Introduction
As the telecommunications industry transitions toward 5G and beyond, two pivotal technologies are emerging to meet the demands for high spectral efficiency and low latency: Non-Orthogonal Multiple Access (NOMA) and Fog-Radio Access Networks (F-RAN). NOMA enhances capacity by allowing multiple users to share the same frequency and time resources through power-domain multiplexing. Concurrently, F-RAN architectures shift cloud computing capabilities to the network edge—specifically to Fog Access Points (FAPs)—to minimize latency and alleviate the burden on the central cloud.
However, the integration of these technologies introduces significant optimization challenges. Maximizing network performance requires simultaneous decisions on user scheduling, power allocation across resource blocks (RBs), and power splitting between NOMA users, all while operating under limited fronthaul capacity. This paper addresses these challenges by proposing a joint optimization framework that decouples this complex mixed-integer non-convex problem into a manageable three-step hybrid approach.
System Model and Problem Formulation
The research considers a downlink F-RAN where a central Baseband Unit (BBU) pool is connected to multiple FAPs via fronthaul links. Each FAP serves multiple users by employing NOMA to pair "strong" and "weak" users on the same RB.
The optimization objective is to maximize network utility, typically defined as a weighted sum-rate, subject to constraints on total FAP power, individual RB power, and fronthaul capacity. The complexity arises from the interdependencies between user pairing, power distribution, and the interference inherent in NOMA systems.
Proposed Methodology
The authors propose a three-step decoupled solution that balances centralized coordination with distributed edge processing:
- User-to-RB Assignment: Two approaches are evaluated. A centralized method uses the Hungarian algorithm, while a distributed approach employs a novel Auction-based Multiple Choice Knapsack Problem (MCKP) algorithm. In the auction model, FAPs bid on users, allowing the network to satisfy fronthaul constraints while distributing decision-making.
- Power Allocation across RBs: Following user assignment, the total power for each RB is optimized using the Alternating Direction Method of Multipliers (ADMM). ADMM is selected for its numerical stability and effectiveness in handling non-convex rate equations.
- NOMA Power Splitting: The final step involves determining the power split between the two users on a single RB. A critical finding is that this step is separable per FAP, allowing it to be performed entirely at the network edge, consistent with the F-RAN philosophy.
Key Insights and Results
The study utilized extensive simulations based on a 7-FAP grid network with 3GPP urban micro-cell path-loss models. Key findings include:
- Separability of NOMA: Handling NOMA power splitting locally at the FAP significantly reduces the signaling burden on the fronthaul, which is often a bottleneck in dense deployments.
- Fronthaul and Fairness: Interestingly, lower fronthaul capacity can drive better fairness. When capacity is limited, the system is incentivized to use NOMA more effectively to serve weaker users rather than prioritizing only the strongest users.
- Performance Gains: The proposed "K-PA-PS" (Knapsack + Power Allocation + Power Split) scheme achieved a 16% gain in network utility compared to baseline schemes. Furthermore, NOMA schemes significantly outperformed Orthogonal Multiple Access (OMA) in terms of Jain’s fairness index, particularly for edge users.
- Computational Efficiency: The ADMM-based optimization converged in fewer than five iterations, demonstrating its feasibility for real-time scheduling in 5G environments.
Conclusion
This research demonstrates that joint scheduling and power adaptation in NOMA-based F-RANs can significantly enhance spectral efficiency and user fairness. While the current model assumes perfect Successive Interference Cancellation (SIC) and focuses on two-user pairings, it provides a robust foundation for future work. Extensions to include Device-to-Device (D2D) communication, relaying, and edge caching are promising directions for creating more intelligent and efficient fog environments.
Summary updated