This paper, recently published in IEEE Transactions on Communications, explores how to optimize wireless networks that use a combination of Reconfigurable Intelligent Surfaces (RIS) and Semi-Grant-Free (SGF) NOMA. By applying Lyapunov stability theory, the authors move beyond "snapshot" optimizations to create a long-term resource allocation strategy that balances power savings with the need to keep data queues stable and delays low.
TL;DR
The researchers developed a long-term resource allocation framework for RIS-assisted SGF-NOMA systems that minimizes power consumption while guaranteeing queue stability. By using Lyapunov optimization and a multi-stage alternating algorithm, their approach significantly outperforms traditional NOMA and OMA baselines in both energy efficiency and delay management.
What Problem Does This Paper Address?
In the push toward 6G, we face a conflict between two types of users: Grant-Based (GB) users, who have high reliability but suffer from high latency due to scheduling "handshakes," and Grant-Free (GF) users, who transmit immediately to save time but often collide with one another.
Semi-Grant-Free (SGF) NOMA allows GF users to "piggyback" on the spectrum already assigned to GB users. However, this creates two major headaches:
- Uncontrollable Channels: NOMA relies on a "channel gain gap" between users to distinguish signals. If users are physically close, this gap disappears, and the system fails.
- Bursty Traffic: Real-world data doesn't arrive in a steady stream; it arrives in bursts. Most existing research only looks at optimizing a single "time slot," ignoring the fact that a power-saving decision now might cause a massive data backlog (and thus, infinite delay) later.
Methods
The authors tackle a complex Mixed-Integer Non-Linear Programming (MINLP) problem using a three-pronged strategy:
- Lyapunov Optimization: This is the "secret sauce" for long-term stability. It transforms the long-term objective into a "drift-plus-penalty" function for each time slot. This allows the system to make real-time decisions that account for current queue lengths, ensuring the network never becomes "unstable" (where data arrives faster than it can be sent).
- Alternating Optimization (AO) Framework: Since you can't optimize everything at once, the paper breaks the problem into three sub-tasks:
- Channel Assignment: Uses a many-to-one matching algorithm with "swap" operations to decide which GF users share which channels while minimizing interference.
- Power Allocation: Employs fractional programming (specifically Quadratic Transform) to turn non-convex rate equations into solvable convex problems.
- RIS Configuration: Uses matrix lifting and a penalty-based method to design the RIS reflection coefficients. This effectively "shapes" the radio environment to create the necessary channel gaps for NOMA to work.
Datasets
The paper relies on numerical simulations rather than a static dataset. The setup includes:
- Network Topology: A single base station, 2 GB users, and 10 GF users randomly distributed.
- Traffic Model: Data arrivals follow a Poisson distribution, simulating the bursty nature of real-world applications.
- Channel Models: A mix of Rayleigh fading (for direct links) and Rician fading (for RIS-assisted links) to reflect realistic urban propagation.
Key Insights and Ideas
- RIS as a "Channel Sculptor": The most powerful insight is that RIS isn't just for boosting signals; it’s used here to artificially create the channel conditions required for Successive Interference Cancellation (SIC). It forces a disparity between users so the base station can peel their signals apart.
- The Power-Delay Trade-off: The system includes a tunable parameter ($\tau$). A high $\tau$ prioritizes power saving (leading to longer queues), while a low $\tau$ prioritizes low latency (consuming more power). This gives network operators a "dial" to balance energy bills against user experience.
- SIC Order Matters for Stability: The simulation revealed that users decoded first in the NOMA chain (those with the "best" channels) actually have less stable queues because they are more sensitive to potential interference.
Key Results
- Fast Convergence: The proposed AO algorithm is highly efficient, typically reaching a stable solution in fewer than 6 iterations.
- Superior Efficiency: The RIS-assisted system achieved significantly lower power consumption compared to "Random RIS" or "No RIS" baselines.
- Stability Thresholds: While traditional Orthogonal Multiple Access (OMA) systems saw their queues explode (become unstable) as data arrival rates increased, the proposed SGF-NOMA system maintained a steady, finite queue length.
- Impact of RIS Elements: Increasing the number of RIS elements ($K$) showed a diminishing but clear return in power reduction, proving that even modest surfaces provide massive gains in SGF scenarios.
Limitations and Future Work
- Perfect CSI Assumption: The paper assumes the base station has perfect Channel State Information (CSI). In practice, estimating the channel for a passive RIS is notoriously difficult and overhead-heavy.
- Computational Complexity: While the AO algorithm is fast, the "matrix lifting" technique used for RIS optimization scales poorly as the number of reflecting elements grows into the thousands.
- Future Directions: The authors suggest investigating algorithm acceleration (perhaps via machine learning) to handle large-scale deployments where decisions must be made in milliseconds.
Reference: Jia, J., et al. (2024). "Delay-Aware Resource Allocation for RIS Assisted Semi-Grant-Free NOMA Systems." IEEE Transactions on Communications. DOI: 10.1109/TCOMM.2024.3439443
Summary updated