A Multiple Measurement Vector-Based Deep Unfolded Network for One-bit DOA Estimation

Abstract

This paper introduces a new direction-of-arrival (DOA) estimation method for multi-snapshot narrowband signals. To reduce the system cost, we adopt one-bit compressed sensing in the process of sampling and quantization for analog signals. We propose a deep unfolded network (DUN) based on multiple measurement vectors (MMVs), known as the learned MMV-based binary iteration soft threshold (L-MMV-BIST) network, to estimate the DOAs from the one-bit measurements. This new DUN is designed by unfolding each update of the binary iterative soft threshold algorithm (ISTA) into a layer of a deep neural network, thus it has the ability to learn soft threshold and other iteration parameters adaptively. Our simulation results show that the L-MMV -BIST network can estimate DOA information from the one-bit measurements. In addition, this network outperforms traditional BIST algorithm in both computational complexity and recovery accuracy.