Deep Learning based Multi-objective Admission Control: A Novel O-RAN Compliant Approach

Significant advances in wireless network technologies have given rise to use cases with unprecedented data rates, number of devices as well as low-latency constraints that thrive on such agility of networks. However, it also means that network management through appropriate resource allocation, coverage, and continuity of connection has become significantly more challenging compared to previous generations of networks. More recently, disaggregated network architectures have received increased attention through virtualization of important network functions and through the promise of significant value creation leveraging data-driven intelligent network control policies. In order to leverage the benefits of those versatile network technologies it is imperative that a fine balance be maintained between the resources allocated to a given set of users and the number of users that gain access to the network through intelligent admission control (AC). In this work, we propose and evaluate an admission control policy that leverages data-driven optimization to admit the maximum number of users while being acutely aware of quality of service (QoS) constraints of the users. This novel approach is achieved within the emerging O-RAN framework leveraging RAN Intelligent Controllers (RICs) to recommend appropriate policies enabling operators to maintain an efficient operating network. Our simulation results show that by intelligent tradeoff between the QoS and incoming UE request rejections, we can reduce the rejections by 2-3x in congested networks while maintaining acceptable QoS.