Reinforcement Learning Based Sensor Encryption and Power Control for Low-Latency WBANs

Healthcare sensing data in wireless body area networks are vulnerable to active eavesdropping that simultaneously performs sniffing and jamming attacks to raise the sensor transmit power and thus steal more data. In this paper, we propose a reinforcement learning based sensor encryption and power control scheme to resist active eavesdropping for low-latency wireless body area networks. This scheme enables the coordinator to jointly optimize the sensor encryption key size and the transmit power based on the sensing data priority, the jamming power and the channel states of the sensor. We design a safe exploration algorithm based on the Dyna architecture to avoid choosing the encryption and power control policies that result in data transmission failure or data leakage. A secure sensing data transmission game between the coordinator and the eavesdropper is formulated to analyze the performance bound of our proposed scheme in terms of the signal-to-interference-plus-noise ratio of sensor signals, the eavesdropping rate, the energy consumption and the transmission latency based on the Nash Equilibrium of the game. Simulation results show that this scheme significantly decreases the eavesdropping rate and the transmission latency, and saves the sensor energy compared with the benchmark against active eavesdropping.