Scalable Multi-Agent Learning for Situationally-Aware Multiple-Access and Grant-Free Transmissions


SOURCE IEEE SPAWC’19, Cannes, France,Jul. 2-5,2019

Published Date: Jul. 2-5,2019


In this paper, we present Situationally-aware Multiple-Access and gRant-free Transmissions (SMART) to address the low-latency multiple access issue in wireless uplinks with massive machine-type communications (mMTC). SMART is smart in a distributed manner, as terminals are trained to be situationally aware. The solution is based on a distributed reinforcement learning framework which is capable of dealing with  diversified quality-of-service requirements. As such, terminals can make informed transmission decisions by themselves, e.g., by taking into account the urgency of their packets and system load. In this way, the effective number of concurrent access terminals is significantly reduced while maintaining the system performance. Compared with conventional contention-based random access schemes, SMART has significant advantages. Building upon our previous work [1], this work presents the first SMART algorithm that is scalable to massive terminals (hundreds) and stable with near-optimal performance.

This entry was posted in Publications and tagged . Bookmark the permalink.

Leave a Reply