We are making on-site experiments of a new source-receiver system for seismic exploration, named ACROSS (Accurately Controlled Routinely Operated Signal System). The system uses precisely controlled sinusoids, which gives a best way to achieve high signal-to-noise ratio without destroying surrounding ground. The system is deployed near Nojima fault which ruptured during 1995 Kobe earthquake to monitor the temporal variation of the coupling state. We also check the temporal stability of the system to establish a practical technique for continuous monitoring of seismic velocity. In our experimental site the sources which generate 20ton-f with centrifugal force are deployed on the ground surface where granite bedrock is exposed. The emitted elastic wave is received with two borehole seismometers deployed at 800m and 1800m deep near the fracture zones of Nojima fault. We made many runs during the last 4 years, in which the longest lasted about one month. During each run the sources were continuously operated with frequency modulation around 20Hz and 30Hz producing plural components of sinusoids with constant frequency interval. In the earlier experiments we found that the elastic signal emitted from the source is strongly varies with time although the force of the source is very precisely controlled. Surface and shallow borehole seismometers are installed just around the ACROSS source to monitor and to reduce the effect of the temporal variation of the signal. We obtained the temporal variation of travel time using the waveform which are transformed from the data in frequency domain. The results of these experiments show that the travel time of P phase varies with time as large as 1 millisecond per month in nearly the same trend for 800, 1600 and 1700m borehole sensors. This indicates that such a large variation can be attributed to the variation of the rock shallower than 800m. Subtracting the travel time variation at 800m from those of 1600m and 1700, we can get the variation between the shallower and the deeper two borehole sensors. Although the variation is still more than 100 microseconds during the one-month experiment, showing the present