Robust DOA Estimation Method Based on Fully Connected Neural Networks

Abstract

This invention specifically relates to a robust DOA estimation method based on a fully connected neural network, comprising: utilizing the far-field narrowband signals received by each element in a sensor array; constructing an observation data vector based on the far-field narrowband signals received by each element; determining a steering vector based on the angle of arrival of the far-field narrowband signals; constructing a covariance matrix based on the observation data vector; using the covariance matrix as a model output parameter and inputting it into a trained phase error estimation model based on a fully connected neural network to obtain the phase error estimate output by the model; determining the corresponding phase error estimation matrix based on the phase error estimate; compensating the steering vector using the phase error estimation matrix to obtain a calibrated steering vector; constructing a MUSIC spatial spectrum function by combining the calibrated steering vector and the noise subspace; searching for spectral peaks of the MUSIC spatial spectrum function within the angle search range; and determining the DOA estimate of the signal source based on the spectral peak search results. This method can maintain high DOA estimation accuracy.

Description

Technical Field

[0001] This invention relates to the field of array signal processing technology, and more specifically to a robust DOA estimation method based on a fully connected neural network. Background Technology

[0002] In related technologies, Direction of Arrival (DOA) estimation is a major research topic in the field of array signal processing. Commonly used DOA estimation methods include traditional beamforming methods, subspace-based methods, maximum likelihood methods, and array correction methods. Among these, the Multiple Signal Classification (MUSIC) algorithm based on subspace methods has been widely studied due to its high resolution. The MUSIC algorithm performs eigenvalue decomposition on the covariance matrix of the array received signal, dividing the eigenvector space into signal and noise subspaces based on the magnitude of the eigenvalues, constructing a spatial spectrum based on their orthogonality, and thus obtaining the DOA estimate of the signal source. This algorithm has excellent angular resolution under ideal conditions, but it is quite sensitive to array errors; for example, its performance will significantly degrade when there are channel gain and phase errors. To address the DOA estimation problem in scenarios with channel errors, Weiss and Friedlander proposed a joint iterative array self-correction method (WF algorithm). The WF algorithm simultaneously achieves signal arrival direction estimation and array correction by alternately updating the DOA estimate and array parameters. This algorithm achieves good estimation performance when the array amplitude and phase errors are not too large, but it may get stuck in local optima when the error is large, leading to a decrease in estimation performance. With the development of deep learning technology, neural network-based DOA estimation methods have attracted widespread attention due to their powerful feature extraction and nonlinear mapping capabilities. Among them, fully connected deep neural networks (FCDNNs) are applied to DOA estimation due to their simplicity and effectiveness. However, existing neural network-based methods mainly focus on improving the accuracy of DOA estimation, with less consideration given to the robustness of the algorithm under different amplitude and phase error scenarios.

[0003] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] This invention provides a robust DOA estimation method based on a fully connected neural network, a computer program product, and an electronic device, which can maintain high DOA estimation accuracy and have good robustness, thereby overcoming the defects existing in the prior art to a certain extent.

[0005] Other features and advantages of the invention will become apparent from the following detailed description, or may be learned in part by practice of the invention.

[0006] According to a first aspect of the present invention, a robust DOA estimation method based on a fully connected neural network is provided, the method comprising:

[0007] The system utilizes the far-field narrowband signals received by each element in the sensor array to construct an observation data vector based on the far-field narrowband signals received by each element; and determines the corresponding steering vector based on the angle of arrival of the far-field narrowband signals.

[0008] Construct the covariance matrix based on the observed data vectors;

[0009] The covariance matrix is ​​used as the model output parameter and input into the trained phase error estimation model based on a fully connected neural network to obtain the phase error estimate output by the model.

[0010] The corresponding phase error estimation matrix is ​​determined based on the phase error estimate; and the steering vector is compensated using the phase error estimation matrix to obtain the calibrated steering vector.

