Deep Learning-Based Optimal Relay Selection Scheme for Underlay Cognitive NOMA Networks With Incremental Relaying

Abstract

In this research work, we propose a deep learning (DL)-based optimal relay selection (ORS) scheme to select the relay in the underlay cognitive NOMA (CNOMA) networks with incremental relaying (CNOMA-IR).We initially propose an ORS scheme in which the optimal relay is chosen in the secondary network by considering the residual energy of the relay, interference from the primary network, and the channel gain. We derive closed-form expressions for outage probabilities of secondary users (SUs) and system throughput (ST) of the CNOMA-IR network with ORS by considering imperfect successive interference cancellation (i-SIC). The ORS using Monte-Carlo simulations is time-consuming and involves higher computational complexity. In order to resolve this challenge, we propose a DL framework for the ORS.With the proposed ORS scheme, we obtain the dataset, tune the hyperparameters of the DL models, train different DL models, and compare the performance. With the best performing gated recurrent unit (GRU) DL model, the results show that the proposed DL framework is able to select the optimal relay accurately in various network scenarios with minimal computation time, which can significantly enhance the throughput and outage performance of the underlay CNOMA-IR networks, compared to the conventional cooperative relaying-based CNOMA (CR-CNOMA) networks.