Joint Trajectory and Resource Optimization of MEC-Assisted UAVs in Sub-THz Networks: A Resources-Based Multi-Agent Proximal Policy Optimization DRL With Attention Mechanism

The use of Terahertz (THz) technology in sixth-generation (6G) networks will bring high-speed and capacity data services. But limitations like molecular absorption, rain attenuation, and limited coverage range cause communication losses. To overcome these losses and improve coverage in rural areas, a high number of base stations are required. Hence, an aerial communication platform, which uses line-of-sight (LoS) communication to avoid losses, is needed. To address this, we study the deployment and optimization of multi-access edge computing (MEC)-powered unmanned aerial vehicle (UAV) for sub-THz communication in remote areas. To this end, we solve an optimization problem to minimize energy consumption and delay for MEC-UAV and mobile users. The formulated problem is a mixed-integer non-linear programming (MINLP) problem. As the problem is an MINLP, we decompose the main problem into two subproblems. Due to its convex nature, we solve the first subproblem with a standard optimization solver, i.e., CVXPY. To solve the second subproblem, we design a resources-based multi-agent proximal policy optimization (RMAPPO) deep reinforcement learning (DRL) algorithm with an attention mechanism. The considered attention mechanism is utilized for encoding a diverse number of observations. This is designed by the network coordinator to provide a differentiated fit reward to each agent in the network. The simulation results show that the proposed algorithm outperforms the benchmark and yields a network utility that is 2.22%, 15.55%, and 17.77% more than the benchmarks.