Low-snapshot DOA (direction of arrival) estimation method, device and equipment for covariance matrix enhancement

Abstract

The invention relates to a covariance matrix enhanced low-snapshot DOA estimation method, device and equipment, and the method comprises the steps: capturing important space correlation information in a signal through a covariance matrix attention mechanism, and mapping a covariance matrix of an observation signal into an enhanced covariance matrix which can be used for DOA estimation through a subspace method through an automatic encoder. The problem that a traditional subspace method depends on an accurately estimated covariance matrix and incoherent signals is solved, DOA estimation is conducted through Root-MUSIC, model driving and data driving are combined on the whole, the whole system still has certain interpretability, compared with the traditional subspace method, the advantages of data driving can be fully utilized, and the DOA estimation efficiency is improved. The performance and robustness of DOA estimation are remarkably improved, meanwhile, the inherent interpretability of a classical subspace method is reserved, and good performance is shown under the conditions of low snapshots and coherent signal sources.

Description

Technical Field

[0001] This invention belongs to the field of radar array signal technology, and relates to a method, apparatus and equipment for low snapshot DOA estimation with covariance matrix enhancement. Background Technology

[0002] Direction of Arrival (DOA) estimation is one of the core research directions in the field of array signal processing, with wide applications in radar, sonar, wireless communication, and medical imaging. Its main purpose is to accurately determine the spatial orientation information of the incident signal source using the signal received by the sensor array. Traditional DOA estimation methods mainly include peak search methods and subspace-based methods. Peak search methods, such as beamforming (CBF), scan the entire space to find the direction with the highest signal energy, but their resolution is limited by the Rayleigh limit, making it difficult to effectively distinguish multiple signal sources when they are close together. Subspace-based methods, such as multiple signal classification (MUSIC) and rotation-invariant subspace parameter estimation (ESPRIT), separate the signal space from the noise space by decomposing the covariance matrix of the received signal, and use the orthogonality between the two to estimate the DOA. Theoretically, these methods can achieve super-resolution.

[0003] However, traditional subspace-based DOA estimation algorithms rely on accurately estimated covariance matrices and require incoherent echo signals. Typically, obtaining a high-precision covariance matrix requires a large amount of independent, identically distributed snapshot data. In practical applications, especially in complex low-altitude environments where targets move at high speeds, channels change rapidly, or resources are limited, it is often difficult to obtain sufficient snapshot data, resulting in the low snapshot problem. Furthermore, multipath effects and other deception interference sources often exist during signal reception, leading to fixed phase and amplitude relationships between multiple signals, thus presenting a signal coherence problem. In such cases, the estimation of the sample covariance matrix will have significant errors, even exhibiting rank defects, resulting in a significant deterioration in the DOA estimation performance of traditional subspace algorithms, or even rendering them ineffective. Summary of the Invention

[0004] To address the problems existing in the above-mentioned traditional technologies, this invention proposes a low-tempo DOA estimation method enhanced by covariance matrix, a low-tempo DOA estimation device enhanced by covariance matrix, and a computer device, which can significantly improve the performance and robustness of DOA estimation.

[0005] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:

[0006] On the one hand, a low-speed DOA estimation method with enhanced covariance matrix is ​​provided, including the following steps:

[0007] The raw low-shot signal data received by the uniform linear array composed of antennas is preprocessed to generate the sampling space covariance matrix.

[0008] A preliminary spatial feature enhancement of the sampling space covariance matrix is ​​performed using a covariance attention mechanism.

[0009] The enhanced sampling space covariance matrix is ​​flattened into a sequence and then input into a multi-head self-attention-based autoencoder to generate an enhanced covariance matrix.

[0010] The enhanced covariance matrix is ​​solved using the subspace method Root-MUSIC to obtain the estimated DOA angle of the original low-speed snapshot signal.

