Research On 1-Bit DOA Estimation For Two L-shaped Arrays

Direction of arrival(DOA)estimation is a basic problem in radar,sonar and many other fields.The DOA method based on compressed sensing(CS)uses the sparsity of the parameters to be estimated to reduce the complexity.The DOA method based on deep learning adopts the idea of data-driven to improve the accuracy of parameter estimation.We proposed an expectation maximization generalized approximate message passing with block sparse structure(BSS-EM-GAMP)algorithm in this thesis.Aiming at the DOA estimation problem of high-dimensional 1-Bit quantization in sparse two L-shaped arrays,we first establish the model of 1-Bit quantization analog-to-digital conversion of the received signal,then construct the block sparse structure of the source signal,and then use BSS-EM-GAMP algorithm to recover the sparse signal,so as to improve the estimation performance.Due to the one-to-one correspondence between DOA and the position of non-zero elements in sparse signal,the estimated DOA can be obtained by estimating the position of non-zero elements in sparse signal.The proposed BSS-EM-GAMP algorithm can process wideband signals and non-circular signals.It still has good performance in the case of unknown number of sources,and can further improve the estimation performance under multiple snapshots case.Simulation results show that the proposed BSS-EM-GAMP algorithm achieves good performance in the case of low SNR and large sparse array.In this thesis,we also designed a learned complex generalized approximate message passing(LCGAMP)network based on deep learning,and dealt with the DOA estimation problem of high-dimensional 1-Bit quantization in sparse two L-shaped arrays.Firstly,the iterative process of generalized approximate message passing(GAMP)algorithm is unrolled into a network.Secondly,the network structure is designed to separate the real part and the imaginary part of the source signal to achieve the purpose of processing complex values.Finally,the parameters of the network are trained by using the randomly generated training data.The proposed LCGAMP network combines the advantages of data-driven and model-driven to improve the performance of DOA estimation.The simulation results show that the LCGAMP network can maintain good performance under low SNR when the source is unknown,and can correctly estimate the angle when the DOA value is close to±90°.