A radio transmitter fingerprinting system

Abstract

This thesis is concerned with an approach to record, process, and classify radio transmitter transients. A radio transmitter transient is emitted by the transmitter when the transmitter's push-to-talk button is depressed. This action engages the transmitter's frequency synthesizer which generates the carrier frequency. The generation of the carrier frequency by the frequency synthesizer is not instantaneous, thus a transient behaviour is exhibited during the carrier frequency acquisition. The capturing of such transient events is achieved by recording the discriminator output of an ICOM R7100 communication receiver. The recording is performed by a Sound Blaster sound card at a sampling rate of 44,100 samples per second and 16 bits per sample accuracy. The recording contains a noise component followed by a transient. The transient is separated from noise by a variance fractal dimension trajectory analysis. Once the transient has been localized, multiresolution wavelet analysis and genetic algorithms select the critical features of the transient used to classify the transient. Multiresolution analysis provides a set of wavelet coefficients that represents the transient features independently, allowing the genetic algorithm to select those that are most critical. The features are classified by a multilayer neural network with 64 inputs and 12 hidden neurons. The average classification rate achieved is 96% for an experimental set of 6 transmitters consisting of four Kenwood transmitters and two Yaesu transmitters. For this experimental transmitter set, the results show that this system classifies transients generated by transmitters of the same manufacturer and model, as well as transients generated by transmitters from different manufacturers.