ADAPTIVE MULTI-RESOLUTION SUPPORT REFINEMENT VIA MODEL-DRIVEN DEEP UNFOLDING FOR PILOT-EFFICIENT MASSIVE MIMO CHANNEL ESTIMATION

Abstract

The ability to estimate channel accurately is critical in 5G Massive MIMO systems because they have very large antenna arrays which result in an increased complexity of the channel estimation and pilot overhead due to the limited number of pilot symbols available to conduct the estimation process. Many of the conventional methods for compressive sensing will fail to recover the support of the estimated channel when the SNR is very low. In addition, current deep unfolding (model-driven) methods for estimating channels are generally based upon fixed-resolution support estimates. The framework that will be proposed in this paper, known as Adaptive Multi-Resolution Support Refinement (AMRSR), uses deep unfolding methods (model-driven) to estimate the channel with pilot-efficient 5G massive MIMO channel estimations through an adaptive multi-resolution process. More specifically, the proposed method employs a hierarchical coarse-to-fine approach to estimate support accurately, thereby refining the support and coefficients for the estimated channel across the deep unfolding stages. The proposed method will be evaluated using a 5G massive MIMO channel model under different SNR levels and pilot overhead scenarios and is compared to conventional OMP/Block OMP and existing deep unfolding techniques by using NMSE, support recovery accuracy, spectral efficiency, and pilot overhead as metrics. The results of the simulations demonstrate the proposed method outperforms existing techniques and is a feasible solution for future wireless communication systems.