AeroLOBE: A Deep Learning-Based Novel Load Balancer for Aerial Communication to Enhance the Performance of 6G Networks

While the low-altitude platform (LAP)-based aerial cells help improve the coverage and capacity for the telecom operator, the deployment and management of the aerial fleet is a non-trivial problem from both a capital expenditure (CAPEX) and an operational expenditure (OPEX) perspective. On the one hand, it is critical to keep the fleet size to a minimum to reduce CAPEX, while on the other hand, it is critical to optimally associate user equipment (UE) with aerial cells in order to maximize the use of aerial cell resources and serve more pieces of UE. Existing research on balancing UE load among aerial cells discusses mechanisms like coverage and capacity optimization. To the best of our knowledge, this is the first time we have treated the forecasted data traffic volume for each user as well as inter-UE traffic consideration to jointly optimize capacity maximization for aerial cells, latency minimization in inter-UE communication, and aerial fleet size minimization. To this end, we present a deep learning-based novel aerial load balancer (AeroLOBE) for aerial communication using a novel constraint fractional group multiple knapsack problem (F-GMKP) formulation and the knapsack optimization (KO) for associating users to LAPs, thereby enhancing performance for the network. Through mathematical modelling and extensive simulation, we show that AeroLOBE reduces the latency of inter-UE communication by over 39% and improves the resource utilization by over 10%, while keeping the blocking rate and fleet reduction targets similar or marginally better than the available load balancing schemes in the literature.