Array signal processing method and device, electronic equipment and storage medium

Abstract

The invention provides an array signal processing method and device, electronic equipment and a storage medium, and relates to the technical field of radar signal processing, receiving signals of a non-uniform array are mapped into receiving data of a virtual uniform array, and a covariance matrix is reconstructed by using a neural network model based on the data, so that the array signal processing efficiency is improved. And finally, estimating the direction of arrival of the signal through subspace decomposition and polynomial rooting operation, so that the problems of low estimation precision and narrow application range caused by the fact that a non-uniform array structure does not meet application conditions of a traditional algorithm, covariance matrix estimation is greatly influenced by signal conditions and subspace discrimination is insufficient in the prior art can be solved. The technical effects of adapting to a non-uniform array scene, improving the accuracy and robustness of covariance matrix estimation under a wide signal-to-noise ratio, enhancing the distinction degree of a signal and a noise subspace, further improving the direction-of-arrival estimation precision and resolving power and having physical interpretability are achieved.

Description

Array signal processing methods and apparatus, electronic devices and storage media Technical Field

[0001] This disclosure relates to the field of radar signal processing technology, and in particular to an array signal processing method and apparatus, electronic equipment and storage medium. Background Technology

[0002] Early digital beamforming methods were constrained by the Rayleigh limit, making it impossible to distinguish multiple targets within the same beamwidth. To overcome this limitation, subsequent methods, such as multiple signal classification and polynomial root-finding, emerged. These methods utilize the orthogonality of the signal and noise subspaces to achieve high-resolution estimation. The latter simplifies the spectral peak search step through polynomial root-finding, improving computational efficiency. However, these traditional subspace methods are model-driven, relying on strict assumptions such as narrowband signals, incoherent sources, a sufficient number of snapshots, and a high signal-to-noise ratio. In real-world scenarios, these assumptions are often difficult to satisfy, leading to a decrease in the discriminative power between the signal and noise subspaces after eigenvalue decomposition of the sample covariance matrix, directly weakening the angle-of-arrival estimation performance.

[0003] In recent years, neural network-based angle-of-arrival (AOA) estimation methods have demonstrated potential under complex conditions such as low snapshot counts and coherent sources due to their data-driven advantages. Among these, the correlation network method, which uses a convolutional neural network to learn the covariance matrix and combines it with an improved version of polynomial root-finding for estimation, outperforms traditional subspace methods and most deep learning baseline models while retaining the interpretability of subspace methods. However, this method still has significant limitations: it can only be trained under a fixed signal-to-noise ratio (SNR), failing to adapt to dynamic changes in SNR in real-world scenarios; using a fixed learning rate may cause it to skip the optimal solution; and due to the dependence of the improved polynomial root-finding method on specific structures, it is only applicable to uniform linear arrays and difficult to extend to non-uniform arrays, limiting its practical application scope. Summary of the Invention

[0004] This disclosure provides an array signal processing method and apparatus, electronic device, and storage medium. Its main objective is to at least partially solve one of the technical problems in the related art.

[0005] According to a first aspect of this disclosure, an array signal processing method is provided, comprising: mapping a received signal from a non-uniform array to received data from a virtual uniform array; reconstructing a covariance matrix based on the received data from the virtual uniform array using a neural network model; performing signal subspace and noise subspace decomposition on the reconstructed array covariance matrix, and performing root-finding operations based on a polynomial constructed from the noise subspace and the array manifold to estimate the direction of arrival of the signal.

[0006] Optionally, mapping the received signal of the non-uniform array to the received data of the virtual uniform array includes: calculating the array manifold transformation matrix using a least squares algorithm based on a preset angle observation interval and the configuration of the virtual uniform array; and using the array manifold transformation matrix to perform a linear transformation on the received signal of the non-uniform array to obtain the received data of the virtual uniform array.

[0007] Optionally, the step of reconstructing the covariance matrix using a neural network model based on the received data from the virtual uniform array includes: calculating a sample covariance matrix for the received data from the virtual uniform array and converting it into input data in real number form; inputting the input data in real number form into an autoencoder network containing an encoder and a decoder; the autoencoder network performing feature extraction and reconstruction on the input data and outputting a real number matrix for synthesizing a substitute covariance matrix.

