Thursdays at 12:15 PST = 3:15 EST

Call in numbers:

Access Toll Free From USA	866 546 3377
Access Toll Free From Canada	866 546 3377
France	0800915011
United Kingdom	08-08 1013382
Japan	00531160959
Participant Passcode	353457#

Agenda for March 5th

• Reprocessing 8.2.1 noise data and averaging the PSD to look for tilted noise spectrum. I edited the plotting portion of the figure generation to average the PSD data into bins of 500 points. This does not affect the noise calculation or measurement, just the the plotting of the PSD. We have data for ~120 bolometers and about 10% show what I would call a positive tilt (e.g. page 12) and about 10% show a negative tilt (e.g. page 7). I have not tried to quantify the level of tilt for these bolometers. (Full data, compare to the full data below in Feb 25th agenda.)
• Nuller current in Tc measurements. Taken with amplitude = 0.1 and gain = 0. I calculate the current through the squid to be 0.45 uA. There should be significantly less current through the bolometer until it goes superconducting, at which point the current shouldn't heat the bolometer at all.
• Thoughts on parameter storage and use for autotuning. I've added some thoughts on storage of parameters in the HWM and how to retrieve an use them for tuning a large system on the Hardware Mapping wiki page here.

Agenda for Feb 25th

• Noise reprocessed with Tcs and G_ave. Full data, Summary. Using these numbers does not affect the results much.
• A couple things have come to light from taking a closer look at the Tc data. They are suspect for a few reasons:
  1. For a single motherboard there seems to be a strong correlation between bolometer channel and Tc. Except for a few exceptions, for each motherboard Tcs for a given bolometer channel are identical.
     • I have not come up with a model of how to make this happen. One possibility was that there was an error in the python code that writes the data and it was writing the same data for all four muxes on a motherboard. If this were the case then the current data would be identical between bolometers A1, B1, C1 and D1. We checked and found that though the data is very similar (as expected from the Tc analysis) they are not identical.
  2. I believe that we did not account for the significantly lower heat capacity between KT7 and 8.2.1. That means the rate of temperature change is faster than intended and the temperature measurements are less reliable. For our measurements we only looked at the transitions from cooling because the warming rate was way too fast. This would bias our temperatures low. Looking at the Temperature vs. time plot below we can see that the Tcs should all be above 440 mK and definitely below 510 mK. We see four peaks with four cooling slopes and three warming slopes. We used cooling slopes 1 and 2 for our Tc analysis. We saw warming slope 1, but it's temperature was very high because of the thermal lag. We did not see warming slopes 2 or 3, nor did we see cooling slopes 3 or 4. That means that peaks 3 and 4 are below Tc. It is not readily visible in the plot, but peak 3 is just about 440 mK. So, 440 mK is below Tc for all bolometers. The PID for the cryoboard is being rewritten to remove the issue of wafer holder heat capacity on future measurements.
  3. Bolometers show two types of behavior when going through transition. We expect the current vs. time plot to start with a constant level of ADC counts. Once transition occurs the current should drop. On page 2 we see bolometers that follow this general shape. There is a spike and some noodling around before the current drops to basically 0. We have chalked this up to some strange behavior as the bolometer goes through transition and not worried about it. However, on page 10 we see bolometers that start with roughly the same ADC level at the beginning, jump up to roughly the same level as the spike from page 2, but don't drop back down. They stay at the elevated level of ADC counts until the next transition. This would point to something going on that we don't really understand about our measurement and the transition.

Y axis [Temperature in K], X axis [time in seconds]

• Tcs will need to be retaken on the DC system.

• As an aside I've developed code that will take a list of frequencies and based on the expected average separation fill in the frequencies of missing bolometers. My thought is that this will be required for the mapping of NA frequencies to bolometers.

