Machine Learning Assisted Capacity Optimization for B5G/6G Integrated Access and Backhaul Networks

The cross-layer design on the routing of traffic and scheduling of wireless backhaul links in the beyond 5G (B5G)/6G integrated access and backhaul (IAB) networks has continued to draw attention, owing to the stringent requirement of ultra-reliable low latency communication (URLLC). Efforts to split the total bandwidth or allocate a static partition to some backhaul links have not improved the sum rate, and the use of approximation algorithms for resource scheduling in the shared spectrum have posed significant computation overhead. In this paper, we propose a two-stage machine learning (ML) framework for capacity optimization, where the scheduling structure of past optimization instances in a simulated IAB multi-hop network are explored and exploited to accelerate the solution of a new optimization instance with linear programming. We evaluate the ML method on different multi-commodity flow (MCF) deployments in uplink and downlink, and achieved up to 94% average throughput in comparison with the delayed column generation (DCG) benchmark algorithm. Further to that, our ML method significantly reduces the computation time by at least 95 %.