Session Management for URLLC in 5G Open Radio Access Network: A Machine Learning Approach

Supporting ultra-reliable and low latency communication (URLLC) has been a mandatory function for the International Mobile Telecommunications 2020 (IMT-2020) systems and so as 3GPP New Radio (NR). Conventionally, methods for URLLC primarily focus on performance enhancement on the air interfaces, which ignore a fact that data transmissions through the core network (CN) and the backhaul data network (DN) may invoke considerable latency and such latency may not be addressed solely by a local base station (BS). In this case, before the event of unacceptable latency occur, a BS should not accept the request of a new session creation, so as not to violate the latency and reliability requirements of the existing serving sessions and the new session. For this purpose, the critical challenge lies in how to proactively detect/cognize that the latency/reliability requirement violation event is going to occur, which relies on an effective experience update and process. To tackle this challenge, we particularly note the feature of event prediction in machine learning (ML) methods through experience training, especially the capability of sequential decision making to interact with an unknown environment in reinforcement learning (RL). In this paper, an intelligent session management is therefore proposed. Based on the recent innovation of Open Radio Access Network (O-RAN) to sustain the proposed RL scheme for intelligent session management, an O-RAN based BS is able to effectively configure/admin the resources for each existing serving sessions and the new session. Our simulation results fully demonstrate the practicability of the proposed approach in supporting URLLC in O-RAN, to justify the potential of our approach in the design for 3GPP NR.