Identifying radio frequency interference with hidden Markov models

Radio frequency interference (RFI) is a significant concern for radio astronomy. Identifying unintentional RFI signals (for example, from equipment operating in the vicinity of radio telescopes) is a challenging topic due to the highly non-ergodic nature of such signals. Another non-ergodic signal type which has been very well researched is human speech, for which hidden Markov model-based approaches have led to some of the best performing classification algorithms. Inspired by this, in this work, we propose the use of HMMs to identify transient RFI events. We train HMMs to distinguish between the sources of several different types of RFI in a previously recorded dataset. We demonstrate that basic HMMs can be used to classify different RFI events according to their sources in the time-domain, providing useful levels of accuracy.