Research Article
Machine Learning Based RATs Selection Supporting Multi-connectivity for Reliability (Invited Paper)
@INPROCEEDINGS{10.1007/978-3-030-25748-4_3, author={Haeyoung Lee and Seiamak Vahid and Klaus Moessner}, title={Machine Learning Based RATs Selection Supporting Multi-connectivity for Reliability (Invited Paper)}, proceedings={Cognitive Radio-Oriented Wireless Networks. 14th EAI International Conference, CrownCom 2019, Poznan, Poland, June 11--12, 2019, Proceedings}, proceedings_a={CROWNCOM}, year={2019}, month={8}, keywords={RAT selection Multi-connectivity Machine learning URLLC}, doi={10.1007/978-3-030-25748-4_3} }- Haeyoung Lee
Seiamak Vahid
Klaus Moessner
Year: 2019
Machine Learning Based RATs Selection Supporting Multi-connectivity for Reliability (Invited Paper)
CROWNCOM
Springer
DOI: 10.1007/978-3-030-25748-4_3
Abstract
While ultra-reliable and low latency communication (uRLLC) is expected to cater to emerging services requiring real-time control, such as factory automation and autonomous driving, the design of uRLLC of stringent requirements would be very challenging. Among novel solutions to satisfy uRLLC’s requirements, interface diversity is widely regarded as an efficient enabler of ultra-reliable connectivity. When mobile devices are connected to multiple base stations (BSs) of different radio access technologies (RATs) and same data is transmitted via multiple links simultaneously, the transmission reliability can be improved. However, duplicate transmission of same data causes an increase in the traffic loads, leading to radio resource shortage. Considering it, efficient configuration of multi-connectivity (MC) for mobile devices is important. In this paper, the RAT selection scheme including efficient MC configuration is proposed. By adopting distributed reinforcement learning (RL), each device could learn the policy for efficient MC configuration and select appropriate RATs. Simulation results show that 20.8% reliability improvements over the single connectivity scheme is observed. Comparing to the method to configure MC for devices all the time, 37.6% improvement is achieved at high traffic loads.
