阅读说明：本技术 基于深度学习的任意阵列波达角估计方法 (Method for estimating wave arrival angle of any array based on deep learning ) 是由 王杰贵 刘方正 刘有军 韩振中 于 2020-04-08 设计创作，主要内容包括：本发明公开了基于深度学习的任意阵列波达角估计方法,具体步骤如下：对各阵元接收到的辐射源信号下变频和数字采样处理,获得采样数据；对采样数据进行信号处理,提取各阵元采样数据的特征信息,并将得到的采样数据和提取的特征信息一同作为深度学习的输入数据；基于已知的辐射源信号来建立训练样本集,并将其作为深度学习的训练数据；深度学习模型初始化,即卷积神经网络参数初始化；利用优化算法进行深度学习训练,得到优化分类器；利用深度学习模型和优化分类器,基于采样数据和提取的特征信息,来估计辐射源信号的波达角；本发明能够适应任意的多元阵列,无需进行通道校正,于深度学习的基础上,实现对辐射源信号的快速高精度测向。(The invention discloses an arbitrary array wave arrival angle estimation method based on deep learning, which comprises the following specific steps: carrying out down-conversion and digital sampling processing on radiation source signals received by each array element to obtain sampling data; carrying out signal processing on the sampling data, extracting characteristic information of the sampling data of each array element, and taking the obtained sampling data and the extracted characteristic information as input data of deep learning; establishing a training sample set based on known radiation source signals, and taking the training sample set as deep learning training data; initializing a deep learning model, namely initializing parameters of a convolutional neural network; carrying out deep learning training by using an optimization algorithm to obtain an optimized classifier; estimating the arrival angle of the radiation source signal based on the sampled data and the extracted feature information by using a deep learning model and an optimization classifier; the method can adapt to any multi-element array, does not need to carry out channel correction, and realizes quick and high-precision direction finding of the radiation source signal on the basis of deep learning.)

1. The method for estimating the wave arrival angle of any array based on deep learning is characterized by comprising the following specific steps of:

the method comprises the following steps: carrying out down-conversion and digital sampling processing on radiation source signals received by each array element to obtain sampling data;

step two: carrying out signal processing on the sampling data, extracting characteristic information of the sampling data of each array element, and taking the obtained sampling data and the extracted characteristic information as input data of deep learning;

step three: establishing a training sample set based on known radiation source signals, and taking the training sample set as deep learning training data;

step four: initializing a deep learning model, namely initializing parameters of a convolutional neural network;

step five: carrying out deep learning training by using an optimization algorithm to obtain an optimized classifier;

step six: and estimating the arrival angle of the radiation source signal based on the sampling data and the extracted characteristic information by using a deep learning model and an optimization classifier.

2. The method for estimating any array wave arrival angle based on deep learning of claim 1, wherein the deep learning model initialization, that is, the convolutional neural network parameter initialization, is expressed as training a convolutional neural network by using a training sample set, and automatically extracting and classifying features associated with a target arrival angle in the training sample set, in a specific manner as follows:

the first step is as follows: for the formed training sample set P, the total training sample amount is M, and the ith sample is P (X)_i，Y_i) Wherein X is_iFor sample data and extracted feature information, Y_iAn angle of arrival label for the ith training sample, i 1, 2.. M;

the second step is that: the connection weight initialization of the network adopts an Xavier initialization mode, the uniform distribution is obeyed, and the bias term of the network is initialized to be 0;

the third step: the convolutional neural network structure for estimating the signal arrival angle consists of 3 convolutional layers, 3 pooling layers, 1 full-connection layer and 1 Softmax classification layer; each convolution layer is respectively followed by an averagepool layer, and the pool layers respectively adopt small windows of 1 x 4, 1 x 4 and 1 x 2 to carry out non-overlapping down-sampling on the convolved features; each convolution layer adopts a Sigmoid activation function, and the step length of each convolution layer is 1; inputting original data with the size of 1 × 500, outputting 6 characteristic graphs with the size of 1 × 488 after convolution by a convolution kernel with the size of 1 × 13, changing the size of the characteristic graphs into 1 × 122 after average pooling, sending the characteristic graphs to a second convolution layer, wherein the size of the convolution kernel is 1 × 11, outputting 12 characteristic graphs with the size of 1 × 112, changing the size of the characteristic graphs into 1 × 28 after second pooling, then obtaining 30 characteristic graphs with the size of 1 × 12 after third pooling, connecting the characteristic graphs into a one-dimensional vector, fully connecting the vector with nodes of a fully-connected layer, and finally obtaining an arrival angle of a radiation source after Softmax classification, wherein the classification layer adopts a cross entropy loss function.

