Generative Model for Joint Resource Management in Multi-Cell Multi-Carrier NOMA Networks

TL;DR

This work tackles joint resource management in downlink multi-cell multi-carrier NOMA networks by formulating a sum-rate maximization problem and solving it with a novel multi-task transformer (MTT). Key contributions include decomposing the optimization into user scheduling, beamforming, and power allocation subproblems with SDR, S-Lemma, and SPCA techniques, and designing an MTT architecture with inception-residual features, projection/position embeddings, and multi-low-rank attention to enable real-time decisions. The dataset for training is generated from established mathematical solutions, and extensive simulations show that MTT achieves near-baseline performance with substantially lower memory and computational requirements, enabling real-time radio-map-assisted resource management. The work demonstrates the practical viability of GAI-based joint optimization for 5G/B5G wireless systems and points to extensions in dynamic network contexts such as LEO satellites and UAV-aided networks.

Abstract

In this work, we design a generative artificial intelligence (GAI) -based framework for joint resource allocation, beamforming, and power allocation in multi-cell multi-carrier non-orthogonal multiple access (NOMA) networks. We formulate the proposed problem as sum rate maximization problem. Next, we design a novel multi-task transformer (MTT) framework to handle the problem in real-time. To provide the necessary training set, we consider simplified but powerful mathematical techniques from the literature. Then, we train and test the proposed MTT. We perform simulation to evaluate the efficiency of the proposed MTT and compare its performance with the mathematical baseline.

Figures (3)

• Figure 1: Structure of the proposed MTT framework.
• Figure 2: Training and validation accuracy of the proposed model.
• Figure 3: Performance comparison between the proposed framework and the baseline.