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

Abstract

The invention discloses a DOA (direction of arrival) estimation method based on a neural network and a large electromagnetic vector sensor, which belongs to the technical field of signal processing, and comprises the following steps of: establishing a six-component separation type large electromagnetic vector sensor, receiving to obtain signal source receiving data, calculating to obtain a steering vector estimation value of a signal by using an ESPRIT (Empirical Shortest Path Redundancy Time) signal parameter estimation method, and performing pre-processing to obtain the steering vector estimation value of the signal. And obtaining a signal data set, training a DOA estimation calculation model by using the signal data set, and completing DOA estimation of the large electromagnetic vector sensor by using the trained DOA estimation calculation model. According to the method, neural network deep learning is introduced into DOA estimation of the polarization sensitive array, a DOA estimation calculation model is established, data driving is combined, DOA estimation can be performed on the large electromagnetic vector sensor, and the problems of complex response and high nonlinearity of a space phase factor and the large electromagnetic vector sensor are solved.

Description

Technical Field

[0001] The invention belongs to the technical field of signal processing, and in particular relates to a DOA estimation method based on a neural network and a large electromagnetic vector sensor. Background Art

[0002] Compared with scalar sensors, electromagnetic vector sensors have more receiving channels under the same aperture conditions, and can obtain more information, mainly polarization information. This will lead to greater advantages for electromagnetic vector sensors in certain specific applications.

[0003] At present, for the DOA estimation algorithm of large electromagnetic vector sensors, the radiation efficiency of large electromagnetic vector sensors is much greater and easier to implement in engineering. Affected by the large electromagnetic response, there are only phase and amplitude comparison methods. The above-mentioned separated electromagnetic vector sensors can achieve 6-component, 5-component, 4-component, 3-component, and 2-component methods. The above methods all require a reasonable arrangement of the array structure, and the requirements for the array structure are very strict. Due to the response of the large electromagnetic vector sensor itself and the influence of the spatial phase factor brought by the arbitrary structure, the DOA estimation problem of the separated large electromagnetic vector sensor with arbitrary structure is currently difficult to solve. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a DOA estimation method based on a neural network and a large electromagnetic vector sensor, which solves the problem of difficulty in DOA estimation due to the complex and highly nonlinear response of the spatial phase factor and the large electromagnetic vector sensor itself.

[0005] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a DOA estimation method based on a neural network and a large electromagnetic vector sensor, comprising the following steps: S1: Establish a 6-component separated large electromagnetic vector sensor; S2: According to the 6-component separated large electromagnetic vector sensor, the signal source receiving data is received; S3: Based on the data received from the signal source, the signal parameter estimation method is used to calculate the estimated value of the signal steering vector; S4: preprocessing the steering vector estimation value of the signal to obtain a signal data set; S5: construct a DOA estimation calculation model, and use the signal data set for training to obtain a trained DOA estimation calculation model; S6: Using the trained DOA estimation calculation model, the signal source signal of the large electromagnetic vector sensor is processed to obtain the DOA estimation value, thereby completing the DOA estimation of the large electromagnetic vector sensor.

[0006] The beneficial effects of the present invention are as follows: the present invention introduces neural network deep learning into the DOA estimation of the polarization sensitive array. By establishing a DOA estimation calculation model and combining it with data driving, it can perform DOA estimation on a large electromagnetic vector sensor, solve the problem of complex and highly nonlinear response of the spatial phase factor and the large electromagnetic vector sensor itself, and can complete the separation of multiple targets and the DOA estimation after separation.

[0007] Furthermore: the 6-component separated 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 separated in space and are in arbitrary positions.

[0008] The beneficial effects of the above further scheme are: designing a separate large electromagnetic vector sensor, improving the DOA estimation accuracy, reducing electromagnetic coupling problems, improving the accuracy of the data set, and being able to obtain electromagnetic vector DOA estimation results of arbitrary structures.

