Then DeepClean model is trained by the witness channel above from 1368982000 s to 1368983024 s (1024-second of science mode data from O4a). Then I used the witness channel to clean the noise in the band [110Hz, 135Hz] of K1 strain data from 1368983024 s to 1368987120 s (4096-second of science mode data from O4a).
The noise can be recognized and reduced and the inspiral range is improved by 0.001248% (Fig. 2).
Fig. 1 shows the coherence between the raw strain and the IMMT1 channel before and after cleaning.
Fig. 2 shows the difference of the coherence between the cleaned strain and the IMMT1 channel before and after cleaning.
The subtracted noise is bandpass filtered by [110Hz, 135Hz].