Technical Field

The invention relates to the technical field of electronic reconnaissance, in particular to an arbitrary array DOA estimation method based on deep learning.

Background

The estimation of the arrival angle of the radiation source signal is to estimate the arrival direction of the radiation source signal by utilizing the radiation source signal received by electronic reconnaissance and through signal processing; the existing method for estimating the angle of arrival of a radiation source signal mainly adopts modes of amplitude method direction finding, phase method direction finding, space spectrum estimation direction finding and the like.

The direction finding by the amplitude method is to estimate the arrival angle by utilizing the received signal amplitude information, the direction finding precision of the method is lower, the direction finding by the single-vibration-element amplitude method adopts a search method, the instantaneous coverage range is small, the direction finding by the multi-vibration-element amplitude method has certain requirements on the antenna arrangement, namely the relative position and the relative angle of each array element are required, and the requirement on the amplitude consistency of a multi-array-element antenna directional diagram and multiple channels is higher.

The phase method direction finding is to estimate the arrival angle by using the phase difference information received by a plurality of array elements, typically, a multi-baseline phase interferometer direction finding is provided, the method has certain requirements on the arrangement of antennas, namely, certain requirements on the space between the array elements, higher requirements on the phase consistency of multiple channels, different arrangements, different direction finding resolving models, direction finding ambiguity and the need of ambiguity resolution.

The space spectrum estimation direction finding is the estimation of the angle of arrival by using the modern spectrum estimation technology and taking MUSIC as a representative, and the method theoretically has high estimation precision, but has high requirements on the array element arrangement and the channel consistency, large operation amount, difficult meeting the real-time processing requirement and less engineering application.

Namely, the existing wave arrival angle estimation method mainly has the following defects: 1. certain requirements are required for the arrangement of the antenna array, and the array and the direction-finding model are different; 2. the requirement on the consistency of the channels is high, and the channel correction is usually required; 3. the direction-finding precision of the amplitude method is low; 4. direction finding ambiguity exists in the direction finding of the phase method, and ambiguity resolution processing is needed; 5. the spatial spectrum estimation hardly meets the real-time requirement;

in order to solve the above-mentioned drawbacks, a technical solution is now provided.

Disclosure of Invention

The invention aims to provide an arbitrary array DOA estimation method based on deep learning, which is irrelevant to array arrangement, can adapt to arbitrary multi-element arrays, has low requirement on channel consistency, does not need channel correction, and realizes quick and high-precision direction finding of radiation source signals on the basis of deep learning.

The technical problems to be solved by the invention are as follows:

how to solve the problems that the existing wave arrival angle estimation method has certain requirements on the arrangement of an antenna array, and the array is different and the direction-finding model is different according to an effective mode; the requirement on the consistency of the channels is high, and the channel correction is usually required; the direction-finding precision of the amplitude method is low; the phase method direction finding has the problems of direction finding ambiguity, and the real-time requirement is difficult to meet due to the need of ambiguity resolution processing and spatial spectrum estimation.

The purpose of the invention can be realized by the following technical scheme:

the method for estimating the wave arrival angle of any array based on deep learning comprises the following specific steps:

the method comprises the following steps: carrying out down-conversion and digital sampling processing on the radiation source signals received by each array element to obtain sampling data, namely carrying out filtering amplification and down-conversion on the radiation source signals received by each array element, and carrying out A/D conversion on the radiation source signals of intermediate frequency to obtain digital signals;

step two: carrying out signal processing on the sampling data, extracting characteristic information of the sampling data of each array element, wherein the characteristic information comprises signal frequency, phase difference between the array elements (fuzzy solution is not needed) and the like, and using the obtained sampling data and the extracted characteristic information as input data of deep learning;

step three: establishing a training sample set based on known radiation source signals, and taking the training sample set as deep learning training data, namely respectively placing the known radiation sources at different angles, obtaining sampling data, extracting associated characteristic information to establish the training sample set, and taking the training sample set as the deep learning training data;

step four: initializing a deep learning model, namely initializing parameters of a convolutional neural network;

step five: deep learning training is carried out by utilizing an optimization algorithm to obtain an optimized classifier, namely a training sample set P is input into a convolutional neural network for network training, training is carried out according to a preset epoch number, and in the training process, an Adam optimization algorithm is adopted to train the convolutional neural network;

step six: and estimating the arrival angle of the radiation source signal based on the sampling data and the extracted characteristic information by using a deep learning model and an optimization classifier.

