[Sbs_gems] [EXTERNAL] Question on pedestal plots

[Andrew Puckett]

Hi Kondo,

I did want to make one more point regarding your point #1: One feature of the data acquisition that we are not yet using in the analysis but which can help with recovering the “baseline” information for the online zero suppressed data is that we are always writing the so-called “MPD debug headers” to the raw data stream, which tells us exactly what common-mode values were subtracted from each time sample in the data. So in principle we know what “baseline” was subtracted from the data, so although we cannot recover data that were zero-suppressed, we can add back the baseline that was subtracted online and/or apply some post hoc correction to the ADC samples.

Cheers,
Andrew

From: Gnanvo, Kondo (kg6cq) <kg6cq at virginia.edu>
Date: Sunday, January 23, 2022 at 2:24 PM
To: Andrew Puckett <puckett at jlab.org>
Cc: sbs_gems at jlab.org <sbs_gems at jlab.org>
Subject: RE: [EXTERNAL] Question on pedestal plots
Hi Andrew,
Thanks for your answers. I fully agree with your two points and that is actually where I was going.

  *   Point #1 was my reservation with Danning method from the start because it lacks the flexibility that the sorting method has to adjust to unexpected situation. For high rate data, I think it is going to be hard to recover it offline because the “baseline” information is lost from online CM correction / zero suppression even if one every 100 events are non-zero suppressed



  *   Point #2: I agree that the negative pulse impact on the efficiency can manifest itself in two ways. Through the common mode correction bias as you described (can still hope to recover offline)and one can clearly see its impact on the V-strips of layer #1 where a larger CM (negative shift) lead to higher occupancy for these strips than the U strip for example. But I suspect that the negative pulses also lead to suppression of real signal (data is lost forever), specially in high rate. But this is far more difficult to evaluate at this point.
I am trying to compile a set of basic plots from Sean GEM studies with beam current yesterday. I will share the plots later tonight.

Best regards
Kondo

From: Andrew Puckett <puckett at jlab.org>
Sent: Sunday, January 23, 2022 1:54 PM
To: Gnanvo, Kondo (kg6cq) <kg6cq at virginia.edu>
Cc: sbs_gems at jlab.org
Subject: Re: [EXTERNAL] Question on pedestal plots

Hi Kondo:

1-) The sagging for Danning method is significantly larger than for the sorting algorithm, this should impact the occupancy calculation and ultimately performance. I am not sure I understand what you mean by “more downside bias for the Danning method” compared to the sorting algorithm

What I mean by this is exactly what you say; i.e., the sag of the common-mode relative to the “zero current” mean value is larger for the Danning method. This is because the Danning method averages in all signals (even large negative ones) that are below essentially (mean + n*sigma), where n = 3 or 5 or something in between, while the sorting method throws out the highest and lowest 20 strips (or other user-configurable number) per time sample per APV card. Therefore the Danning method is more affected by negative pulses.

2-) Sagging is worse for the U/V layers than for the X/Y layers ==> is it just because of the background rate, I guess we can study this by looking at the low beam currents GEM studies runs. I can’t think of anything intrinsic to the U/V layers that would explain that.

Indeed this is related to the higher background rate in the U/V layers. I do not think it is something intrinsic to the GEM, although if the strips are somewhat longer I suppose they might be slightly noisier and perhaps more likely to show negative pulses under high rate conditions.

As far as the effect of the common-mode sag on the occupancy and the efficiency, assuming that the “true” signal baseline is unaffected by the high rate (a significant assumption), the sag of the calculated common mode would tend to result in the application of a lower effective threshold to the signals, whether the zero suppression is done online or offline. In other words, the calculation will result in an estimated baseline that is below the true one, so the zero suppression will allow in more background/noise, on average, and also lead to a positive bias to the (baseline-subtracted) ADC values. This in turn would lead to the apparent “raw” occupancy being higher than the “true” occupancy. Fortunately, applying a lower effective threshold means we should not be losing any data we would otherwise want to keep, so in principle it is an effect we could measure and correct for in software, but the beam current dependence of the effect complicates the implementation of such a correction. I assume that the 1/100 full readout events would play a big role in the implementation.



Andrew





From: Gnanvo, Kondo (kg6cq) <kg6cq at virginia.edu<mailto:kg6cq at virginia.edu>>
Date: Sunday, January 23, 2022 at 11:26 AM
To: Andrew Puckett <puckett at jlab.org<mailto:puckett at jlab.org>>
Cc: sbs_gems at jlab.org<mailto:sbs_gems at jlab.org> <sbs_gems at jlab.org<mailto:sbs_gems at jlab.org>>
Subject: RE: [EXTERNAL] Question on pedestal plots
Hi Andrew,
Thanks, This all make sense. I was also suspecting that sag will have to do with the negative pulses.
There are 3 different features that I can see here that might be worth investigating

1-) The sagging for Danning method is significantly larger than for the sorting algorithm, this should impact the occupancy calculation and ultimately performance. I am not sure I understand what you mean by “more downside bias for the Danning method” compared to the sorting algorithm

2-) Sagging is worse for the U/V layers than for the X/Y layers ==> is it just because of the background rate, I guess we can study this by looking at the low beam currents GEM studies runs. I can’t think of anything intrinsic to the U/V layers that would explain that.

3-) We can also look at the negative pulse rate as a function of the beam current and consequently look at te correlation between  the common mode sagging amplitude and the  negative pulse rate

Best regards
Kondo

From: Andrew Puckett <puckett at jlab.org<mailto:puckett at jlab.org>>
Sent: Saturday, January 22, 2022 7:28 PM
To: Gnanvo, Kondo (kg6cq) <kg6cq at virginia.edu<mailto:kg6cq at virginia.edu>>
Cc: Sbs_gems at jlab.org<mailto:Sbs_gems at jlab.org>
Subject: Re: [EXTERNAL] Question on pedestal plots

Hi Kondo, We’ve seen this phenomenon since the beginning of the experiment, it was worse for high Q2 kinematics. The negative sag of the common mode for either method  is attributable to the wrong polarity fluctuations. The difference between Danning and sorting method calculations is something we’ve also seen from the beginning. The Danning method as implemented offline is subject to more downside bias as a result of this compared to the sorting method. I suppose if we see a larger difference in some modules it could affect the efficiency, but is hopefully correctable in software.

Andrew
puckett.physics.uconn.edu



On Jan 22, 2022, at 6:58 PM, Gnanvo, Kondo (kg6cq) <kg6cq at virginia.edu<mailto:kg6cq at virginia.edu>> wrote:

Hi Andrew,
I was looking at some pedestal plots and I noticed two things,

1-) The difference between sorting and Danning method has this curve-like shape for the U-V strip layer that is not seen for the X-Y layers and the difference is quite significant compare to the common mode fluctuation itself. I looked at some earlier plots (run 13394) when we had all 4 U-V layers as well as the most recent one and it seems quite consistent.

https://logbooks.jlab.org/files/2022/01/3975625/summaryPlots_13455_BBGEM_ped_and_commonmode_50k_printme.pdf

https://logbooks.jlab.org/files/2022/01/3974078/summaryPlots_13394_BBGEM_ped_and_commonmode_50k_printme.pdf

2-) I also see for the V stripe of layer#1, the difference is more pronounced and I am thinking this might partly explain the lower efficiency we are seeing with this layer

I was wondering if there is any explanation for the discrepancy. Maybe there is something here that we can learn something from

Best regards
Kondo
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/sbs_gems/attachments/20220123/730bd9dd/attachment-0001.html>