Unsupervised Learning for Detection of Radio Frequency Interference Signals

In order to ensure that wireless communication systems operate at their best, it's crucial to develop reliable methods for detecting RF interference. This study compares two unsupervised anomaly detection algorithms, Support Vector Data Description (SVDD) and One-Class Support Vector Machine (SVM), using a small training database. We evaluated these algorithms by detecting various types of RF interference signals, including Continuous Interference (CI), Continuous Wave Interference (CWI), and Modulated Continuous Wave Interference (MCWI). While previous works have made strides in RF interference detection, they have some limitations. For example, some methods may struggle with accurately detecting RF interference in dynamic environments or when faced with varying interference patterns. Additionally, certain algorithms may exhibit reduced performance when dealing with low signal-to-noise ratios or in the presence of complex interference sources. To address these challenges, both SVDD and One-Class SVM (OCS) algorithms are used and described the methodology for evaluating their performance. Throughout the experiments, how well SVDD and OCS could identify abnormal interference signals are measured using the precision metric. The findings revealed that employing normalization techniques like feature scaling over a range or robust normalization significantly improved the accuracy of RF interference detection. Importantly, SVDD consistently outperformed OCS in this regard.