[0008] Optionally, the encoder and decoder of the autoencoder network each contain at least one convolutional layer, and the network uses an activation function that can simultaneously preserve positive and negative feature information of the input data.

[0009] Optionally, the root-finding operation based on the construction of a polynomial using the noise subspace and the array manifold includes: constructing a polynomial with complex exponents as variables using the orthogonal projection matrix of the noise subspace and the guiding vector of the uniform linear array; solving for the roots of the polynomial and selecting the roots whose modulus is closest to the unit circle; and calculating the corresponding spatial angle based on the phase information of the roots as an estimate of the direction of arrival.

[0010] Optionally, the neural network model is trained using a hybrid dataset covering multiple signal-to-noise ratio conditions, and a dynamically adjusted learning rate strategy is employed during training.

[0011] According to a second aspect of this disclosure, an array signal processing apparatus is provided, comprising: a mapping unit for mapping received signals from a non-uniform array to received data from a virtual uniform array; a reconstruction unit for reconstructing a covariance matrix based on the received data from the virtual uniform array using a neural network model; and a computation unit for performing signal subspace and noise subspace decomposition on the reconstructed array covariance matrix, and performing root-finding operations based on a polynomial constructed from the noise subspace and the array manifold to estimate the direction of arrival of the signal.

[0012] Optionally, the mapping unit is further configured to: calculate the array manifold transformation matrix using a least squares algorithm based on a preset angle observation interval and the configuration of the virtual uniform array; and perform a linear transformation on the received signal of the non-uniform array using the array manifold transformation matrix to obtain the received data of the virtual uniform array.

[0013] Optionally, the reconstruction unit is further configured to: calculate the sample covariance matrix of the received data from the virtual uniform array and convert it into input data in real number form; input the input data in real number form to an autoencoder network containing an encoder and a decoder; the autoencoder network performs feature extraction and reconstruction on the input data and outputs a real number matrix for synthesizing a substitute covariance matrix.

[0014] Optionally, the encoder and decoder of the autoencoder network each contain at least one convolutional layer, and the network uses an activation function that can simultaneously preserve positive and negative feature information of the input data.

[0015] Optionally, the computation unit is further configured to: construct a polynomial with complex exponents as variables using the orthogonal projection matrix of the noise subspace and the steering vector of the uniform linear array; solve for the roots of the polynomial and select the roots whose modulus is closest to the unit circle; and calculate the corresponding spatial angle based on the phase information of the roots as an estimate of the direction of arrival.

[0016] Optionally, the neural network model is trained using a hybrid dataset covering multiple signal-to-noise ratio conditions, and a dynamically adjusted learning rate strategy is employed during training.

[0017] According to a third aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect above.

[0018] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are configured to cause the computer to perform the method described in the first aspect above.

[0019] According to a fifth aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the method described in the first aspect above.

[0020] The array signal processing method, apparatus, electronic device, and storage medium disclosed herein map the received signal of a non-uniform array to the received data of a virtual uniform array, then reconstruct the covariance matrix based on this data using a neural network model, and finally estimate the signal direction of arrival (DOA) through subspace decomposition and polynomial root-finding operations. Therefore, it can solve the problems of low estimation accuracy and narrow applicability in the prior art caused by the non-uniform array structure not meeting the applicable conditions of traditional algorithms, the large influence of signal conditions on covariance matrix estimation, and insufficient subspace discrimination. It achieves the technical effects of adapting to non-uniform array scenarios, improving the accuracy and robustness of covariance matrix estimation under wide signal-to-noise ratio, enhancing the discrimination between signal and noise subspaces, thereby improving the accuracy and resolution of DOA estimation, while also having physical interpretability.

[0021] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0022] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Specifically: Figure 1 is a flowchart illustrating an array signal processing method according to an embodiment of this disclosure; Figure 2 is a structural diagram illustrating an array signal processing device according to an embodiment of this disclosure; and Figure 3 is a schematic block diagram illustrating an example electronic device according to an embodiment of this disclosure. Detailed Implementation

[0023] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0024] The array signal processing method, apparatus, electronic device, and storage medium of the present disclosure are described below with reference to the accompanying drawings.

