Energy-Efficient Resource Allocation Scheme Based on Reinforcement Learning in Distributed LoRa Networks.

Area of Science:

• Wireless Communications
• Internet of Things (IoT)
• Machine Learning

Background:

• Rapid expansion of Long Range (LoRa) devices causes network congestion, diminishing spectrum and energy efficiency.
• Existing methods struggle to balance performance and power consumption in dense LoRa deployments.

Purpose of the Study:

• To develop an energy-efficient, distributed reinforcement learning method for LoRa networks.
• To enable individual LoRa devices to autonomously optimize transmission parameters (channel, transmission power, bandwidth).

Main Methods:

• Utilized the Upper Confidence Bound (UCB)1-tuned algorithm for parameter selection.
• Integrated energy consumption metrics into the reinforcement learning reward function.
• Designed a lightweight algorithm suitable for resource-constrained IoT devices.

Main Results:

• Achieved significant reductions in power consumption compared to baseline methods.
• Demonstrated high transmission success rates even in dense network scenarios.
• Outperformed fixed allocation, ADR-Lite, and epsilon-greedy approaches.

Conclusions:

• The proposed reinforcement learning method effectively enhances energy efficiency and transmission success in LoRa networks.
• This lightweight solution is practical for real-world, resource-limited IoT applications.
• The method offers a superior alternative to existing parameter allocation strategies for LoRa devices.