Sensing-Assisted Channel Prediction in Complex Wireless Environments: An LLM-Based Approach

TL;DR

This paper introduces an LLM-based method for sensing-assisted channel prediction in complex wireless environments, leveraging shared scatterers and CSI data to improve prediction accuracy in multi-antenna OFDM systems.

Contribution

It proposes a novel LLM-based approach that adapts pre-trained text models to handle complex CSI data for enhanced channel prediction.

Findings

LLM-based approach outperforms conventional deep learning methods

Significant improvement over benchmark schemes without sensing

Effective utilization of shared environment information for prediction

Abstract

This letter studies the sensing-assisted channel prediction for a multi-antenna orthogonal frequency division multiplexing (OFDM) system operating in realistic and complex wireless environments. In this system,an integrated sensing and communication (ISAC) transmitter leverages the mono-static sensing capability to facilitate the prediction of its bi-static communication channel, by exploiting the fact that the sensing and communication channels share the same physical environment involving shared scatterers. Specifically, we propose a novel large language model (LLM)-based channel prediction approach,which adapts pre-trained text-based LLM to handle the complex-matrix-form channel state information (CSI) data. This approach utilizes the LLM's strong ability to capture the intricate spatiotemporal relationships between the multi-path sensing and communication channels, and thus efficiently predicts upcoming communication CSI based on historical communication and sensing CSI data. Experimental results show that the proposed LLM-based approach significantly outperforms conventional deep learning-based methods and the benchmark scheme without sensing assistance.