Summary
Q: What is the relationship between the strength of violin mode ring-ups and the number of narrow spectral artifacts around the violin modes? Is there a clear cut-off at which the contamination begins?
A: The answer depends on the time period analyzed. There was an unusual time period spanning from mid-June 2023 through (very approximately) August 2023. During this time period, the number lines during ring-ups was much greater than in the rest of O4, and the appearance of the contamination may have begun at lower violin mode amplitudes.
What to keep in mind when looking at the plots.
1. These plots use the Fscan line count in a 200-Hz band around each violin mode region, which is a pretty rough metric, and not good for picking up small variations in the line count. It's the best we've got at the moment, and it can show big-picture changes. But on some days, contamination is present, but only in the form of ~10 narrow lines symmetrically arranged around a high violin mode peak. (Example in the last figure, fig 7) This small jump in the line count may not show up above the usual fluctuations. However, in aggregate (over all of O4) this phenomenon does become an issue for CW data quality. These "slight contamination" cases are also particularly important for answering the question "at what violin mode amplitude does the contamination just start to emerge?" In short, we shouldn't put too much faith in this method for locating a cut-off problematic violin mode height.
2. The violin modes may not be the only factor in play, so we shouldn't necessarily expect a very clear trend. For example, consider alog 79825 . This alog showed that at least some of the contamination lines are violin mode + calibration line intermodulations. Some of them (the weaker ones) disappeared below the rest of the noise when the violin mode amplitude decreased. Others (the stronger ones) remained visible at reduced amplitude. Both clusters vanished when the temporary calibration lines were off. If we asked the question "How high do the violin modes need to be...?" using just these two clusters, we'd get different apparent answers depending on (a) which cluster we chose to track (weak or strong), and (b) which time period we selected (calibration lines on or off). This is because at least some of the contamination is dependent on the presence & strength of a second line, not a violin mode.
Looking at the data
First, let's take a look at a simple scatter plot of the violin mode height vs the number of lines identified. This is figure 1. It's essentially an updated version of the scatter plots in alog 71501. It looks like there's a change around 1e-39 on the horizontal axis (which corresponds to peak violin mode height).
However, when we add color-coding by date (figure 2), new features can be seen. There's a shift at the left side of the plot, and an unusual group of high-line-count points in early O4.
The shift at the left side of the plot is likely due to an unrelated data quality issue: combs in the band of interest. In particular, the 9.5 Hz comb, which was identified and removed mid O4, contributes to the line count. Once we subtract out the number of lines which were identified as being part of a comb, this shift disappears (figure 3).
With the distracting factor of comb counts removed, we still need to understand the high-line-count time period. This is more interesting. I've broken the data down into three epochs: start of O4 - June 21, 2023 (figure 4); June 21, 2023 - Sept 1 2023 (figure 5); and Sept 1 2023 - present (figure 6). As shown in the plots, the middle epoch seems notably different from the others.
These dates are highly approximate. The violin mode ring-ups are intermittent, so it's not possible to pinpoint the changes sharply. The Sept 1 date is just the month boundary that seemed to best differentiate between the unusual time period and the rest of O4. The June 21 date is somewhat less arbitrary; it's the date on which the input power was brought back to 60W (alog 70648), which seems a bit suspicious. Note that, with this data set, I can't actually differentiate between a change on June 21 and a change (say) on June 15th, so please don't be misled by the specificity of the selected boundary.
Kiet, Sheila
We recently started looking into the whether nonlinearity of the ADC can contribute to this by looking at the ADC range that we were using in O4a.
They are showed in the H1:OMC-DCPD_A_WINDOW_{MAX,MIN} that sum the 4 DC photodiodes (DCPD). They are 18 bits DCPD, so that channel should saturate at 4* 2^17 ~520,000 counts.
Now there are instances that agree with Ansel report when there are violin mode ring up that we can see a shift in the count baseline.
Jun 29 - Jun 30, 2023 when the baseline seems to shift up and stay there for >1 months, Detchar summary page show significant higher violin mode ring up in the usual 500-520Hz region as well as the nearby region (480-500 Hz)
Oct 9, 2023 is when the temporary calibration lines are turned off 72096, the down shift happened right after the lines are off (after 16:40 UTC)
During this period, we were using a~5% of the ADC range (difference between max and min channel divided by the total range - 500,000 to 500,000 counts), and it went down to ~2.5 % once the shift happenned on Oct 9, 2023. We want to do something similar with Livingston, using the L1:IOP-LSC0_SAT_CHECK_DCPD_{A,B}_{MAX,MIN} channels to see the ADC range and the typical count values of those channels.
