Fast ARMA modelling of power spectral density functions

A method for real-time estimation of ARMA models from power spectral density functions is proposed. It is based upon the combination of LPC methods to estimate both numerator and denominator polynomial of the transfer function and an iterative scheme to attribute the appropriate parts of the power spectral density function to pole or zero modelling of the transfer function. This approach yields an estimate to the data which approximately corresponds to a quadratic fit on a logarithmic scale. As further distinctive features we mention that the estimation is fast compared to more traditional methods and that it allows for frequency-warped modelling.