Reports until 19:28, Wednesday 24 February 2016
H1 ISC (DetChar, ISC)
sheila.dwyer@LIGO.ORG - posted 19:28, Wednesday 24 February 2016 - last comment - 15:39, Sunday 28 February 2016(25703)
Moved OMC dither frequency by 0.21 Hz, 16 Hz comb on PZT1

Today Keita and I spent some time thinking about OMC length noise, there will be an update coming soon with more information and a noise projection.  

We spent some time looking at some nonlinear behavoir noise in the drive to PZT1.  Our dither frequency is 4100 Hz, and looking at the low voltage PZT monitor we can see a small 8 Hz and a larger 16 Hz comb.  There is also other non stationary noise in the monitor, and a broad peak at 12.7 kHz.  We have moved the dither line frequency to 4100.21 Hz, so if this was the cause of the 16 Hz comb in DARM we would now expect it to be more like a 16.84 Hz comb.  Evan Goetz tells us that we need 15 minutes or more of data in low noise to evaluate if this has changed any combs in DARM. 

We have just reached nominal low noise at 3:14:34 UTC Feb 25th, although the low frequency noise (below 50 Hz) is worse than normal.  I've temporarily changed the dither frequency in the OMC guardian, so if there is a longer lock later tonight it should also have this changed dither frequency.    (If anyone wants to double check what the dither frequency is, the channel is   H1:OMC-LSC_OSC_FREQ

 

 

Comments related to this report
keith.riles@LIGO.ORG - 10:59, Thursday 25 February 2016 (25717)DetChar
To quote Bill Murray in Groundhog Day, "Anything different is good." (at least in this context)

The 16-Hz comb does indeed appear to have changed into a 16.84-Hz comb. 
A DARM spectrum from two hours (400-sec coherence time) last night in the 150-250 Hz band is
attached, along with one from January on a day when the 16-Hz comb
was strong. New lines are seen in this band at

151.56 Hz
168.40 Hz
185.24 Hz
202.08 Hz
218.92 Hz
235.76 Hz

Some 1-Hz zooms are shown for a couple of the new lines.

So...can we fix this problem?
Images attached to this comment
sheila.dwyer@LIGO.ORG - 20:36, Thursday 25 February 2016 (25739)DetChar, ISC

Keita, Sheila

So that we can keep the OMC dither small while driving a reasonable level of counts out of the DAC, we added a voltage divider (somewhat creatively built) to the D-sub from the DAC to the driver chassis.  This is a 11k/110 Ohm divider on pins 1 +6.  We have increased the dither amplitude from 6 cnts to 600 cnts, so the round off errors will now be 100 times smaller compared to our signal.  

The attached screenshots show the PZT1 AC monitor before and after this change.  The lines below 1 kHz are always there, (even when there is 0 coming out of the DAC) and are not present on the analog signal coming into the driver chassis for the monitor.  

We have reverted the frequency to 4100 Hz.  If we get a long enough low noise lock tonight we can hope that the 16 Hz comb will be better.  If things look good we should upgrade our voltage divider. 

Images attached to this comment
keita.kawabe@LIGO.ORG - 09:43, Friday 26 February 2016 (25741)

Thing is, our dither line used to be 12 counts pk-pk, so the rounding error was actually significant (signal/error ratio is something like 10 in RMS), and the error showed up as lots of lines because we're sending in only one sinusoidal signal. These lines actually drive the PZT length.

Making the dither bigger, the round off error RMS doesn't change much so RIN will become smaller.

We inserted two sets of 11k-110 Ohm resistive divider, one each for positive and negative input of the low voltage pzt driver input because it was easy.  This is a temporary non-solution. A permanent solution is TBD.

The first attachment shows the spectrum of the DAC IOP channel for the dither, i.e. the very last stage of the digital, before we increased the amplitude. RMS of the forest of lines is about a factor of 10 below the RMS of the dither.

The second plot is after increasing the amplitude by a factor of 100, the rounding error RMS is still at the same level though you cannot tell from the plot, the dither to error RMS ratio should be more like 1000 now.

Three large lines in the second plot are not round off errors but imaging peaks that were previously buried in the round off errors: 12283.8kHz=16384-4100.2Hz, 20484.2Hz=16384+4100.2Hz, and 28667.8=32768-4100.2Hz.

Images attached to this comment
sheila.dwyer@LIGO.ORG - 23:48, Thursday 25 February 2016 (25747)

Actually, the dither line is still at 4100.21 Hz for tonight, (I had forgotten that I put this into the guardian).  We will revert it tomorow.

keith.riles@LIGO.ORG - 15:13, Friday 26 February 2016 (25758)
Splendid -- no 16-Hz or 16.84-Hz comb seen in a 2-hour stretch from last night!

See attached spectrum 150-250 Hz spectrum for comparison
with above plots, along with 0-1000 Hz spectra from last night
and from the night before, with the 16.84-comb present.

Many thanks from the CW group as we look ahead to O2.

Images attached to this comment
sheila.dwyer@LIGO.ORG - 17:22, Friday 26 February 2016 (25761)
Images attached to this comment
keith.riles@LIGO.ORG - 15:39, Sunday 28 February 2016 (25771)
Sheila pointed out that the noise floor for the 2nd spectrum
is considerably higher than for the 1st and could be hiding
residual lines due to the dither. So I tried shifting the 2-hour 
time window a half hour earlier, to avoid the hellacious glitch
seen in the inspiral range (see 1st figure) near the end of
the original Feb 26 interval. The 2nd figure shows the resulting
spectrum with a noise floor closer to that on Feb 25. The
16.84-Hz comb still does not appear. 

So I think it's safe to say that the 100 times multiplication / divide
trick did indeed suppress the 16.84 Hz (originally 16 Hz) comb a
great deal, but of course, we will need long coherence times and
long integrations to see if what's left causes residual trouble for CW searches.

Images attached to this comment