Large Language Model-based CSI Feedback for Massive MIMO Systems

Recently, large language models (LLMs) have gained much attention in next-generation wireless communications due to their adaptability to the dynamic wireless environment. One application of LLM in wireless communication is channel state information (CSI) feedback enhancement. While existing studies for CSI feedback suffer from generalization and user mobility issues due to the fixed network parameters, LLM can effectively solve the problems by leveraging the generalization capability. In this paper, we propose an LLM-based CSI feedback (LLM-CF) that utilizes the versatility of LLM to achieve CSI feedback performance enhancement in various communication scenarios. One key component of LLM-CF is the wireless configuration prompt (WCP), which explicitly provides LLM with the specifications on wireless scenarios and a detailed guideline for CSI feedback. By following the instructions in WCP, LLM-CF can obtain input characteristics, scenario details, and step-by-step logic for reconstructing CSI. Using this information, LLM-CF can adaptively reconstruct downlink CSI in response to the dynamic wireless environment. From the simulation results, we show that LLM-CF outperforms the conventional CSI feedback techniques by more than 2.5 dB gain in terms of the normalized mean square error (NMSE).