A probability-density-based autoregressive model using support vector method and higher-order spectra estimation

When the characteristics of non-Gaussian time series such as biological signals are described, not only power spectra but also higher-order spectra are required. In order to obtain estimated values with few statistical fluctuations, some parametric estimation method has to be established. As the parametric model, regression-function-based autoregressive models Such as the neural network autoregressive model have been Studied so far. On the other hand, probability-density-based autoregressive models in which the correlation information of the time series is represented by the conditional probability density function have been proposed. However, in the existing probability-density-based autoregressive models, higher-order spectral estimation is not assumed. So, if we adopt the probability-density-based autoregressive model for the estimation of higher-order spectra, some problems such as the stationarity arise. In this paper, we proposed a new probability-density-based autoregressive model using the support vector method. Further, we estimated higher-order spectra of the time series by the proposed model. (C) 2006 Wiley Periodicals, Inc.