LLM-inspired decomposition multi-objective evolutionary algorithm for joint BDS-3 resource allocation

Abstract

With the BeiDou-3 Global Navigation Satellite System (BDS-3) now attaining worldwide coverage, the rapid expansion of digital services has triggered an unprecedented surge in operational demands. However, limited spectral resources and onboard energy reserves are insufficient to meet this growing demand, which restricts further improvement in system efficiency. To optimally allocate constrained resources and elevate system performance, this paper introduces a coordinated power and bandwidth allocation strategy for BDS-3, grounded in a large language model (LLM)-inspired decomposition multi-objective evolutionary algorithm (MOEA). The proposed method adaptively tunes power and bandwidth assignments according to the channel conditions and service requirements of each link, with the dual objective of harmonizing energy efficiency and communication rate within the system. Specifically, we develop an LLM-inspired decomposition multi-objective evolutionary algorithm (LEDMA) to tackle the highly non-convex multi-objective optimization problem. A decomposition approach is first employed to partition the original problem into multiple scalar subproblems. Subsequently, the LLM serves as a black-box search operator and is seamlessly incorporated into the MOEA framework via carefully designed prompt engineering, enabling concurrent resolution of these subproblems. Simulation results validate that LEDMA effectively reconciles energy efficiency and communication rate, yielding Pareto fronts and convergence characteristics that surpass those of benchmark multi-objective evolutionary algorithms.