Throughput improvement under device clustering and power allocation in NOMA-MTC system

The explosive growth of massive Machine-Type Communications (mMTC) has introduced significant challenges in efficient spectrum utilisation and scalable resource allocation due to massive connectivity and heterogeneous traffic demands. Orthogonal schemes struggle with dense deployments, making NOMA a promising yet complex solution due to its pairing and power allocation challenges. This paper proposes K-Means++ clustering for its stability and Q-Learning power allocation for its adaptability to efficiently utilise spectrum in a NOMA-enabled MTC system. The proposed Machine Learning (ML) framework leverages the potential of NOMA-MTC to enhance system throughput under power constraints by dynamically adapting power allocation to the number of devices within each cluster without imposing restrictions on cluster size. Simulation results demonstrate that the proposed K-Means++ with QL approach achieves approximately 20% higher throughput compared to Gaussian Mixture Model (GMM) clustering, while maintaining higher spectral efficiency and ensuring faster convergence of power allocation coefficients compared to traditional schemes.