Beyond Conventional Power Allocation: A Critical Review of Intelligent Optimization in NOMA-Based 5G and 6G Networks

Non-orthogonal multiple access (NOMA) is a promising multiple-access technique for improving spectral efficiency and supporting heterogeneous connectivity in fifth-generation (5G), 5G-Advanced, and emerging sixth-generation (6G) wireless networks. However, its performance depends substantially on effective power allocation under competing requirements for throughput, energy efficiency, fairness, reliability, and security. Conventional allocation strategies may be insufficient for dynamic and complex network environments, motivating the development of intelligent optimization techniques. This review critically examines metaheuristic algorithms, machine learning, deep reinforcement learning, and hybrid optimization frameworks for NOMA power allocation. It compares their operating principles, strengths, limitations, computational complexity, convergence characteristics, and suitability for single-cell, multi-cell, cooperative, distributed, and mobile network scenarios. Particular attention is given to imperfect channel state information, residual successive interference cancellation (SIC) errors, channel uncertainty, physical-layer security, and the computational constraints associated with real-time deployment. The review identifies persistent challenges involving limited generalization across network configurations, conflicting optimization objectives, high training costs, inconsistent benchmarking, and insufficient experimental validation. It further highlights the need for uncertainty-aware optimization, risk-sensitive learning, distributed intelligence, lightweight edge-based implementation, explainable decision-making, and joint optimization of power allocation with other radio-resource management functions. The review concludes that no single intelligent optimization method is universally superior; its effectiveness depends on network conditions, objective formulation, information availability, and computational resources. Standardized evaluation and realistic experimental validation are therefore essential to establishing the reliability, scalability, and practical applicability of intelligent NOMA power allocation for future wireless networks.