WB-CPRSN algorithm for mainlobe maintaining wideband beamforming

Abstract

Aiming at the problem that most existing mainlobe interference suppression wideband beamforming algorithms fail to properly capture and process the complex-valued information of interference signals, which leads to a decrease in their mainlobe interference suppression performance, an effective mainlobe maintaining wideband beamforming algorithm, named WB-CPRSN, is proposed based on a complex-valued processing residual shrinkage network (CPRSN). Initially, Eigen-projection Processing and Focusing Reconstruction (EPFR) algorithm is introduced to generate the dataset for the proposed neural network. Subsequently, a CPRSN model is proposed, in which a Complex-valued Attention Module (CAM) is incorporated to extract the mainlobe interference features. It is used to perceive global features and fuse them with the original data to enhance more important channel features, thereby improving the prediction performance of the network. The above model effectively mitigates the impact of interference on the mainlobe array gain by fully leveraging the phase information of complex-valued data and extracting mainlobe interference features to obtain an optimal beamforming weight vector. Finally, the well-trained CPRSN can rapidly predict near-optimal mainlobe maintaining wideband beamforming weight vectors. Simulation results indicate that the proposed WB-CPRSN algorithm has satisfactory performance against mainlobe interference and modest computational time. For example, the WB-CPRSN algorithm can form an accurate and narrow mainlobe, and achieve nulls lower than -70 dB in the two side lobe interference directions.