[0025] Figure 1 is a schematic flowchart of an array signal processing method provided in an embodiment of this disclosure.

[0026] As shown in Figure 1, the method includes the following steps: Step 101, mapping the received signal of the non-uniform array to the received data of the virtual uniform array.

[0027] In the embodiments of this disclosure, a preset array transformation strategy is used to map the received signal acquired by a non-uniform array to the received data of a virtual uniform array. The core objective is to address the compatibility issue between the non-uniform array structure and subsequent subspace-based angle estimation algorithms, ensuring that the transformed virtual uniform array data meets the array manifold structure requirements of these algorithms. Specifically, the transformation process first clarifies the structural parameters of the target virtual uniform array. Then, based on the element distribution characteristics of the non-uniform array and the angle observation range of the target signal, a signal mapping relationship is established between the non-uniform array and the virtual uniform array. This mapping relationship achieves an equivalent conversion of the received data from the two arrays, ensuring that the converted data accurately retains the angle information in the original signal. As one implementation method, this can be achieved by dividing the angle observation interval, constructing the array manifold matrices of the non-uniform array and the virtual uniform array, solving the interpolation matrix using the least squares method, and then using this interpolation matrix to transform the received signal of the non-uniform array to obtain the received data of the virtual uniform array.

[0028] Step 102: Based on the received data from the virtual uniform array, reconstruct the covariance matrix using a neural network model.

[0029] In the embodiments of this disclosure, based on the received data from the aforementioned virtual uniform array, the covariance matrix is ​​reconstructed using a preset neural network model. The core objective is to address the problem that traditional covariance matrix estimation methods are affected by the number of snapshots, signal-to-noise ratio, and signal coherence, leading to insufficient estimation accuracy and poor robustness. This process utilizes the feature extraction and data reconstruction capabilities of neural networks to mine deep correlation information of the signal from the received data of the virtual uniform array, filter redundant noise interference, and then output an optimized covariance matrix adapted to subsequent angle estimation algorithms. This ensures that the reconstructed matrix retains the angular feature information in the original signal while possessing good numerical characteristics and subspace separability. As one implementation method, the received data of the virtual uniform array can be preprocessed to convert it into an input format adapted to the neural network. Then, a neural network containing encoding and decoding structures is used to extract signal features and reconstruct the output matrix. After post-processing, a complex form of the alternative covariance matrix is ​​synthesized, providing reliable data support for subsequent subspace decomposition.

[0030] By leveraging the adaptability and feature learning capabilities of neural network models, the accuracy and robustness of covariance matrix estimation are significantly improved. It does not rely on strict assumptions such as a sufficient number of snapshots or a high signal-to-noise ratio, and can be adapted to coherent or incoherent signal scenarios. This provides a high-quality data foundation for subsequent subspace decomposition and angle estimation, thereby enhancing the overall method's adaptability and estimation accuracy under complex signal conditions.

[0031] Step 103: Decompose the reconstructed array covariance matrix into signal subspace and noise subspace, and perform root-finding operations on the polynomial constructed based on the noise subspace and array manifold to estimate the direction of arrival of the signal.

[0032] In the embodiments of this disclosure, the reconstructed array covariance matrix is ​​decomposed into signal and noise subspaces. Then, a polynomial is constructed based on the orthogonality of the noise subspace and the array manifold. The direction of arrival (DOA) of the signal is estimated through polynomial root-finding operations. The core objective is to address the limitations of traditional angle estimation methods in terms of resolution and susceptibility to noise interference. The core principle of this process is to utilize the inherent separation characteristics of the signal and noise subspaces, combined with the array manifold's ability to represent signal angle information, to transform the angle estimation problem into a polynomial root-finding mathematical problem. By solving for the polynomial roots that satisfy the orthogonality condition, the incident angle of the signal is derived in reverse. As one implementation method, eigenvalue decomposition can be used to separate the reconstructed covariance matrix into subspaces. An orthogonal polynomial is constructed based on the noise subspace and the array manifold vectors. The roots of this polynomial near the unit circle are solved, and the DOA estimation result is calculated through the mapping relationship between the roots and the angle.

