Leveraging Transformers for Enhanced Resource Allocation in Multi-Carrier NOMA Wireless Communication Systems

This study introduces an approach that utilizes a transformer model to enhance power allocation and channel assignment in multi-carrier non-orthogonal multiple access (NOMA) downlink systems, with the goal of maximizing the sum data rate. Designed to accommodate two users per channel, our method employs an encoder-only transformer to analyze the channel-gain-to-noise ratio (CNR) matrix and determine the optimal channel-selection matrix. We refine the optimization process with a custom loss function, tailored to the system’s specific constraints. To address the scalability challenges posed by large CNR matrices, we implement a dimensional expansion technique. This technique enables efficient training on downscaled simulated data without altering the transformer’s core architecture. Simulation results confirm the effectiveness of our strategy, showing over 90% accuracy in identifying optimal channel assignments and highlighting the potential of integrating transformers and deep learning into dynamic resource allocation for wireless communication systems.

I. Introduction

In the rapidly evolving field of wireless communications, non-orthogonal multiple access (NOMA) is recognized for enhancing throughput and connectivity. Resource allocation in NOMA focuses on optimizing channel assignment and power allocation to maximize the sum data rate while ensuring user quality-of-service and fairness. For the downlink of a multi-carrier NOMA system, a three-step resource allocation framework has been proposed to navigate the intricacies of the sum rate maximization [1]. Sub-channel assignment in downlink NOMA can be viewed as a many-to-many two-sided user-subchannel matching game [2]. In practice, a dual-user per channel strategy is recommended, which analytically defines the optimal power allocation with a predetermined channel assignment, followed by leveraging a matching algorithm for refined channel assignment [3]. For the uplink, strategies like many-to-many matching models address the subchannel-user mapping challenge, along with the implementation of water-filling and geometric programming for power allocation [4]. The application of many-to-one two-sided matching theory enhances NOMA in device-to-device (D2D) communication schemes [5], and pairing algorithms facilitate joint D2D group associations and channel assignments in uplink NOMA networks [6].