DOA estimation method based on neural network and large electromagnetic vector sensor

Abstract

The application discloses a DOA estimation method based on a neural network and a large electromagnetic vector sensor and belongs to the technical field of signal processing, which comprises the following steps: a 6-component separation type large electromagnetic vector sensor is established, received data of a signal source is obtained, a signal parameter estimation method of ESPRIT is used to calculate the estimated value of a signal steering vector, after preprocessing, a signal data set is obtained, a DOA estimation calculation model is trained by using the signal data set, and the DOA estimation of the large electromagnetic vector sensor is completed by using the trained DOA estimation calculation model. The application introduces the neural network deep learning into the DOA estimation of the polarization sensitive array, establishes a DOA estimation calculation model, and combines data driving, so that the DOA estimation of the large electromagnetic vector sensor can be performed, and the problems of the spatial phase factor and the complex and highly nonlinear response of the large electromagnetic vector sensor itself are solved.

Description

Technical Field

[0001] This invention belongs to the field of signal processing technology, specifically relating to a DOA estimation method based on neural networks and large electromagnetic vector sensors. Background Technology

[0002] Compared to scalar sensors, electromagnetic vector sensors, under the same aperture conditions, have more receiving channels and thus acquire more information, primarily polarization information. This gives electromagnetic vector sensors a greater advantage in certain specific applications.

[0003] Currently, for DOA estimation algorithms of large electromagnetic vector sensors, the radiation efficiency of these sensors is much higher, making them easier to implement in engineering. However, due to the influence of their large electromagnetic response, only phase and amplitude comparison methods are currently available. The aforementioned discrete electromagnetic vector sensors can implement 6-component, 5-component, 4-component, 3-component, and 2-component methods. All of these methods require a reasonable arrangement of the array structure, placing very strict requirements on the array structure. Due to the inherent response of the large electromagnetic vector sensor and the influence of the spatial phase factor introduced by arbitrary structures, the DOA estimation problem for discrete large electromagnetic vector sensors with arbitrary structures is currently a difficult problem to solve. Summary of the Invention

[0004] To address the aforementioned shortcomings in existing technologies, this invention provides a DOA estimation method based on neural networks and large electromagnetic vector sensors, which solves the problem of difficulty in DOA estimation due to the complex and highly nonlinear response of spatial phase factors and large electromagnetic vector sensors.

[0005] To achieve the aforementioned objectives, the present invention employs the following technical solution: a DOA estimation method based on neural networks and large electromagnetic vector sensors, comprising the following steps:

[0006] S1: Establish a 6-component discrete large electromagnetic vector sensor;

[0007] S2: Based on the 6-component discrete large electromagnetic vector sensor, receive the signal source received data;

[0008] S3: Based on the data received by the signal source, the estimated value of the signal steering vector is calculated using the ESPRIT signal parameter estimation method;

[0009] S4: Preprocess the estimated steering vector of the signal to obtain the signal dataset;

[0010] S5: Construct a DOA estimation calculation model and train it using a signal dataset to obtain a trained DOA estimation calculation model;

[0011] S6: Using the trained DOA estimation calculation model, process the signal source signal of the large electromagnetic vector sensor to obtain the DOA estimate value, and complete the DOA estimation of the large electromagnetic vector sensor.

[0012] The beneficial effects of this invention are as follows: This invention introduces neural network deep learning into DOA estimation of polarization sensitive arrays. By establishing a DOA estimation calculation model and combining it with data-driven methods, it can perform DOA estimation on large electromagnetic vector sensors, solving the problems of spatial phase factor and the complex and highly nonlinear response of large electromagnetic vector sensors. It can also complete the separation of multiple targets and the DOA estimation after separation.

[0013] Furthermore: the 6-component split-type large electromagnetic vector sensor includes three mutually orthogonal electric dipoles and three mutually orthogonal large magnetic rings; the electric dipoles and large magnetic rings are spatially separated and their positions are arbitrary.

[0014] The advantages of the above-mentioned further scheme are: designing a split large electromagnetic vector sensor, improving DOA estimation accuracy, reducing electromagnetic coupling problems, improving the accuracy of the dataset, and being able to obtain electromagnetic vector DOA estimation results of arbitrary structures.

[0015] Furthermore, the specific steps of S4 are as follows:

[0016] S401: Matrix the estimated steering vector of the signal to obtain the steering vector matrix;

[0017] S402: Perform imaginary part extraction, real part extraction, and angle extraction on the guide vector matrix to obtain the imaginary part, real part, and angle, respectively.

