Deep Learning Based Power Optimizing for NOMA Based Relay Aided D2D Transmissions

Abstract

The future generation of wireless communication networks demands for high spectral efficiency to accommodate a large number of devices over the limited available frequency spectrum. Device to device (D2D) systems exploit channel reuse to offer high spectral efficiency and reduce the burden on the communication infrastructure by facilitating communication between devices without involving the base station. We can further enhance the efficiency of D2D systems by employing non-orthogonal multiple access (NOMA) for the transmission of the signals. In NOMA the signals of multiple users are transmitted on the same channel, simultaneously. Deployment of relays can assist the users that do not have a reliable link of communication. A combination of these advanced technologies may offer very high spectral efficiency and a robust communication system. This article aims to design efficient resource allocation techniques for the future communication systems. We consider sum rate maximization problem subject to limited power budget at different transmitting nodes and necessary transmit power gap among users for successful NOMA implementation. Under decode and forward relaying protocol, the problem turns out to be a unique joint uplink-downlink NOMA optimization. We then propose a deep neural networks (DNN) framework to acquire a joint power loading solution at source and relaying nodes. To obtain reliable data for DNN training and testing, we also derive an optimal solution of the problem through convex optimization paradigm, which is used later as a bench mark to verify the performance of proposed DNN based solution. It is observed that DNN provides promising results both in terms of sum rate and the computational complexity.