Automation of Pulse Repetition Interval Modulation Classification

Sammanfattning: Radar systems are essential in the field of electronic warfare and radar signals are used for estimating speed, location and direction of an object. A radar system sends out small transmissions called pulses. The interval between the start of one pulse to the start of the next pulse is called pulse repetition interval (PRI). Each pulse series is called a pulse train and it can be viewed as a continuous time series of data. The pattern that defines the PRI is often referred to as the PRI modulation. PRI is an important parameter because the functional purpose of the radar can in many cases be identified if someone manages to identify the modulation of the PRI. An incorrect guess of the PRI modulation can lead to an inaccurate guess of who is emitting the signal, i.e., assuming an enemy is a friend. This can in the worst case be a fatal consequence. This thesis investigates if it is possible to semi-automate the manual process of classifying PRI images using machine learning methods. The results indicate that the proposed convolutional neural network (CNN) performs better than the manual classification across all tested noise levels, and the CNN also speeds up the process of classifying PRI images. The model can be used as a tool for operators when they want to classify different PRI images, thus making the process semi-automated. Future work includes to investigate if it is possible to add a third dimension to the images. This dimension would be added as a time interval or to use colors in the existing images used in this thesis to distinguish the number of PRI’s that has the same value. Moreover, a valuable addition to the work would be to use real data in the proposed CNN instead of the synthetically generated data that was used in this thesis.