Multiuser detection in hybrid Non-Orthogonal Multiple Access (NOMA) using machine learning

Abstract

The growing popularity of network applications has created a higher need for limited radio spectrum resources due to the increasing number of users varying for access to these finite resources. Efficiently managing the utilization of spectrum becomes increasingly crucial as demand rises. One effective strategy to tackle this challenge involves employing multiple access techniques. Among these approaches Non-Orthogonal Multiple Access (NOMA) emerges as a method, in this context. NOMA enables several users to utilize the frequency band through power domain multiplexing, which greatly boosts spectrum efficiency when compared to traditional orthogonal methods. However as more users access the limited radio spectrum, the task of multiple user detection (MUD) becomes more complicated leading to challenges in effectively distinguishing and decoding signals from various users. Dealing with these complexities requires techniques that can address non linearities and interference present, in these systems. Artificial Neural Networks (ANNs) are one of the ways to handle these problems in an efficient manner. ANNs have the capacity of handling complex non-linear relationships can substantially enhance the accuracy for Multiple User (MU) in NOMA systems. This dissertation presents an ANN model explicitly for MUD in NOMA networks. Our method capitalizes on the power of deep neural networks to discern multi-user signals and filters them out which alleviated disadvantages in classical detection methods. Through training, the ANN recognizes the patterns and characteristics of received signals. This new method improves detection accuracy and spectrum utilization. The integration of ANNs with NOMA systems is a good innovation in wireless communication technology providing better performance for massive user scenarios. The ANN-based multi-user detector performance has been carefully tested using simulated experiments. The results show that at epoch 2, the proposed technique has a mean square error (MSE) of 0.023103, thereby proving the efficiency of the neural network in solving the multi-user identification problem. This level of performance will give an idea of the capability of ANNs in solving the complexity introduced by NOMA systems and will act as a benchmark for future improvements. The efficiency of ANN in this context therefore shows that it can boost detection accuracy and optimize spectrum use, hence rendering a valuable edge over other previous methods. The research outcome obtained in support of the efficiency of ANN-based approaches and has established a concrete platform for further enhancement and use.