Channel Prediction with Liquid Time-Constant Networks: An Online and Adaptive Approach

Abstract

Accurate channel state information (CSI) prediction and estimation are critical to the communication system to adapt to the rapid change of wireless channels. The CSI feedback from the receiver may become outdated and inaccurate due to the compression and transmission delay, especially for the multiple-input multiple-output (MIMO) system. Deep learning-based algorithms for channel prediction have been widely used, however, traditional recurrent neural network (RNN) based methods may incur unstable behavior in the dynamic system. In this paper, we propose a novel MIMO channel prediction method based on a liquid time constant (LTC) network, which provides more stable and bounded performance in the CSI prediction task. An online prediction structure is also introduced to better cope with current architecture and reduce the computational requirement on the device. Results reveal that our proposed method outperforms the traditional RNN based algorithm and auto regressive (AR) models in prediction accuracy by 10% - 40% on both simulation data and measurement data.