[0011] The MUSIC spatial spectrum function is constructed by combining the calibrated steering vector and the noise subspace; and the spectral peaks of the MUSIC spatial spectrum function are searched within the angle search range to determine the DOA estimate of the signal source based on the spectral peak search results.

[0012] In some exemplary embodiments, constructing the observation data vector based on the far-field narrowband signals received by each array element includes:

[0013] The far-field narrowband signals received by each array element are combined to construct the initial observation data;

[0014] The initial observation data is transposed to obtain the observation data vector.

[0015] In some exemplary implementations, constructing a covariance matrix based on observed data vectors includes:

[0016] The covariance matrix is ​​determined based on the observed data vector and the conjugate transpose of the observed data vector.

[0017] In some exemplary embodiments, the method further includes:

[0018] Take the top-right element of the covariance matrix, arrange them to form a dimension of Complex vectors;

[0019] The real and imaginary parts of the complex vector are further concatenated to obtain an M(M-1) dimensional intermediate vector; the intermediate vector is then configured as the input parameter of the model.

[0020] In some exemplary embodiments, determining the covariance matrix based on the observed data vector and the conjugate transpose of the observed data vector includes:

[0021] The covariance matrix is ​​constructed by combining the diagonal matrix with eigenvalues ​​arranged in descending order, the signal subspace, the noise subspace, the eigenvector matrix, the signal eigenvalues, and the noise eigenvalues.

[0022] In some exemplary embodiments, the phase error matrix includes the phase error corresponding to each array element; wherein the first array element is configured as the reference array element.

[0023] In some exemplary embodiments, determining the DOA estimate of the signal source based on the spectral peak search results includes:

[0024] The source DOA estimate is determined based on the angle corresponding to the spectral peak.

[0025] In some exemplary embodiments, the phase error estimation model based on a fully connected neural network includes: an input layer, an intermediate layer, and an output layer; the intermediate layer includes multiple consecutive hidden layers and a dropout layer; wherein each hidden layer is configured with a tanh nonlinear activation function; and the number of neurons in adjacent hidden layers is different.

[0026] In some exemplary embodiments, the method further includes: pre-training a phase error estimation model based on a fully connected neural network, including:

[0027] The far-field narrowband signals of samples received by the sensor array in multiple different scenarios are used to construct the corresponding sample observation data vector;

[0028] Sample data-label pairs are constructed based on sample observation data vectors and the actual phase errors of the corresponding uncalibrated array elements; a training data set is constructed based on multiple sample data-label pairs.

[0029] The initial phase error estimation model is iteratively trained using the training dataset to obtain a fully connected neural network-based phase error estimation model.

[0030] According to a second aspect of the present invention, a computer program product is provided, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the above-described robust DOA estimation method based on a fully connected neural network.

[0031] According to a third aspect of the present invention, an electronic device is provided, comprising:

[0032] Processor; and

[0033] Memory for storing the executable instructions of the processor;

[0034] The processor is configured to implement the robust DOA estimation method based on a fully connected neural network as described above when executing the executable instructions.

[0035] According to a fourth aspect of the present invention, a storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the robust DOA estimation method based on a fully connected neural network described above.

[0036] The robust DOA estimation method based on a fully connected neural network provided in the embodiments of the present invention estimates and compensates for phase errors in the array channels using a phase error estimation model based on a fully connected neural network. Then, it uses the compensated array steering vector to perform a MUSIC spatial spectrum search, thereby achieving high-precision DOA estimation. This method maintains high DOA estimation accuracy even with large phase errors in the array channels, demonstrating good robustness.

[0037] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0038] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0039] Figure 1 The illustration shows a schematic diagram of a robust DOA estimation method based on a fully connected neural network, an exemplary embodiment of the present invention.

[0040] Figure 2 The diagram illustrates the flow of a robust DOA estimation method based on a fully connected neural network, an exemplary embodiment of the present invention.

[0041] Figure 3 The diagram illustrates the spatial spectrum of DOA estimation when the number of information sources is 1, 2, and 3, respectively, according to an exemplary embodiment of the present invention.

