Machine Learning Aided Resource Allocation in a Downlink Multicarrier NOMA network with Coordinated Direct and Relay Transmission

Abstract

In this paper we propose an Artificial Neural Network (ANN) based approach to reduce the computational complexity on solving the combinatorial optimization problem of resource allocation in a downlink multicarrier non-orthogonal multiple access (MC-NOMA) network aided by coordinated direct and relay transmission (CDRT) in the presence of underlay cognitive radio (CR) users. The combinatorial optimization involves optimal user pairing, relay selection, subcarrier pairing and assignment which, when solved by exhaustive search, incurs a high computational complexity and processing delay. We show that an ANN trained by stochastic gradient descent (SGD) based supervised learning algorithm can do the same with low complexity and can provide more than 50% reduction in processing delay.