[0018] S403: Establish three-dimensional matrix data based on the imaginary part, real part, and angle;

[0019] S404: Based on the three-dimensional matrix data, add labels to obtain the signal dataset; the label is a vector with a single non-zero element.

[0020] The beneficial effects of the above-mentioned further scheme are as follows: by extracting the imaginary part, real part, and angle, the accuracy of DOA estimation can be improved, the computational complexity can be reduced, the dependence on the number of samples can be reduced, and adding labels can improve the ability to distinguish the source of the signal.

[0021] Furthermore, the expression for the signal dataset is as follows:

[0022]

[0023]

[0024]

[0025]

[0026]

[0027]

[0028]

[0029] in, For signal datasets, For tags, It is a three-dimensional matrix data. The total number of samples, For the real part, The virtual part, For angle, To perform the real part operation, To perform the imaginary part operation, To perform the angle measurement operation, For the first The guiding vector matrix of each sample. For the first The estimated steering vector value for each sample. This is the conjugate transpose. For data dimensions, For the first The azimuth angle of each sample For the first The pitch angle of each sample For the first The polarization angle of each sample. For the first The polarization phase difference of each sample.

[0030] The beneficial effects of the above-mentioned further solutions are as follows: the signal dataset established by the present invention can provide the amplitude, phase and spatial information of the signal, enhance the signal recognition accuracy, reduce the data training cost and improve the generalization ability of the algorithm.

[0031] Furthermore: the DOA estimation calculation model includes an input layer, a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a fifth convolutional layer, a sixth convolutional layer, a first fully connected layer, a second fully connected layer, and an output layer connected in sequence;

[0032] Both the first and second convolutional layers contain 256 convolutional kernels and the kernel size is [missing value]. ;

[0033] The third, fourth, fifth, and sixth convolutional layers each contain 256 convolutional kernels, and the kernel size is [missing value]. ;

[0034] The first, second, third, fourth, fifth, and sixth convolutional layers are all connected to a batch normalization layer and a ReLU activation function;

[0035] The first fully connected layer is connected to a ReLU activation function and a dropout layer; the second fully connected layer is connected to a sigmoid activation function.

[0036] The beneficial effects of the above-mentioned further scheme are as follows: the DOA estimation calculation model, through a multi-layer convolutional and fully connected layer structure, combined with batch normalization layers and ReLU activation function, can effectively extract signal features and accelerate the training process, and uses Dropout to prevent overfitting, while the Sigmoid activation function outputs a probability distribution, enhancing the model's generalization ability and real-time performance.

[0037] Furthermore, the expression for the DOA estimate is as follows:

[0038]

[0039]

[0040]

[0041]

[0042]

[0043]

[0044]

[0045]

[0046] in, This is the estimated value of DOA. Let be the dimension of the weight matrix of the second fully connected layer. The bias vector of the second fully connected layer. This is the output of the first fully connected layer. Let be the dimension of the weight matrix of the first fully connected layer. The bias vector of the first fully connected layer. This is the output of the sixth convolutional layer. The dimension of the sixth convolutional layer. This is the bias vector of the sixth convolutional layer. This is the output of the fifth convolutional layer. The dimension of the fifth convolutional layer. This is the bias vector for the fifth convolutional layer. This is the output of the fourth convolutional layer. The dimension of the fourth convolutional layer. This is the bias vector for the fourth convolutional layer. This is the output of the third convolutional layer. The dimension of the third convolutional layer. This is the bias vector of the third convolutional layer. This is the output of the second convolutional layer. The dimension of the second convolutional layer. This is the bias vector for the second convolutional layer. This is the output of the second convolutional layer. The dimension of the first convolutional layer. This is the bias vector of the first convolutional layer. It is a three-dimensional matrix data. This is a convolution operation.

[0047] The beneficial effects of the above-mentioned further scheme are: by obtaining the DOA estimate value through the DOA estimation calculation model and in the form of an expression, the interpretability and generalization ability of the model can be enhanced, and the calculation process of the model can be easily understood.

[0048] Furthermore, the expression for the loss function of the DOA estimation calculation model is as follows:

[0049]

[0050] in, The loss value. For real labels, To estimate the output labels of the computational model for DOA, For real numbers, The total number of samples.