Agenda for Feb 18th

• Nulled IV curves for bolometers
  • I am happy to report that I was wrong when I said that we don't save the bolometer tuning data. We do, which means that we have nulled IV curves for all the bolometers that we have noise for. I am reworking the analysis code to handle the output. The units are uV and not ADC and also the data is noisier than a carrier only IV so the region near turnaround needs to be fit as a whole, we can't just look for the minimum point.
  • Nulled IVs from gentle bolometer tuning for noise.
    • The nulled IVs show P_turns 5-10 pW higher than the the original values, with the difference increasing for some high frequency channels. This may not be surprising as we can see that high frequency channels were not sufficiently overbiased, though I have not had a chance to compare carefully.
    • Nulled IVs also show more scatter on the points than the standard IVs.
  • Original IVs for reference.
• ETF data for responsivity
  • Data taken at turnaround and at 89% of turnaround (~80% R_ob).
  • If anyone would like/needs the raw data for analysis let me know.
• Calculate Gs for bolometers and use in noise analysis
  • Work on getting the nulled IV curves analyzed is important for this, but I have not made any other movement.

We open on Thursday.

Agenda for Feb 11th

• Noise
  • Noise comparisons (all noise here taken with 8.2.1):
    • Between OB noise and in transition noise: Posted here is the summary of the first OB noise (demod gain = 2). The peak at 0 comes from a bug where data from bolometer channel 2 doesn't get recorded properly. The squid for comb O has been identified as a bad actor. Comb B shows similar high noise levels while OB. Cross referencing those bolometer noise levels to the histogram revealed that ignoring those two combs removes all of the bolometers with noises at factors of 2-5. This supports our model that this data was taken with some bad actor squids as well as issues tuning bolometers. If you wish to see the full noise results they can be found here.
    • Between noise taken with 1 bolometer in transition and full comb (demod gain = 2): Posted here is the summary of the 1 detector in transition measurement (M5) and here is the summary for the full combs in transition (M6). There is some movement in the histogram, but the overall distribution is about the same. If anything the M6 noise clusters more closely to a factor of 2 too high. If you are interested here are the full results for the noise data from 1 bolo in transition (M5) and all bolometers in transition (M6).
    • Between noise taken by Peter Hyland and Francois Aubin:
      • The most interesting data that Francois took was with all bolometers at 0.9 R_ob and with demod gain = 2. The summary shows that there are a good number of bolometers with noise within 20% of the expected value and the vast majority of detectors and within a factor of 2. It also has some interesting plots of the noise vs. various parameters. In Francois' data there is a correlation between noise level and frequency. Both the summary and the full data have more combs than I had originally shown.
      • Francois' results point to some crucial difference between the procedure that he and I use. I'll be sitting down with him and going through our procedures to pinpoint the differences and see if I can figure out what it is in my procedure that is resulting in higher noise.
  • Noise taken with bad actors left over biased
    • Bad actors were selected from Francois' 0.9 R_ob data. Bolometers were tuned with the standard tuning script (turn off nuller, lower carrier, etc.) and I saw no difference between tuning down to 0.9 and 0.8 R_ob. The noise at 0.9 R_ob (full data) is clustered around a ratio of 1 (summary data), which backs up what Francois saw. However, the 0.8 data (full data) shows noise that had increased with most sitting just below a factor of 5 (summary data). Based on Ziggy's measurements of the relative achieved depth in transition for standard tuning and gentle tuning, this may not be surprising as we could be as low 0.72 R_ob or even lower. That could latch, or nearly latch, the bolometers causing the high noise.
  • Noise taken with bolos tuned with 'gentle' tuning
    • Worked with a single comb, D, that looked good to limit influence from bad squids etc. Took noise at turnaround (full data, summary) and at 0.86 R_ob (full data, summary). Bolometers tuned using the gentle tuning script. Bolometers were overbiased between measurements. Noise found to be nominal at both depths in transition. ETF data was taken at both depths (turnaround, 86%) as well. Bolometers show a 40% increase in speed.
    • Noise from the entire wafer at 0.83 R_ob (Full data, Summary). Noise is distributed around 130% of expected level, with some outliers. This seems to show that the gentle tuning reduces the noise, which naively would point to the standard tuning dropping the bolometers significantly farther into transition.
    • Ziggy already posted some data that shows that the regular tuning drops the bolometer by ~7% more than was desired.
      • Ziggy plans on taking more of this data when the black dewar is free to see just how much lower the high frequency channels get.
  • Taking ETF at 0.8 and 0.9 show a factor of 2-3 in the speed of the bolometers.
• ETF data out past 5 kHz.
  • PB_8.2.1_etf_Normal.pdf: This is data taken with a modified ETF script. It was done on comb D, with the stage warm and is closer to network analysis. I read back the amplitude of the helper, not a side band, and ran frequencies out to 35000 away from the carrier. That number was chosen as half the separation between the two closest peaks in the NA. The result shows the roll off from the L/C.
• Tcs: I'm working with a graduate student (James) to automate the Tc determination and creating a convenient output. Checking the data by hand though shows that most Tcs seem to lie in a range between 430 - 445 mK.
  • Tcs for the entire wafer: Wedge_821_Tc.pdf. The first page is a histogram of the Tcs. You can see from the plots of the data that we had a glitch that shifted the temperature around in an unphysical way. We weren't able to automate the Tc determination because of that, but James was able to automate the rest of the pdf production.
  • Here is an Open Office spreadsheet of the Tcs as well: Tc_table_821.ods.

