Research On DOA Estimation Algorithm Of Neural Network In Antenna Arra

As one of the key technologies in the field of array signal processing,Direction of arrival Direction of arrival(DOA)estimation is widely used in the fields of communication,navigation,radio astronomy and so on.In recent ten years,some researchers have introduced machine learning technology into the field of DOA estimation.Artificial neural network has become a promising tool because of its good nonlinear characteristics,adaptive characteristics,ability to learn from the environment in a supervised or unsupervised way and extensive global approximation ability It is a natural tool for solving nonlinear signal processing problems.Its good fault tolerance,powerful parallel implementation ability and distributed information storage and computing ability make it suitable for solving the problems of high complexity and large amount of data signal processing.Compared with the traditional DOA estimation algorithm,the DOA algorithm based on neural network has stronger robustness under low signal-to-noise ratio(SNR)and higher estimation accuracy under the condition of array defects.This paper mainly studies the application of neural network in DOA estimation,the main contents are as follows:1.Aiming at the problem that most conventional DOA estimation algorithms are poorly adapted to the defects of antenna arrays,DOA estimation under extreme noise is studied from the perspective of deep learning.In this work,this paper first proposes a denoising automatic encoder network based on artificial neural network,which can predict the covariance matrix closer to the theory,the angle can be estimated by any covariance based method.Then,combining the idea of single bit quantization with it,a DOA estimation algorithm for single bit denoising automatic encoder is proposed.The experimental results show that the covariance based DOA estimator has significant performance improvement in root mean square error(RMSE)at low SNR.However,the proposed method is independent of DOA estimator and opens up new feasibility for the testing of other methods.2.Aiming at the problem that most traditional DOA estimation algorithms have poor adaptability to the defects of antenna array,a DOA estimation algorithm based on residual neural network is studied.For the problem that the number of signal sources needs to be known in advance in the traditional DOA estimation based on neural network,a spatial signal classification network is proposed.The network can be regarded as the preprocessing step of the proposed method,which can divide the incoming signals in different directions into corresponding spatial regions,so there is no need to know the number of sources in advance,and it can also reduce the generalization burden of the whole network.In order to solve the meshing problem of compressed sensing theory,linear interpolation method is introduced.The linear interpolation algorithm can be used to calculate the DOA value closer to the real value.In order to solve the problem that the traditional depth learning method locates the signal source on a very rough grid with a grid spacing of 5° or even 10°,this rough DOA estimation can not meet the accuracy requirements in most general DOA estimation applications.The neural network structure proposed in this paper is improved by setting the grid spacing to 1°,which greatly improves the recognition accuracy and resolution.Simulation results show that the proposed method is more robust than other DOA algorithms based on neural network in the case of array defects.In addition,in different cases of array defect degree,it also has higher estimation accuracy than the traditional DOA algorithm.