[0051] The beneficial effects of the above-mentioned further scheme are as follows: using the cross-entropy loss function as the loss metric of the model can effectively quantify the difference between the model's prediction and the actual arrival direction, promote the model to learn accurate direction classification, accelerate the training process, and improve the accuracy and generalization ability of the estimation. Attached Figure Description

[0052] Figure 1 This is a flowchart of a DOA estimation method based on neural networks and large electromagnetic vector sensors.

[0053] Figure 2 This is a graph showing the error variation of the loss function;

[0054] Figure 3 These are the output tag values ​​under different signal-to-noise ratio conditions. Detailed Implementation

[0055] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0056] Example 1

[0057] like Figure 1 The flowchart shown illustrates a DOA estimation method based on neural networks and large electromagnetic vector sensors, comprising the following steps:

[0058] S1: Establish a 6-component discrete large electromagnetic vector sensor;

[0059] S2: Based on the 6-component discrete large electromagnetic vector sensor, receive the signal source received data;

[0060] S3: Based on the data received by the signal source, the estimated value of the signal steering vector is calculated using the ESPRIT signal parameter estimation method;

[0061] S4: Preprocess the estimated steering vector of the signal to obtain the signal dataset;

[0062] S5: Construct a DOA estimation calculation model and train it using a signal dataset to obtain a trained DOA estimation calculation model;

[0063] S6: Using the trained DOA estimation calculation model, process the signal source signal of the large electromagnetic vector sensor to obtain the DOA estimate value, and complete the DOA estimation of the large electromagnetic vector sensor.

[0064] In S1, the 6-component split-type large electromagnetic vector sensor includes three mutually orthogonal electric dipoles and three mutually orthogonal large magnetic rings; the electric dipoles and large magnetic rings are spatially separated and their positions are arbitrary;

[0065] In one embodiment of the invention, the electric dipole includes , and The coordinates are as follows: , , ; Magnetic ring includes , and The position coordinates are: , , The coordinate positions are not restricted in any way and can be arranged arbitrarily, where the length of the electric dipoles is... Much greater than 0.1 wavelengths, that is At this point, the electric dipole is a long electric dipole, resulting in high radiation efficiency; the radius of the magnetic ring... The range is At this point, the magnetic ring is a large magnetic ring with high radiation efficiency; the long electric dipole and the large magnetic ring constitute a large electromagnetic vector sensor.

[0066] In S2, assume the expression for the transmitted signal from the signal source is as follows:

[0067]

[0068] in, For signal amplitude, For the frequency of the signal, For time, The imaginary unit, Assuming a uniformly randomly distributed phase, the expression for the time-delayed signal can be obtained as follows:

[0069]

[0070] The transmitted signal and the time-delayed signal have the following invariant relationship:

[0071]

[0072] Therefore, based on the rotational invariance of the large electromagnetic vector sensor, the expression for the data received by the signal source is constructed as follows:

[0073]

[0074] Furthermore, neglecting noise, the rotation-invariant relationship is as follows:

[0075]

[0076] in, For time delay;

[0077] The traditional ESPRIT method for estimating signal parameters is used to calculate the steering vector of each signal from the received data. .

[0078] In one embodiment of the present invention, the specific steps of S4 are as follows:

[0079] S401: The steering vector estimate of the signal is matrixed to obtain the steering vector matrix. The expression of the steering vector matrix is ​​as follows:

[0080]

[0081]

[0082] in, For the first The guiding vector matrix of each sample. For the first The estimated steering vector value for each sample. This is the conjugate transpose. For data dimensions, For the first The azimuth angle of each sample For the first The pitch angle of each sample For the first The polarization angle of each sample. For the first The polarization phase difference of each sample.

[0083] S402: Perform imaginary part extraction, real part extraction, and angle extraction on the guide vector matrix to obtain the imaginary part, real part, and angle, respectively. The expressions for the imaginary part, real part, and angle are as follows;

[0084]

[0085]

[0086]

[0087] in, For the real part, The virtual part, For angle, To perform the real part operation, To perform the imaginary part operation, To perform the angle measurement operation, For the first The guiding vector matrix of each sample;

[0088] S403: Based on the imaginary part, real part, and angle, establish three-dimensional matrix data. The expression for the three-dimensional matrix data is as follows:

[0089]

[0090] in, It is a three-dimensional matrix data;

[0091] S404: Based on the three-dimensional matrix data, labels are added to obtain the signal dataset. The expression for the signal dataset is as follows:

[0092]

[0093] in, For signal datasets, The label for the first sample. The label for the second sample. For the first The label of each sample, This is the three-dimensional matrix data of the first sample. This is the three-dimensional matrix data of the second sample. For the first The data consists of a three-dimensional matrix of samples; where the label is a vector with a single non-zero element, representing the category of the signal. For example, if the search angle of the signal is divided into 10 categories and the target angle is in the 7th category, then the label is... If there is no target, then label .