[0009] Further: the specific steps of S4 are as follows: S401: Matrixing the steering vector estimation value of the signal to obtain a steering vector matrix; S402: performing imaginary part processing, real part processing and angle processing on the steering vector matrix to obtain the imaginary part, real part and angle respectively; S403: Create three-dimensional matrix data according to the imaginary part, the real part and the angle; S404: adding a label according to the three-dimensional matrix data to obtain a signal data set; the label is a vector of a single non-zero element.

[0010] The beneficial effects of the above further scheme are: by extracting the imaginary part, the real part and the 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.

[0011] Further: the expression of the signal data set is as follows:

[0012]

[0013]

[0014]

[0015]

[0016]

[0017]

[0018] in, is the signal data set, For labels, is a three-dimensional matrix data, is the total number of samples, is the real part, is the imaginary part, is the angle, To take the real part operation, To take the imaginary part operation, To take the angle operation, For the The steering vector matrix of samples, For the The estimated value of the steering vector for samples, is the conjugate transpose, is the data dimension, For the The azimuth of the samples, For the The pitch angle of each sample, For the The polarization angle of the samples, For the The polarization phase difference of the samples.

[0019] The beneficial effect of the above further scheme is that the signal data set established by the present invention can provide the amplitude, phase and spatial information of the signal, enhance the accuracy of signal recognition, reduce the cost of data training, and improve the generalization ability of the algorithm.

[0020] Further: 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 convolution layer and the second convolution layer both include 256 convolution kernels and the convolution kernel size is ; The third convolution layer, the fourth convolution layer, the fifth convolution layer and the sixth convolution layer all include 256 convolution kernels and the convolution kernel size is ; The first convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer, the fifth convolutional layer and the sixth convolutional layer are all connected with 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.

[0021] The beneficial effects of the above further scheme are as follows: the DOA estimation calculation model can effectively extract the characteristics of the signal and accelerate the training process through a multi-layer convolution and fully connected layer structure, combined with a batch normalization layer and a ReLU activation function, and uses Dropout to prevent overfitting. The Sigmoid activation function outputs a probability distribution, which enhances the generalization ability and real-time performance of the model.

[0022] Further: The expression of the DOA estimation value is as follows:

[0023]

[0024]

[0025]

[0026]

[0027]

[0028]

[0029]

[0030] in, is the estimated value of DOA, is the dimension of the weight matrix of the second fully connected layer, is the bias vector of the second fully connected layer, is the output of the first fully connected layer, is the dimension of the weight matrix of the first fully connected layer, is the bias vector of the first fully connected layer, is the output of the sixth convolutional layer, is the dimension of the sixth convolutional layer, is the bias vector of the sixth convolutional layer, is the output of the fifth convolutional layer, is the dimension of the fifth convolutional layer, is the bias vector of the fifth convolutional layer, is the output of the fourth convolutional layer, is the dimension of the fourth convolutional layer, is the bias vector of the fourth convolutional layer, is the output of the third convolutional layer, is the dimension of the third convolutional layer, is the bias vector of the third convolutional layer, is the output of the second convolutional layer, is the dimension of the second convolutional layer, is the bias vector of the second convolutional layer, is the output of the second convolutional layer, is the dimension of the first convolutional layer, is the bias vector of the first convolutional layer, is a three-dimensional matrix data, is the convolution operation.

[0031] The beneficial effect of the above further scheme is that the DOA estimation value is obtained through the DOA estimation calculation model in the form of an expression, which can enhance the interpretability and generalization ability of the model and facilitate the understanding of the calculation process of the model.

[0032] Further: The expression of the loss function of the DOA estimation calculation model is as follows:

[0033] in, is the loss value, is the true label, The output labels of the model are calculated for DOA estimation, is a real number, is the total number of samples.

