Deep Reinforcement Learning for Optimizing Restricted Access Window in IEEE 802.11ah MAC Layer.

Area of Science:

• Wireless Communication Networks
• Internet of Things (IoT)
• Network Performance Optimization

Background:

• The IEEE 802.11ah standard addresses the increasing scale of IoT applications.
• The Restricted Access Window (RAW) mechanism in the MAC layer manages numerous stations to reduce contention and improve energy efficiency.
• Optimizing RAW parameters (groups, slots, duration) is crucial for network performance.

Purpose of the Study:

• To optimize RAW parameters configuration for uplink IEEE 802.11ah-based IoT networks.
• To enhance network throughput by addressing the RAW parameters optimization problem.
• To propose a robust method for determining preferable RAW parameters under complex and dynamic network conditions.

Main Methods:

• Formulated a RAW parameters optimization problem to improve network throughput.
• Proposed a Deep Reinforcement Learning (DRL) approach, specifically Proximal Policy Optimization (PPO), to determine optimal RAW parameters.
• Validated the PPO-based algorithm in an NS-3 simulator with periodic and random traffic patterns.

Main Results:

• The PPO-based DRL algorithm successfully obtained optimized RAW parameters across various network conditions.
• Significant improvements in network throughput were demonstrated using the proposed optimization method.
• The algorithm showed effectiveness in enhancing learning efficiency and stability.

Conclusions:

• The PPO-based DRL approach provides an effective solution for optimizing RAW parameters in IEEE 802.11ah IoT networks.
• Optimized RAW parameters lead to substantial gains in network throughput.
• This method offers a promising strategy for managing contention and improving performance in large-scale IoT deployments.