Multidimensional Classification of Radar Signals : A comparison between unidimensional and multidimensional classification models for pulsed radar signals

Abstract: Radar is a technique used by many different types of remote sensing systems to keep track of their surroundings. The transmitted radar signals may carry information that could be used to infer the type of transmitter. Multiple papers have investigated the classification of pulse repetition intervals produced by radar systems. The aim of this thesis is to investigate whether multidimensional machine learning models can provide an advantage in the classification of radar signals. In this thesis, we compare the accuracy and performance between unidimensional and multidimensional generative classifiers. To facilitate this, we have 94 time series sequences of radar signal data, each representing a unique type of radar. The time series sequences are split up into a training set and a test set. The classifiers consists of 94 Bayesian Gaussian Mixture Models that are trained on one class each from the training set. The classifiers are then evaluated by their ability to classify the test set. Simulation results show that the multidimensional classifiers yield a significant increase in accuracy, albeit with slower inference and training times.