CQCNN-SV algorithm for wideband space–time adaptive processing

Abstract

This paper presents a wideband robust beamforming algorithm based on a complex quantized convolutional neural network (CQCNN) for solving the steering vector (SV) mismatch problem, named as CQCNN-SV algorithm. Firstly, the CQCNN is constructed by the complex convolution layers, quantization assistance layers, and normalization layers, respectively. Specially, the network channel filtering threshold function is used to construct the quantization assistance layer with the functions of network weight pruning. The CQCNN structure is suitable for wideband beamforming in space–time two-dimensional signal processing, which can improve the feature extraction ability and convergence speed of complex-valued data. Subsequently, the mismatched desired signal SV is corrected by solving the quadratic programming problem, and the corrected SV is treated as the training label. Finally, the space–time two-dimensional covariance matrix and the training label are fed into the CQCNN model. The wideband beamforming weight vector in the space–time antenna structure is given by the desired signal SV, which is predicted by the well-trained CQCNN. Theoretical analysis and simulation experiments show that the proposed algorithm not only has good real-time performance but also has stable system output performance.