[0042] Figure 4 This diagram illustrates the variation curve of RMSE of DOA estimation with SNR when the number of information sources is 1 in an exemplary embodiment of the present invention.

[0043] Figure 5 This diagram illustrates the variation curve of RMSE of DOA estimation with SNR when the number of information sources is 2 in an exemplary embodiment of the present invention.

[0044] Figure 6 This diagram illustrates the variation curve of RMSE of DOA estimation with SNR when the number of information sources is 3 in an exemplary embodiment of the present invention.

[0045] Figure 7 The diagram illustrates the composition of an electronic device according to an exemplary embodiment of the present invention. Detailed Implementation

[0046] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the invention will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0047] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0048] In related technologies, traditional DOA estimation algorithms suffer from a significant performance degradation when there is phase error in the array channels.

[0049] To address the shortcomings and deficiencies of existing technologies, this example implementation provides a robust DOA estimation method based on a fully connected neural network. (Reference) Figure 1 As shown, the method may specifically include the following steps:

[0050] Step S11: Utilize the far-field narrowband signals received by each element in the sensor array, construct an observation data vector based on the far-field narrowband signals received by each element; and determine the corresponding steering vector according to the angle of arrival of the far-field narrowband signals.

[0051] Step S12: Construct the covariance matrix based on the observed data vector;

[0052] Step S13: Use the covariance matrix as the model output parameter and input it into the trained phase error estimation model based on a fully connected neural network to obtain the phase error estimate value output by the model.

[0053] Step S14: Determine the corresponding phase error estimation matrix based on the phase error estimate; and use the phase error estimation matrix to compensate the steering vector to obtain the calibrated steering vector.

[0054] Step S15: Construct the MUSIC spatial spectrum function by combining the calibrated steering vector and noise subspace; and search for the spectral peaks of the MUSIC spatial spectrum function within the angle search range to determine the DOA estimate of the signal source based on the spectral peak search results.

[0055] The following will describe in more detail each step of the robust DOA estimation method based on a fully connected neural network in this exemplary embodiment, with reference to the accompanying drawings and embodiments.

[0056] In step S11, the far-field narrowband signal is received by each element in the sensor array, and an observation data vector is constructed based on the far-field narrowband signal received by each element; and the corresponding steering vector is determined according to the angle of arrival of the far-field narrowband signal.

[0057] For example, the above method can be executed on a smart terminal device or a server. For instance, data collected by a sensor array can be sent to a smart terminal device; the terminal device can create a processing task for the current data and execute the task to output a DOA estimate of the signal source.

[0058] The step of constructing the observation data vector based on the far-field narrowband signals received by each array element includes: combining the far-field narrowband signals received by each array element to construct initial observation data; and transposing the initial observation data to obtain the observation data vector.

[0059] Specifically, the sensor array consists of a uniform linear array of M sensors. K far-field, narrowband signal sources are incident on the array, and the sensor array receives K far-field, narrowband signals; the direction of arrival of the incident signal sources is θ, and the angle of arrival of the k-th far-field narrowband signal is defined as θ. k , k = 1, 2, ..., K; the direction finding range of the sensor array is [θ min ,θ max ), where θ min With θ max These are the minimum and maximum angles of the direction finding range, respectively.

[0060] The signal received by the array element is x m (t), m=1,2,…,M; the observation data vector of M array elements is x(t)=[x1(t),x2(t),…,xM (t)] T ;in,(·) T This indicates the transpose operation.

[0061] For example, suppose the angles of arrival of the received K far-field narrowband signals are θ1, θ2, ..., θ K s(t) is a vector composed of K information sources, s(t) = [s1(t), s2(t), ..., s2(t)]. K (t)] T , A=[a(θ1),a(θ2),…,a(θ K )],a(θ i () is the steering vector at the i-th angle. The steering vector is represented as:

[0062]

[0063] Where, d m Let λ be the distance between the m-th array element and the first array element, λ be the signal wavelength, and j be the imaginary unit.

