Broadband source localization using an adaptive technique

The first step is a frequency dependent ARMA modeling of the signals coming from an array of sensors in presence of additive white noise. The purpose is to estimate an ARMA model at a single frequency f₀ which takes into account the information available in the whole frequency band. The idea is then to exploit the frequency variation of the ARMA models. This frequency evolution appears in the state space equation of a Kalman filter indexed by frequency where the state vectors are the ARMA models. The observation equation at each frequency is given by the ARMA modeling of signals. Filtering is followed by smoothing so that the model at frequency f₀ integrates the information of all the frequency band