Energy efficiency optimization and DQN-based power allocation in UAV-IRS assisted 6 G system

Abstract

With the growing number of devices, applications, and services, energy-efficient communication has become critical in next-generation wireless networks. Terahertz (THz) bands provide vast bandwidth for high data rates, but their poor penetration causes user blockage. To address this, we consider a massive MIMO (mMIMO) THz non-orthogonal multiple access (NOMA) system assisted by an IRS-mounted UAV that reflects signals toward blocked users. We formulate the UAV trajectory design as an NP-hard problem and propose a low-complexity Track Selection Scheme (TSS) that minimizes UAV energy consumption. We develop a Deep Q-Network (DQN)-based NOMA power allocation scheme to further enhance performance and formulate an energy-efficiency maximization problem. The latter is solved using a combination of the Dinkelbach method and the Concave–Convex Procedure (CCCP). Simulation results demonstrate that the proposed schemes significantly reduce UAV energy consumption and improve system sum-rate and energy efficiency compared with benchmark algorithms.