DNN-Based RV-Adaptive HARQ for Low-Latency Communications

Abstract

In comparison to Ultra-Reliable Low-Latency Communications (URLLC) in 5G mobile communication systems, the next generation URLLC (xURLLC) requires even lower latency and higher reliability. To address these requirements, we propose a novel redundancy version (RV)-adaptive hybrid automatic repeat request (HARQ) scheme. Our approach builds upon the existing 5G HARQ framework and 5G low-density parity-check (LDPC) codes, introducing an adaptive selection of the RV for the subsequent transmission based on the current decoding states. To determine the optimal RV, we use a modified version of the reciprocal channel approximation (RCA) algorithm to compare the decoding thresholds of all available RVs and select the one with the lowest threshold to maximize the likelihood of successful decoding. To further reduce processing delay, we apply a deep neural network (DNN) to predict the optimal RV for the next transmission. This prediction is made prior to the decoding process, ensuring that no additional delay is introduced. Since the selection is made before the decoding starts, our proposed method can seamlessly integrate with other adaptive-HARQ and fast HARQ schemes.