A Deep Reinforcement Learning-Based Adaptive Transmission Design for Trustworthy URLLC Support in 5G and Beyond Wireless Systems

The natural tension between achieving high reliability and guaranteeing low latency is a key challenge in supporting ultra-reliable low-latency communications (URLLC) service in current and future wireless systems. To achieve the most effective solution, various reliability and latency mechanisms must be jointly designed. In this paper, we propose to optimally select the numerology, slot size, modulation and coding schemes for each transmission or retransmission attempt for a packet transmission over 5G and beyond wireless systems. We formulate this sequential decision-making problem into an Markov decision process (MDP) and find the optimal policy using a deep reinforcement learning algorithm. Through selected numerical examples, we demonstrate that the proposed joint design can achieve significant performance gains over the conventional schemes. We also show that such joint selection is especially important when the latency budget is stringent and/or channel quality is poor.