craig.cahillane@LIGO.ORG - posted 15:16, Wednesday 30 September 2020 - last comment - 13:04, Wednesday 26 May 2021(56942)
Animation of August 2020 correlated noise plot
Here is a visualization of the correlated noise averaging down over a nine-hour no-squeezing period taken on August 29th, 2020. Animation Blue is the mean-averaged correlated noise, vulnerable to glitches. Orange is the bias-corrected median-averaged correlated noise, robust to glitches. Green is the mean-averaged DARM. Red is the bias-corrected median-averaged DARM. The number of averages increases ~quadratically for every frame. Results What I was looking for was whether we had hit the correlated noise floor visually, or if it was changing throughout the lock. The correlated noise (orange line) appears relatively stable everywhere after about 5 seconds, or 15000 averages. This includes at low frequency, below 100 Hz, aka the "mystery noise" region. Above 1 kHz, the intensity noise is increasing slowly. At 1 kHz, the "correlated mystery noise" seems stable. The glitches appear to ring up the violin modes at 500 Hz. The 500 Hz violins lines start very thin, but spread wider, indicating spectral leakage from large lines. Other lines appear to start high and fall over the course of the lock. The last large glitch on the final frame may be from the lockloss. Details We have a total of 67156 averages at the end, corresponding to 33578 seconds: gps_start = 1282730359 # August 29 2020 09:59:01 gps_stop = 1282763937 # August 29 2020 19:18:39 averages = 67156 binwidth = 2 # Hz overlap = 0 Previous alogs This is related to the median-averaged correlated noise work in alog 56896 Sheila acquired the no-squeeze data in alog 56632
Non-image files attached to this report
Comments related to this report
Here is another version of the animation. This time we perform a sliding average of 15000 CSDs across the 67156 total CSDs, to try and capture changes over the nine-hour period. In total there are 100 frames in the animation. The first frame of the animation represents the result from CSDs 1 - 15000. The second frame represents CSDs 521 - 15521, the third 1043 - 16043, etc, to the last frame, 52155 - 67156. Each frame represents 7500 seconds of data. Most of the conclusions from the alog above remain: - Violin modes at 550 Hz and 850 Hz and at decay much faster, since this time the slider "slides away" from the CSDs with the highest lines there. - The modes at 500 Hz again grow worse, as evidence by the increased spectral leakage around there. - Intensity noise at 5 kHz still grows slightly worse - "correlated mystery noise" at 1 kHz still never changes, - Regular "mystery noise" below 100 Hz does change, ever so slightly, but definitely on the 10% level. Hover your mouse at 70 Hz or 50 Hz at the beginning of the animation and watch the noise breathe. Interesting that its changing at all, and with 15000 averages we can be sure that it isn't a fluke. 15000 averages is overkill for the correlated noise below 100 Hz, but required for the noise at 1 kHz. Could potentially get more temporal resolution by reducing the sliding number of averages, and ignoring the results at frequencies where the coherence is lower.
EDIT: Sheila alerted me that my DARM trace was a factor of sqrt(2) too high. Fixed animation now attached.
Non-image files attached to this comment
It's confusing upon reviewing this alog, but the Mean PSD and Median PSDs in this movie refer to the PSD of a single DCPD.
This is confusing and should not be conflated with the true DARM noise.
True DARM = <A + B, A + B> = <A,A> + <B,B> + 2 Re(<A,B>).
What is plotted here is just <A,A>.