AI/ML-Based Wireless Channel Estimation and Prediction for Downlink Precoding

We develop an artificial intelligence/machine learning (AI/ML)-based approach which performs a simultaneous channel estimation and time-domain prediction relying on uplink sounding reference signals (SRSs). Key features of the proposed solution include i) its intrinsic amicability for training based on real-world data, and ii) its adaptability to system parameters such as communication bandwidth and number of base station antenna ports. We evaluate the link-level throughput performance of the proposed approach when integrated in a 3GPP-compliant 5G new radio (NR) simulator. The results demonstrate that AI/ML-based solutions have the potential to enhance the downlink precoding performance of time division duplex (TDD) systems, especially at low signal-to-noise ratios (SNRs).