[0034] The beneficial effects of the above further scheme are: using the cross entropy loss function as the loss metric of the model can effectively quantify the difference between the model prediction and the actual arrival direction, promote the model to learn accurate direction classification, accelerate the training process, and improve the estimation accuracy and generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a flow chart of a DOA estimation method based on neural network and large electromagnetic vector sensor; Figure 2 It is the error change graph of the loss function; Figure 3 is the output label value under different signal-to-noise ratio conditions. DETAILED DESCRIPTION

[0036] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0037] Example 1 like Figure 1 As shown, a flow chart of a DOA estimation method based on a neural network and a large electromagnetic vector sensor includes the following steps: S1: Establish a 6-component separated large electromagnetic vector sensor; S2: According to the 6-component separated large electromagnetic vector sensor, the signal source receiving data is received; S3: Based on the data received from the signal source, the signal parameter estimation method is used to calculate the estimated value of the signal steering vector; S4: preprocessing the steering vector estimation value of the signal to obtain a signal data set; S5: construct a DOA estimation calculation model, and use the signal data set for training to obtain a trained DOA estimation calculation model; S6: Using the trained DOA estimation calculation model, the signal source signal of the large electromagnetic vector sensor is processed to obtain the DOA estimation value, thereby completing the DOA estimation of the large electromagnetic vector sensor.

[0038] In S1, the 6-component separated large electromagnetic vector sensor includes three mutually orthogonal electric dipoles and three mutually orthogonal large magnetic rings; the electric dipoles and the large magnetic rings are separated in space and have arbitrary positions; In one embodiment of the present invention, the electric dipole comprises , and , the coordinates are: , , ;Magnetic ring includes , and , the coordinates of the positions are: , , , the coordinate position is not restricted and can be arranged arbitrarily, where the length of the electric dipole Much larger than 0.1 wavelength, that is , at this time the electric dipole is a long electric dipole with high radiation efficiency; the radius of the magnetic ring The range is At this time, 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.

[0039] In S2, it is assumed that the expression of the transmitted signal of the signal source is as follows:

[0040] in, is the signal amplitude, is the frequency of the signal, For time, is an imaginary unit, For a uniformly randomly distributed phase, the expression of the time-delay signal can be obtained as follows:

[0041] The transmitted signal and the delayed signal have an invariant relationship as follows:

[0042] Therefore, according to the rotation invariance relationship of the large electromagnetic vector sensor, the expression of the signal source receiving data is constructed as follows:

[0043] And without considering the noise, there is a rotation invariant relationship as follows:

[0044] in, For time delay; The traditional ESPRIT signal parameter estimation method is used to calculate the signal source receiving data and solve the steering vector of each signal. .

[0045] In one embodiment of the present invention, the specific steps of S4 are as follows: S401: Matrix the steering vector estimation value of the signal to obtain a steering vector matrix. The steering vector matrix is ​​expressed as follows:

[0046]

[0047] in, For the The steering vector matrix of samples, For the The estimated value of the steering vector for samples, is the conjugate transpose, is the data dimension, For the The azimuth of the samples, For the The pitch angle of each sample, For the The polarization angle of the samples, For the The polarization phase difference of each sample.

[0048] S402: performing imaginary part processing, real part processing and angle processing on the steering vector matrix to obtain the imaginary part, real part and angle respectively. The expressions of the imaginary part, real part and angle are as follows respectively;

[0049]

[0050]

[0051] in, is the real part, is the imaginary part, is the angle, To take the real part operation, To take the imaginary part operation, To take the angle operation, For the The steering vector matrix of samples; S403: Create three-dimensional matrix data according to the imaginary part, the real part and the angle. The expression of the three-dimensional matrix data is as follows:

[0052] in, is three-dimensional matrix data; S404: According to the three-dimensional matrix data, labels are added to obtain a signal data set. The expression of the signal data set is as follows:

[0053] in, is the signal data set, is the label of the first sample, is the label of the second sample, For the The labels of samples, is the three-dimensional matrix data of the first sample, is the three-dimensional matrix data of the second sample, For the The three-dimensional matrix data of samples; where the label is a vector of a single non-zero element, which can represent the category of the signal. For example, the search angle of the signal is divided into 10 categories and the target angle is in the 7th category, then the label , if there is no target, then the label .

