Distributed Multichannel Access in High-Frequency Diversity Networks: A Multi-Agent Learning Approach With Correlated Equilibrium

Wen Li, Yuhua Xu, Yunpeng Cheng, Yang Yang, Xueqiang Chen, Meng Wang, Dianxiong LiuIEEE Access 2019 · OpenReview

Introduction

High-frequency (HF) communication, operating in the 3–30 MHz range, remains a critical backbone for long-distance military and emergency communications. However, the HF spectrum is notoriously difficult to manage; it is narrow, subject to deep fading, and frequently congested. To improve reliability, "diversity techniques" are often employed, where a single user equipment (UE) transmits across multiple channels simultaneously.

While diversity improves individual link quality, it introduces a significant challenge in distributed networks: the "wild west" of channel access. Without a central controller, multiple UEs attempting to use several channels at once lead to a high probability of collisions. Traditional solutions often fail to balance efficiency with fairness, frequently allowing aggressive users to dominate the best spectrum.

System model of the HF diversity network, showing the core, access, and UE layers where collisions occur.

The Proposed Solution: Correlated Equilibrium

This paper proposes a distributed learning algorithm based on Correlated Equilibrium (CE) to manage multi-user multichannel access. Unlike the standard Nash Equilibrium, which can lead to inefficient outcomes in non-cooperative games, CE allows agents to coordinate their actions based on a shared, external signal.

The Multichannel Access Algorithm (MAA)

The core of the approach is a Multichannel Access Strategy Table (MAST) maintained by each UE. The system operates through several key mechanisms:

  1. Coordination Signals: A central station broadcasts a random integer at the start of each time slot. This signal acts as a "common clock" or "traffic light," providing a basis for coordination without requiring direct communication between UEs.
  2. Distributed Learning: UEs start with random strategies. Successful transmissions reinforce the current mapping in the MAST, while collisions trigger a probabilistic update to explore different channels.
  3. Intelligent Monitoring: When not transmitting, UEs "listen" to idle channels. An adjustment factor ($\beta$) is used to bias this monitoring toward channels with historically better communication quality.
Visualization of the time-slot and channel allocation scheme used in the proposed algorithm.

Performance and Insights

The authors evaluated the MAA using numerical simulations in MATLAB, comparing it against Multichannel Slotted-ALOHA and other learning variants.

  • Rapid Convergence: The system typically reaches a stable, collision-free state in fewer than 400 iterations.
  • Throughput and Fairness: The MAA achieves approximately 60% normalized throughput—a massive improvement over the 10% seen with Slotted-ALOHA. Furthermore, it maintains a Jain Fairness Index (JFI) near 0.98, ensuring nearly perfectly equal resource distribution.
  • Robustness: The algorithm remains effective even with a 5% error rate in interpreting coordination signals, making it suitable for high-interference HF environments.

Conclusion

By framing channel access as a game of Correlated Equilibrium, this research demonstrates that independent users can learn to share a complex spectrum efficiently and fairly. Future work will focus on testing these algorithms against more complex ionospheric channel models and moving toward practical hardware implementations.

Summary updated

Abstract

This paper investigates the problem of multi-user multichannel access in distributed high-frequency (HF) diversity communication networks using a game-theoretic learning algorithm which is based on correlation equilibrium (CE). We formulate the channel access problem in the HF networks as a non-cooperative game. In the access game, each user equipment (UE) optimizes its access strategy without the information about other UEs, which makes the channel access problem challenging. It is proved that there is at least one CE point that makes all UEs' access strategy efficient and fair. We propose a distributed learning algorithm based on CE to achieve multi-user access with low cost and fairly in the distributed HF networks. We use coordination signals to help each UE learn the access strategy by themselves. When each UE receives and recognizes the right coordination signals, UEs will learn to transmit data on right channels without further collisions after the learning phase. The simulation results show that the proposed learning algorithm can not only completely avoid interference and get optimal throughput but also guarantee fairness among all UEs.
BibTeX
@inproceedings{li2019distributed,
  author = {Wen Li and Yuhua Xu and Yunpeng Cheng and Yang Yang and Xueqiang Chen and Meng Wang and Dianxiong Liu},
  title = {{Distributed Multichannel Access in High-Frequency Diversity Networks: A Multi-Agent Learning Approach With Correlated Equilibrium}},
  booktitle = {IEEE Access},
  year = {2019},
  url = {https://openreview.net/forum?id=b57HWFg4NT},
}
Figure