[0033] By organically combining subspace decomposition and polynomial root finding, the physical interpretability and mathematical precision of the subspace method are fully utilized, effectively improving the resolution and accuracy of direction of arrival estimation. It can stably distinguish nearby signal sources, while reducing noise interference on the estimation results and ensuring the reliability of estimation in complex signal environments.

[0034] The array signal processing method disclosed herein maps the received signal of a non-uniform array to the received data of a virtual uniform array, then reconstructs the covariance matrix based on this data using a neural network model, and finally estimates the signal direction of arrival (DOA) through subspace decomposition and polynomial root-finding operations. Therefore, it can solve the problems of low estimation accuracy and narrow applicability in the prior art caused by the non-uniform array structure not meeting the applicable conditions of traditional algorithms, the large influence of signal conditions on covariance matrix estimation, and insufficient subspace discrimination. It achieves the technical effects of adapting to non-uniform array scenarios, improving the accuracy and robustness of covariance matrix estimation under wide signal-to-noise ratio, enhancing the discrimination between signal and noise subspaces, thereby improving the accuracy and resolution of DOA estimation, while also having physical interpretability.

[0035] As a specific embodiment of this disclosure, based on the basic scheme, the method of mapping the received signal of the non-uniform array to the received data of the virtual uniform array is further defined as follows: based on the preset angle observation interval and the configuration of the virtual uniform array, the array manifold transformation matrix is ​​calculated by the least squares algorithm; the received signal of the non-uniform array is linearly transformed using the array manifold transformation matrix to obtain the received data of the virtual uniform array.

[0036] Specifically, firstly, the preset angle observation range is defined, and the left boundary θ of this range is set according to the actual radar angle measurement requirements. l Right boundary θ The angle observation range is refined into a series of continuous angle point sets by dividing the grid with a spacing Δθ. Simultaneously, the configuration parameters of the virtual uniform array are determined, including the number of elements, element spacing, and array aperture, ensuring that its array manifold meets the structural requirements of subsequent subspace estimation algorithms. For each angle point within this angle observation range, the guiding vectors corresponding to the non-uniform array and the virtual uniform array are calculated, thereby constructing the array manifold matrix A of the non-uniform array and the array manifold matrix Ā of the virtual uniform array. With the objective of minimizing the error between the transformed array manifold and the target array manifold, a cost function min8||B is constructed. Given A - Ā||², the analytical solution B=(AA) of the cost function is obtained using the least squares algorithm. ) - ¹AĀ This yields the array manifold transformation matrix B. Finally, this array manifold transformation matrix B is linearly multiplied with the received signal from the non-uniform array. The form X completes the equivalent mapping from the non-uniform array received signal to the virtual uniform array received data, where X is the received signal matrix of the non-uniform array.

[0037] Solving the array manifold transformation matrix using the least squares algorithm can accurately establish the signal correspondence between the non-uniform array and the virtual uniform array, preserving the angular feature information in the original received signal to the maximum extent and avoiding signal distortion. At the same time, the linear transformation process is computationally efficient, ensuring that the received data of the transformed virtual uniform array strictly meets the Vandermonde structure requirements of subsequent subspace algorithms, providing a reliable guarantee for the accuracy of subsequent covariance matrix reconstruction and angle estimation.

[0038] As a specific embodiment of this disclosure, based on the basic scheme, the method of reconstructing the covariance matrix using the received data based on the virtual uniform array using a neural network model is further defined as follows: calculating the sample covariance matrix of the received data of the virtual uniform array and converting it into input data in real number form; inputting the input data in real number form into an autoencoder network containing an encoder and a decoder; the autoencoder network performing feature extraction and reconstruction on the input data and outputting a real number matrix for synthesizing a substitute covariance matrix.

