Autonomous non-linear classification of LPI radar signal modulations

Abstract

In this thesis, an autonomous feature extraction algorithm for classification of Low Probability of Intercept (LPI) radar modulations is investigated. A software engineering architecture that allows a full investigation of various preprocessing algorithms and classification techniques is applied to a database of important LPI radar waveform modulations including Frequency Modulation Continuous Waveform (FMCW), Phase Shift Keying (PSK), Frequency Shift Keying (FSK) and combined PSK and FSK. The architecture uses time-frequency detection techniques to identify the parameters of the modulation. These include the Wigner-Ville distribution, the Choi-Williams distribution and quadrature mirror filtering. Autonomous time-frequency image cropping algorithm is followed by a feature extraction algorithm based on principal components analysis. Classification networks include the multilayer perceptron, the radial basis function and the probabilistic neural networks. Lastly, using image processing techniques on images obtained by the Wigner-Ville distribution and the Choi-Williams distribution, two autonomous extraction algorithms are investigated to derive the significant modulation parameters of polyphase coded LPI radar waveform modulations.