Learning Sparse Array Capon Beamformer Design Using Deep Learning Approach

Abstract

The paper considers receive beamforming for designing sparse array configurations. We consider to maximize signal-to-interference plus noise ratio (MaxSINR) for a desired point source in a narrowband interferening environment. The sparse array design methods can either be data driven or rely entirely on the prior knowledge of the interference DOAs and respective powers. In this paper, we propose a design which is essentially data-driven and conceived by training the Deep Neural Network (DNN). Towards this goal, the training scenarios are generated through enumeration to learn an effective representation that can perform well in a downstream task. The input to the DNN is the received correlation matrix and output is the corresponding sparse configuration with superior interference mitigation capability. The performance of the DNN is evaluated by the ability to learn the implementation of enumerated design. It is shown through design examples that DNN effectively learns the enumerated algorithm and, as such, paves the way for efficient real-time implementation.