Transient analysis and genetic algorithms for classification

This paper presents a method of generating unique fingerprints of radio transmitter turn-on transients. The fingerprinting system consists of the application of multiresolution wavelet analysis used to characterize the features contained in the transient followed by the use of a genetic algorithm to extract the wavelet coefficients that represent critical features of the transient. To measure the ability of the system to generate efficient and unique fingerprints, a neural network is used to classify the transients by their fingerprints. To test the noise sensitivity of the system, noisy transients were applied to a trained neural network, the network was able to positively classify noisy transients with 20 dB signal to noise ratios (SNR) and up. Experiments with real radio transients show that the system is able generate uniqiue fingerprints for absolute classification by a neural network for radios of differing model type as well as radios of the same model type.