[0011] On the other hand, a low-speed DOA estimation device with enhanced covariance matrix is ​​also provided, comprising:

[0012] The preprocessing module is used to preprocess the low-shot raw signal data received by the uniform linear array composed of antennas to generate the sampling space covariance matrix.

[0013] The spatial enhancement module is used to perform preliminary spatial feature enhancement on the sampled spatial covariance matrix using a covariance attention mechanism;

[0014] The encoding enhancement module is used to flatten the enhanced sampling space covariance matrix into a sequence and input it into a multi-head self-attention-based autoencoder to generate an enhanced covariance matrix.

[0015] The output estimation module is used to solve the enhanced covariance matrix using the subspace method Root-MUSIC to obtain the estimated DOA angle of the original low-shot signal.

[0016] In another aspect, a computer device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the aforementioned low snapshot DOA estimation method with enhanced covariance matrix.

[0017] One of the above technical solutions has the following advantages and beneficial effects:

[0018] The aforementioned method, apparatus, and device for low-shot DOA estimation with enhanced covariance matrix utilizes a covariance matrix attention (CA) mechanism to capture important spatial correlation information in the signal. Then, an autoencoder maps the covariance matrix of the observed signal into an enhanced covariance matrix that can be used for DOA estimation using subspace methods. This solves the problem of traditional subspace methods' dependence on accurately estimated covariance matrices and incoherent signals. Root-MUSIC is then used for DOA estimation. Overall, the system combines model-driven and data-driven approaches, maintaining a degree of interpretability. Compared to traditional subspace methods, this approach not only fully leverages the advantages of data-driven methods, significantly improving the performance and robustness of DOA estimation, but also retains the inherent interpretability of classical subspace methods, exhibiting good performance in low-shot and coherent source scenarios. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the conventional technology, the drawings used in the description of the embodiments or the conventional technology will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating a low-speed DOA estimation method with enhanced covariance matrix in one embodiment.

[0021] Figure 2 This is a schematic diagram of the CMET algorithm provided by the present invention in one embodiment;

[0022] Figure 3 This is a schematic diagram comparing the MUSIC spectrum of the present invention and the traditional subspace method in one embodiment;

[0023] Figure 4 This is a schematic diagram of RMSPE for processing coherent sources under different signal-to-noise ratios of the present invention and other algorithms in one embodiment;

[0024] Figure 5 This is a schematic diagram of RMSPE for processing coherent sources under different snapshots of the present invention and other algorithms in one embodiment;

[0025] Figure 6 This is a block diagram of a low-speed DOA estimation device with enhanced covariance matrix in one embodiment. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention.

[0027] It should be noted that, in this document, the reference to "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The presentation of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will understand that the embodiments described herein can be combined with other embodiments. The term "and / or" as used herein refers to any combination of one or more of the associated listed items, and all possible combinations, including such combinations.

[0028] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0029] In one embodiment, such as Figure 1 As shown, a low-speed DOA estimation method with enhanced covariance matrix is ​​provided, which may include the following steps S12 to S18:

[0030] S12, preprocess the low-shot raw signal data received by the uniform linear array composed of antennas to generate the sampling space covariance matrix;

[0031] S14, using the covariance attention mechanism to perform preliminary spatial feature enhancement on the sampling space covariance matrix;

[0032] S16, flatten the enhanced sampling space covariance matrix into a sequence and input it into a multi-head self-attention-based autoencoder to generate the enhanced covariance matrix;

[0033] S18. The enhanced covariance matrix is ​​solved using the subspace method Root-MUSIC to obtain the DOA estimation angle of the original low-speed snapshot signal.

