Adaptive IR-HARQ System With Deep Learning-Based Early Prediction for URLLC Service

Ultra-high reliability and low latency communication (URLLC) is a crucial concept in the realm of 5G and beyond. URLLC has a demand for short packets with an end-to-end latency of 1 ms with a success probability of 99.999%. The stringent reliability and latency requirements of URLLC significantly affect the hybrid automatic repeat request (HARQ) mechanism at the PHY/MAC layer. In this paper, we propose an adaptive incremental redundancy HARQ (IR-HARQ) scheme that leverages deep learning techniques to predict the decoding reliability of a received codeword before the actual decoding process. Additionally, we propose novel packet transmission schemes suitable for IR-HARQ. In particular, the proposed concept enables the transmitter to respond more quickly by receiving early feedback (FB) on decoding reliability and simultaneously transmitting the required incremental redundancy versions (RVs) using the proposed transmission schemes. We also propose an early prediction method based on long short-term memory (LSTM) using novel features to accurately predict decoding reliability. This method achieves more than 98% prediction accuracy for a code rate of 2/5, and over 96% for code rates of 5/6 and 3/4 in quasi-static fading channel environment. The proposed adaptive IR-HARQ, combined with the packet transmission schemes, significantly improves throughput and bandwidth efficiency compared to conventional IR-HARQ, while meeting the latency requirements of URLLC.