Deep-Learning-Based Resource Allocation for Transmit Power Minimization in Uplink NOMA IoT Cellular Networks

Abstract

For Internet of Things (IoT) networks, it is important to develop energy-efficient communication schemes to extend the operating life of battery-powered IoT devices. Additionally, non-orthogonal multiple access (NOMA) can utilize frequency resources more efficiently than orthogonal multiple access, making it more suitable to support massive connectivity of IoT users. Motivated by these facts, we consider uplink NOMA IoT cellular networks and develop two novel algorithms that jointly optimize sub-band allocation and transmit power control to minimize the total transmit power of all users and the maximum transmit power among all users' transmit power, respectively, while meeting the minimum required data rate for all users. Specifically, we propose a novel two-step approach that sequentially performs sub-band assignment and transmit power control for each IoT user, in which a genetic algorithm-based method is applied for sub-band assignment whereas unsupervised learning (USL) implemented as deep neural network (DNN) models is utilized for transmit power control. Moreover, we propose loss functions that can achieve an appropriate balance between power minimization and rate constraint satisfaction in the process of training. Extensive simulations are performed to evaluate the performance of the proposed algorithm in various aspects, and we show that our proposed two-step algorithm can approach the optimal performance achievable through an exhaustive search with much lower computational complexity.