Deep-Unrolling-Based Gridless Channel Estimation in Massive SIMO Systems With Antenna Failures

Abstract

In massive single input multiple output (SIMO) systems, parametric channel estimation is generally modeled as the parameter estimation problem including angles of spatial paths and powers of spatial paths. However, the antenna element failure (AEF) of the array and the angle discretization error caused by gridding degrade the parameter estimation accuracy. In this letter, a deep-unrolling-based gridless AEF channel estimation network (DU-GACE) is proposed to solve these issues. The framework of DU-GACE is designed as a gridless deep-unrolling network version of alternating projection algorithm. In each layer of DU-GACE, an AEF estimation sub-network and an eigenvalue-based sub-network are put forward, where the former estimates the AEF term and the latter is developed to detect the accurate number of spatial paths to assist the gridless angle estimation. Simulation results are provided to show that our network outperforms existing methods in terms of achievable spectrum efficiency, channel and angle estimation accuracy.