Intelligent Offloading for NOMA-Assisted MEC via Dual Connectivity

Abstract

Multiaccess edge computing (MEC), which brings computing capability close to the user equipment (UE) within a radio access network (RAN), is a promising technique to meet the challenges of explosive growth in edge data traffic and the number of connected devices. Nevertheless, MEC offloading failures, i.e., interruption of task offloading due to UE mobility and limited battery capacity, are still an important issue. In this article, we aim to minimize the total energy consumption in dual connectivity (DC)- and nonorthogonal multiple access (NOMA)-assisted computation offloading systems. This problem is formulated as a mixed-integer nonlinear programming (MINLP) problem with nonconvex constraints, which is a nondeterministic polynomial-time (NP)-hard and perplexing problem. To address this problem, we propose an iterative optimization algorithm to obtain a suboptimal time scheduling and task allocation solution. Specifically, we solve the nonconvex constrained MINLP problems by exploiting the successive convex approximation (SCA) algorithm and the nonlinear optimization algorithm with the mesh adaptive direct search (MADS) (NOMAD). Furthermore, a deep learning-based intelligent offloading scheme is introduced, called a cyclic branch network, to fully utilize data traffic and the computing capability on the network edge. Numerical results show that the proposed intelligent offloading scheme has comparable performance and dramatically reduces the inference time compared with the traditional optimization algorithm.