Further, the deep learning model initialization, that is, the convolutional neural network parameter initialization is expressed as training a convolutional neural network by using a training sample set, and automatically extracting and classifying features associated with a target arrival angle in the training sample set, specifically, the following method is adopted:

the first step is as follows: for the formed training sample set P, the total training sample amount is M, and the ith sample is P (X)_i，Y_i) Wherein X is_iFor sample data and extracted feature information, Y_iAn angle of arrival label for the ith training sample, i 1, 2.. M;

the second step is that: the connection weight initialization of the network adopts an Xavier initialization mode, the uniform distribution is obeyed, and the bias term of the network is initialized to be 0;

the third step: the convolutional neural network structure for estimating the signal arrival angle consists of 3 convolutional layers, 3 pooling layers, 1 full-connection layer and 1 Softmax classification layer; each convolution layer is respectively followed by an averagepool layer, and the pool layers respectively adopt small windows of 1 x 4, 1 x 4 and 1 x 2 to carry out non-overlapping down-sampling on the convolved features; each convolution layer adopts a Sigmoid activation function, and the step length of each convolution layer is 1; inputting original data with the size of 1 × 500, outputting 6 characteristic graphs with the size of 1 × 488 after convolution by a convolution kernel with the size of 1 × 13, changing the size of the characteristic graphs into 1 × 122 after average pooling, sending the characteristic graphs to a second convolution layer, wherein the size of the convolution kernel is 1 × 11, outputting 12 characteristic graphs with the size of 1 × 112, changing the size of the characteristic graphs into 1 × 28 after second pooling, then obtaining 30 characteristic graphs with the size of 1 × 12 after third pooling, connecting the characteristic graphs into a one-dimensional vector, fully connecting the vector with nodes of a fully-connected layer, and finally obtaining an arrival angle of a radiation source after Softmax classification, wherein the classification layer adopts a cross entropy loss function.

The invention has the beneficial effects that:

the method disclosed by the invention is irrelevant to array arrangement, can be suitable for any multi-element array, has low requirement on channel consistency, does not need to carry out channel correction, and realizes quick and high-precision direction finding on radiation source signals on the basis of deep learning, wherein the direction finding precision is superior to 0.2 degree.

Drawings

In order to facilitate understanding for those skilled in the art, the present invention will be further described with reference to the accompanying drawings;

fig. 1 is a schematic diagram of an antenna array arrangement according to the present invention;

FIG. 2 is a flow chart of the steps of the present invention;

FIG. 3 is a diagram of a convolutional neural network architecture of the present invention.

Detailed Description

As shown in fig. 1-3, the method for estimating the wave arrival angle of any array based on deep learning, the mathematical model:

the traditional direction-finding system adopts linear arrangement or circular arrangement, which is respectively shown in fig. 1(a) and 1(b), and is not suitable for irregular arrangement shown in fig. 1 (c);

the method of the invention is irrelevant to array arrangement, can be suitable for any multi-element array, and takes irregular arrangement as shown in figure 1(c) as an example to establish a one-dimensional wave arrival angle estimation model;

the direction-finding array is composed of L antenna elements, the k-th element position vector X_k＝(x_k，y_k) If the incident signal source is a narrow-band far-field signal, the number of the signal sources is p, p is smaller than L, the incident direction of the signal is θ i, i is 1, 2.

Thereinλ is signal wavelength, dk (θ i) is wave path difference of the kth array element and the reference array element in the ith signal direction, aki is phase adjustment factor of the ith signal to the kth array element relative to the reference array element, and nk (t) is noise;

the above equation is written in matrix form as:

Z＝AS+N

and A, S, Z, N and a therein_iThe explanation is as follows:

A＝[a₁a₂… a_p]

S＝[s₁(t) s₂(t) … s_p(t)]^T

Z＝[z₀(t) z₁(t) … z_L-1(t)]^T

N＝[n₀(t) n₁(t) … n_L-1(t)]^T

a_i＝[a_0ia_1i… a_(L-1)i]^T；

the method for estimating the wave arrival angle of any array based on deep learning comprises the following specific steps:

the method comprises the following steps: carrying out down-conversion and digital sampling processing on the radiation source signals received by each array element to obtain sampling data, namely carrying out filtering amplification and down-conversion on the radiation source signals received by each array element, and carrying out A/D conversion on the radiation source signals of intermediate frequency to obtain digital signals;

step two: carrying out signal processing on the sampling data, extracting characteristic information of the sampling data of each array element, wherein the characteristic information comprises signal frequency, phase difference between the array elements (fuzzy solution is not needed) and the like, and using the obtained sampling data and the extracted characteristic information as input data of deep learning;

the extraction of the phase difference between the array elements adopts a frequency domain phase discrimination algorithm, and takes the first path of signal and the second path of signal as an example, the phase difference of the two paths of signals comprises the phase difference caused by the wave path difference of the signals and the phase difference caused by the inconsistency of the channels;

let the first path signal be x₁(t), the second signal is x₂(t), the two signals are conventional signals, and the Fourier transform of the two signals is respectively represented as:

wherein, τ₀₁The time delay of the signal reaching two array elements;

and the conjugate multiplication of the first path of signal and the second path of signal is used on the frequency domain to obtain:

Y(f)＝X₁(f)[exp(-j2πfτ₀₁)X₁(f)]^*＝exp(j2πfτ₀₁)|X₁(f)|²

i.e. the measured phase difference is expressed as:

estimating the carrier frequency of the signal, and solving the position of the maximum value corresponding to the carrier frequency on the digital frequency spectrum to obtain the accurate phase difference;

then, considering the phase inconsistency of the channels, that is, the phase difference of the two actually obtained signals is expressed as:

wherein phi₀₁Is the phase difference caused by the inconsistency of the two channels;

the extracted characteristic data actually contains phase difference information obtained by array element arrangement and phase difference information caused by channel inconsistency;

step three: establishing a training sample set based on known radiation source signals, and taking the training sample set as deep learning training data, namely respectively placing the known radiation sources at different angles, obtaining sampling data, extracting associated characteristic information to establish the training sample set, and taking the training sample set as the deep learning training data;

step four: initializing a deep learning model, namely initializing parameters of a convolutional neural network;

the convolutional neural network is trained by utilizing a training sample set, and the features associated with the target arrival angle in the training sample set are automatically extracted and classified, wherein the specific mode is as follows:

the first step is as follows: for the formed training sample set P, the total training sample amount is M, and the ith sample is P (X)_i，Y_i) Wherein X is_iFor sample data and extracted feature information, Y_iAn angle of arrival label for the ith training sample, i 1, 2.. M;

the second step is that: the connection weight initialization of the network adopts an Xavier initialization mode, the uniform distribution is obeyed, and the bias term of the network is initialized to be 0;

the third step: the convolutional neural network structure for estimating the signal arrival angle consists of 3 convolutional layers, 3 pooling layers, 1 full-connection layer and 1 Softmax classification layer; each convolution layer is respectively followed by an averagepool layer, and the pool layers respectively adopt small windows of 1 x 4, 1 x 4 and 1 x 2 to carry out non-overlapping down-sampling on the convolved features; each convolution layer adopts a Sigmoid activation function, and the step length of each convolution layer is 1; inputting original data with the size of 1 × 500, outputting 6 characteristic graphs with the size of 1 × 488 after convolution by a convolution kernel with the size of 1 × 13, changing the size of the characteristic graphs into 1 × 122 after average pooling, sending the characteristic graphs to a second convolution layer, wherein the size of the convolution kernel is 1 × 11, outputting 12 characteristic graphs with the size of 1 × 112, changing the size of the characteristic graphs into 1 × 28 after second pooling, then obtaining 30 characteristic graphs with the size of 1 × 12 after third pooling, connecting the characteristic graphs into a one-dimensional vector, fully connecting the vector with nodes of a fully-connected layer, and finally obtaining an arrival angle of a radiation source through a Softmax classification layer, wherein the classification layer adopts a cross entropy loss function;

step five: deep learning training is carried out by utilizing an optimization algorithm to obtain an optimized classifier, namely a training sample set P is input into a convolutional neural network for network training, training is carried out according to a preset epoch number, and in the training process, an Adam optimization algorithm is adopted to train the convolutional neural network;

step six: estimating the arrival angle of the radiation source signal based on the sampled data and the extracted feature information by using a deep learning model and an optimization classifier;

experimental analysis: the frequency of the radiation source signal is 9600MHz, the radiation source is placed at 0, 25, 50, 75, 100, 125 and 150 degrees respectively in the antenna array form as in fig. 1(c), direction finding is performed based on a single pulse by using the direction finding method proposed herein, and the obtained experimental results are shown in table 1:

TABLE 1 simulation results of single pulse direction finding

The experimental result shows that the arbitrary array wave arrival angle estimation method based on deep learning can adapt to an arbitrary multi-element array, and the direction finding precision is still better than 0.2 degree under the condition of not carrying out channel correction; the method of the invention is irrelevant to array arrangement, has low requirement on channel consistency, does not need to carry out channel correction, and realizes quick and high-precision direction finding of radiation source signals on the basis of deep learning.

The foregoing is merely exemplary and illustrative of the present invention and various modifications, additions and substitutions may be made by those skilled in the art to the specific embodiments described without departing from the scope of the invention as defined in the following claims.