Spatial sensitivity synthesis based on alternate projection for the machine-learning-based coding digital receiving array

Abstract

Recently, a novel low-cost coding digital receiving array based on machine learning (ML-CDRA) has been proposed to reduce the required radio frequency channels in modern wireless systems. The spatial sensitivity of ML-CDRA is studied which describes the spatial accumulation gain in different directions. It is demonstrated that the spatial sensitivity is determined by the encoding network, decoding network, and beamforming criterion. To obtain the desired spatial sensitivity, a spatial sensitivity synthesis method is proposed based on the alternate projection by optimising the encoding network with the constraint of amplitude-phase quantisation. Simulation results show that the proposed method can significantly improve the spatial sensitivity of ML-CDRA. Furthermore, in the directions of interest, the spatial accumulation gain of ML-CDRA can exceed the full-channel digital receiving array.