[0064] When phase errors exist, a phase error matrix is ​​constructed based on the phase errors of each array element, and is expressed as: in, The phase error of the i-th element Phase error The value range is [-L, L] degrees, where L is the known phase error range.

[0065] Define the noise vector as: n(t) = [n1(t), n2(t), ..., n M (t)] T At this time, the array received signal vector is represented as: x(t)=ΓAs(t)+n(t).

[0066] In step S12, a covariance matrix is ​​constructed based on the observed data vector.

[0067] For example, constructing a covariance matrix based on an observation data vector includes: determining the covariance matrix based on the observation data vector and the conjugate transpose of the observation data vector.

[0068] Specifically, a covariance matrix is ​​constructed based on the observation data vectors of each array element; the specific formula may include:

[0069]

[0070] in,(·) H This represents the conjugate transpose operation, where N is the number of samples; and x(t) is the observed data vector.

[0071] In step S13, the covariance matrix is ​​used as the model output parameter and input into the trained phase error estimation model based on a fully connected neural network to obtain the phase error estimate value output by the model.

[0072] For example, a phase error estimation model can be pre-trained on the terminal device side; this can be built based on a fully connected neural network. The covariance matrix can be used as the model output parameter to obtain the estimated phase error value output by the model.

[0073] In one exemplary embodiment, the method further includes: taking the top-right element of the covariance matrix and arranging it to form a dimension of... Complex vectors;

[0074] The real and imaginary parts of the complex vector are further concatenated to obtain an M(M-1) dimensional intermediate vector; the intermediate vector is then configured as the input parameter of the model.

[0075] Specifically, this can be done by analyzing the covariance matrix. Arrange the elements in the upper right triangle to form a A vector b cor As a complex vector, it is represented as:

[0076] b cor =[R 1,2 ,R 1,3 ,…,R 1,M ,R 2,3 ,…,R 2,M ,…,R M-1,M ] T .

[0077] Current b cor It is a complex vector containing real and imaginary parts. We need to split the real and imaginary parts and concatenate them to obtain an M(M-1) dimensional vector as an intermediate vector. Specifically, it can be represented as:

[0078] b = [Real{b cor T},Imag{b cor T}]

[0079] Real(·) is the operation to extract the real part, and Imag(·) is the operation to extract the imaginary part.

[0080] Then, b is standardized to obtain a vector:

[0081]

[0082] Where mean(·) represents taking the mean, and ||·||2 represents the L2 norm.

[0083] For example, the method further includes: performing eigenvalue decomposition on the covariance matrix to obtain: signal eigenvalues, a diagonal matrix of eigenvalues ​​arranged in descending order, a signal subspace, a noise subspace, an eigenvector matrix, signal eigenvalues, and noise eigenvalues ​​to construct the covariance matrix.

[0084] Specifically, for the covariance matrix Eigenvalue decomposition can be performed, and the corresponding formulas may include:

[0085]

[0086] Among them, Λ=diag{λ1, λ2,…,λ m Let} be a diagonal matrix with eigenvalues ​​arranged in descending order, and U be the corresponding eigenvector matrix; U s The signal subspace is composed of the eigenvectors corresponding to the K largest eigenvalues; U n The noise subspace is composed of the eigenvectors corresponding to the remaining MK eigenvalues; Λ s =diag{λ1,λ2,…,λ K} represents the signal characteristic value, Λ n =diag{λ K+1 ,λ K+2 ,…,λ M} represents the noise characteristic value.

[0087] In step S14, the corresponding phase error estimation matrix is ​​determined based on the phase error estimate; and the steering vector is compensated using the phase error estimation matrix to obtain the calibrated steering vector.

[0088] For example, the phase error matrix includes the phase error corresponding to each array element; wherein the first array element is configured as the reference array element.

