Online Scheduling and Optimality Analysis for NOMA-Based MEC Systems

Compared to conventional cloud computing, non-orthogonal multiple access (NOMA) based mobile edge computing (MEC) can provide scalable and low-latency computing services for the next generation network. However, considering the prevalent stochasticity of wireless networks and the sophisticated signal processing of NOMA, it is critical but challenging to design an efficient and low-complexity task offloading strategy for NOMA-based MEC, especially in large-scale networks. This paper investigates the offloading scheduling and resource allocation problem to maximize the long-term system utility (a measure of throughput and fairness) under complete and partial device knowledge, respectively. Due to the combinatorial property of the temporary coupling and non-convexity, we first apply the Lyapunov optimization to optimally decouple the stochastic optimization over a long time horizon into a series of per-slot deterministic subproblems without a prior knowledge of network dynamics. Then, we propose to equivalently transform the nonconvex per-slot subproblems into convex ones by introducing a set of auxiliary variables. The designed online algorithm is proved to be asymptotically optimal, even under the partial knowledge of device states. Simulation results validate the superiority of the online scheduling algorithm in terms of system utility and stability improvement compared with benchmarks.

I. Introduction

With the proliferation of novel applications, such as autonomous driving, online games and augmented reality, large amounts of computation-intensive and latency-sensitive tasks are generated on mobile devices [1]. However, mobile terminals are of constrained computational resources, resulting in a degradation of user experience. Confronted with such challenges, mobile edge computing (MEC), which migrates computation tasks from mobile devices to more resourceful computing facilities deployed at the network edge, is considered as a promising paradigm to enhance the communication and computation capacity of the next generation network [2].