RL-TSCH: A Reinforcement Learning Algorithm for Radio Scheduling in TSCH 802.15.4e
| Autor: | Sonxay Luongoudon, Hung Nguyen-Duy, Thu Ngo-Quynh, Fumihide Kojima, Toan Nguyen-Duc, Tien Pham-Van |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
021103 operations research
Computer science business.industry Network packet Testbed 0211 other engineering and technologies Q-learning 020206 networking & telecommunications 02 engineering and technology Scheduling (computing) IPv6 Protocol stack 0202 electrical engineering electronic engineering information engineering Reinforcement learning Markov decision process 6LoWPAN business Computer network |
| Zdroj: | ICTC |
| DOI: | 10.1109/ictc46691.2019.8939833 |
| Popis: | In order to apply Internet of Things technologies into real world, the low-power low-cost protocol stack CoAP/UDP/IPv6/RPL/6LoWPAN/802.15.4-802.15.4e is a good solution and appropriate to low-power low-cost applications. The standard IEEE 802.15.4e utilizes scheduled-based multiple access approach in this, at the beginning of each time slot, the nodes will always turn on the radio for a specified period of time to check if packets are sent or not and the process is repeated in the next time slot (Minimal Scheduling Algorithm). Thus, it leads to unnecessary waste of energy if for a while the node does not transmit or send a packet. To overcome this disadvantage, we propose in this paper a reinforcement learning solution, called RL-TSCH for scheduling nodes (controlling the node’s radio on or off) of TSCH 802.15.4e based on Markov Decision Process associated with Q-Learning algorithm. We evaluate the performance of our RL-TSCH algorithm and compare with Minimal Scheduling Algorithm in simulation and real testbed with Contiki operating system. The initial result shows that, our mechanism achieves similar packet delivery rate (96%) while saving power consumption up to 30% compared to Minimal Scheduling. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|