Abstract
We study the problem of interference management in large-scale small cell networks, where each user equipment (UE) needs to determine in a distributed manner when and at what power level it should transmit to its serving small cell base station (SBS) such that a given network performance criterion is maximized subject to minimum quality of service (QoS) requirements by the UEs. We first propose a distributed algorithm for the UE-SBS pairs to find a subset of weakly interfering UE-SBS pairs, namely the maximal independent sets (MISs) of the interference graph in logarithmic time (with respect to the number of UEs). Then we propose a novel problem formulation which enables UE-SBS pairs to determine the optimal fractions of time occupied by each MIS in a distributed manner. We analytically bound the performance of our distributed policy in terms of the competitive ratio with respect to the optimal network performance, which is obtained in a centralized manner with NP (non-deterministic polynomial time) complexity. Remarkably, the competitive ratio is independent of the network size, which guarantees scalability in terms of performance for arbitrarily large networks. Through simulations, we show that our proposed policies achieve significant performance improvements (from 150% to 700%) over the existing policies.
Introduction
As mobile data demand continues to explode, providers are increasingly deploying "small cells" (low-power base stations like picocells and femtocells) in offices and homes. While this boosts capacity, it creates a complex environment of mutual interference. Traditional cellular networks rely on a central controller to coordinate transmissions, but in dense, heterogeneous environments with thousands of small cells, centralized coordination is computationally impossible (NP-hard) and creates massive signaling overhead. The challenge addressed here is how individual user equipments (UEs) and base stations (SBSs) can decide when to transmit and at what power level using only local information, while ensuring all users meet their minimum throughput requirements.
Proposed Methodology
The authors introduce a scalable, four-step distributed framework for managing this interference:
- Neighbor Identification: Each UE-SBS pair identifies "strong interferers" in its immediate vicinity to build a local view of an interference graph.
- Distributed MIS Generation: The network identifies Maximal Independent Sets (MISs)—groups of users who do not strongly interfere and can transmit simultaneously. The proposed distributed algorithm finds these sets in logarithmic time relative to the number of users, a significant speedup over previous linear-time methods.
- Optimization via ADMM: Determining the optimal timing for each MIS is normally a coupled problem requiring global knowledge. The authors reformulate this into a Decoupled Problem (DP) and use the Alternating Direction Method of Multipliers (ADMM). This allows users to reach a consensus on optimal timing through simple local message exchanges with neighbors.
- Transmission: Users transmit during their assigned slots at maximum power, exploiting the "weak" interference allowed within an MIS.
Key Insights and Theoretical Guarantees
- Spatial Reuse vs. Power Control: The paper argues that scheduling (turning users on/off via MISs) is more effective at eliminating the "strong" interference that degrades performance in dense deployments than traditional constant power control.
- The Power of Decoupling: By introducing auxiliary variables and using ADMM, the authors prove that local consensus is mathematically equivalent to the global optimum.
- Scalability: The authors analytically prove a constant competitive ratio, meaning the gap between this distributed solution and a theoretical "perfect" centralized solution does not grow as the network expands. This ensures performance scalability for arbitrarily large networks.
- System Feasibility: For the system to meet everyone's minimum requirements, the interference graph must have a bounded degree; if one user is overwhelmed by too many neighbors, no algorithm can guarantee their connection.
Performance Evaluation
The framework was evaluated using extensive numerical simulations, including small-scale 4-cell scenarios and large-scale models with up to 105 UEs in a 12-room building. Standard path-loss and Rayleigh fading models were used to simulate realistic wireless environments.
- Throughput Gains: The proposed policy outperformed state-of-the-art STDMA and power control policies by 150% to 700%.
- Fairness and Convergence: Under max-min fairness criteria, the policy achieved significantly higher minimum throughput for the "worst-off" users. The distributed algorithms converged in logarithmic time, making them practical for real-time hardware implementation.
- Robustness: The system maintained high performance even under dynamic channel conditions and frequent user churn.
Conclusion and Limitations
While the framework offers massive gains, practical implementation assumes time-slotted synchronization across the network and requires a reliable backhaul (like the X2 interface) for local message exchanges. Future work could integrate beamforming (MIMO) by treating it as a method to "delete edges" in the interference graph, and extend the model to handle mobile base stations.
Summary updated