Design of Deterministic Grant-Free Access with Deep Reinforcement Learning

Abstract

Relying on the special mathematical properties among the access patterns of users, deterministic grant-free access can achieve ultra-high reliability within a finite time duration, and thus is favorable in coping with the challenge in ultra-reliable low-latency communications (URLLC) for 5G. Recently, interference canceling (IC) codes, proposed for access under the successive interference cancellation technique at the physical layer, have been becoming a hot research topic. However, it is difficult to obtain IC codes by the current mathematical tools or traditional search algorithms. To fill this gap, we put forth a deep reinforcement learning (DRL) based algorithm to search IC codes, with carefully designed metrics and reward functions as per the underlying mathematical constraints. The search results indicate that the algorithm can efficiently discover IC codes, while the simulation results indicate that the discovered IC codes yield significantly lower failure probability than the random access protocol given the same latency requirements, and thus are more suitable for URLLC.