Approximate reinforcement learning to control beaconing congestion in distributed networks.

Area of Science:

• Vehicular Communications
• Network Congestion Control
• Wireless Networking

Background:

• Excessive periodic messages (beacons) in vehicular communications increase channel load, impacting safety applications.
• Current congestion control solutions often embed data in message payloads, risking failure in poor channel conditions due to packet loss.

Purpose of the Study:

• To develop a non-cooperative, distributed beaconing allocation strategy for vehicular networks.
• To address the challenge of controlling beaconing rates to ensure reliable communication for safety and driver-assistance systems.

Main Methods:

• Formulated the beaconing rate control problem as a Markov Decision Process (MDP).
• Utilized approximate reinforcement learning to find optimal beaconing actions.
• Compared the proposed solution (SSFA) against traditional congestion control methods.

Main Results:

• The proposed SSFA approach maintains available channel capacity, ensuring emergency notification delivery.
• SSFA demonstrated faster convergence compared to other existing proposals.
• Achieved favorable packet delivery and collision ratios.

Conclusions:

• The developed distributed beaconing allocation strategy effectively manages channel load in vehicular networks.
• Reinforcement learning provides an efficient method for optimizing beaconing rates in a non-cooperative environment.
• The SSFA method enhances the reliability of safety-critical communications by guaranteeing message delivery.