Deep Learning based Channel Estimation for Massive MIMO with Hybrid Transceivers

TL;DR

This paper introduces a deep learning framework for uplink channel estimation in hybrid massive MIMO systems, leveraging angular space segmentation and neural networks to improve accuracy and reduce complexity.

Contribution

It proposes a novel angular space segmentation method and region-specific neural networks for efficient and accurate channel estimation in HAD massive MIMO systems.

Findings

Outperforms state-of-the-art CS algorithms in estimation accuracy

Reduces computational complexity significantly

Enhances channel estimation capability through joint optimization

Abstract

Accurate and efficient estimation of the high dimensional channels is one of the critical challenges for practical applications of massive multiple-input multiple-output (MIMO). In the context of hybrid analog-digital (HAD) transceivers, channel estimation becomes even more complicated due to information loss caused by limited radio-frequency chains. The conventional compressive sensing (CS) algorithms usually suffer from unsatisfactory performance and high computational complexity. In this paper, we propose a novel deep learning (DL) based framework for uplink channel estimation in HAD massive MIMO systems. To better exploit the sparsity structure of channels in the angular domain, a novel angular space segmentation method is proposed, where the entire angular space is segmented into many small regions and a dedicated neural network is trained offline for each region. During online testing, the most suitable network is selected based on the information from the global positioning system. Inside each neural network, the region-specific measurement matrix and channel estimator are jointly optimized, which not only improves the signal measurement efficiency, but also enhances the channel estimation capability. Simulation results show that the proposed approach significantly outperforms the state-of-the-art CS algorithms in terms of estimation performance and computational complexity.