[0034] It is understandable that the low-tempo DOA estimation method with covariance matrix enhancement borrows the idea of ​​the multi-head self-attention (MHSA) mechanism in the ViT (Vision Transformer) encoder, and proposes a low-tempo DOA estimation technique based on the CMET (Covariance Matrix Enhancement Transformer) network, such as... Figure 2 As shown, this method utilizes the Covariance Matrix Attention (CA) mechanism to capture important spatial correlation information in the signal. Then, an autoencoder maps the covariance matrix of the observed signal to an enhanced covariance matrix that can be used for DOA estimation via subspace methods, thus addressing the dependence of traditional subspace methods on accurately estimated covariance matrices and incoherent signals. Root-MUSIC is then used for DOA estimation. Overall, the system combines model-driven and data-driven approaches, maintaining a degree of interpretability and addressing the lack of interpretability. Compared to traditional subspace methods, this method not only fully leverages the advantages of data-driven approaches, significantly improving the performance and robustness of DOA estimation, but also retains the inherent interpretability of classic subspace methods, exhibiting better performance in low-shot and coherent source conditions.

[0035] Specifically, the covariance matrix-enhanced low snapshot DOA estimation method described above may include the following processing steps:

[0036] The first step is to preprocess the received low-speed original signal.

[0037] Consider M narrowband far-field source signals (i.e., low-shot raw signals) incident from different directions onto a uniform linear array (ULA) consisting of N antennas. The signal model of the input signal received by the uniform linear array can be expressed as:

[0038]

[0039] Where θ=[θ1,...,θ M [ ] represents the signal incident angle; θ is the steering matrix; s(t) and n(t) are the transmitted signal and additive Gaussian noise vectors, respectively; T is the number of received snapshots. For a uniform linear array, a(θ) m This can be represented as:

[0040]

[0041] Where λ = c / f is the signal wavelength, and d is the distance between arrays. The sampling spatial covariance matrix (SCM) can be calculated as:

[0042]

[0043] The second step is to use the covariance attention mechanism to perform preliminary spatial feature enhancement on the input signal.

[0044] As can be understood, firstly, by separating the real and imaginary parts of the sampling space covariance matrix and concatenating them along the channel dimension, the input signal becomes... Where H, W, and C represent the height, width, and number of channels of the sampling space covariance matrix, respectively. Let be the real part of the sampling space covariance matrix. is the imaginary part of the sampling space covariance matrix.

[0045] Optionally, the activation function used when extracting features from the input signal using depthwise separable convolution is the inversely modified activation function. Specifically, the output features after extracting features from the input signal using depthwise separable convolution are:

[0046]

[0047] Among them, Z W and Z H The features of the spatial covariance matrix in the row and column directions are sampled respectively, F h (·) and F w (·) represents the channel convolution transform, F b1 (·),F b2 (·),F b3 (·),F b4 (·) represents the batch normalization layer, F c1 (·) and F c2 (·) is a 1×1 convolution operation, and AR(·) is the inverse correction activation function, defined as:

[0048] AR(x)=[ReLU(x),ReLU(-x)] (5) AR(·) ensures that both positive and negative values ​​can be propagated by splicing the positive and negative activation parts. Compared with the ReLU activation function, it is more suitable for processing data of the sample space covariance matrix.

[0049] Specifically, a perceptual attention matrix is ​​generated by utilizing the correlation information in the row and column directions of the sampling space covariance matrix aggregated by one-dimensional convolutional kernels. Then, Z... W and Z H Feature compression is performed to generate attention mechanism weights. The output of the attention mechanism weights is:

[0050]

[0051] Among them, g w and g hLet F(·) be the attention weight, F(·) be the fully connected layer, and σ(·) be the sigmoid activation function. The final overall output enhancement signal is:

[0052] Y = X × g h ×(g w ) T (7)

[0053] The superscript T indicates transpose.

[0054] The third step involves using an autoencoder based on multi-head self-attention to generate an enhanced covariance matrix.

[0055] Furthermore, the multi-head self-attention-based autoencoder includes a learnable positional encoder, a multi-head attention module, a feedforward neural network (FFN), and a deconvolutional decoder. The learnable positional encoder is used to preserve the spatial location information of the input features. The multi-head attention module is used to capture the complex relationships within the input sequence and then integrate the global information of the input covariance matrix. The feedforward neural network is used to capture the positive and negative correlation characteristics of the covariance matrix. The deconvolutional decoder is used to map the features back to the original spatial size output.

