Abstract
This paper presents a comprehensive stochastic geometry analysis of downlink Non-Orthogonal Multiple Access (NOMA) in multi-cell networks. By modeling base stations and users as Poisson Point Processes, the authors investigate the impact of intercell interference and imperfect Successive Interference Cancellation (SIC). They propose novel spatial user clustering models and compare distance-based versus interference-aware user ordering, demonstrating that simpler ordering can sometimes outperform complex real-time metrics. The study also introduces low-complexity resource allocation algorithms that achieve near-optimal performance.
TL;DR
This research provides a stochastic geometry framework for downlink NOMA in multi-cell environments, accounting for intercell interference and imperfect SIC. Key findings include the surprising effectiveness of simple distance-based user ordering over interference-aware metrics and the significant gains offered by interference-aware spatial clustering.
What Problem Does This Paper Address?
Non-Orthogonal Multiple Access (NOMA) is a cornerstone spectrum-sharing technology for 5G and beyond. Unlike traditional Orthogonal Multiple Access (OMA), which isolates users in time or frequency, NOMA multiplexes multiple users in the power domain on the same resource block. Receivers then employ Successive Interference Cancellation (SIC) to decode their respective signals.
However, realistic deployment faces significant hurdles. Most existing research ignores intercell interference or uses fixed power allocation that neglects user fairness. In a large-scale network, determining optimal cluster sizes, selecting users (clustering), establishing decoding orders, and allocating power are complex, non-convex problems. This paper addresses these by modeling a multi-cell NOMA network using Poisson Point Processes (PPP) to capture the stochastic nature of interference and imperfect SIC.
Methods
The authors utilize stochastic geometry to model the spatial distribution of base stations and User Equipments (UEs). Their methodology introduces several innovations:
- Spatial Clustering Models: Three models are proposed for NOMA user selection. Model 1 selects users from a full disk around the base station. Model 2 selects users from a half-disk oriented away from the nearest interfering base station to minimize interference. Model 3 acts as a theoretical upper bound by selecting users on a line segment opposite the interferer.
- User Ordering Techniques: For the SIC decoding chain, the study compares MSP (Mean Signal Power, distance-based) against ISINR (Instantaneous Signal-to-Intercell-Interference-and-Noise-Ratio, which accounts for real-time fading and interference).
- Resource Allocation (RA) Algorithms: Two optimization problems are formulated: maximizing cell sum rate with minimum throughput constraints ($\mathcal{P}1$) and maximizing symmetric throughput ($\mathcal{P}2$). To solve these non-convex problems, the authors developed low-complexity iterative algorithms that avoid exhaustive searches.
Key Insights and Ideas
- Simplicity vs. Precision in Ordering: Counterintuitively, ordering users by ISINR (incorporating real-time fading and interference) does not always outperform MSP (distance-based) ordering. Because ISINR ordering is highly precise, its "weakest" user often faces a much worse channel state than the weakest user in MSP ordering. This creates a bottleneck in the SIC chain that can degrade overall cell performance, particularly in larger clusters.
- The "Goldilocks" Cluster Size: There is an optimal number of users ($N$) per NOMA cluster. Beyond this point, the base station's power budget is spread too thin, preventing the weakest users from meeting minimum throughput requirements.
- The Critical Threshold of Imperfect SIC: The authors model imperfect SIC using a residual interference factor ($\beta$). They demonstrate a critical threshold for $\beta$; if residual interference is too high, traditional OMA (like TDMA) becomes more efficient than NOMA.
- Interference-Aware Clustering Gains: Geographically shielding users from the nearest interferer (Models 2 and 3) drastically improves performance compared to naive full-disk clustering (Model 1).
Key Results
- Algorithmic Efficiency: The proposed RA algorithms find near-optimal solutions with significantly lower complexity—roughly 1,000 times faster than exhaustive searches in tested scenarios.
- Clustering Performance: Interference-aware clustering (Model 2) outperformed the baseline (Model 1) by 15.5% in sum rate. The theoretical upper bound (Model 3) showed a 129% improvement.
- Network Capacity: Better clustering allows for larger NOMA groups. Under identical constraints, Model 2 and Model 3 supported maximum cluster sizes of 7 and 9 users, respectively, compared to 6 for Model 1.
Limitations and Future Work
Limitations: The proposed resource allocation algorithms are heuristic and do not guarantee global optimality due to the non-convex nature of the problems. Additionally, Model 3 is a theoretical benchmark rather than a practical deployment strategy.
Future Work: Potential research directions include exploring intermediate spatial sectors (e.g., $\pi/2$ sectors) to balance user availability with interference mitigation. Extending these models to uplink NOMA or MIMO-NOMA systems also remains a key area for future investigation.
Summary updated