (SE) algorithm in suppressing reverberation is verified by computer simulation. In the simulation, a linear frequency modulation (LFM) waveform is selected as the waveform of the FDA-MIMO sonar system, which has high pulse compression ratio, low correlation sidelobe level and good Doppler tolerance characteristics. Gaussian noise is added to the signal received by each array element to make the signal-to-noise ratio 0dB.

[0172] Figure 3 The figure shows a comparison of the space-time spectra and profiles obtained by the DAS algorithm (without reverberation suppression), the traditional direct data domain D3 algorithm, the STAP algorithm and the Weighted-SE algorithm provided by the present invention under the condition of a signal-to-noise ratio of 0.

[0173] Figure 3(a), (b), (c) and (d) are the spatial-temporal spectrum estimation diagrams of two targets in a reverberant environment using the DAS algorithm, D3 algorithm, STAP and Weighted-SE algorithm respectively.

[0174] like Figure 3 As shown in (a), without reverberation suppression, the target signal is very weak and blurred in the reverberation background, making it difficult to be effectively detected. Therefore, it is necessary to perform effective reverberation suppression on moving targets. Figure 3 (c) Figure 3 (b) has better reverberation suppression effect and the target signal is clearer, but due to the waveform correlation of FDA-MIMO sonar, there is still residual reverberation. Figure 3 (b) with Figure 3 (c) Both can clearly detect targets with a Doppler frequency of about -3 Hz, but have poor detection effects on slow-moving targets with a Doppler frequency of about -1 Hz. Compared with the above results, the algorithm proposed in the present invention can effectively suppress reverberation and waveform correlation and clearly detect two targets.

[0175] like Figure 3 (e) and (f) show the normalized Doppler estimation profiles of the two targets.

[0176] DAS has no reverberation suppression effect. By comparison, it can be seen that the Doppler estimation obtained by the D3 algorithm is only greatly suppressed near the target signal, and the suppression amplitude does not exceed 60dB. The Doppler estimation obtained by the STAP algorithm has a good reverberation suppression effect, and the reverberation amplitude is suppressed by more than 100dB, but there is still a target residual signal, and the target cannot be effectively detected under the background of multi-target interference. The reverberation amplitude of the Doppler estimation obtained by the Weighted-SE of the present invention is suppressed by more than about 100dB and 70dB respectively, and the target residual signal is also suppressed. The amplitudes of the two targets are basically the same, so the target can be accurately detected under the background of multi-target interference.

[0177] Figure 3 (g) and (h) are normalized angle estimation profiles of the two targets.

[0178] The results show that the DAS method, the D3 method and the STAP method all fail to achieve a good sidelobe suppression effect, while the method of the present invention has a higher angle estimation accuracy and suppresses the sidelobe by 70 dB.

[0179] The present invention also provides a sparse reverberation suppression device for frequency diversity MIMO sonar, comprising:

[0180] A signal processing unit, used for filtering and sparse processing the received signal, expressing the amplitude matrix as a sparse coefficient a, and expressing the received signal as: s = Ga + c + n;

[0181] A weight vector calculation unit, used to estimate the reverberation space-time spectrum and covariance matrix, and calculate the weight vector of the weighted sparse algorithm;

[0182] The optimization problem conversion unit is used to solve the sparse coefficient a based on the optimization problem of weighted compressed sensing, and convert the NP-hard l0-norm optimization problem into a solvable l1-norm optimization problem;

[0183] The optimization problem solving unit is used to optimize the regularization parameter and initialize a, solve the l1 norm optimization problem, and calculate the distance, angle and speed information of the target signal model from the optimization results.

[0184] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A reverberation suppression method for MIMO sonar, characterized in that: include: Transmit and receive echo signals using a MIMO sonar, filter and sparsify the received signals, represent the amplitude matrix as a sparse coefficient a, and represent the received signal as: s = Ga + c + n; where, is the result of vectorizing the amplitude matrix A column by column, represents the response vector at each point in the angle-Doppler grid, and represent reverberation and additive white Gaussian noise respectively. The grid on the angle-Doppler plane is H×K, where MNK < HK, that is, the number of targets is much smaller than the number of grids on the angle-Doppler plane, and the a is the sparse coefficient; The sparse STAP algorithm is used to estimate the reverberation covariance matrix through the range unit signal without the target, and the weight vector W of weighted compressed sensing is calculated according to the reverberation covariance matrix, that is, W = R -1 G; the reverberation covariance matrix is ​​used to describe the characteristics of the reverberation signal, and the weight vector W is a measurement matrix adjusted using the reverberation covariance matrix, which is used to filter the reverberation when restoring the sparse coefficient a; The sparse coefficient a is solved by using the weighted W norm optimization problem, which transforms the NP-hard l0 norm optimization problem into a solvable l1 norm optimization problem; the l1 norm optimization problem is Solve the l1 norm optimization problem, and estimate the distance, angle and speed based on the solved a.

