A Method Based on Frequent Pattern Mining to Predict Spectral Availability of HF

The HF radio communication has long been a big problem in channel selection since the spectrum environment is dynamic. To verify the feasibility of detecting idle channels by spectrum prediction, the data in this paper are based on real-world measurements collected by USRP in different time periods. The received signal power is converted to continuous sequences through a new channel state model reflecting spectrum availability. We then develop a prediction algorithm using simplified frequent pattern mining which can predict channel availability based on past channel states with considerable accuracy. The experimental results show that the measured data are more fluctuant in the afternoon which increase the predicted difficulty, nevertheless, the proposed algorithm is superior to neural network and Markov model in this situation, and the larger samples the better prediction performance.