On SE and Delay Trade-Off in UAV-Enabled IoT Networks Underlying HetNets with Optimal Spectrum Partitioning

In this study, we propose an innovative method for optimal spectrum partitioning and resource allocation in unmanned aerial vehicle (UAV)-assisted heterogeneous networks (HetNets). We divide the HetNets architecture into two communication tiers: the uplink for UAVs and IoT devices, and the downlink for the macro base station (MBS) and user equipment (UE). Our unified resource allocation framework includes power allocation, subchannel association along with introducing a novel spectrum partitioning parameter that are jointly optimized for both tiers. We introduce a new utility function as a multi-objective optimization (MOO) problem which simultaneously maximize spectral efficiency (SE) in MBS-UE networks and minimize transmission delay in UAV-IoT networks while maintaining quality of service (QoS). Due to the nonconvex nature and inherent complexity of the problem which is complex and intractable, we transform the multi-objective formulation into a single-objective one using a scalarization method. We then develop two sub optimal effective algorithms: an analyticalbased approach and a learning-based approach. In the analytical-based approach, we decompose the multiobjective problem into several subproblems using block coordinate descent (BCD); each nonconvex subproblem is subsequently transformed into a convex one via the successive convex approximation (SCA) algorithm. In the learning-based approach, we propose a lowcomplexity framework that leverages both supervised and unsupervised deep learning strategies. Simulation results indicate that both algorithms outperform baseline schemes, with the learning-based approach showing particularly promising performance compared to the analytical-based approach.