Time domain classification of transient radio frequency interference

Short, transient radio-frequency interference (RFI) events could threaten the quality of astronomical observations made by new and planned radio telescopes such as MeerKAT, the SKA and HERA in the radio quiet reserve in South Africa. Because they are so short, often of the order of microseconds long, these events are difficult to detect and identify in the time-frequency plots typically produced by RFI monitoring systems. In this paper, we record and analyse a dataset of the time domain RFI signals of nine typical transient RFI sources. We show that it is possible to classify such transient signals in the time domain according to their source using Principal Components Analysis (PCA) and Kernel PCA. Using an adapted measure of cluster separation, we show that Kernel PCA is significantly better than standard PCA at distinguishing transient RFI sources from one another.