2. The reverberation suppression method according to claim 1, characterized in that: The filtering and sparse processing of the received signal comprises: Perform matched filtering on the received signal; The received signal is non-uniformly sampled and represented as a received data matrix Among them, A(i,R,θ,f d ) is the received signal amplitude of the target at position (R,θ) in the kth pulse, A vector for receiving a signal; The received signal is represented by an H×K angle Doppler grid, and the sparse coefficient a is extracted. The received signal is represented as s=Ga+c+n, where a is the result of vectorizing the amplitude matrix A by column. represents the response vector at each point in the angle-Doppler grid, and represent reverberation and additive white Gaussian noise respectively, where H>MN.

3. The reverberation suppression method according to claim 1, characterized in that: The weight vector W of weighted compressed sensing calculated according to the reverberation covariance matrix includes: The sparse coefficient a is solved using the l0 norm optimization problem for the distance units on both sides of the tested unit: min||a||0,st||s-Ga||2≤ε Solve the l0 norm optimization problem to obtain the angle Doppler spectrum of the range unit on both sides of the tested unit; The reverberation covariance matrix is ​​estimated using the signals of the distance units on both sides of the unit under test: R=ss H =Gaa H G H Calculate the weight vector W based on the reverberation covariance matrix: W=R -1 G 4. The reverberation suppression method according to claim 3, characterized in that: The l0-norm optimization problem can be solved by using methods including but not limited to greedy algorithms and iterative threshold algorithms.

5. The reverberation suppression method according to claim 1, characterized in that: The transformation of the NP-hard l0-norm optimization problem into a solvable l1-norm optimization problem includes: The sparsity of Wa is expressed using the Laplace distribution density function: Denotes s with respect to a, G, σ 2 The probability density function of is: Substitute the probability density function of W and the probability density function of s into the problem of maximizing a using the maximum a posteriori probability method: The l0 optimization problem can be transformed into an l1 optimization problem:

6. The reverberation suppression method according to claim 1, characterized in that: The solution to the l1 norm optimization problem includes: The regularization parameter is estimated in the angle Doppler spectrum using the CFAR detector and the ML estimator, where the regularization parameter is η=2σ 2 γ, where σ 2 is the variance of the additive noise, and γ is the scale coefficient of the Laplace density function about Wa, which is: The initialization of a is completed through the delay superposition algorithm and fast Fourier transform; Solve the norm optimization problem using sparse recovery algorithm; The dynamic threshold method is used to detect multiple target peaks. and Obtain the detected target signal angle and speed parameters by solving Estimate the target's position.

7. The reverberation suppression method according to claim 6, characterized in that: The sparse recovery algorithm for solving the norm optimization problem may be any one of the OMP algorithm, the FOCUSS algorithm, the adaptive moment estimation algorithm and the BFGS method.

8. A reverberation suppression device for MIMO sonar, characterized in that: include: A signal processing unit, used for filtering and sparse processing the received signal, expressing the amplitude matrix as a sparse coefficient a, and expressing the received signal as: s = Ga + c + n; A weight vector calculation unit, used to estimate the reverberation space-time spectrum and covariance matrix, and calculate the weight vector of the weighted sparse algorithm; The optimization problem conversion unit is used to solve the sparse coefficient a based on the optimization problem of weighted compressed sensing, and convert the NP-hard l0-norm optimization problem into a solvable l1-norm optimization problem; The optimization problem solving unit is used to optimize the regularization parameter and initialize a, solve the l1 norm optimization problem, and calculate the distance, angle and speed information of the target signal model from the optimization results.

Citation Information

Patent Citations

  • Object detection method based on distribution type Phased-MIMO combined processing

    CN103713289A

  • Underwater space-time adaptive processing method and system based on sparse recovery

    CN111650574A

  • NMF active sonar anti-reverberation method based on sparse representation

    CN115097428A

  • Multi-frame joint detection method, system and terminal for moving target under deep sea reverberation background

    CN118151157A

  • Sensor system and method for determining target location using sparsity-based processing

    US20140111372A1

Cited By

  • Noise and interference suppression method for underwater acoustic communication transmission link

    CN121643933A

  • A method for noise and interference suppression for underwater acoustic communication transmission links

    CN121643933B