Detection of signals corrupted by nonstationary random noise via Kalman filter-based stationarization approach

In this paper, a method of stationarization of nonstationary data is proposed in the signal detection problem. The signal to be detected is corrupted in a nonstationary random noise whose model is given by an ARMA(p, q) model. The time-varying coefficient parameters of the ARMA model are estimated by the Kalman filter. The stationalization of nonstationary observation data based on the estimated coefficient parameters leads us to the conventional binary hypothesis-testing for signals in stationary random noise.