Application of Machine Learning Techniques to HF Propagation Prediction

The High Frequency (HF) radio frequency band, from 3-30 MHz, is an important military communications medium as it can provide Beyond Line-of-Sight communications connectivity without the use of relays or intermediate equipment. Regarding radio signal propagation in the HF band, it is well known that the best choice of carrier frequency to connect any two points on earth at a given time depends on many factors including time of day, time of year, year of the 11-year solar cycle, local noise and interference, Earth's geomagnetic activity, and solar weather. The current state of the art includes 1) HF propagation programs, which generate Signal-to-Noise Ratio (SNR) predictions at the radio receiver, based on long term monthly median data, as well as 2) near real time propagation data gleaned from active HF signaling. This paper applies Machine Learning (ML) to these two different approaches. First, a deep learning model is trained on the predictions of the Voice of America Coverage Analysis Program (VOACAP). A similar model is then trained on near real time propagation prediction SNR reports, that combine Weak Signal Propagation Reporter (WSPR)-like data with additional observed parameters from the National Oceanic and Atmospheric Administration (NOAA) Space Weather Prediction Center (SWPC). Finally, the resulting predictions are combined with Automatic Link Establishment (ALE) and potentially other propagation tools, to provide a hybrid prediction model that leverages the advantages of each technique.