Learning-based Block-wise Planar Channel Estimation for Time-Varying MIMO OFDM

TL;DR

This paper introduces a learning-based block-wise planar channel estimator for MIMO OFDM systems that models the channel efficiently and leverages deep learning to improve estimation accuracy in time-varying environments.

Contribution

It proposes a novel block-wise planar channel model combined with a 3D dilated residual CNN for enhanced, low-complexity channel estimation in MIMO OFDM systems.

Findings

Outperforms state-of-the-art estimators in accuracy.

Maintains low computational complexity.

Effectively captures channel correlations across domains.

Abstract

In this paper, we propose a learning-based block-wise planar channel estimator (LBPCE) with high accuracy and low complexity to estimate the time-varying frequency-selective channel of a multiple-input multiple-output (MIMO) orthogonal frequency-division multiplexing (OFDM) system. First, we establish a block-wise planar channel model (BPCM) to characterize the correlation of the channel across subcarriers and OFDM symbols. Specifically, adjacent subcarriers and OFDM symbols are divided into several sub-blocks, and an affine function (i.e., a plane) with only three variables (namely, mean, time-domain slope, and frequency-domain slope) is used to approximate the channel in each sub-block, which significantly reduces the number of variables to be determined in channel estimation. Second, we design a 3D dilated residual convolutional network (3D-DRCN) that leverages the time-frequency-space-domain correlations of the channel to further improve the channel estimates of each user. Numerical results demonstrate that the proposed significantly outperforms the state-of-the-art estimators and maintains a relatively low computational complexity.