Energy-Efficient Relay Transmission for WBAN: Energy Consumption Minimizing Design With Hybrid Supervised/Reinforcement Learning

Energy-efficient transmission is essential to wireless body area networks (WBAN) as most biosensors in WBAN have limited energy supply. In this paper, we study the energy consumption minimization problem for each amplify-and-forward (AF) relay transmission session while satisfying a certain reliability requirement in WBAN. To minimize the energy consumption during successful transmission and wasted energy due to failed transmission attempts, over finite blocklength (FBL) regime, we design an intelligent agent that can determine: i) whether to transmit or not for given current channel state information (CSI) and available resources and ii) the best power levels and blocklength values for the current transmission session if the agent decides to transmit. To perform these two tasks simultaneously, we propose a novel hybrid supervised/reinforcement learning solution. Specifically, we design a classification network following the supervised learning approach to determine whether to transmit or not based on predicted minimal packet error probability. We then develop a deep reinforcement learning (DRL)-based solution that determines the optimal values of the transmission parameters. We also propose a DRL-based online parameter tuning (DRL-OPT) algorithm to minimize the impact of model inaccuracy and/or environment changes. Simulation results reveal that the performance of the proposed hybrid solution is almost identical to that of the exhaustive search. The DRL-OPT algorithm can follow environment variation and maintain a good performance with low computational complexity. Moreover, we numerically analyze the effect of slot duration on energy consumption and develop a guideline for practical WBAN design.