Predicting HF Propagation with Machine Learning and DX Cluster Data

This presentation was given by Alan Spindel, AG4WK, at Hamvention 2025. The presenter shared the groundbreaking work of Robert Griffin, KC2JJM, a professional software developer and avid amateur radio experimenter based in Long Island, New York. Unable to attend in person, Griffin entrusted the presentation of his innovative project, which uses real-world DX cluster data and machine learning to predict HF propagation. His system collects over 30,000 DX spots per day, building a dataset of millions of entries. With Python and tools like pandas, Redis, piHAMtools, and systemd, he created a pipeline to clean, store, and process this vast amount of data, enabling high-speed analysis and real-time querying for radio propagation conditions. Griffin's approach integrates solar weather data and DX spots to train predictive models capable of identifying ideal operating bands and times. Using a high-core server and massive in-memory processing, the system allows near-instant retrieval of complex propagation patterns. His future roadmap includes global DX cluster integration, reverse beacon network data, enhanced machine learning models, and a conversational AI interface for natural language propagation queries. Additionally, Griffin is working on a real-time HF speech-to-text black box with noise reduction and speaker identification, to be integrated into upcoming Ten-Tec transceivers.