Modelling of chaotic time series using a variable-length windowing approach

A multidimensional feature extraction method for chaotic time series is presented. The method uses a variable-length windowing approach. Mackey–Glass delay-difference equation is used to create chaotic sample signals. Among many possible alternatives, the length of the data segment having the smallest variance fractal dimension (VFD) value is found and used as the feature value. Multidimensional feature vectors are formed to model each sample signal. A probabilistic neural network (PNN) is trained and tested with these vectors. It is shown that the application of the new feature extraction method improves the classification performance of the PNN as compared to a VFD based feature extraction method using a fix-length windowing approach.

[1]  W. Kinsner,et al.  Multifractal modelling of radio transmitter transients for classification , 1997, IEEE WESCANEX 97 Communications, Power and Computing. Conference Proceedings.

[2]  Iwao Sasase,et al.  Weak radar signal detection based on wavelet transform , 1994, Proceedings of ICASSP '94. IEEE International Conference on Acoustics, Speech and Signal Processing.

[3]  Schuster,et al.  Generalized dimensions and entropies from a measured time series. , 1987, Physical review. A, General physics.

[4]  Jack Sklansky,et al.  An overview of mapping techniques for exploratory pattern analysis , 1988, Pattern Recognit..

[5]  O. H. Tekbas,et al.  An experimental performance evaluation of a novel radio-transmitter identification system under diverse environmental conditions , 2004, Canadian Journal of Electrical and Computer Engineering.

[6]  W. Kinsner,et al.  Speech segmentation by variance fractal dimension , 1994, 1994 Proceedings of Canadian Conference on Electrical and Computer Engineering.

[7]  John C. Platt.  A Resource-Allocating Network for Function Interpolation , 1991, Neural Computation.

[8]  Donald F. Specht,et al.  Probabilistic neural networks , 1990, Neural Networks.