[0039] Specifically, the sample covariance matrix is ​​calculated based on the received data from the virtual uniform array, according to the formula... Execution, in which Let T be the received signal vector of the virtual uniform array at the t-th snapshot, and T be the number of snapshots. This calculation yields a complex two-dimensional matrix whose dimension is related to the number of array elements. To adapt to the processing requirements of the neural network, the complex covariance matrix undergoes real-to-imaginary separation processing. The real and imaginary parts of the matrix are separated and concatenated row-wise, converting it into real-number input data of dimension 2N×N (N is the number of array elements in the virtual uniform array), ensuring that the data format meets the subsequent network input requirements. This real-number input data is fed into a preset autoencoder network, which consists of an encoder and a decoder. The encoder contains two layers of convolutional neural networks with a uniform kernel size of 2×2 and an increasing number of channels layer by layer, used to extract deep features of the signal from the input data and filter redundant noise. The decoder correspondingly sets two layers of deconvolutional neural networks with a decreasing number of channels layer by layer, used to reconstruct and restore the extracted features. The network training process uses the inverse rectifier activation function (AReLU), which concatenates the positive and negative parts of the input vector after taking ReLU, thus preserving the positive and negative numerical components in the input data. Finally, the decoder outputs a real matrix with a dimension of 2N×N, which is used to synthesize the subsequent replacement covariance matrix.

[0040] By first calculating the sample covariance matrix and converting it to a real number input format, the data size is effectively compressed, avoiding the problem of inefficient network training caused by the excessive dimensionality of the original signal. The combination of the autoencoder's convolution-deconvolution structure and the AReLU activation function can accurately extract signal features and retain key information, filter noise interference, and the output real number matrix can accurately support the synthesis of the replacement covariance matrix, providing a high-quality data foundation for subsequent angle estimation and significantly improving the accuracy and robustness of covariance matrix reconstruction.

[0041] As a specific embodiment of this disclosure, based on the basic scheme, the encoder and decoder of the autoencoder network are further defined to each contain at least one convolutional layer, and the network uses an activation function that can simultaneously preserve the positive and negative feature information of the input data.

[0042] Specifically, the encoder of the autoencoder network is constructed using at least one convolutional layer, and the decoder is constructed using at least one deconvolutional layer (i.e., a transposed convolutional layer). The kernel size of both the encoder and decoder is uniformly set to 2×2, a size that accurately captures the local phase correlation features of adjacent array element signals in the covariance matrix of the virtual uniform array. The number of channels in the encoder's convolutional layers increases layer by layer along the signal propagation direction to progressively extract deep, effective features from the input data and filter out noise interference. Conversely, the number of channels in the decoder's deconvolutional layers decreases layer by layer along the signal output direction to progressively reconstruct the extracted features into output data that conforms to the covariance matrix structure requirements. The network uses the inverse rectifier activation function (AReLU), which processes the positive and negative features in the input data separately. Specifically, it performs a ReLU operation on the positive part of the input vector (preserving positive values ​​and suppressing negative values), and simultaneously performs a ReLU operation on the negative part of the input vector after inverting it (preserving negative values ​​and suppressing positive values). The two processing results are then concatenated into an output vector, thus completely preserving the positive and negative features in the input data and adapting to the characteristic that sensor signals contain both positive and negative values.

[0043] By combining convolutional and deconvolutional layers, key angular features in the covariance matrix can be efficiently extracted and reconstructed. The 2×2 convolutional kernel can accurately capture local signal correlations. The AReLU activation function avoids the problem of losing negative feature information in traditional activation functions, making the network more suitable for the numerical characteristics of radar received signals. This significantly improves the accuracy and completeness of covariance matrix reconstruction, providing core support for the subsequent synthesis of alternative covariance matrices and the accuracy of angle estimation.

[0044] As a specific embodiment of this disclosure, based on the basic scheme, the root-finding operation based on the construction of a polynomial using the noise subspace and the array manifold is further defined as follows: constructing a polynomial with complex exponents as variables using the orthogonal projection matrix of the noise subspace and the guiding vector of the uniform linear array; solving for the roots of the polynomial and selecting the roots whose modulus is closest to the unit circle; and calculating the corresponding spatial angle based on the phase information of the roots as an estimate of the direction of arrival.

