HyperCEUNet: Parameter-Aware Hypernetwork-Driven UNet for Channel Estimation

TL;DR

HyperCEUNet introduces a parameter-aware deep learning framework that leverages channel parameters and correlation information to enhance channel estimation accuracy in wireless systems.

Contribution

It proposes a hypernetwork-driven UNet architecture that adaptively incorporates channel parameters for improved estimation performance.

Findings

HyperCEUNet outperforms conventional methods in simulation accuracy.

The adaptive front-end convolutional layer improves estimation by exploiting channel parameters.

Correlation-aware initialization enhances the effectiveness of the deep learning model.

Abstract

Deep learning-based channel estimation has been recognized as a promising technique for sixth-generation wireless systems. However, most existing approaches rely solely on least-squares estimates obtained from demodulation reference signals, which fail to explicitly exploit channel time-frequency correlation parameters. Inspired by the independent channel parameter estimation enabled by semi-static reference signals in modern wireless systems, this letter presents a parameter-aware deep learning-based channel estimation framework termed HyperCEUNet. Specifically, the proposed hypernetwork generates an adaptive front-end convolutional layer based on estimated channel parameters, serving as a pre-filtering stage before the UNet-based estimator. In addition, the Wiener-filtered channel estimates are adopted to provide a correlation-aware initialization for data resources. Simulation results demonstrate that our proposed HyperCEUNet effectively improves channel estimation accuracy compared with its conventional counterparts.