[0089] Specifically, the phase error estimate based on the current model output. A phase error estimation matrix can be constructed, and the formula can be expressed as:

[0090]

[0091] in, Let m be the phase error of the m-th array element. For reference array elements.

[0092] Based on the obtained phase error estimation matrix The steering vector can be compensated to obtain a compensated steering vector, thus achieving steering vector calibration. The formula can be expressed as:

[0093]

[0094] In step S15, the MUSIC spatial spectrum function is constructed by combining the calibrated steering vector and the noise subspace; and the spectral peaks of the MUSIC spatial spectrum function are searched within the angle search range to determine the DOA estimate of the signal source based on the spectral peak search results.

[0095] For example, determining the DOA estimate of the signal source based on the spectral peak search result includes: determining the DOA estimate of the signal source based on the angle corresponding to the spectral peak.

[0096] Specifically, based on the compensated steering vector and the extracted noise subspace U n Construct the MUSIC spatial spectral function, the formula includes:

[0097]

[0098] in, To find the guide vector at the search angle θ and after compensating for phase error, U n This is the noise subspace.

[0099] The DOA estimate of the signal source is obtained by searching for the spectral peaks of the MUSIC spatial spectral function within the angular search range.

[0100] For example, the phase error estimation model based on a fully connected neural network includes: an input layer, an intermediate layer, and an output layer; the intermediate layer includes multiple consecutive hidden layers and a dropout layer; wherein each hidden layer is configured with a tanh nonlinear activation function; and the number of neurons in adjacent hidden layers is different.

[0101] Specifically, fully connected neural networks are trained using a regression training task. For example, the model can consist of an input layer, two hidden layers, and an output layer. The input layer has M×(M-1) neurons, and the hidden layers have 70 and 30 neurons respectively. Each hidden layer is followed by a tanh non-linear activation function, and dropout is added after the second hidden layer to prevent overfitting during training. The number of neurons in the output layer is the same as the number of elements with phase errors; for example, the output layer includes two neurons. The corrected spatial spectrum is constructed using the MUSIC algorithm, and the DOA estimate of the target signal is obtained by searching for spectral peaks in the spatial spectrum.

[0102] For example, the method further includes: pre-training a phase error estimation model based on a fully connected neural network, including:

[0103] Step S21: Utilize the far-field narrowband signals of samples received by the sensor array in multiple different scenarios, and construct the corresponding sample observation data vector;

[0104] Step S22: Construct sample data-label pairs based on the sample observation data vector and the actual phase error of the corresponding uncalibrated array elements; construct a training data set based on multiple sample data-label pairs;

[0105] Step S23: Iteratively train the initial phase error estimation model using the training dataset to obtain a trained phase error estimation model based on a fully connected neural network.

[0106] For example, the sensor array could be a uniform linear array consisting of 6 elements; the target signal could be a far-field, narrow-band signal; and the signal-to-noise ratio could be... For the target signal power, Given noise power, training data is generated under Gaussian noise conditions; the number of Monte Carlo experiments is defined as 100; the direction finding range is [θ]. min = -60°, θ max =60°).

[0107] For example, when constructing a training dataset, data simulation can be performed in a dual-target scenario, adding phase error and random noise to each set of target angles, and labeling the data as y. i Where the subscript i is the data sequence number and the corresponding output label is z. i Because deep neural networks have good nonlinear mapping capabilities, they can effectively improve the phase error estimation accuracy in dual-target scenarios. Combining them with the MUSIC algorithm for correction can further improve DOA estimation performance.

