Traffic capacity of multi-cell WLANS

Thomas Bonald, Ali. M. A. Ibrahim, James W. Roberts

TL;DR

This paper moves beyond the study of isolated WiFi cells to analyze the "traffic capacity" of multi-cell WLANs under dynamic load. The authors demonstrate that inter-cell interference creates complex dependencies that can be modeled as coupled queuing systems, revealing that network capacity can actually grow significantly faster than the number of available frequency channels—a counterintuitive and vital insight for network planning.

What Problem Does This Paper Address?

Most research on IEEE 802.11 (WiFi) performance focuses on a single Access Point (AP) with a static number of users. However, real-world deployments involve multiple APs with users constantly starting and finishing data transfers (dynamic traffic).

In these environments, APs using the same frequency channel interfere with each other. To manage this, WiFi uses the CSMA/CA protocol and the RTS/CTS (Request-To-Send/Clear-To-Send) mechanism, which creates an "exclusion region" around a transmission. If an AP or a user is within this region, they must remain silent. While this prevents collisions, it also limits how much data the network can carry. The authors set out to define and calculate the traffic capacity: the maximum data intensity (bits per second) a multi-cell network can support before the number of active downloads explodes to infinity.

Methods

The authors employ a flow-level stochastic model to represent the network:

  • Coupled Queuing System: Each AP is treated as a queue. Unlike standard queues, the service rate of one AP depends on the activity of its neighbors due to the overlapping exclusion regions.
  • Stability Analysis: They derive the mathematical conditions under which these coupled queues remain stable (i.e., the number of users doesn't grow indefinitely).
  • Analytical Topologies: They apply the model to 1D (linear) and 2D (grid) network layouts to derive closed-form expressions for capacity.
  • Fixed-Point Approximation: To estimate the user experience (mean transfer rate), they develop an approximation that decouples the queues, making the complex math more tractable.
  • Validation: The theoretical results are validated using NS-2 simulations, a standard tool in network research, to ensure the simplified mathematical model holds up against more realistic protocol behavior.

Datasets

This is a theoretical and simulation-based paper; it does not use external real-world datasets. Instead, the authors evaluate their models using:

  • Synthetic Topologies: Infinite linear networks (1D) and infinite grid networks (2D).
  • NS-2 Simulations: Used as a ground-truth benchmark to validate the analytical "toy" models.

Key Insights and Ideas

  • Non-Linear Scaling: The most surprising finding is that using $M$ non-interfering channels can increase capacity by much more than a factor of $M$. For example, in a 2D grid, capacity can scale closer to $M^2$ (or $4M$ in specific configurations). This happens because more channels allow APs to be placed closer together, shrinking the individual cell size and reducing the relative impact of the interference "footprint."
  • The "Exclusion Region" Discontinuity: Capacity doesn't change smoothly as you move APs. There are "jumps" or discontinuities at critical distances (e.g., when an AP moves just far enough away that it no longer hears its neighbor's RTS/CTS signals).
  • Indirect Interference: APs can interfere even if they can't "hear" each other directly, provided their respective users are close enough to cause collisions. This "hidden node" interference significantly shapes the network's total capacity.
  • Fairness vs. Capacity: The model assumes a "processor-sharing" approach, meaning all active users at an AP share the available bandwidth equally, which is a reasonable approximation of how WiFi protocols behave under load.

Key Results

  • 1D Networks: In a linear network with 3 channels (like the common 802.11b/g setup), the capacity increases 4.23-fold compared to a single-channel network when APs are placed optimally.
  • 2D Networks: For a 2D grid with $M$ channels, the optimal traffic density is approximately $4(\sqrt{M}-1)^2$. For 4 channels, the capacity is roughly 4 times that of a single-channel network; for 100 channels, it jumps to nearly 324 times.
  • User Experience: The mean transfer rate seen by a user decreases almost linearly as the network load increases, reaching zero exactly at the calculated traffic capacity.
  • Optimal Placement: The results provide a theoretical "sweet spot" for AP spacing: placing APs just beyond their mutual transmission range maximizes the reuse of frequency channels.

Limitations and Future Work

  • Simplified Radio Model: The authors assume a "disk" model for interference and neglect complex real-world effects like fading, shadowing, or varying bit rates based on signal strength (link adaptation).
  • Slotted Time: The model assumes synchronized, slotted transmissions, which simplifies the math but doesn't perfectly capture the asynchronous nature of real WiFi.
  • Downlink Focus: The study primarily looks at downlink traffic (AP to user). In modern networks, uplink traffic (especially for video calls) is increasingly important and would add another layer of complexity.
  • RTS/CTS Overhead: The analytical model slightly overestimates capacity because it ignores the finite time required to transmit the RTS/CTS control frames themselves, a discrepancy noted during NS-2 validation.

Future work suggested by the authors includes relaxing these assumptions to account for more realistic radio environments and exploring opportunistic scheduling schemes that could further boost capacity.

Summary updated

Abstract

Performance of WLANs has been extensively studied during the past few years. While the focus has mostly been on isolated cells, the coverage of WLANs is in practice most often realised through several cells. Cells using the same frequency channel typically interact through the exclusion region enforced by the RTS/CTS mechanism prior to the transmission of any packet. In this paper, we investigate the impact of this interaction on the overall network capacity under realistic dynamic traffic conditions. Specifically, we represent each cell as a queue and derive the stability condition of the corresponding coupled queuing system. This condition is then used to calculate the network capacity. To gain insight into the particular nature of interference in multi-cell WLANs, we apply our model to a number of simple network topologies and explicitly derive the capacity in several cases. The results notably show that the capacity gain obtained by using M frequency channels can grow significantly faster than M, the rate one might intuitively expect. In addition to stability results, we present an approximate model to derive the impact of network load on the mean transfer rate seen by the users.
BibTeX
@article{bonald2008traffic,
  author = {Thomas Bonald and Ali. M. A. Ibrahim and James W. Roberts},
  title = {{Traffic capacity of multi-cell WLANS}},
  journal = {ACM SIGMETRICS Performance Evaluation Review},
  year = {2008},
  doi = {10.1145/1384529.1375504},
}

Loading related content...

Figure