[0094] In S5, a DOA estimation calculation model is constructed. The DOA estimation calculation model includes an input layer, a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a fifth convolutional layer, a sixth convolutional layer, a first fully connected layer, a second fully connected layer, and an output layer connected in sequence.

[0095] Both the first and second convolutional layers contain 256 convolutional kernels, and the kernel size is [missing value]. ;

[0096] The third, fourth, fifth, and sixth convolutional layers each contain 256 convolutional kernels, and the kernel size is [missing value]. ;

[0097] The first, second, third, fourth, fifth, and sixth convolutional layers are all connected to a batch normalization layer and a ReLU activation function;

[0098] The first fully connected layer is connected to the ReLU activation function and the dropout layer; the second fully connected layer is connected to the sigmoid activation function to induce classification probability.

[0099] In one embodiment of the present invention, the network parameters of the convolutional layer are set to... ,in The dimension is convolution kernel, Indicates the number of input channels. This represents the length and width of the convolution kernel. This indicates the number of convolution kernels, i.e., the number of output channels. The dimension is The bias vector;

[0100] The network parameters of a fully connected network are ,in, The dimension is The weight matrix, Indicates the number of input channels. This indicates the number of neurons in the fully connected layer, i.e., the number of output channels. The dimension is The bias vector, This refers to matrix multiplication.

[0101] The first convolutional layer extracts features from the input data, using the ReLU activation function. , Its expression is as follows:

[0102]

[0103] in, This is the output of the second convolutional layer. The dimension of the first convolutional layer. This is the bias vector of the first convolutional layer. It is a three-dimensional matrix data. This is a convolution operation.

[0104] The second, third, fourth, and fifth convolutional layers are all used for non-linear mapping of features, employing the ReLU activation function. , Their expressions are as follows:

[0105]

[0106]

[0107]

[0108]

[0109] in, This is the output of the fifth convolutional layer. The dimension of the fifth convolutional layer. This is the bias vector for the fifth convolutional layer. This is the output of the fourth convolutional layer. The dimension of the fourth convolutional layer. This is the bias vector for the fourth convolutional layer. This is the output of the third convolutional layer. The dimension of the third convolutional layer. This is the bias vector of the third convolutional layer. This is the output of the second convolutional layer. The dimension of the second convolutional layer. This is the bias vector for the second convolutional layer;

[0110] The sixth convolutional layer is used to reconstruct the data from the output of the fifth convolutional layer, with an output dimension of . The dataset, , Its expression is as follows:

[0111]

[0112] in, This is the output of the sixth convolutional layer. The dimension of the sixth convolutional layer. This is the bias vector of the sixth convolutional layer;

[0113] The first fully connected layer integrates the data output from the convolutional layers and uses the ReLU activation function. , Its expression is as follows:

[0114]

[0115] in, This is the output of the first fully connected layer. Let be the dimension of the weight matrix of the first fully connected layer. This is the bias vector of the first fully connected layer;

[0116] The second fully connected layer, used to induce and output classification probabilities, employs the sigmoid activation function. , Its expression is as follows:

[0117]

[0118] in, This is the estimated value of DOA. Let be the dimension of the weight matrix of the second fully connected layer. This is the bias vector of the second fully connected layer.

[0119] In S5, the DOA estimation calculation model is trained using the signal dataset to obtain the trained DOA estimation calculation model; the expression of the loss function of the DOA estimation calculation model is as follows:

[0120]

[0121] in, The loss value. For real labels, To estimate the output labels of the computational model for DOA, For real numbers, The total number of samples.

[0122] In one embodiment of the present invention, during the training process of the DOA estimation computation model, the learning rate is gradually reduced as the number of iterations increases, so that the network parameters gradually converge to the optimal or near-optimal. At the same time, a Dropout strategy is adopted to randomly remove neurons with a certain probability, which alleviates the dependence of the DOA estimation computation model on the distribution of training set data and avoids overfitting. The parameters of the DOA estimation computation model are trained by minimizing the loss function through the backpropagation algorithm, and the Adam optimizer is used to update the network parameters. After each forward propagation of the neural network, the error is backpropagated and the network weights are updated. This process is repeated many times until the network objective function converges.