[0045] Specifically, the noise subspace obtained based on subspace decomposition Calculate its orthogonal projection matrix ,in This is a matrix composed of the eigenvectors of the noise subspace obtained after eigenvalue decomposition. It incorporates the guiding vector of a uniform linear array. (i.e., the specific representation of an array manifold, in which...) , d is the element spacing of the virtual uniform array. For the signal wavelength, (For the direction of arrival to be estimated), construct a complex exponential curve. polynomials with variables The polynomial is constructed based on the orthogonality principle between the steering vector and the noise subspace, and its roots have a direct mapping relationship with the direction of arrival. All complex roots of this polynomial are solved numerically. Since, ideally, the root modulus corresponding to the direction of arrival is 1 (i.e., it lies on the unit circle), and noise interference causes some roots to deviate from the unit circle, the roots with the smallest difference between their modulus and 1 (i.e., closest to the unit circle) are selected as the valid roots. Finally, the phase information of the valid roots is extracted. Substitute into the angle calculation formula The corresponding spatial angle is calculated, and this angle is the estimated value of the signal wave direction of arrival.

[0046] By constructing a polynomial using the orthogonal projection matrix of the noise subspace and the steering vector, the direction-of-arrival estimation is transformed into a precise mathematical root-finding problem, significantly improving the logicality and interpretability of the estimation. Filtering roots near the unit circle effectively eliminates invalid roots caused by noise interference, ensuring the reliability of the computational basis. Directly mapping angles based on the phase information of the roots simplifies the calculation process while ensuring estimation accuracy, enabling the method to maintain good resolution and stability even in complex noise environments.

[0047] As a specific embodiment of this disclosure, based on the basic scheme, the neural network model is further limited to being trained using a mixed dataset covering multiple signal-to-noise ratio conditions, and a dynamically adjusted learning rate strategy is adopted during the training process.

[0048] Specifically, the training dataset for the neural network model is a hybrid dataset covering various signal-to-noise ratio (SNR) conditions. This dataset contains both coherent and incoherent signal samples, with SNR conditions covering typical scenarios such as 10dB and 20dB to simulate the dynamic changes in signal and noise levels in real-world applications. During sample generation, the signal power and noise power are dynamically adjusted according to a set SNR (satisfying the formula...). Coherent signals are constructed by setting a fixed proportional relationship between the source signals, while each element of an incoherent signal follows a complex Gaussian distribution. The noise term also follows a complex Gaussian distribution. A total of 180,000 training samples were generated for the entire mixed dataset, with 45,000 samples corresponding to each signal-to-noise ratio condition. During training, the dataset was divided into training and validation sets at a 9:1 ratio for iterative optimization of model parameters and performance validation. The training process employed a stochastic gradient descent (SGD) optimizer, along with a dynamically adjusted learning rate strategy: the initial learning rate was set to... To ensure rapid convergence in the early stages of training; after every 40 training epochs, the current learning rate is reduced to 0.2 times the original value, and the total number of training epochs is set to 80. The model parameters are precisely optimized through fine-tuning of the learning rate in the later stages.

[0049] Hybrid datasets covering multiple signal-to-noise ratios enable neural network models to learn signal features under different noise levels, significantly improving the model's adaptability and generalization ability to wide signal-to-noise ratio scenarios, and allowing it to maintain stable performance even under unseen signal-to-noise ratio conditions. The dynamic learning rate strategy avoids the training oscillations caused by an excessively large fixed learning rate or the slow convergence caused by an excessively small fixed learning rate, while also balancing the convergence speed in the early stages of training with the optimization accuracy in the later stages. This effectively improves the stability of model training and the quality of the final covariance matrix reconstruction, providing a reliable guarantee for the accuracy of angle estimation.

[0050] It should be noted that the embodiments of this disclosure may include multiple steps. For ease of description, these steps are numbered, but these numbers are not a limitation on the execution time slots or execution order between the steps; these steps can be implemented in any order, and the embodiments of this disclosure do not limit this.

[0051] Corresponding to the array signal processing method described above, this disclosure also proposes an array signal processing apparatus. Since the apparatus embodiments of this disclosure correspond to the method embodiments described above, details not disclosed in the apparatus embodiments can be referred to the method embodiments described above, and will not be repeated here.