[0108] Consider training the network simultaneously using data from scenarios with different signal-to-noise ratios (SNRs). The SNR values ​​are set to [-4dB, -2dB, 0dB]. This approach improves the algorithm's robustness under low SNR conditions while also allowing for better generalization to high SNR scenarios. It also avoids introducing a large amount of high SNR training data with low marginal benefits. The number of snapshots is set to 400. The dual-target angle search interval is [-60°, 60°), and the set of the two target angle intervals is Δ = {3°, 6°, ..., 60°}. The first target angle range is θ1 ∈ [-60°, -60° + Δ]. s ], where Δ s This represents the angle between two targets selected from the Δ set, where the second angle θ2 = θ1 + Δ s Using the first array element as the reference element, and assuming that elements 2 and 3 are uncalibrated, phase errors are applied to elements 2 and 3 respectively, with a phase error range of [-40°, 40°] and a step size of 2°. By calculating the covariance matrix, the upper right corner element is extracted as the input feature, and the corresponding phase error is used as the output label, forming a data-label set {(y1, z1), (y2, z2), ..., (y...}. F,z F )}.

[0109] During model training, samples of the covariance matrix are input into the initial model to obtain the phase error estimate. The difference between the phase error estimate and the phase error label value is calculated, and the model hyperparameters are adjusted based on this difference. Backpropagation training is then performed on the model. The training is iterated until a preset number of times, or until the model converges, to obtain the trained phase error estimation model.

[0110] When testing the model, refer to Figure 2 As shown, this method obtains input data that is not in the training dataset. The input is fed into a trained fully connected neural network to obtain an estimated phase error value.

[0111] Construct a phase error matrix based on the obtained phase error estimate. Inverting it and compensating for the steering vector, the calibrated steering vector becomes The spatial spectrum is obtained using the calibrated steering vector and the MUSIC algorithm, and spectral peak search is performed to obtain K spectral peak data. The angles corresponding to these peaks are the estimated source directions.

[0112] For example, the data label corresponding to a fully connected neural network is A vector z i ,in The number of array elements, therefore This yields a data-label pair (y i ,z i ), where the subscript i is the sequence number, z i This represents the true phase error of the uncalibrated array elements.

[0113] Obtain observation data x(t) under different scenarios, and based on the obtained corresponding... And obtain data-label pairs (y i ,z i Finally, the training data-label set {(y1,z1),(y2,z2),…,(y...}} of the fully connected neural network, consisting of F sets of data-label pairs, is obtained. F ,z F )}.

[0114] Using training data - label set {(y1,z1),(y2,z2),…,(y F ,z F Training a fully connected neural network, assuming the network has L layers (excluding the input layer), using net... l Represents the output of layer l, net l =W l,l-1h l-1 +b l l = 1, 2, ..., L, where W l,l-1 b represents the weight matrix between layer (l-1) and layer l. l h represents the bias vector of the l-th layer. l-1 This represents the activation value of the (l-1)th layer; for each intermediate layer, its output is: h l =g l (net l ), l=1,2,…,L-1, where g(·) is the activation function; for the output layer z out =net L .

[0115] For example, this embodiment examines the performance of the method of the present invention (i.e., FCDNN-MUSIC) and compares it with the traditional DOA estimation methods MUSIC algorithm and WF algorithm. The performance index is the root mean square error (RMSE) of the spatial spectrum and DOA estimation.

[0116] This embodiment examines the DOA estimation performance of the method under the conditions of K=1, 2, and 3. Unless otherwise specified, the phase errors of antenna 2 and antenna 3 are -25.5° and 30.5°, respectively, the number of snapshots N=400, and the signal-to-noise ratio SNR=10dB. The spatial spectrum and RMSE of the algorithm are examined. When K=1, for a single target, unless otherwise specified, the single-target DOA is θ1=-20°; when K=2, for a dual target, unless otherwise specified, the dual-target DOAs are (θ1=-20°, θ2=12°); when K=3, for a triple target, unless otherwise specified, the triple-target DOAs are (θ1=-20°, θ2=0°, θ3=12°).

