Optimization of User Pairing and Power Allocation for NOMA Downlink Systems

NOMA (Non-Orthogonal Multiple Access) technology enables multi-user sharing of identical time-frequency resources through power-domain multiplexing. The performance of NOMA downlink systems critically depends on user pairing strategies and power allocation schemes. Optimization objectives typically focus on enhancing system throughput, ensuring user fairness, or reducing energy consumption. The core challenge lies in resolving the coupling relationship between inter-user interference and resource allocation.

User Pairing Optimization - Channel Disparity Utilization: NOMA leverages differences in user channel conditions, typically pairing users with significant channel gain disparities (e.g., near-user and cell-edge user) to enhance Successive Interference Cancellation (SIC) effectiveness. Code implementation often involves sorting users by channel gain and implementing pairing algorithms with O(n²) complexity. - Dynamic Grouping Strategy: Based on real-time Channel State Information (CSI), optimal user combinations can be achieved through greedy algorithms or graph theory models (e.g., bipartite graph matching). Implementation typically uses Hungarian algorithm (O(n³)) or heuristic approaches to avoid strong interference pairings.

Power Allocation Optimization - Proportional Fair Allocation: While guaranteeing minimum rate requirements, power is dynamically adjusted according to channel quality. Common implementations include water-filling algorithms or convex optimization methods (e.g., using CVX toolbox in MATLAB) to maximize weighted sum-rate. - Hierarchical Optimization Framework: Problems are often decomposed into two layers—first optimizing power with fixed pairing (solved via Lagrange multiplier methods), then adjusting pairing based on power results to form an iterative closed-loop. Code structure typically involves nested while-loops with convergence checks.

Extended Considerations - Machine Learning Assistance: Deep reinforcement learning can handle dynamic environments, avoiding high computational complexity of traditional optimization methods. Implementation may use Q-learning or policy gradient methods with state-action reward frameworks. - Hybrid Multiple Access Scenarios: When combining with OMA (Orthogonal Multiple Access), joint optimization of resource block allocation and power allocation dimensions becomes necessary, requiring mixed-integer programming solutions.