[0123] In S6, the trained DOA estimation calculation model is used to process the signal source signal of the large electromagnetic vector sensor to obtain the DOA estimate value, thus completing the DOA estimation of the large electromagnetic vector sensor.

[0124] The beneficial effects of this invention are as follows: This invention introduces neural network deep learning into DOA estimation of polarization sensitive arrays. By establishing a DOA estimation calculation model and combining it with data-driven methods, it can perform DOA estimation on large electromagnetic vector sensors, solving the problems of spatial phase factor and the complex and highly nonlinear response of large electromagnetic vector sensors. It can also complete the separation of multiple targets and the DOA estimation after separation.

[0125] Example 2

[0126] In one embodiment of the present invention, the Deep Learning Toolbox in Matlab R2020b is used to train the DOA estimation computation model.

[0127] S1: Establish a 6-component discrete large electromagnetic vector sensor; the coordinate settings of the electric dipole are as follows: 、 、 Magnetic ring , and The coordinates are: 、 、 Electric dipole far greater than One wavelength, for a long electric dipole, magnetic ring radius far greater than It belongs to a large magnetic ring, and the other parameters are: .

[0128] S2: Based on a 6-component discrete large electromagnetic vector sensor, receive data from the signal source; the signal-to-noise ratio (SNR) of the received data is set to 20dB, the number of snapshots is set to 10, the azimuth angle is 42 degrees, and the elevation angle is... The range of variation is set to 0.1°-10°, and it is evenly divided into discrete angles with an interval of 0.1°. The signal traverses each positioning range, and training and test sets are randomly sampled in a ratio of 8:2. The number of samples is 100, and each sample forms 2000 data points. The total dataset size is 200,000, and the ratio of training to test set data is 8:2.

[0129] S3: Based on the data received by the signal source, the estimated value of the signal steering vector is calculated using the ESPRIT signal parameter estimation method;

[0130] S4: Preprocess the estimated steering vector of the signal to obtain the signal dataset;

[0131] S5: Construct a DOA estimation calculation model and train it using a signal dataset to obtain a trained DOA estimation calculation model;

[0132] A DOA estimation computation model is constructed based on convolutional layers and fully connected layers. The hyperparameter settings of the DOA estimation computation model are shown in Table 1 below:

[0133] Table 1

[0134]

[0135] In embodiments of the present invention, the decision to stop training can be based on the number of training rounds, or the decision to end training can be based on the loss value of the loss function. Figure 2 As shown, the error variation graph of the loss function shows that the loss amount of the loss function of the DOA estimation calculation model in this invention decreases rapidly, achieving a fast convergence effect.

[0136] S6: Using the trained DOA estimation calculation model, process the signal source signal of the large electromagnetic vector sensor to obtain the DOA estimate value, and complete the DOA estimation of the large electromagnetic vector sensor.

[0137] In one embodiment of the present invention, a trained DOA estimation calculation model is used to obtain the DOA estimate; select To validate the model, since the signal-to-noise ratio (SNR) of the training set is 20dB, four SNR values ​​(SNR=-15dB, SNR=-5dB, SNR=5dB, and SNR=15dB) were selected to verify the results of the DOA estimation calculation model. Figure 3 The figure shows the output label values ​​under different signal-to-noise ratio conditions, where... Figure 3(a) Calculate the label and true values ​​of the model for DOA estimation with SNR = -15dB. Figure 3 (b) Calculate the label and true values ​​of the DOA estimation model for SNR=-5dB. Figure 3 (c) Calculate the label and true values ​​of the DOA estimation model for an SNR of 5dB. Figure 3 (d) Outputs the label value and true value of the DOA estimation calculation model with SNR=15dB; The DOA estimation calculation model of the present invention can correctly estimate the angle of the target and has high accuracy.

[0138] The beneficial effects of this invention are as follows: This invention introduces neural network deep learning into DOA estimation of polarization sensitive arrays. By establishing a DOA estimation calculation model and combining it with data-driven methods, it can perform DOA estimation for large electromagnetic vector sensors. This solves the problems of spatial phase factor and the complex and highly nonlinear response of large electromagnetic vector sensors. It can complete DOA estimation of large electromagnetic vector sensors with high accuracy.

