Personalized Federated Learning-Based Frequency Selection Over HF Space-Ground Integrated Networks

Given the increasing complexity and fierce competition in the electromagnetic environment, there is a growing focus on the promptness and reliability of HF (High-Frequency) communication. Conventional HF communication equipment suffers from low frequency selection efficiency, leading to inefficient link establishment and significant channel fading. Additionally, the limited coverage of ionosphere sounding poses a challenge to integrating HF communication and ionosphere sounding across different locations to improve HF frequency selection efficiency. To address this issue, this article presents an innovative personalized federated learning framework called the high-frequency space-ground integrated network (HFSGIN). This framework utilizes a significant amount of ionospheric data collected from vertical sounding stations and satellites to enhance the integration capability of HF communications with ionosphere sounding. As a result, the efficiency and stability of building HF communication links are improved. The article outlines the basic structure, rationale, and characteristics of HFSGIN. By incorporating personalized federated learning, spatio-temporal correlation analysis, and accounting for the uneven distribution of ionospheric parameters, accurate prediction of these parameters is ensured. This framework takes into account the specific requirements of fine-grained regions for HF user tasks, as well as the general demands of coarse-grained regions for overall network tasks. Additionally, the feasibility of using federated learning in HFSGIN is confirmed, and typical evaluation metrics are provided to facilitate comprehensive training evaluation and adjustment from distributed nodes to the network. Finally, the article suggests various promising research directions.