Large Language Model Empowered CSI Feedback in Massive MIMO Systems

TL;DR

A novel LLM-based framework for CSI feedback to exploit the potential of LLMs, and designs an information-theoretic mask selection strategy based on self-information, which ensures that masked tokens correspond to elements with lower self-information, while visible tokens correspond to elements with higher self-information, thus maximizing the accuracy of LLM predictions.

Abstract

Despite the success of large language models (LLMs) across domains, their potential for efficient channel state information (CSI) compression and feedback in frequency division duplex (FDD) massive multiple-input multiple-output (mMIMO) systems remains largely unexplored yet increasingly important. In this paper, we propose a novel LLM-based framework for CSI feedback to exploit the potential of LLMs. We first reformulate the CSI compression feedback task as a masked token prediction task that aligns more closely with the functionality of LLMs. Subsequently, we design an information-theoretic mask selection strategy based on self-information, identifying and selecting CSI elements with the highest self-information at the user equipment (UE) for feedback. This ensures that masked tokens correspond to elements with lower self-information, while visible tokens correspond to elements with higher self-information, thus maximizing the accuracy of LLM predictions.

Figures (13)

• Figure 1: The proposed LLMCsiNet framework for CSI compression and feedback.
• Figure 2: Design details of the self-information encoder $f_{\mathrm{EN}}(\cdot)$.
• Figure 3: Design details of the preliminary decoder $f_{\mathrm{PD}}(\cdot)$.
• Figure 4: Design details of the masked token prediction module $f_{\mathrm{TP}}(\cdot)$.
• Figure 5: Two-Stage Training Procedure for LLMCsiNet