Learning to allocate: a delay and temperature-aware slot allocation framework for WBAN with TDMA-MAC

Abstract

Data aggregation in the Wireless Body Area Networks (WBAN) is a multidimensional problem. It can be addressed at different levels of the network. The proposed work identifies the scheduling problem for the slots in a Time Division Multiple Access Medium Access Control (TDMA-MAC) superframe. A resource-constrained single channel WBAN is considered, and the proposed work models the set of nodes and slots of one subframe as a bipartite graph and aims to obtain a globally optimal matching solution for one subframe with end-to-end latency and temperature rise minimization as prime goals. Later, it extends the solution to the entire superframe, which consists of several subframes. The proposed Learning to Allocate (LTA) framework uses a Multi-Agent Reinforcement Learning (MARL)-based dynamic bipartite weight update. The proposed Reinforcement Learning-Optimized Delay and Temperature Aware Scheduling (RL-ODTAS) algorithm deployed on a WBAN co-ordinator was tested on a custom-made simulation testbed with heterogeneous nodes that handle two categories of data. The simulation results indicate an average of 14.15% end-to-end delay improvement for emergency data. Also, at the end of 1500 superframes, a temperature rise reduction of up to 0.43 °C is seen compared to Hungarian Matching without a learning component.