Multifractal modelling of radio transmitter transients for classification

This paper presents a method of extracting and modelling radio transmitter transients for classification. It is motivated by the real possibility of identifying radio transmitters used in violation of federal and international regulations. A system has been developed which takes the beginning of a radio transmission and separates it from ambient channel noise using a multifractal segmentation technique. Then, significant features are extracted from the transient and a more compact multifractal model is obtained. Finally, this model is analysed by a neural network for classification. Preliminary results indicate that classification using multifractal models is feasible. More specifically, a probabilistic neural network has been trained using 160 out of 415 available transients. Testing the system with the remaining 255 transients produced results in which 92.5% of the transients were classified correctly.