Robust Hybrid Convolutional Beamspace for High Resolution DOA Estimation

Abstract

Hybrid beamforming realizes a combination of analog and digital processing of sensor data using a reduced number of digital channels. The analog combiner in the analog stage decreases the number of analog-to-digital converters (ADCs), resulting in minimized hardware cost and power consumption. Recently, a hybrid convolutional beamspace (CBS) beamformer was proposed to achieve lower computational complexity, higher direction-of-arrival (DOA) resolution, and a smaller mean-squared-error (MSE). However, the hybrid CBS method assumes a well-calibrated uniform linear array (ULA) for DOA estimation, which is vulnerable to gain and phase mismatches in the sensor elements that degrade the performance of the hybrid beamformer. To address this limitation, in this paper, a robust hybrid convolutional beamspace method is proposed for DOA estimation, where calibration and high-resolution estimation of source directions are carried out simultaneously in the beamspace using variational inference.