Performance Monitoring for AI-based CSI Feedback via Proxy
Jiajia Guo, Shaodan Ma · 2023
Artificial intelligence (AI) has shown great potential in channel state information (CSI) feedback problems. However, the existing works focus on further improving the performance of AI-enabled autoencoder-based CSI feedback but ignore some new problems caused by the introduction of AI. In this work, the performance monitoring for AI-based CSI feedback is considered and a proxy-based performance monitoring framework is proposed. This is the first work that considers performance monitoring. Specifically, we use the knowledge distillation technique to transfer the knowledge learned by the complicated decoder at the base station to the lightweight proxy decoder at the user. The lightweight proxy decoder is then used to predict the CSI reconstruction accuracy. Simulation on a public channel dataset shows that the performance monitoring method proposed in this work can predict feedback performance with high quality, and the classification accuracy is over 95%.