[0054] In S5, a DOA estimation calculation model is constructed, and 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 and second convolutional layers both include 256 convolution kernels and the convolution kernel size is ; The third, fourth, fifth and sixth convolutional layers all include 256 convolution kernels and the convolution kernel size is ; The first convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer, the fifth convolutional layer, and the sixth convolutional layer are all connected with batch normalization layers and ReLU activation functions; The first fully connected layer is connected with a ReLU activation function and a dropout layer; the second fully connected layer is connected with a sigmoid activation function to induce classification probability.

[0055] In one embodiment of the present invention, the network parameters of the convolutional layer are set to ,in The dimension is The convolution kernel, Indicates the number of input channels, represents the length and width of the convolution kernel, Represents the number of convolution kernels, that is, the number of output channels, The dimension is The bias vector of The network parameters of the fully connected network are ,in, The dimension is The weight matrix of Indicates the number of input channels, represents the number of neurons in the fully connected layer, that is, the number of output channels, The dimension is The bias vector of is the matrix multiplication operation; The first convolutional layer is used to extract features from the input data and uses the ReLU activation function. , , which is expressed as follows:

[0056] in, is the output of the second convolutional layer, is the dimension of the first convolutional layer, is the bias vector of the first convolutional layer, is a three-dimensional matrix data, is the convolution operation.

[0057] The second convolutional layer, the third convolutional layer, the fourth convolutional layer, and the fifth convolutional layer are all used to perform nonlinear mapping on the features, using the ReLU activation function. , , and their expressions are as follows:

[0058]

[0059]

[0060]

[0061] in, is the output of the fifth convolutional layer, is the dimension of the fifth convolutional layer, is the bias vector of the fifth convolutional layer, is the output of the fourth convolutional layer, is the dimension of the fourth convolutional layer, is the bias vector of the fourth convolutional layer, is the output of the third convolutional layer, is the dimension of the third convolutional layer, is the bias vector of the third convolutional layer, is the output of the second convolutional layer, is the dimension of the second convolutional layer, is the bias vector of the second convolutional layer; The sixth convolutional layer is used to reconstruct the output of the fifth convolutional layer. The output dimension is Datasets, , , which is expressed as follows:

[0062] in, is the output of the sixth convolutional layer, is the dimension of the sixth convolutional layer, is the bias vector of the sixth convolutional layer; The first fully connected layer is used to integrate the data output by the convolutional layer and uses the ReLU activation function. , , which is expressed as follows:

[0063] in, is the output of the first fully connected layer, is the dimension of the weight matrix of the first fully connected layer, is the bias vector of the first fully connected layer; The second fully connected layer is used to induce the classification probability and output it, using the sigmoid activation function. , , which is expressed as follows:

[0064] in, is the estimated value of DOA, is the dimension of the weight matrix of the second fully connected layer, is the bias vector of the second fully connected layer.

[0065] In S5, the DOA estimation calculation model is trained using the signal data set to obtain a trained DOA estimation calculation model; the loss function of the DOA estimation calculation model is expressed as follows:

[0066] in, is the loss value, is the true label, The output labels of the model are calculated for DOA estimation, is a real number, is the total number of samples.

[0067] In one embodiment of the present invention, during the training process of the DOA estimation calculation model, as the number of iterations increases, the learning rate is gradually reduced so that the network parameters gradually converge to the optimal or near-optimal state; at the same time, the Dropout strategy is adopted to randomly remove neurons with a certain probability to alleviate the dependence of the DOA estimation calculation model on the distribution of training set data and avoid overfitting problems; the parameters of the DOA estimation calculation model are trained by minimizing the loss function through the back-propagation algorithm, and the network parameters are updated using the Adam optimizer. After each forward propagation of the neural network, the error is back-propagated and the network weights are updated, and the network objective function is converged after repeated updates.

[0068] In S6, the signal source signal of the large electromagnetic vector sensor is processed using the trained DOA estimation calculation model to obtain a DOA estimation value, thereby completing the DOA estimation of the large electromagnetic vector sensor.

