DiNAMAC: A disruption tolerant, reinforcement learning-based Mac protocol for implantable body sensor networks

Ongoing advancements in Body Sensor Networks (BSN) have enabled continuous health monitoring of chronically ill patients, with the use of implantable and body worn sensor nodes. Inevitable day-to-day activities such as walking, running, and sleeping cause severe disruptions in the wireless link among these sensor nodes, resulting in temporary shadowing of wireless signals. These disruptions in the wireless link not only reduce the reliability of the network but also increase the power consumption. Both signal disruption and power consumption must be reduced in order to achieve long term monitoring of physiological signals in chronic patients. In this paper we propose a MAC protocol called DiNAMAC (Disruption tolerant reiNforcement leArning-based MAC), which is not only aware of the wireless link quality but also is aware of network resource availability and application requirements. DiNAMAC uses reinforcement learning to adapt the scheduling based on channel conditions and to prioritize data transmission and availability according to the application requirements. In addition, we design DiNAMAC based on a model-free learning technique to make it more practical in real-world applications. Our simulation results show that DiNAMAC performs better than conventional MAC protocols in terms of latency and throughput even with when the wireless link quality is challenged by large temporal variations.