A computationally efficient procedure was developed for the fitting of a locally stationary autoregressive model. The amount of computations is bounded by a function of the data length and the model order only and does not depend on the number of possible arrival times. This facilitates automatic determination of arrival time by an on-line system. The on-line system FUNIMAR (fast univariate case of minimum AIC method of AR model fitting) was developed to implement the procedure. The method was

more »

... hecked by applying it to weak seismic signals from earthquakes which were superimposed with background noises such as traffic noise, electronic hum noise, and heavy microtremor. Experimental results confirm the effectiveness of the method for determining the onset time of P-wave and suggest that an AR model of the fifth order is sufficient to determine the onset time of the wave even if it is buried in considerably heavy noises. It is also shown that the onset time of S-wave can also be successfully determined by the method.