[0069] The beneficial effects of the present invention are as follows: the present invention introduces neural network deep learning into the DOA estimation of the polarization sensitive array. By establishing a DOA estimation calculation model and combining it with data driving, it can perform DOA estimation on a large electromagnetic vector sensor, solve the problem of complex and highly nonlinear response of the spatial phase factor and the large electromagnetic vector sensor itself, and can complete the separation of multiple targets and the DOA estimation after separation.

[0070] Example 2 In one embodiment of the present invention, the Deep Learning Toolbox in Matlab R2020b is used to implement the training of the DOA estimation calculation model.

[0071] S1: Establish a 6-component separated large electromagnetic vector sensor; the coordinates of the electric dipole are set as follows: 、 、 ; Magnetic ring , and The coordinates are: 、 、 ; Electric dipole , much larger than wavelengths, is the long electric dipole, and the radius of the magnetic ring , much larger than , belongs to the large magnetic ring, the other parameters are .

[0072] S2: Based on the 6-component separated large electromagnetic vector sensor, the signal source receiving data is received; the signal-to-noise ratio (SNR) of the signal source receiving data is set to 20dB, the number of snapshots is set to 10, the azimuth angle is 42 degrees, and the pitch angle is The variation range of is set to 0.1°-10°, which is evenly divided into discrete angles with an interval of 0.1°. The signal traverses each positioning range and is randomly selected in a ratio of 8:2 to produce training and test sets. The number of samples is 100, and each sample forms 2000 data. The total data set size is 200000, and the data ratio of the training set and the test set is 8:2.

[0073] S3: Based on the data received from the signal source, the signal parameter estimation method is used to calculate the estimated value of the signal steering vector; S4: preprocessing the steering vector estimation value of the signal to obtain a signal data set; S5: construct a DOA estimation calculation model, and use the signal data set for training to obtain a trained DOA estimation calculation model; Based on the convolutional layer and the fully connected layer, a DOA estimation calculation model is constructed. The hyperparameter settings of the DOA estimation calculation model are as follows in Table 1: Table 1

[0074] In the embodiment of the present invention, it can be determined whether to stop training according to the number of training rounds, or it can be determined whether training is finished according to the loss value of the loss function, such as Figure 2 As shown, it is an error variation diagram of the loss function. The loss amount of the loss function of the DOA estimation calculation model in the present invention is rapidly reduced, and the effect of rapid convergence can be achieved.

[0075] S6: Using the trained DOA estimation calculation model, the signal source signal of the large electromagnetic vector sensor is processed to obtain the DOA estimation value, thereby completing the DOA estimation of the large electromagnetic vector sensor.

[0076] In one embodiment of the present invention, a DOA estimation value is obtained by using a trained DOA estimation calculation model; For verification, since the signal-to-noise ratio (SNR) of the training set is 20 dB, four signal-to-noise ratios (SNR=-15 dB, SNR=-5 dB, SNR=5 dB, and SNR=15 dB) are selected to verify the results of the DOA estimation calculation model, as shown in Figure 2. Figure 3 As shown, it is the output label value under different signal-to-noise ratio conditions, where Figure 3 (a) Output label value and true value of the DOA estimation calculation model for SNR=-15dB, Figure 3 (b) Output label value and true value of the DOA estimation calculation model for SNR=-5dB, Figure 3 (c) Output label value and true value of the DOA estimation calculation model for SNR=5dB, Figure 3 (d) The label value and the true value output by 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 with high accuracy.

[0077] The beneficial effects of the present invention are as follows: the present invention introduces neural network deep learning into the DOA estimation of the polarization sensitive array, and by establishing a DOA estimation calculation model and combining it with data driving, it can perform DOA estimation on a large electromagnetic vector sensor, thereby solving the problems of complex and highly nonlinear spatial phase factors and the response of the large electromagnetic vector sensor itself, and can complete the DOA estimation of the large electromagnetic vector sensor with high accuracy.

