Detection of Multipath Propagation Interference in Pulsed Radar Signals in a Non-Coherent Receiver with Convolutional Neural Network Models

Sammanfattning

This thesis researches if convolutional neural network models can be used to detect and classify multipath signal propagation in radar signals. Specifically, a convolutional neural network model is used to detect multipath propagation in a simulated frequency modulated pulsed radar signal. A combined determin- istic and stochastic signal propagation software model is used to generate a labeled data set that emulates pulses detected by radar warning system located on a vessel in a marine environment. The simulated data set is created and quantified with MATLAB and is represented in the form of in- phase and quadrature signals. The model is trained and evaluated on data that represent no multipath signal propagation, multipath signal propagation, and non-multipath interference (two separate emitters that interfere without multipath) in two separate scenarios for a wide range of signal-to-noise ratios. The machine learning model has an average accuracy of 91.78% on the test set and produces robust results for high signal-to-noise environments for the multipath classes.