Transformer-Driven Resource Allocation for Enhanced Multi-Carrier NOMA Downlink

This paper presents a transformer-driven resource allocation strategy to optimize channel assignment and power allocation in multi-carrier non-orthogonal multiple access (NOMA) downlink systems, aiming to maximize the sum data rate. Our approach configures the system to support two users per channel. We employ an attention-based deep learning model, specifically an encoder-only transformer, to process the channel-gain-to-noise ratio matrix and derive the optimal channel-assignment matrix. The transformer model includes a tailored loss function that addresses the requirements of the output channel-assignment matrix, precisely steering the optimization to meet the system demands. Simulation results demonstrate the superior performance of our method, confirming its effectiveness in enhancing wireless communication optimization through the integration of deep learning techniques.