Reinforcement Learning-Enabled Cross-Layer Optimization for Low-Power and Lossy Networks under Heterogeneous Traffic...

Area of Science:

• Computer Science
• Electrical Engineering
• Network Engineering

Background:

• Internet of Things (IoT) networks face congestion due to massive sensor deployment and heterogeneous traffic.
• Existing protocols struggle with high collision probability, impacting efficiency in IoT networks.
• IEEE 802.15.4 medium access control and network layer ranking mechanisms are key to wireless channel access and packet routing.

Purpose of the Study:

• To intelligently leverage cooperation between multiple communication layers in IoT networks.
• To optimize sensor node performance by addressing network congestion and collision probability.
• To develop an efficient mechanism for next-generation IoT networks.

Main Methods:

• Utilized Q-learning (QL), a machine learning algorithm, for intelligent collision probability inference.
• Integrated channel collision probability and network layer ranking states into the QL algorithm.
• Employed an accumulated reward function to optimize sensor node performance.

Main Results:

• The proposed QL-based scheme achieved a higher packet reception ratio.
• Significantly lower control overheads were produced compared to existing mechanisms.
• Reduced energy consumption was observed for sensor nodes.

Conclusions:

• The QL-based approach effectively optimizes IoT network performance by managing collision probability.
• Cooperation between communication layers, facilitated by machine learning, enhances IoT network efficiency.
• The proposed algorithm offers a promising solution for energy and computationally constrained IoT devices.