[0056] Specifically, the enhanced signal output in the second step The real and imaginary parts are connected, and then flattened into N sequences of length 2N as input to the Transformer encoder. Therefore, the input features... Then add learnable positional encoding (PE) to preserve spatial location information:

[0057] Z0 = Z + E pos (8)

[0058] Among them, E pos This is for positional encoding. Multi-head self-attention (MHSA) can improve a model's ability to capture complex relationships within a sequence; its basic form is as follows:

[0059]

[0060] Among them, Q i =W i Q Z0, K i =W i K Z0, V i =W i V Z0, i = 1, ..., h; W Q W K W V and W O Let d be a learnable weight matrix. kHere, is the dimension of the key vector, `head` is the attention head, `softmax` is the normalized exponential function, and `Concat()` is the connection function. The MHSA output is fed into a feed-forward network (FFN), which typically contains two linear transformations (fully connected layers) and a non-linear activation function. Similarly, an inverse activation function is used to better capture the positive and negative correlation characteristics of the covariance matrix.

[0061] FFN(Z)=Linear2(AR(Linear1(Z)))(10)

[0062] Ultimately, the Transformer encoder outputs a high-dimensional and abstract embedding vector, which does not directly correspond to the physical structure of the sampling space covariance matrix. The deconvolution decoder then maps the features back to the original N×2N spatial size output, and reconstructs a complex matrix by taking the first N rows as the real part and the remaining N rows as the imaginary part.

[0063] The fourth step is to use the subspace method to estimate the DOA output.

[0064] The subspace method employed is the existing Root-MUSIC algorithm, a direction estimation algorithm based on the root-MUSIC method. It utilizes the characteristics of the signal subspace to achieve direction estimation; the signal direction corresponds to the peak value of the MUSIC spectrum, i.e., the denominator is zero. Therefore, the goal of the Root-MUSIC algorithm is to find the root of the denominator polynomial P(z) of the MUSIC spectrum. The denominator term of the MUSIC spectrum can be expanded into the following form:

[0065]

[0066] in, It is a Hermitian matrix, U n g is the matrix formed by the eigenvectors of the noise subspace. k These are the combination coefficients of the elements in matrix G. To convert a polynomial into a form containing only positive powers, we can multiply it by z. N-1 The denominator polynomial of the MUSIC spectrum becomes:

[0067]

[0068] From all roots of the denominator polynomial P(z) of the MUSIC spectrum, select M roots (M being the number of signal sources) whose moduli are closest to the unit circle and lie inside the unit circle. These roots correspond to the signal's... θ is calculated using the following formula:

[0069]

[0070] Where arg(z) is the phase angle of the root z.

[0071] In some implementations, four simulation experiments are used to illustrate the effectiveness of the present invention in enhancing the covariance matrix and the performance of DOA estimation.

[0072] In the experimental design, the uniform linear array receiver model is a ULA composed of 8 array elements, and the number of incident signals is randomly distributed between [2,4]. The input signals are generated according to formula (1) and are all narrowband far-field source signals. The signal incident angle (DOA) is uniformly generated within (-π / 2, π / 2). The root mean square periodic error (RMSPE) is selected as the evaluation index for DOA estimation performance in the simulation experiment. The smaller the RMSPE, the better the DOA estimation performance.

