Latency Optimization for Computation Offloading With Hybrid NOMA–OMA Transmission

Abstract

The Internet-of-Things (IoT) platform is faced with critical challenges posed by the conflict between resource-hungry IoT applications and resource-constrained IoT devices. Mobile-edge computing provides a promising solution by allowing IoT devices to offload their computation to nearby edge servers to enable fast and energy-efficient data processing. In this article, we study a scenario, where two IoT users (IoT devices) offload their computation workloads to an edge server with hybrid nonorthogonal multiple access (NOMA)–orthogonal multiple access (OMA) transmission. The hybrid multiple access transmission incorporates three offloading methods, namely, hybrid NOMA, pure NOMA, and pure OMA. The offloading-method selection, together with user selection, which determines the roles played by different IoT users in data transmission, comprises our offloading strategy and is optimized to minimize the maximal offloading latency of the two IoT users. By exploiting the method of successive convex approximation, we design an efficient algorithm to solve the complicated nonconvex problem and rigorously prove the convergence of our algorithm. Extensive numerical tests show that our scheme can always help IoT users to flexibly choose the best offloading strategy. Inspired by experimental observations, we analytically establish the criteria for the three offloading methods. We show that pure OMA transmission is never the best offloading method, except in some extreme cases that rarely occur in practice, while pure NOMA transmission is the most desirable offloading method in terms of latency minimization. We then propose detection approaches for the best offloading strategy with both offloading-method selection and user selection under certain system settings. The user selection is applied to avoid the pure OMA transmission and encourage the pure NOMA transmission.