Virgo Logbook

Here continues from the previous entry.

This kind of analysis is very useful to understand how much of the noise non-stationarities can be explained by other uncorrelated channels, but unfortunately it is quite hard to extract information about which are the auxiliary channels more relevant. To obtain this information I tried a slightly different approach: I take the BRMS and perform a linear regression with the auxiliary channels one by one. I choose the one which fit best, or in other words the one that gives lower standard deviation of the residuals. Then I subtract this contribution from the BRMS and repeat again. In this way I am not able to find the best possible linear combination, but I can obtain a list of relevant channels reasonably sort in order of importance.

In the reports at the address

http://wwwcascina.virgo.infn.it/DataAnalysis/Noise/MultiLinearNonStat/
the result of this analysis is shown in the third plot: in red the real BRMS, in green the one predicted using 30 channels and in blue the one using only the first 5 channels. The accuracy of the prediction is not as good as the one obtained with the full multidimensional analysis, but at least the slower trends are well reproduced, even using only 5 channels.

The table that follows this plot gives the list of channels in order of importance: therefore the blue curve is obtained using the first five of them. The meaning of the columns is: you take the signal written in the first column and you call it X. You normalize it as shown in the last column and then you multiply by the weight. This gives one addendum to the BRMS estimate.

Trying to summarize which are the more relevant channels, I can say that
- bands with frequencies above 1 kHz are modulated often like B1p_DC, q1ph / v, IB_tx / ty RMS, Oo_B5_q2_y. Other important channels are alpha and beta (which here are only indicators of some other effect, since they should not affect noise at so high frequency), also some temperature, like NE building ans NI suspension and the vertical position of NI and WI
- the noise in the mysterious band (100-1000 Hz) is often modulated by NI/WI/NE vertical position, IB_tx / ty / tz RMS, q81v, NE_tx / ty RMS. Interesting dependencies also with beta, MICH and PRCL gains, injection and MC suspension temperature, q51v, q1pv, seismic activity
- the noise below 100 Hz is minly modulated like seismic activity, vertical position of OB and NE, MC_tx RMS, beta, alpha

This list is for sure a bit confusing, however I hope the report pages can be more useful to understand how the noise in different band varies.

The most interesting point in my opinion is that this analysis allows to reproduce very well the non-stationarity of the mystery noise and might be a step forward in understanding it.