[0117] The above experimental results show the effectiveness of the method of the present invention under different experimental conditions:

[0118] The spatial spectrum of the DOA estimate when K = 1, 2, 3 is as follows: Figure 3As shown, when the number of sources is 1, both the method of this invention (FCDNN-MUSIC) and the MUSIC algorithm can estimate the DOA of the source well. However, since the WF algorithm needs to use the steering vectors of multiple sources to construct an invertible optimization matrix Q to estimate the phase error during the correction process, this matrix Q is not invertible in the single source scenario, causing the WF algorithm to fail, resulting in severe distortion of the spatial spectrum. When the number of sources is 2, all three algorithms can effectively estimate the source direction. The method of this invention and the WF algorithm both show narrower spectral peaks and lower sidelobe levels. When the number of sources is 3, the angle resolution capability of the traditional MUSIC algorithm decreases significantly, while the method of this invention and the WF algorithm still maintain good resolution performance and can clearly distinguish the directions of the three sources. It is noted that the WF algorithm has estimation bias at θ1 = -20° and θ2 = 0°, while the method of this invention can accurately estimate the angles of the three targets.

[0119] When K=1, in the case of a single objective, such as Figure 4 As shown, the RMSE of both the method of this invention and the MUSIC algorithm is less than 1°, and the RMSE of the method of this invention is slightly lower than that of the MUSIC algorithm, showing higher estimation accuracy. WF fails in the single source scenario, causing its RMSE to remain at around 15°.

[0120] When K=2, in the dual-objective case, such as Figure 5 As shown, the RMSE of the three algorithms all show a decreasing trend with the increase of SNR, and the RMSE of the method of this invention is always the lowest.

[0121] When K=3, in the case of three objectives, such as Figure 6 As shown, under low signal-to-noise ratio (-5dB) conditions, the RMSE of the method of this invention is about 2.5°, which is close to the RMSE of the WF algorithm. With the increase of SNR, the performance of the three algorithms is improved, but the degree of improvement is different: the RMSE of the method of this invention drops rapidly and stabilizes below 0.2° after SNR>5dB, the RMSE of the WF algorithm stabilizes at about 1.2°, and the performance of the MUSIC algorithm is improved, with its RMSE decreasing from 9° to about 0.8° and then stabilizing.

[0122] The method provided in this invention estimates and compensates for the phase error of the array using a fully connected neural network (FCDNN), and then obtains an accurate DOA estimate using the MUSIC method, maintaining high estimation accuracy under different phase error scenarios. This invention aims to simultaneously improve the DOA estimation accuracy and robustness of the algorithm, providing a more reliable solution to the DOA estimation problem in practical applications. Under the condition of phase error in the array channels, this method estimates and compensates for the phase error using a fully connected neural network, and then uses the compensated array steering vector to perform a MUSIC spatial spectrum search, thereby achieving high-precision DOA estimation. This method maintains high DOA estimation accuracy even when there is a large phase error in the array channels, demonstrating good robustness.

[0123] It should be noted that the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Furthermore, it is readily understood that these processes may, for example, be executed synchronously or asynchronously in multiple modules.

[0124] It should be noted that although several modules or units of the device for performing actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0125] Figure 7 A schematic diagram of an electronic device suitable for implementing embodiments of the present invention is shown.

[0126] It should be noted that, Figure 7 The illustrated electronic device 1000 is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0127] like Figure 7As shown, the electronic device 1000 includes a Central Processing Unit (CPU) 1001, which can perform various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) 1002 or programs loaded from Storage Unit 1008 into Random Access Memory (RAM) 1003. The RAM 1003 also stores various programs and data required for system operation. The CPU 1001, ROM 1002, and RAM 1003 are interconnected via a bus 1004. An Input / Output (I / O) interface 1005 is also connected to the bus 1004.

[0128] The following components are connected to I / O interface 1005: an input section 1006 including a keyboard, mouse, etc.; an output section 1007 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to I / O interface 1005 as needed. Removable media 1011, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 1010 as needed so that computer programs read from them can be installed into storage section 1008 as needed.

[0129] In particular, according to embodiments of the present invention, the processes described below with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a storage medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 1009, and / or installed from removable medium 1011. When the computer program is executed by central processing unit (CPU) 1001, it performs various functions defined in the system of this application.

