Rethinking offload: How to intelligently combine WiFi and small cells?

Meryem Simsek, Mehdi Bennis, Mérouane Debbah, Andreas Czylwik

TL;DR

This paper tackles the challenge of efficiently integrating Wi-Fi and small cell (e.g., 3G/LTE) networks to handle surging mobile data demands. They propose a novel "cross-system learning" framework where small cells intelligently offload traffic between Wi-Fi and licensed bands, optimizing performance based on user needs and network conditions. This approach significantly outperforms existing methods, especially under heavy load.

What Problem Does This Paper Address?

The mobile data explosion is creating "pain-point" situations where demand outstrips network capacity. Future small cell base stations (SCBSs) are expected to be multi-mode, capable of operating on both licensed (like LTE) and unlicensed (like Wi-Fi) bands. The challenge lies in cost-effectively integrating these technologies to manage peak traffic and improve user experience. While Wi-Fi offers free spectrum, it suffers from congestion and uncontrolled access. Licensed spectrum offers better management but has capacity limitations. This paper addresses how SCBSs can intelligently leverage both to optimize traffic offloading and overall network performance, considering diverse user Quality of Service (QoS) requirements.

Methods

The core of the paper is the cross-system learning framework. This is a self-organizing approach where SCBSs learn to steer traffic between Wi-Fi and licensed bands. It's inspired by reinforcement learning, where SCBSs aim to optimize their long-term utility (throughput) by learning optimal transmission strategies.

Key aspects of the method include:

  • Reinforcement Learning (RL) based Behavioral Rule: SCBSs learn a probability distribution over their actions (power allocation and subband selection) to minimize their "regret" – the difference between the utility they could have achieved and what they actually did. This balances exploration (trying new actions) and exploitation (using known good actions).
  • Cross-System Learning: Unlike previous RL approaches that treated licensed and unlicensed bands independently, this framework learns jointly. The learning on the faster Wi-Fi timescale influences the slower licensed band learning.
  • Traffic-Aware Scheduler: A novel scheduler is introduced that goes beyond just channel conditions. It considers users' traffic requirements, completion times (delay), and service classes to make proactive scheduling decisions. This scheduler has two phases: sorting users by priority and then calculating a cost metric for scheduling.
  • Wi-Fi Access Mechanism: For Wi-Fi, SCBSs use a sensing mechanism to find idle periods for transmission, with defined attempt intervals, transmission durations, and sensing durations.

Datasets

Not discussed in the paper. The evaluation is based on simulations.

Key Insights and Ideas

  • Synergy is Key: Treating Wi-Fi and licensed bands as separate entities limits performance. Intelligent, cross-system learning unlocks significant gains by exploiting the interplay between them.
  • Learning to Offload: SCBSs can learn to dynamically offload delay-tolerant traffic to the potentially congested but free Wi-Fi, while keeping delay-sensitive traffic on the managed licensed spectrum.
  • Traffic-Awareness Matters: A simple RL approach isn't enough. Incorporating user traffic requirements and QoS into the scheduling decision is crucial for maximizing overall network performance, especially for diverse traffic types.
  • Faster Convergence: The proposed cross-system learning approach converges much faster than independent learning methods, which is vital for dynamic network environments.
  • Beyond Standard Schedulers: The proposed traffic-aware scheduler, when combined with the cross-system learning, significantly outperforms traditional schedulers like Proportional Fair (PF) and Earliest Deadline First (EDF).

Key Results

  • Convergence Speed: The cross-system learning algorithm converges in less than 50 iterations, while a standard independent RL approach requires hundreds.
  • Throughput Gains:
    • The proposed HetNet+WiFi (load-based) solution shows the best performance in terms of average UE throughput.
  • The overall cell throughput can be improved by up to 200% when deploying 6 small cells, compared to traditional PF and EDF schedulers, by using the proposed cross-system learning with a traffic-aware scheduler.
  • The average UE throughput is significantly higher with the proposed approach, especially as the number of UEs increases.
  • Performance Comparison: The proposed solution consistently outperforms benchmark algorithms (Macro-only, HetNet, HetNet+WiFi with random access) across various scenarios.

Limitations and Future Work

  • Limitations: The paper focuses on a single band scenario for femtocells and Wi-Fi interference. The current model assumes femtocells and Wi-Fi hotspots are in different houses, thus not significantly interfering.
  • Future Work: The authors plan to extend the model to:
  • Consider high-mobility users.
  • Investigate the interplay between mobility, cell association, and interference management.

Summary updated

Abstract

As future small cell base stations (SCBSs) are set to be multi-mode capable (i.e., transmitting on both licensed and unlicensed bands), a cost-effective integration of both technologies coping with peak data demands is crucial. Using tools from reinforcement learning, a distributed cross-system traffic steering framework is proposed whereby SCBSs leverage WiFi, to autonomously optimize their long-term performance over the licensed spectrum band, as a function of the traffic load and users' heterogeneous Quality of Service (QoS) requirements. The proposed traffic steering solution is validated in a Long-Term Evolution (LTE) simulator augmented with WiFi hotspots. Remarkably, it is shown that the proposed cross-system learning-based approach outperforms several benchmark algorithms and traffic steering policies, with gains reaching up to 200% when using a traffic-aware scheduler as compared to the classical proportional fair (PF) scheduler.
BibTeX
@inproceedings{simsek2013rethinking,
  author = {Meryem Simsek and Mehdi Bennis and Mérouane Debbah and Andreas Czylwik},
  title = {{Rethinking offload: How to intelligently combine WiFi and small cells?}},
  booktitle = {ICC},
  year = {2013},
  doi = {10.1109/icc.2013.6655411},
}
Figure