Optimal-Capacity, Shortest Path Routing in Self-Organizing 5G Networks using Machine Learning

Abstract

Machine learning is expected to be a key enabler in 5G wireless self-organizing networks (SONs) that will be significantly more autonomous, smarter, adaptable and user-centric than current networks. This paper proposes a methodology, User Specific-Optimal Capacity Shortest Path (US-OCSP) routing, that uses machine learning to determine the resource-based optimum-capacity shortest path for a user between source and destination. The methodology takes into account two primary metrics, available capacity at network nodes (eNodeBs/gNodeBs) and distance, that are critical in determining the optimal path for an end-user. An ns-3 simulation determines the capacity, which is measured by the availability of resources [i.e., Physical Resource Blocks (PRBs)] at all possible serving network nodes between the source and destination, that is followed by implementation of Q-learning, a reinforcement type of machine learning algorithm that determines the shortest path avoiding congested network nodes so as to achieve the required throughput and/or bit rate. The ability to determine the optimal-capacity shortest path route will facilitate effective resource allocation that will optimize end-user satisfaction in a 5G SON network.