Claims

1. A DOA estimation method based on neural network and large electromagnetic vector sensor, characterized in that: The following steps are involved: S1: Establish a 6-component separated large electromagnetic vector sensor; S2: According to the 6-component separated large electromagnetic vector sensor, the signal source receiving data is received; S3: Based on the data received from the signal source, the signal parameter estimation method is used to calculate the estimated value of the signal steering vector; S4: preprocessing the steering vector estimation value of the signal to obtain a signal data set; S5: construct a DOA estimation calculation model, and use the signal data set for training to obtain a trained DOA estimation calculation model; S6: Using the trained DOA estimation calculation model, the signal source signal of the large electromagnetic vector sensor is processed to obtain the DOA estimation value, thereby completing the DOA estimation of the large electromagnetic vector sensor.

2. The DOA estimation method based on neural network and large electromagnetic vector sensor according to claim 1 is characterized in that: The 6-component separated large electromagnetic vector sensor comprises three mutually orthogonal electric dipoles and three mutually orthogonal large magnetic rings; the electric dipoles and large magnetic rings are separated in space and are in arbitrary positions.

3. The DOA estimation method based on neural network and large electromagnetic vector sensor according to claim 1 is characterized in that: The specific steps of S4 are as follows: S401: Matrixing the steering vector estimation value of the signal to obtain a steering vector matrix; S402: performing imaginary part processing, real part processing and angle processing on the steering vector matrix to obtain the imaginary part, real part and angle respectively; S403: Create three-dimensional matrix data according to the imaginary part, the real part and the angle; S404: adding a label according to the three-dimensional matrix data to obtain a signal data set; the label is a vector of a single non-zero element.

4. The DOA estimation method based on neural network and large electromagnetic vector sensor according to claim 3 is characterized in that: The expression of the signal data set is as follows: in, is the signal data set, For labels, is a three-dimensional matrix data, is the total number of samples, is the real part, is the imaginary part, is the angle, To take the real part operation, To take the imaginary part operation, To take the angle operation, For the The steering vector matrix of samples, For the The estimated value of the steering vector for samples, is the conjugate transpose, is the data dimension, For the The azimuth of the samples, For the The pitch angle of each sample, For the The polarization angle of the samples, For the The polarization phase difference of each sample.

5. The DOA estimation method based on neural network and large electromagnetic vector sensor according to claim 1 is characterized in that: 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 convolution layer and the second convolution layer both include 256 convolution kernels and the convolution kernel size is ; The third convolution layer, the fourth convolution layer, the fifth convolution layer and the sixth convolution layer all include 256 convolution kernels and the convolution kernel size is ; The first convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer, the fifth convolutional layer and the sixth convolutional layer are all connected with 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.

6. The DOA estimation method based on neural network and large electromagnetic vector sensor according to claim 5 is characterized in that: The expression of the DOA estimate is as follows: in, is the estimated value of DOA, is the dimension of the weight matrix of the second fully connected layer, is the bias vector of the second fully connected layer, is the output of the first fully connected layer, is the dimension of the weight matrix of the first fully connected layer, is the bias vector of the first fully connected layer, is the output of the sixth convolutional layer, is the dimension of the sixth convolutional layer, is the bias vector of the sixth convolutional layer, is the output of the fifth convolutional layer, is the dimension of the fifth convolutional layer, is the bias vector of the fifth convolutional layer, is the output of the fourth convolutional layer, is the dimension of the fourth convolutional layer, is the bias vector of the fourth convolutional layer, is the output of the third convolutional layer, is the dimension of the third convolutional layer, is the bias vector of the third convolutional layer, is the output of the second convolutional layer, is the dimension of the second convolutional layer, is the bias vector of the second convolutional layer, is the output of the second convolutional layer, is the dimension of the first convolutional layer, is the bias vector of the first convolutional layer, is a three-dimensional matrix data, is the convolution operation.

7. The DOA estimation method based on neural network and large electromagnetic vector sensor according to claim 5 is characterized in that: The loss function of the DOA estimation calculation model is expressed as follows: in, is the loss value, is the true label, The output labels of the model are calculated for DOA estimation, is a real number, is the total number of samples.

Images