Statistical classification of radio frequency interference (RFI) in a radio astronomy environment

Abstract

© 2016 IEEE. We present the application of statistical classifiers to the problem of automatic identification of radio frequency interference (RFI) in radio astronomy. RFI can corrupt measurements made by radio telescopes and it is therefore very important that such interference can be identified. We compile a dataset of RFI signals gathered at the SKA site near Carnavon, South Africa, and use this data to train and evaluate some statistical classifiers. We find the best performing system to use the k-nearest-neighbour (knn) classifier and achieve an accuracy of 93%. Since our dataset was limited by the capturing equipment in terms of record length, we feel that there is scope to improve on this figure in the future.