[0052] Figure 2 is a schematic diagram of an array signal processing device provided in an embodiment of this disclosure. As shown in Figure 2, it includes: a mapping unit 21, used to map the received signal of a non-uniform array to the received data of a virtual uniform array; a reconstruction unit 22, used to reconstruct the covariance matrix based on the received data of the virtual uniform array using a neural network model; and a computation unit 23, used to decompose the reconstructed array covariance matrix into signal subspace and noise subspace, and to perform root-finding operations based on the noise subspace and array manifold to construct polynomials to estimate the direction of arrival of the signal.

[0053] The array signal processing method disclosed herein maps the received signal of a non-uniform array to the received data of a virtual uniform array, then reconstructs the covariance matrix based on this data using a neural network model, and finally estimates the signal direction of arrival (DOA) through subspace decomposition and polynomial root-finding operations. Therefore, it can solve the problems of low estimation accuracy and narrow applicability in the prior art caused by the non-uniform array structure not meeting the applicable conditions of traditional algorithms, the large influence of signal conditions on covariance matrix estimation, and insufficient subspace discrimination. It achieves the technical effects of adapting to non-uniform array scenarios, improving the accuracy and robustness of covariance matrix estimation under wide signal-to-noise ratio, enhancing the discrimination between signal and noise subspaces, thereby improving the accuracy and resolution of DOA estimation, while also having physical interpretability.

[0054] Furthermore, in one possible implementation of this embodiment, the mapping unit 21 is also used to: calculate the array manifold transformation matrix using a least squares algorithm based on a preset angle observation interval and the configuration of the virtual uniform array; and use the array manifold transformation matrix to perform a linear transformation on the received signal of the non-uniform array to obtain the received data of the virtual uniform array.

[0055] Furthermore, in one possible implementation of this embodiment, the reconstruction unit 22 is also used to: calculate the sample covariance matrix of the received data of the virtual uniform array and convert it into input data in real number form; input the input data in real number form to an autoencoder network containing an encoder and a decoder; the autoencoder network performs feature extraction and reconstruction on the input data and outputs a real number matrix for synthesizing a substitute covariance matrix.

[0056] Furthermore, in one possible implementation of this embodiment, both the encoder and decoder of the autoencoder network contain at least one convolutional layer, and the network uses an activation function that can simultaneously preserve positive and negative feature information of the input data.

[0057] Furthermore, in one possible implementation of this embodiment, the operation unit 23 is also used to: construct a polynomial with complex exponents as variables using the orthogonal projection matrix of the noise subspace and the steering vector of the uniform linear array; solve for the roots of the polynomial and select the roots whose modulus is closest to the unit circle; calculate the corresponding spatial angle based on the phase information of the roots as an estimate of the direction of arrival.

[0058] Furthermore, in one possible implementation of this embodiment, the neural network model is trained using a mixed dataset covering multiple signal-to-noise ratio conditions, and a dynamically adjusted learning rate strategy is employed during the training process.

[0059] It should be noted that the foregoing explanation of the method embodiments also applies to the apparatus of this embodiment, and the principle is the same, so it is not limited in this embodiment.

[0060] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0061] Figure 3 illustrates a schematic block diagram of an example electronic device 300 that can be used to implement embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0062] As shown in Figure 3, the electronic device 300 includes a computing unit 301, which can perform various appropriate actions and processes based on a computer program stored in ROM (Read-Only Memory) 302 or a computer program loaded from storage unit 308 into RAM (Random Access Memory) 303. The RAM 303 can also store various programs and data required for the operation of the electronic device 300. The computing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An I / O (Input / Output) interface 305 is also connected to the bus 304.

[0063] Multiple components in electronic device 300 are connected to I / O interface 305, including: input unit 306, such as keyboard, mouse, etc.; output unit 307, such as various types of displays, speakers, etc.; storage unit 308, such as disk, optical disk, etc.; and communication unit 309, such as network card, modem, wireless transceiver, etc. Communication unit 309 allows electronic device 300 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0064] The computing unit 301 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, CPUs (Central Processing Units), GPUs (Graphics Processing Units), various special-purpose AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, DSPs (Digital Signal Processors), and any suitable processor, controller, microcontroller, etc. The computing unit 301 performs the various methods and processes described above, such as array signal processing methods. For example, in some embodiments, the array signal processing method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 308. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 300 via ROM 302 and / or communication unit 309. When the computer program is loaded into RAM 303 and executed by the computing unit 301, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the computing unit 301 may be configured to perform the aforementioned array signal processing method by any other suitable means (e.g., by means of firmware).

