FERMI: a femtocell resource management system forinterference mitigation in OFDMA networks

Suman Banerjee, Mustafa Y. Arslan, Jongwon Yoon, Karthikeyan Sundaresan, Srikanth V. Krishnamurthy, Suman Banerjee 0001MobiCom 2011 · DOI

TL;DR

This paper introduces FERMI, a novel resource management system designed to tackle the significant interference challenges posed by dense femtocell deployments in OFDMA networks. By decoupling resource allocation from per-frame scheduling and employing frequency-domain isolation and intelligent client categorization, FERMI significantly boosts network throughput and fairness compared to existing approaches.

What Problem Does This Paper Address?

The proliferation of femtocells, small cellular base stations deployed in homes and offices, promises increased data rates and network capacity through spatial reuse. However, this dense deployment, especially when using OFDMA (Orthogonal Frequency Division Multiple Access) technology, creates a complex interference landscape. Unlike macrocells where interference is more predictable, femtocell interference is pervasive and challenging to manage. Existing solutions, often designed for macrocells or Wi-Fi, are inadequate for the synchronized, licensed-spectrum nature of OFDMA femtocells. Specifically, managing interference between femtocells, while ensuring fair resource allocation and high network utilization, is a critical and largely unaddressed problem.

Methods

FERMI tackles this challenge through a multi-pronged approach:

  • Frequency Domain Isolation (FDI): Instead of isolating resources in time (Time Domain Isolation - TDI), FERMI uses orthogonal sub-channels in the frequency domain. This allows for "power pooling," where multiple femtocells can pool their transmit power across their allocated sub-channels, leading to higher effective power per sub-channel and enabling higher modulation and coding schemes (MCS).
  • Client Categorization: FERMI intelligently distinguishes between two types of clients: those that only need link adaptation (rate control) and those that require dedicated resource isolation due to significant interference. This is achieved through proactive, measurement-driven triggers, achieving over 90% accuracy.
  • Zoning: To accommodate both types of clients within the same frame, FERMI divides the OFDMA frame into two zones: a "reuse zone" where clients needing only link adaptation can share spectrum, and a "resource isolation zone" where clients requiring isolation are scheduled. This allows for graceful coexistence and efficient resource utilization.
  • Centralized Resource Allocation (A³ Algorithm): A central controller (CC) manages resource allocation across the network. The A³ algorithm, designed to be near-optimal and scalable, transforms the interference graph into a chordal graph to efficiently compute weighted max-min fair allocations of sub-channels. It then assigns these sub-channels, minimizing fragmentation and overhead.
  • Decoupled Zoning: FERMI allows each femtocell to determine its operational reuse zone size based on the desired reuse zone sizes of its neighbors (local minimum), preventing a single, potentially inefficient, global minimum from propagating across the network.

Datasets

The paper describes experiments conducted on a prototype four-cell WiMAX femtocell testbed. They also mention using a channel model proposed by the IEEE 802.16 Broadband Wireless Access Working Group for their simulation evaluations. No new datasets were introduced.

Key Insights and Ideas

  • FDI's Power Pooling Advantage: The key insight is that frequency-domain isolation, unlike time-domain isolation, enables "power pooling." This allows femtocells to collectively increase their transmit power per sub-channel, leading to significant throughput gains, especially as the number of femtocells increases.
  • Intelligent Client Categorization is Crucial: Not all clients require aggressive resource isolation. Differentiating between clients needing simple link adaptation and those needing isolation is vital for avoiding unnecessary resource underutilization.
  • Zoning Balances Reuse and Isolation: The zoning mechanism is a clever way to allow both types of clients to coexist within the same frame, maximizing the benefits of both spectrum reuse and interference mitigation.
  • Decoupled Zoning for Network-Wide Efficiency: Allowing local minimums for reuse zone sizes, rather than a single global minimum, prevents inefficient propagation of restrictive zoning parameters across the entire network, leading to better overall performance.
  • Chordal Graphs for Scalable Allocation: Transforming the interference graph into a chordal graph is a smart algorithmic trick that allows for efficient, near-optimal resource allocation in polynomial time, overcoming the NP-hard nature of the problem on general graphs.

Key Results

  • Significant Throughput Gains: FERMI, particularly with zoning, demonstrated throughput gains of up to 50% on average compared to a baseline approach without zoning or sophisticated resource management.
  • Near-Optimal Fairness: The A³ algorithm achieved near-optimal weighted max-min fairness, matching the benchmark allocation in over 70% of tested topologies and staying within 10% for the rest.
  • High Client Categorization Accuracy: The client categorization mechanism achieved over 90% accuracy in distinguishing between clients needing link adaptation and those needing resource isolation.
  • Reduced Overhead: The A³ algorithm's approach to allocation and assignment minimizes fragmentation, leading to lower control overhead in WiMAX frames.
  • Superiority over Heuristics: FERMI consistently outperformed other heuristic approaches (like "dist" and "deg") in terms of fairness and throughput, especially in more complex interference scenarios.

Limitations and Future Work

  • Testbed Limitations: The authors acknowledge that their testbed setup did not include a macrocell, and they relied on external GPS modules for synchronization. They also note the challenge of implementing sophisticated features like zoning and categorization on off-the-shelf WiMAX MSs without modification.
  • Scalability of A³: While A³ is designed for scalability, the paper mentions that the underlying problem of finding maximal cliques in general graphs is exponential. However, their approach of transforming to a chordal graph mitigates this for practical scenarios.
  • Future Work: The paper suggests exploring alternative definitions for "load" in resource allocation and further investigating the impact of dynamic network changes (client arrivals/departures) on the system. They also hint at a more detailed technical report [17] for further algorithmic details.

Summary updated

BibTeX
@inproceedings{banerjee2011fermi,
  author = {Suman Banerjee and Mustafa Y. Arslan and Jongwon Yoon and Karthikeyan Sundaresan and Srikanth V. Krishnamurthy and Suman Banerjee},
  title = {{FERMI: a femtocell resource management system forinterference mitigation in OFDMA networks}},
  booktitle = {MobiCom},
  year = {2011},
  doi = {10.1145/2030613.2030617},
}
Figure