Reinforcement learning based energy-neutral operation for hybrid EH powered TBAN

The aging population, outbreak of new infectious diseases and shortage of medical resources promote rapid development of telemedicine. Wireless textile body area network (TBAN), which combines functional textile and wireless body area network (WBAN), is gaining great attention as an efficient medium of remote medical care. This is because of its unique materials and application scenario, as well as its convenience and friendliness to the elderly. Moreover, it is an effective application for integrating edge computing with next generation of wearable technology. Nonetheless, it is unavoidable that TBAN has to deal with reliability and energy issues. Given these deficiencies and challenges, this paper focuses on the feasibility of achieving wearable energy neutral operation (ENO) in TBAN while maintaining robustness. In addition to adding user posture factors regarding network specifics, we combine hybrid energy harvesting (EH) techniques and duty cycle schemes. A hybrid radio frequency (RF) energy and Triboelectric nanogenerator (TENG) EH-assisted TBAN system is built in this work. We analyze and discuss the delay, data rate and packet error rate (PER) under five typical daily activities (standing, sitting, lying, walking, and running). To optimize the ENO problem, two reinforcement learning (Q-learning and Deep Q-Network (DQN)) based algorithms are proposed. According to numerical results, both algorithms ultimately lead to stable power levels compared to the continuous decline of battery power without optimization. DQN-based optimization performs better than Q-Learning. For instance, 14% and 56% improvements in PER and battery power, respectively.