Deep Learning-based Task Offloading and Time Allocation for Edge Computing WBANs

Abstract

The limited energy resource and low computing capability of tiny sensor nodes have been longstanding challenges for the deployment of Wireless Body Area Networks (WBANs). The emergence of Wireless Power Transfer (WPT) and Mobile Edge Computing (MEC) provide a potential solution. Therefore, we devote this paper to developing an efficient task offloading and time allocation scheme based on WPT and MEC for WBANs. Technically, we first propose an MEC-aided system model with multiple WBANs where WBANs can harvest energy from Radio Frequency (RF) signals wirelessly and an edge server can process the tasks offloaded by WBANs. We formulate an non-convex optimization problem with the objective to maximize the computation rate, and propose a deep learning-based task offloading and time allocation algorithm, named DNN-TOTA. DNN-TOTA adopts the Deep Neural Network (DNN) and the Order-Preserving Quantization (OPQ) method to generate the candidate offloading decision sets, and uses the convex optimization to solve the time allocation problem. The simulation results demonstrate that the computation rate of our proposed DNN-TOTA can reach to 98% of the optimal solution.