[0130] It should be noted that the storage medium shown in the embodiments of the present invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, wherein computer-readable program code is carried. Such transmitted data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any storage medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the storage medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0131] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0132] The units described in the embodiments of the present invention can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0133] It should be noted that, as another aspect, this application also provides a storage medium, which may be included in an electronic device or may exist independently without being assembled into the electronic device. The aforementioned storage medium carries one or more programs, which, when executed by an electronic device, cause the electronic device to perform the methods described in the following embodiments. For example, the electronic device may perform... Figure 1 The steps of the method shown.

[0134] In one embodiment, this application provides a computer program product including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0135] Furthermore, the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0136] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the claims.

[0137] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A robust DOA estimation method based on a fully connected neural network, characterized in that, The method includes: The system utilizes the far-field narrowband signals received by each element in the sensor array to construct an observation data vector based on the far-field narrowband signals received by each element; and determines the corresponding steering vector based on the angle of arrival of the far-field narrowband signals. Construct the covariance matrix based on the observed data vectors; The covariance matrix is ​​used as the model output parameter and input into the trained phase error estimation model based on a fully connected neural network to obtain the phase error estimate output by the model. The corresponding phase error estimation matrix is ​​determined based on the phase error estimate; and the steering vector is compensated using the phase error estimation matrix to obtain the calibrated steering vector. The MUSIC spatial spectrum function is constructed by combining the calibrated steering vector and the noise subspace; and the spectral peaks of the MUSIC spatial spectrum function are searched within the angle search range to determine the DOA estimate of the signal source based on the spectral peak search results.

2. The method according to claim 1, characterized in that, The construction of the observation data vector based on the far-field narrowband signals received by each array element includes: The far-field narrowband signals received by each array element are combined to construct the initial observation data; The initial observation data is transposed to obtain the observation data vector.

3. The method according to claim 1, characterized in that, Constructing a covariance matrix based on observed data vectors includes: The covariance matrix is ​​determined based on the observed data vector and the conjugate transpose of the observed data vector.

4. The method according to claim 1 or 3, characterized in that, The method further includes: Take the top-right element of the covariance matrix, arrange them to form a dimension of , where M is the number of array elements in the sensor array; The real and imaginary parts of the complex vector are further concatenated to obtain... An intermediate vector of dimension; the intermediate vector is configured as the input parameter of the model.

5. The method according to claim 1, characterized in that, The method further includes: performing eigenvalue decomposition on the covariance matrix to obtain: signal eigenvalues, a diagonal matrix of eigenvalues ​​arranged in descending order, a signal subspace, a noise subspace, an eigenvector matrix, signal eigenvalues, and noise eigenvalues ​​to construct the covariance matrix.

6. The method according to claim 1, characterized in that, The phase error matrix includes the phase error corresponding to each array element; wherein, the first array element is configured as the reference array element.

7. The method according to claim 1, characterized in that, The step of determining the DOA estimate of the signal source based on the spectral peak search results includes: The source DOA estimate is determined based on the angle corresponding to the spectral peak.

8. The method according to claim 1, characterized in that, The phase error estimation model based on a fully connected neural network includes an input layer, an intermediate layer, and an output layer; the intermediate layer includes multiple consecutive hidden layers and a dropout layer; each hidden layer is configured with a tanh nonlinear activation function; the number of neurons in adjacent hidden layers is different.

9. The method according to claim 1 or 8, characterized in that, The method further includes: pre-training a phase error estimation model based on a fully connected neural network, including: The far-field narrowband signals of samples received by the sensor array in multiple different scenarios are used to construct the corresponding sample observation data vector; Sample data-label pairs are constructed based on sample observation data vectors and the actual phase errors of the corresponding uncalibrated array elements; a training data set is constructed based on multiple sample data-label pairs. The initial phase error estimation model is iteratively trained using the training dataset to obtain a fully connected neural network-based phase error estimation model.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the robust DOA estimation method based on a fully connected neural network as described in any one of claims 1 to 9.

Images