Reinforcement Learning-Based Time-Slotted Protocol: A Reinforcement Learning Approach for Optimizing Long-Range...

Area of Science:

• Wireless Communication
• Internet of Things (IoT)
• Network Protocols

Background:

• Low-Power Wide-Area Networks (LPWANs) are crucial for IoT, with LoRa being a key technology.
• LoRa's random access mode leads to collisions and scalability issues as networks grow.
• Optimizing transmission parameter distribution is vital for LoRa network performance.

Purpose of the Study:

• To introduce a novel Reinforcement Learning-based Time-Slotted (RL-TS) LoRa protocol.
• To enhance the scalability and performance of LoRa networks by optimizing transmission parameters and time slot allocation.
• To autonomously enable nodes to select their time slots using a reinforcement learning algorithm.

Main Methods:

• Developed a Reinforcement Learning-based Time-Slotted (RL-TS) protocol for LoRa.
• Implemented a mechanism for autonomous time slot selection by nodes.
• Utilized simulations to evaluate convergence speed, throughput, and packet delivery ratio (PDR).

Main Results:

• RL-TS demonstrated a significant increase in PDR from 0.45-0.85 (LoRa) to 0.88-0.97.
• Throughput improved from 80-150 packets (LoRa) to 156-172 packets with RL-TS.
• RL-TS achieved an 82% reduction in collisions compared to conventional LoRa.

Conclusions:

• The proposed RL-TS protocol effectively enhances LoRa network performance by reducing collisions and improving PDR and throughput.
• Reinforcement learning is a viable approach for optimizing resource allocation in LPWANs.
• RL-TS offers a scalable and efficient solution for demanding IoT applications.