Another thing for us to maybe take a closer look is the baseline count value increase around May 03 2023. There was a change to the DCPC total photocurrent during that time (69358). Maybe worth checking if there is violin mode contaimination during the period before that.
Kiet, Sheila
More updates related to the ADC range investigation:
- ADC ranges comparison between Hanford and Livingston:
- At Livingston we are using the L1:IOP-LSC0_SAT_CHECK_DCPD_{A,B}_{MAX,MIN} channels, which were not turned on until 1398035799; these channels saturate at 2^17 counts.
- At Hanford we are using the H1:OMC-DCPD_A_WINDOW_{MAX, MIN}, these channels saturate at 4 * 2^17 counts as they are sum over 4 DCPD
- We looked through the data in Feb 2025 when the violin modes at Livingston were somewhat higher than usual.
- The range being used in compatible with Hanford (2-4%), and the count values are also similar as LHO counts/4 and LLO counts are in +- 5000 counts of each others.
- Comparing the comtaimination before the DARM offset change in ER15:
- We saw hint of contaimination even before the change (Using Fscan spectrum of May 2nd)
Further points + investigations:
- Ansel pointed out that it was odd to have a significant shift in the baseline count value + range when the temporary calibration lines turned off as these calibration lines were not that different in height than other calibration lines (see plot in the Alog 83997)
- Joseph from LLO gave us a spectrum comparison between H1, L1 raw ADC count, there is a notable difference in the higher order violin modes
- To do: looking to periods of when the both (1st and 2nd) violin modes ring up and periods of only the first violin mode ring up to see the contaimination caused by the 2nd mode or higher down mixing
- Evan pointed out that during the period of high contaimination (June 30th - Aug 9th; 2023), the range stayed between 7 - 20%; and LHO in general seemed to have higher rate of saturation + intermitten increase of the ADC range than LLO.
- To do: Selecting the periods of stable ADC range in LHO data, and run the average spectrum over those periods to see the level of contaimination, assessing the contribution of the periods with increase ADC range.
- To do: Selecting the periods of stable ADC range in LHO data, and run the average spectrum over those periods to see the level of contaimination, assessing the contribution of the periods with increase ADC range.
Kiet, Sheila
Following up on the investigation into potential intermixing between higher-order violin modes down to the ~500 Hz region:
The Fscan team compiled a detailed summary of the daily maximum peak height (log10 of peak height above noise in the first violin mode region) for the violin modes near 500 Hz (v1) and 1000 Hz (v2). They also tracked line counts in the corresponding frequency bands: 400–600 Hz for v1 and 900–1000 Hz for v2. This data is available in the Google spreadsheet (LIGO credentials required).
- We identified dates when both violin modes were elevated (n1_height > 7; n2_height > 8) and when only the fundamental mode was elevated (n1_height > 7; n2_height < 8). For each case, we computed average PSDs using an FFT length of 1800 s. The study period spans from August 10, 2023, to January 14, 2025, starting when ADC counts stabilized after the temporary calibration lines at 24.4 and 24.5 Hz were turned off (see alog 72096)
- The psds comparisons is shown in vmodes_psds_comparison.png.
- Note that the number of averages differs between the cases; there are significantly fewer days with only v1 elevated, which explains why the [v1 high, v2 low] spectrum appears noisier in some regions. However, similar features are still present in the [v1 high, v2 high] case.
- Notably, there appears to be more spectral content in the 450–550 Hz range when both modes are elevated, with certain lines showing significant power (highlighted in green).
- The psds comparisons is shown in vmodes_psds_comparison.png.
- Daily Fscan data around the violin modes is summarized in Pairwise_scatter_plots.png , where
n1_heightandn2_heightare the max peak heights of v1 and v2, andn1_countandn2_countare the corresponding line counts. There appears to be a threshold in violin mode amplitude beyond which line counts increase (based on {n1_height, n2_height} vs. {n1_count, n2_count} trends). - We also ploted how n1_count varies with n2_height when n1_height is high in n1_count_vs_n2_height_when_v1_high.png.
Next: We plan to further investigate the lines that appear when both modes are high, the goal is to identify possible intermodulation products using the recorded peak frequencies of the violin modes.
