Research On DOA Estimation Method Of Array Signal Based On Machine Learning

Direction of arrival(DOA)estimation is an important branch of array signal processing.It is widely used in wireless communication,public security,environmental monitoring,radar detection and other fields.Machine learning is one of the most intelligent and cutting-edge research fields in artificial intelligence.It is also one of the most practical and convenient methods to process massive data and obtain effective information in the era of big data.With the development of machine learning,using machine learning method to realize high-precision,real-time and intelligent DOA estimation has become a research hotspot in recent years.The main work of this dissertation is as follows:1.Aiming at the problem that the traditional DOA estimation method is only suitable for specific antenna arrays,the received signal model and input parameter extraction model are established to extract the phase difference characteristics,relative amplitude characteristics and covariance matrix characteristics of the received signal,which are suitable for any antenna array,and an evaluation model is established to evaluate the estimation performance of the DOA estimation method.2.A k-nearest neighbor(KNN)algorithm is implemented At the same time,aiming at the problem that it is difficult to measure the sample similarity in the k-nearest neighbor algorithm,an angle of arrival estimation method based on the improved knearest neighbor algorithm is proposed by introducing the information entropy distance.The simulation results show that the angle of arrival estimation method based on the knearest neighbor algorithm can achieve high-precision and real-time angle of arrival estimation,and the angle of arrival based on the improved k-nearest neighbor algorithm The estimation method has further improvement in estimation accuracy.3.A DOA estimation method based on convolutional neural networks(CNN)is implemented.Aiming at the gradient disappearance caused by too many layers in convolutional neural networks,long short term memory(LSTM)is integrated into convolutional neural networks,and two task types of classification and regression are distinguished at the same time,A DOA estimation method based on cnn-lstm is proposed.The simulation results show that the DOA estimation method based on convolutional neural network can achieve high-precision DOA estimation.Compared with the DOA estimation method based on convolutional neural network,the estimation accuracy of the classification model and regression model based on cnn-lstm is improved,and the prediction time of a single sample of the classification model is significantly reduced.4.Aiming at the problem of redundant time cost caused by repeated parameter adjustment in k-nearest neighbor algorithm and convolutional neural network,a DOA estimation method based on width concurrent forest is proposed.The random forest algorithm is combined with concurrent learning and introduced into width concurrent forest model,taking into account the robustness of traditional random forest and the low response time of concurrent learning,The width operation replaces the depth expansion in the neural network,which speeds up the learning and decision-making time.Simulation results show that this method can achieve DOA estimation with high stability and low response time.