[0073] Simulation Experiment 1 presents the DOA estimation results of the traditional subspace method and the covariance matrix-enhanced low-shot DOA estimation method (referred to as the proposed method) under low-shot coherent source conditions. The experiment used three fully coherent sources with a snapshot number of 2, a signal-to-noise ratio of 10 dB, and true angles of [10.32°, 27.75°, -30.64°]. A comparison is made between the original MUSIC spectrum of the signal and the MUSIC spectrum output by the proposed method. Figure 3 As shown, the traditional MUSIC spectrum exhibits blurry and broad peaks, failing to accurately estimate the angular directions of the three coherent signals. In contrast, the CMET-enhanced MUSIC spectrum curve clearly displays three sharp and precise peaks, which accurately correspond to the true signal directions indicated by the red dashed line. This demonstrates that the proposed method can effectively enhance the original covariance matrix and improve the accuracy of subsequent DOA estimation in low-frame-rate coherent source scenarios.

[0074] Simulation Experiment 2 presents the RMSPE of the proposed method and other methods (including traditional subspace methods Root-MUSIC and ESPRIT, and DNN-based estimation methods SubspaceNet and DA-MUSIC) for DOA estimation under different numbers of coherent sources. The experiments were conducted with fully coherent sources, a snapshot count of 2, and a signal-to-noise ratio of 10 dB. As shown in Table 1, under the extremely low snapshot condition with coherent sources, the performance levels of different methods show significant differences. Traditional subspace algorithms exhibit poor estimation performance, even failing to provide effective estimation. DNN-based estimation methods outperform traditional subspace algorithms, but their estimation performance decreases significantly with the increase in the number of sources. The proposed method, combined with the Root-MUSIC method (CMET+Root-MUSIC, abbreviated as CMET+R-MUSIC), achieves the lowest RMSPE under different numbers of sources, demonstrating the significant advantages and robustness of CMET under low snapshot conditions.

[0075] Table 1

[0076] Root mean square periodic error [rad] 2 sources 3 sources 4 sources Root-MUSIC 0.351 0.432 0.399 ESPRIT 0.519 0.453 0.407 DA-MUSIC 0.086 0.195 0.322 SubspaceNet 0.033 0.102 0.222 CMET+ESPRIT 0.039 0.056 0.111 CMET+Root-Music 0.017 0.039 0.106

[0077] Simulation Experiment 3 demonstrates the generalization ability of the proposed method and other methods under different noise levels. The experiment used three fully coherent sources with a snapshot number of 2, and the signal-to-noise ratio varied from 0 to 10 dB. Figure 4 This demonstrates the RMSPE of different algorithms processing coherent sources under different signal-to-noise ratios. For example... Figure 4 As shown, traditional subspace methods do not show significant improvement in DOA estimation performance with increasing signal-to-noise ratio. Compared to DNN-based SubspaceNet and DA-MUSIC, the CMET network proposed in this invention maintains the lowest RMSPE, indicating that the CMET network can effectively extract and enhance signal covariance information from noise under low snapshot conditions, and can provide stable performance even under extremely poor data quality.

[0078] Simulation Experiment 4 presents a performance analysis of the proposed method and other methods for DOA estimation of coherent sources under different snapshot numbers. The experiment used three fully coherent sources with a signal-to-noise ratio of 0 dB and a snapshot number ranging from 2 to 100. Figure 5 This demonstrates the RMSPE of different algorithms processing coherent sources under different snapshots. For example... Figure 5 As shown, traditional subspace algorithms are limited by coherent information sources, and the DOA estimation performance does not improve significantly with the increase of the number of snapshots. Other algorithms will decrease RMSPE with the increase of the number of snapshots until they tend to stabilize. Among them, the method proposed in this invention still has excellent performance at both low and high snapshot counts, demonstrating the universality and robustness of the proposed method under different snapshot count conditions.