Claims

1. A DOA estimation method based on neural networks and large electromagnetic vector sensors, characterized in that, Includes the following steps: S1: Establish a 6-component discrete large electromagnetic vector sensor; S2: Based on the 6-component discrete large electromagnetic vector sensor, receive the signal source received data; S3: Based on the data received by the signal source, the estimated value of the signal steering vector is calculated using the ESPRIT signal parameter estimation method; S4: Preprocess the estimated steering vector of the signal to obtain the signal dataset; S5: Construct a DOA estimation calculation model and train it using a signal dataset to obtain a trained DOA estimation calculation model; The DOA estimation calculation model includes an input layer, a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a fifth convolutional layer, a sixth convolutional layer, a first fully connected layer, a second fully connected layer, and an output layer connected in sequence. The first convolutional layer and the second convolutional layer each include 256 convolutional kernels and the convolutional kernel size is ; The third convolutional layer, the fourth convolutional layer, the fifth convolutional layer and the sixth convolutional layer each include 256 convolutional kernels and the convolutional kernel size is ; The first, second, third, fourth, fifth, and sixth convolutional layers are all connected to a batch normalization layer and a ReLU activation function; The first fully connected layer is connected to a ReLU activation function and a dropout layer; the second fully connected layer is connected to a sigmoid activation function. S6: Using the trained DOA estimation calculation model, process the signal source signal of the large electromagnetic vector sensor to obtain the DOA estimate value, and complete the DOA estimation of the large electromagnetic vector sensor. The expression for the DOA estimate is as follows: in, This is the estimated value of DOA. Let be the dimension of the weight matrix of the second fully connected layer. The bias vector of the second fully connected layer. This is the output of the first fully connected layer. Let be the dimension of the weight matrix of the first fully connected layer. The bias vector of the first fully connected layer. This is the output of the sixth convolutional layer. The dimension of the sixth convolutional layer. This is the bias vector of the sixth convolutional layer. This is the output of the fifth convolutional layer. The dimension of the fifth convolutional layer. This is the bias vector of the fifth convolutional layer. This is the output of the fourth convolutional layer. The dimension of the fourth convolutional layer. This is the bias vector for the fourth convolutional layer. This is the output of the third convolutional layer. The dimension of the third convolutional layer. This is the bias vector of the third convolutional layer. This is the output of the second convolutional layer. The dimension of the second convolutional layer. This is the bias vector for the second convolutional layer. This is the output of the first convolutional layer. The dimension of the first convolutional layer. This is the bias vector of the first convolutional layer. It is a three-dimensional matrix data. This is a convolution operation.

2. The DOA estimation method based on neural networks and large electromagnetic vector sensors according to claim 1, characterized in that, The 6-component split-type large electromagnetic vector sensor includes three mutually orthogonal electric dipoles and three mutually orthogonal large magnetic rings; the electric dipoles and large magnetic rings are spatially separated and their positions are arbitrary.

3. The DOA estimation method based on neural networks and large electromagnetic vector sensors according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Matrix the estimated steering vector of the signal to obtain the steering vector matrix; S402: Perform imaginary part extraction, real part extraction, and angle extraction on the guide vector matrix to obtain the imaginary part, real part, and angle, respectively. S403: Establish three-dimensional matrix data based on the imaginary part, real part, and angle; S404: Based on the three-dimensional matrix data, add labels to obtain the signal dataset; the label is a vector with a single non-zero element.

4. The neural network and large electromagnetic vector sensor based DOA estimation method according to claim 3, characterized in that, The expression for the signal dataset is as follows: in, For signal datasets, For tags, It is a three-dimensional matrix data. The total number of samples, For the real part, The virtual part, For angle, To perform the real part operation, To perform the imaginary part operation, To perform the angle measurement operation, For the first The guiding vector matrix of each sample. For the first The estimated steering vector value for each sample. This is the conjugate transpose. For data dimensions, For the first The azimuth angle of each sample For the first The pitch angle of each sample For the first The polarization angle of each sample. For the first The polarization phase difference of each sample.

5. The neural network and large electromagnetic vector sensor based DOA estimation method according to claim 1, characterized in that, The expression for the loss function of the DOA estimation calculation model is as follows: in, This is the loss value. For real labels, To estimate the output labels of the computational model for DOA, For real numbers, The total number of samples.

Images