Decentralized Load Balancing in Mobile Communication Networks

Future generations of mobile communication networks are envisioned to utilize dense deployments of heterogeneous cell types to fulfill increasing performance requirements. An efficient and optimized utilization of the network resources, including user allocation management and load balancing between cells, is crucial for such networks to maintain high throughput and to handle increasing interferences. Due to the combinatorial nature of user allocation, load balancing is a mixed-integer linear problem that does not scale well will the number of users and cells. Heuristic methods to solve similar problems in this context are available, but they typically require extensive coordination between network entities while achieving highly suboptimal performance. We propose a machine learning based approach that achieves close to optimal performance while requiring very limited local interaction and computational effort during operation.