Session-Specific Design for Wireless Transmission Over WBANs: Navigating Energy Efficiency and Reliability Tradeoff With Multiobjective Reinforcement Learning

In wireless body area networks (WBANs) for healthcare monitoring applications, simultaneously achieving high energy efficiency and high reliability is challenging as it involves conflicting objectives. In this paper, we formulate and solve a multi-objective optimization (MOO) problem to minimize both energy consumption and packet loss rate (PLR) for WBAN transmission with an automatic repeat request (ARQ) mechanism. To effectively balance these conflicting objectives based on user preferences, we aim to design an intelligent agent that optimally configures transmission parameters for each transmission attempt of the session. Specifically, we reformulate the problem as a multi-objective Markov decision process (MOMDP). To approximate the entire Pareto front using a single policy network, we propose a multi-objective reinforcement learning (MORL) solution. The proposed method conditions the policy on preference weights, utilizes accumulated rewards for policy updates, and employs the penalty-based boundary intersection (PBI) method for reward scalarization. Simulation results demonstrate that the proposed MORL solution can provide an improved Pareto front approximation over other MORL algorithms, even when training with a small number of preference weights. Additionally, the agent can adjust transmission parameter values to effectively balance energy efficiency and reliability according to user preferences.