ROMJIST Volume 28, No. 4, 2025, pp. 327-340, DOI: 10.59277/ROMJIST.2025.4.02
Sothearath MENG, Kimchheang CHHEA, Sengly MUY, Jung-Ryun LEE Deep Reinforcement Learning and Metaheuristic Approaches to Maximize Downlink Sum-Rate for Internet of Things Systems in Non-Orthogonal Multiple Access-based Space-Air-Ground Integrated Networks
ABSTRACT: Internet of Things (IoT) device has significantly increased the need for reliable and efficient communication systems. Space–Air–Ground Integrated Network (SAGIN) addresses this need through its hierarchical structure, by integrating Low Earth Orbit (LEO) satellites, High Altitude Platforms (HAPs), and Unmanned Aerial Vehicles (UAVs). This paper focuses on maximizing the downlink sum-rate in a Non-Orthogonal Multiple Access (NOMA)-based SAGIN-IoT system by jointly optimizing the geographical locations and transmit powers of HAPs and UAVs, bandwidth allocation ratio, and link selection between a LEO satellite and IoT devices. The problem is formulated as a complex joint optimization task involving both discrete and continuous variables, reflecting the dynamic and large-scale nature of the SAGIN network. To solve this, two solution algorithms are employed: a deep reinforcement learning (DRL) algorithm and an Alternative Optimization (AO) algorithm. The proposed DRL framework leverages a deep Q-learning (DQL) architecture to efficiently navigate the high-dimensional and dynamic environments of SAGIN. The AO algorithm, on the other hand, decomposes the original optimization problem into two subproblems, iteratively solving them using Differential Annealing (DA). The performance of the proposed DQL and AO algorithms is compared with that of Gradient Search (GS). Simulation results demonstrate that DQL achieves superior performance in terms of overall sum-rate optimization with lower computational complexity. While the AO algorithm provides competitive results, it requires higher computational complexity than both DQL and GS.KEYWORDS: Alternative Optimization (AO), Deep Reinforcement Learning (DRL); High-Altitude Platform (HAP); Internet of Things (IoT); link selection; Unmanned Aerial Vehicle (UAV)Read full text (pdf)