Deep Learning for Beamspace Channel Estimation in Millimeter-Wave Massive MIMO Systems

TL;DR

This paper introduces a novel deep learning-based beamspace channel estimation method for millimeter-wave massive MIMO systems, leveraging Gaussian mixture models to improve accuracy over existing algorithms.

Contribution

The paper proposes a prior-aided GM-LAMP network that incorporates Gaussian mixture priors into the AMP framework for enhanced channel estimation accuracy.

Findings

GM-LAMP outperforms traditional AMP and LAMP in accuracy

Simulation results validate the effectiveness on theoretical and ray-tracing channels

The method reduces estimation error significantly

Abstract

Millimeter-wave massive multiple-input multiple-output (MIMO) can use a lens antenna array to considerably reduce the number of radio frequency (RF) chains, but channel estimation is challenging due to the number of RF chains is much smaller than that of antennas. By exploiting the sparsity of beamspace channels, the beamspace channel estimation can be formulated as a sparse signal recovery problem, which can be solved by the classical iterative algorithm named approximate message passing (AMP), and its corresponding version learned AMP (LAMP) realized by a deep neural network (DNN). However, these existing schemes cannot achieve satisfactory estimation accuracy. To improve the channel estimation performance, we propose a prior-aided Gaussian mixture LAMP (GM-LAMP) based beamspace channel estimation scheme in this paper. Specifically, based on the prior information that beamspace channel elements can be modeled by the Gaussian mixture distribution, we first derive a new shrinkage function to refine the AMP algorithm. Then, by replacing the original shrinkage function in the LAMP network with the derived Gaussian mixture shrinkage function, a prior-aided GM-LAMP network is developed to estimate the beamspace channel more accurately. Simulation results by using both the theoretical channel model and the ray-tracing based channel dataset show that, the proposed GM-LAMP network can achieve better channel estimation accuracy.