[0065] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application-Specific Standard Products), SOCs (System-on-Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0066] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0067] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0068] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0069] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include LANs (Local Area Networks), WANs (Wide Area Networks), the Internet, and blockchain networks.

[0070] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service system that addresses the shortcomings of traditional physical hosts and VPS (Virtual Private Server) services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.

[0071] It's important to note that artificial intelligence (AI) is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It encompasses both hardware and software technologies. AI hardware technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing. AI software technologies primarily include computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graph technologies.

[0072] The various numerical designations such as "first," "second," etc., used in this disclosure are merely for ease of description and are not intended to limit the scope of the embodiments of this disclosure, nor do they indicate a sequential order.

[0073] At least one of the features described in this disclosure can also be described as one or more, and multiple features can be two, three, four or more, and this disclosure does not impose any limitations. In the embodiments of this disclosure, for a technical feature, the technical features in that technical feature are distinguished by "first", "second", "third", "A", "B", "C" and "D", etc., and there is no sequential order or size order among the technical features described by "first", "second", "third", "A", "B", "C" and "D".

[0074] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0075] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. An array signal processing method, characterized in that, include: The received signal of the non-uniform array is mapped to the received data of the virtual uniform array; Based on the received data from the virtual uniform array, the covariance matrix is ​​reconstructed using a neural network model; the reconstructed array covariance matrix is ​​decomposed into signal subspace and noise subspace, and a root-finding operation is performed on a polynomial constructed based on the noise subspace and the array manifold to estimate the direction of arrival of the signal.

2. The method according to claim 1, characterized in that, The step of mapping the received signal of the non-uniform array to the received data of the virtual uniform array includes: calculating the array manifold transformation matrix using a least squares algorithm based on a preset angle observation interval and the configuration of the virtual uniform array; and using the array manifold transformation matrix to perform a linear transformation on the received signal of the non-uniform array to obtain the received data of the virtual uniform array.

3. The method according to claim 1, characterized in that, The method of reconstructing the covariance matrix using a neural network model based on the received data from the virtual uniform array includes: calculating the sample covariance matrix of the received data from the virtual uniform array and converting it into input data in real number form; inputting the input data in real number form into an autoencoder network containing an encoder and a decoder; the autoencoder network performing feature extraction and reconstruction on the input data and outputting a real number matrix for synthesizing a substitute covariance matrix.

4. The method according to claim 3, characterized in that, The encoder and decoder of the autoencoder network each contain at least one convolutional layer, and the network uses an activation function that can simultaneously preserve positive and negative feature information of the input data.

5. The method according to claim 1, characterized in that, The root-finding operation based on the construction of a polynomial using the noise subspace and the array manifold includes: constructing a polynomial with complex exponents as variables using the orthogonal projection matrix of the noise subspace and the guiding vector of the uniform linear array; solving for the roots of the polynomial and selecting the roots whose modulus is closest to the unit circle; and calculating the corresponding spatial angle based on the phase information of the roots as an estimate of the direction of arrival.

6. The method according to any one of claims 1 to 5, characterized in that, The neural network model is trained using a hybrid dataset covering various signal-to-noise ratio conditions, and a dynamically adjusted learning rate strategy is employed during the training process.

7. An array signal processing device, characterized in that, include: The mapping unit is used to map the received signal of the non-uniform array to the received data of the virtual uniform array; The reconstruction unit is used to reconstruct the covariance matrix based on the received data of the virtual uniform array using a neural network model; the operation unit is used to decompose the reconstructed array covariance matrix into signal subspace and noise subspace, and to perform root-finding operations based on the noise subspace and array manifold to construct polynomials to estimate the direction of arrival of the signal.

8. An electronic device, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-6.

9. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1-6.

Images