[0079] It should be understood that, although the above process Figure 1 The steps in the diagram are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order in which these steps are executed; they can be performed in other orders. Furthermore, the above process... Figure 1 At least some of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0080] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0081] In one embodiment, such as Figure 6 As shown, a low-frame DOA estimation device 100 with covariance matrix enhancement is also provided, including a preprocessing module 11, a spatial enhancement module 13, an encoding enhancement module 15, and an estimation output module 17. The preprocessing module 11 preprocesses the original low-frame signal data received by a uniform linear array composed of antennas to generate a sampling spatial covariance matrix. The spatial enhancement module 13 uses a covariance attention mechanism to perform preliminary spatial feature enhancement on the sampling spatial covariance matrix. The encoding enhancement module 15 flattens the enhanced sampling spatial covariance matrix into a sequence and inputs it into an autoencoder based on multi-head self-attention to generate an enhanced covariance matrix. The estimation output module 17 uses the subspace method Root-MUSIC to solve the enhanced covariance matrix to obtain the estimated DOA angle of the original low-frame signal.

[0082] The aforementioned low-shot DOA estimation device 100 with enhanced covariance matrix utilizes a matrix attention mechanism to capture important spatial correlation information in the signal. Then, an autoencoder maps the covariance matrix of the observed signal into an enhanced covariance matrix that can be used for DOA estimation via subspace methods. This solves the problem of traditional subspace methods' dependence on accurately estimated covariance matrices and incoherent signals. Root-MUSIC is then used for DOA estimation. The overall approach combines model-driven and data-driven methods, maintaining a degree of interpretability. Compared to traditional subspace methods, it not only fully leverages the advantages of data-driven methods, significantly improving the performance and robustness of DOA estimation, but also retains the inherent interpretability of classical subspace methods, exhibiting good performance in low-shot and coherent source scenarios.

[0083] In one embodiment, the spatial enhancement module 13 generates a perceptual attention matrix by utilizing the correlation information in the row and column directions of the sampled spatial covariance matrix aggregated by one-dimensional convolution kernels.

[0084] In one embodiment, when the spatial enhancement module 13 performs preliminary spatial feature enhancement on the sampled spatial covariance matrix using the covariance attention mechanism, the activation function used for feature extraction of the input signal through depthwise separable convolution is the inverse correction activation function.

[0085] In one embodiment, the multi-head self-attention-based autoencoder includes a learnable positional encoder, a multi-head attention module, a feedforward neural network, and a deconvolutional decoder. The learnable positional encoder is used to preserve the spatial location information of the input features. The multi-head attention module is used to capture the complex relationships within the input sequence and then integrate the global information of the input covariance matrix. The feedforward neural network is used to capture the positive and negative correlation characteristics of the covariance matrix. The deconvolutional decoder is used to map the features back to the original spatial size output.

[0086] It is understood that the explanations of the features in the covariance matrix-enhanced low-shot DOA estimation device 100 can be understood by referring to the corresponding explanations in the embodiments of the covariance matrix-enhanced low-shot DOA estimation method. Each module in the covariance matrix-enhanced low-shot DOA estimation device 100 can be implemented entirely or partially through software, hardware, or a combination thereof. The components can be embedded in hardware or independently of a device with data processing capabilities, or stored in software in the memory of the aforementioned device, so that the processor can call and execute the operations corresponding to each module. The aforementioned device can be, but is not limited to, various types of computers already existing in the art.

[0087] In one embodiment, a computer device is also provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following processing steps: preprocessing the raw low-frame signal data received by a uniform linear array composed of antennas to generate a sampling spatial covariance matrix; performing preliminary spatial feature enhancement on the sampling spatial covariance matrix using a covariance attention mechanism; flattening the enhanced sampling spatial covariance matrix into a sequence and inputting it into an autoencoder based on multi-head self-attention to generate an enhanced covariance matrix; and solving the enhanced covariance matrix using the subspace method Root-MUSIC to obtain the estimated DOA angle of the raw low-frame signal.

[0088] In one embodiment, when the processor executes the computer program, it can also implement the steps or sub-steps added in the embodiments of the low snapshot DOA estimation method with enhanced covariance matrix described above.

[0089] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), memory bus DRAM (RDRAM), and interface DRAM (DRDRAM), etc.

[0090] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0091] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of protection of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and all such modifications and improvements fall within the scope of protection of the present invention.

