Prediction of Maximum Usable Frequency for HF Communications Based on Federated Learning

Abstract

Spectrum prediction technology can not only improve spectrum utilization rate but also improve communication quality to a certain extent. In order to give full play to the advantages of high-frequency (HF) communication and improve the quality and stability of HF communication, it is urgent to predict the HF spectrum. We propose a prediction model of maximum usable frequency (MUF) of HF based on federated learning. The proposed model has great advantages in protecting user privacy, reducing communication loss, and shortening training time. First, the ionospheric vertical sounding dataset is analyzed and processed to obtain the training dataset and test dataset, and the data are standardized and reconstructed. Second, the pre-processed training data set was used as the input data to enter the long short-term memory (LSTM) network for local model training. The federated learning (FL) server station aggregates all local training models to generate global models, and delivers model parameters to local users for updating local models and carrying out the next round of training. Finally, the experimental results show that the proposed model has obvious advantages in the prediction efficiency and accuracy, compared to traditional centralized way.