An Efficient User Clustering in IRS-Assisted NOMA Systems

The integration of non-orthogonal multiple access (NOMA) and intelligent reflecting surfaces (IRS) has emerged as a promising approach for enhancing spectral efficiency, user connectivity, and fairness in next generation wireless networks.However, the overall system performance is highly contingent on effective user clustering mechanisms, which directly impact successive interference cancellation (SIC) and optimal power allocation.Conventional clustering algorithms, such as K-Means, DBSCAN, and hierarchical clustering, often encounter scalability issues and yield imbalanced user distributions across clusters, adversely affecting system level metrics.In this study, we introduce a Balanced K-Means clustering framework specifically tailored for IRS-assisted NOMA systems.The proposed method addresses the imbalance and overfitting challenges inherent in traditional clustering by enforcing approximately equal cluster sizes, thereby facilitating more equitable resource allocation and robust SIC performance.The simulation results were simulated over a normalized synthetic dataset representing user spatial features and the proposed Balanced K-Means clustering outperforms conventional methods, achieving higher throughput and better bit error rate for varying user densities and IRS configurations.