Sparse Array Configuration Analysis and Deep Learning Classifications for Beamfornming

Sparse arrays offer additional spatial degrees of freedom associated with nonuniform inter-element sensor spacing over the array aperture. This flexibility enables both the array configuration and array weights to play an important role in optimum beamforming. In this paper, we examine the separate role of sparse array configuration when decoupled from the array weight design. In this scenario, the array configuration is optimized to orthogonalize the desired source and the interference subspace, where the array weights are unoptimized and correspond to conventional beamforming, matching the desired source steering vector. We consider two sources with variable angles and analyze specific and most frequently sparse configurations chosen by the orthogonalization criterion, in lieu of maximum signal-to-interference and noise ratio (SINR) beamforming. Identifying these configurations proves beneficial in reducing the number of labels when designing a sparse array configuration common to all desired directions and in a dynamic environment based on deep learning (DL).