Summary

• Going through our procedures together Francois and do not find any major differences in our procedures. We also checked that the parsers we use (Francois uses an EBEX parser) produce the same noise level when streaming data. We came to the conclusion that our noise levels seem consistent. We did not make an effort to exactly replicate each other's conditions (this was a product of not wanting to influence how each made their noise measurements.) as a result we do not have exactly comparable data. The closest data is the overbiased data and those noises look consistent. Also, Francois did not take bolometers as deep into the transition as I did on my first noise measurement. It is quite likely that I had bolometers that were latching at that point (Francois saw latches at 90% and I was a 80% R_ob). This may be a simple reason for the higher noise. Later noise measurements by me were made using the gentle tuning which seems to be more stable. I saw no overt latched bolometers for the data I took at 0.83 R_ob, however I took some data with just comb D and all other squids turned off after the noise data. At that time I just turned off the squids and carriers etc. and did not overbias the bolometers, which is usually the point where a quietly latched bolometer will make itself known. All the data seems to be consistent with the standard bolometer tuning script 1) dropping bolometers deeper into the transition when a nuller is applied after tuning and 2) dropping high frequency channels lower than intended, producing higher noise. The takeaway is that bolometer tuning should be done with the gentle tuning. Noise data from standard tuning shows a correlation with frequency (high frequency channels showing higher noise), while gentle tuned data shows the same noise levels for all frequencies. In this process we have also found that some squids resonate badly contaminating other squids on the squid board with high noise. The current assumption is that these squids are what lead to the ADC overloads seen on the demodulators. This could also skew our data to higher noise, but it seems to have less of an effect than feared.
• Tc data show Tcs ~437 mK
• A modified ETF script show rolloff from the L/C.

Questions:

• Are higher noise channels really deeper in the transition? - Tune with gentle tuning and check again. Preliminary result from gentle tuning noise is that when tuned to the intended depth the noise across the bolometers is constant with frequency. This would imply that increase in noise comes from preferentially tuning our high frequency channels too low.

Issues:

• We are not using the correct method for determining phonon noise. - We assume 1/Vb for our analysis and that is not the case.
• I will pick Francois' brain about the differences in our procedures and also in discrepancies between noise terms calculated at Berkeley and McGill.
• Kam saw something funny with squid loop gain. - Try measurement on many squids to check for pathology? Take finely spaced IVs?

-- PeterHyland - 05 Feb 2010

* Polarbear_8.2.1_20100210-M1_rebinned-500_goodaxis.pdf: Polarbear_8.2.1_20100210-M1_rebinned-500_goodaxis.pdf

---

This topic: BolometerTesting > WebHome > PolarBear > DetectorMeeting Topic revision: r24 - 2010-03-04 - PeterHyland

---