Dynamic Spectrum Access for Overcoming Channel Congestion in WBAN: A Distributed Deep Reinforcement Learning Algorithm

For body area network (BAN) sensors, considering the low power and large quantity of communication nodes, as well as the challenges posed by interference and the multi-user, multi-device scenarios, improving communication quality through simple methods becomes difficult. The common methods include modeling channels using actual measurements and analyzing their channel quality (Aminzadeh et al in IEEE Trans Anten Propag 69(7):4083–4092, 2020 [1]). To address these challenges, we improve dynamic spectrum access methods based on multi-depth reinforcement learning and adapt them for channel allocation in WBAN nodes. Specifically, each node performs simple and necessary state information gathering, which is then aggregated in the cloud for training using a Double Deep Q-Network (DDQN). Through our derivation, we show that aggregated training can avoid the issue of low efficiency in some nodes. Experimental results demonstrate the strong performance of our algorithm in obtaining the optimal solution for spectrum access, achieving a 10–20% increase in channel utilization compared to the slotted-hoc method under optimal transmission probabilities. By multi-depth reinforcement learning, we successfully apply our improved method to the channel allocation problem in WBAN.