Claims

1. A low-speed DOA estimation method with enhanced covariance matrix, characterized in that, Including the following steps: The raw low-shot signal data received by the uniform linear array composed of antennas is preprocessed to generate the sampling space covariance matrix. A preliminary spatial feature enhancement of the sampling space covariance matrix is ​​performed using a covariance attention mechanism. The enhanced sampling space covariance matrix is ​​flattened into a sequence and then input into a multi-head self-attention-based autoencoder to generate an enhanced covariance matrix. The enhanced covariance matrix is ​​solved using the subspace method Root-MUSIC to obtain the estimated DOA angle of the original low-speed snapshot signal.

2. The low-shot DOA estimation method with enhanced covariance matrix according to claim 1, characterized in that, In the step of performing preliminary spatial feature enhancement on the sampling space covariance matrix using the covariance attention mechanism, the perceptual attention matrix is ​​generated by using the correlation information in the row and column directions of the sampling space covariance matrix aggregated by one-dimensional convolution kernels.

3. The low-speed DOA estimation method with enhanced covariance matrix according to claim 2, characterized in that, In the step of performing preliminary spatial feature enhancement on the sampling space covariance matrix using the covariance attention mechanism, the activation function used for feature extraction of the input signal through depthwise separable convolution is the inverse correction activation function.

4. The low-shot DOA estimation method with enhanced covariance matrix according to any one of claims 1 to 3, characterized in that, The multi-head self-attention-based autoencoder includes a learnable positional encoder, a multi-head attention module, a feedforward neural network, and a deconvolutional decoder. The learnable positional encoder is used to preserve the spatial location information of the input features. The multi-head attention module is used to capture the complex relationships within the input sequence and then integrate the global information of the input covariance matrix. The feedforward neural network is used to capture the positive and negative correlation characteristics of the covariance matrix. The deconvolutional decoder is used to map the features back to the original spatial size output.

5. A low-speed DOA estimation device with enhanced covariance matrix, characterized in that, include: The preprocessing module is used to preprocess the low-shot raw signal data received by the uniform linear array composed of antennas to generate the sampling space covariance matrix. The spatial enhancement module is used to perform preliminary spatial feature enhancement on the sampled spatial covariance matrix using a covariance attention mechanism; The encoding enhancement module is used to flatten the enhanced sampling space covariance matrix into a sequence and input it into a multi-head self-attention-based autoencoder to generate an enhanced covariance matrix. The output estimation module is used to solve the enhanced covariance matrix using the subspace method Root-MUSIC to obtain the estimated DOA angle of the original low-shot signal.

6. The low-speed DOA estimation device with enhanced covariance matrix according to claim 5, characterized in that, The spatial augmentation module uses the correlation information in the row and column directions of the sampled spatial covariance matrix aggregated by one-dimensional convolution kernels to generate a perceptual attention matrix.

7. The low-speed DOA estimation device with enhanced covariance matrix according to claim 6, characterized in that, When the spatial augmentation module performs preliminary spatial feature enhancement on the sampled spatial covariance matrix using the covariance attention mechanism, the activation function used for feature extraction of the input signal through depthwise separable convolution is the inverse correction activation function.

8. The low-speed DOA estimation device with enhanced covariance matrix according to any one of claims 5 to 7, characterized in that, The multi-head self-attention-based autoencoder includes a learnable positional encoder, a multi-head attention module, a feedforward neural network, and a deconvolutional decoder. The learnable positional encoder is used to preserve the spatial location information of the input features. The multi-head attention module is used to capture the complex relationships within the input sequence and then integrate the global information of the input covariance matrix. The feedforward neural network is used to capture the positive and negative correlation characteristics of the covariance matrix. The deconvolutional decoder is used to map the features back to the original spatial size output.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the low snapshot DOA estimation method with covariance matrix enhancement as described in any one of claims 1 to 4.

Images