[Sbs_gems] [EXTERNAL] Negative Pulse Analysis Update
Sean Jeffas
sj9ry at virginia.edu
Mon Feb 21 14:56:51 EST 2022
Hi Kondo,
1. Yes, CM is calculated after subtracting the pedestal offset. Sure I
can use a different number of strips, but I think 20 strips is good for
this current analysis, and I will defer my argument on this to point number
3.
2. Yes I am plotting CM (danning) – CM (sorting), and old is left, new
is right. I will actually just update my slides to remove this page, since
it is already irrelevant with Andrew's new ideas.
3. By true baseline I mean the actual real APV baseline for exact time
sample during that trigger. Meaning the "true baseline" is the number we
are trying to calculate with the CM. As I mentioned in my above email,
there is no way to tell how close your calculated CM is to the true
baseline, except to plot the full event, and look at it, as I have done
here https://logbooks.jlab.org/entry/3984280. Then by looking at the
Sorting method result (green line), it visually looks to be within 15 ADC
of where I think the "true baseline" is just using my eyes. Again I have no
other ideas on how to check how accurate the CM calculation is, and I would
be interested if there are any other ideas. So my conclusion, looking
through all beam currents and hundreds of event displays, is that the
Sorting method with 20 strips is always 15 ADC or less from the correct
baseline just looking at it visually. Then if you check my plot "Common
Mode Comparison to Pedestal" here,
https://logbooks.jlab.org/files/2022/02/3984417/neg_pulse_study_3.pdf,
you can see that at 7 uA the Sorting method gives us results that are
shifted by up to -30 ADC. Therefore, I think I have proven that the Sorting
method is accurately calculating the CM with my event display, and the
Sorting method is giving large negative offsets which increase with
occupancy due to the A/C coupling in the APV. *But the important point* is
that I am not saying there is anything wrong with this. Like you say, the
shift exists, and we can correct for it, no bid deal. I am simply pointing
out proof that this exists, since a month ago it was causing confusing with
this negative pulse study, and no one had ever discussed the A/C coupling
before.
Best,
Sean
On Mon, Feb 21, 2022 at 12:07 PM Gnanvo, Kondo (kg6cq) <kg6cq at virginia.edu>
wrote:
> Hi Sean,
>
> Thanks for the detail answers to my questions. I think we overall agree,
> it just some definitions that I think we have different understanding to.
>
> Please see my additional comments inline starting with +++
>
>
>
> *From:* Sean Jeffas <sj9ry at virginia.edu>
> *Sent:* Monday, February 21, 2022 10:41 AM
> *To:* Gnanvo, Kondo (kg6cq) <kg6cq at virginia.edu>
> *Cc:* Sbs_gems at jlab.org
> *Subject:* Re: [Sbs_gems] [EXTERNAL] Negative Pulse Analysis Update
>
>
>
> Hi Kondo,
>
>
>
> 1. Yes I am removing the highest 54 and lowest 54 ADC channels. So yes
> I agree that this removes the negative pulse effect. Is this not what we
> want in order to correctly calculate the CM?
>
> +++ Yes, that the way to calculate the CM taking into account the negative
> pulses. My point here is that even though sorting is far better to address
> the negative pulses than Danning’ method, It is not perfect and the less
> channels you use to calculate the CM (i.e 20) the less precise your
> correction will be. You can check that with a plot of CM (7uA) – CM
> (pedestal data) as a function of how many channels you use (from 20 to 80)
> for example
>
> +++ BTW, here I hope that before you compute CM, you correct for the
> pedestal offset before è This order becomes critical here since you are
> removing both negative and positive pulses
>
> 1. My apologies, on slide 5 I talk about old and new CM calculation
> methods. The left and right plots are supposed to be the same, but compare
> between the old and new methods. But this slide is already outdated, since
> Andrew has developed a new online CM algorithm that seems even better, and
> he will update about this later.
>
> +++ Ok, I understand, but to be precise, what you are plotting on the Y
> axis is it CM (danning) – CM (sorting) and which one is the old (left plot
> or right) and which one is the new?
>
> 1. The only ways I know to definitely prove a baseline shift is to
> plug an oscilloscope into the output and look at it with beam or to
> correctly measure the CM baseline during beam and compare to the pedestal
> levels. It is too late for the oscilloscope option but we can measure the
> baseline from the full readout events. Unless someone has a better
> suggestion, the only way to "prove" the CM is calculated correctly is to
> visually look and see how the result compares to the raw strip data. I have
> done this using the event displays here,
> https://logbooks.jlab.org/entry/3984280. Specifically you can look at
> this run, which I zoomed in the axis to make things even more clear,
> https://logbooks.jlab.org/files/2022/02/3984280/CM_event_display_run13602_zoomed_0.pdf.
> I think from all this data it is clear that the sorting method gives a CM
> result that is within 15 ADC of the true baseline. Then when we compare the
> sorting method results over different beam currents we see the CM offset
> increase up to -35 ADC,
> https://logbooks.jlab.org/files/2022/02/3984417/neg_pulse_study_3.pdf.
> Therefore I see no other explanation than the APV baseline is truly
> shifting downwards while the beam is on.
>
> +++ Here it will be good if you define:
>
> +++ what you mean by true baseline? Is it the baseline you get from
> pedestal data after correcting for the CM?
>
> +++ when you say CM is within 15 ADC of the true baseline, what do you
> mean? is that statement true for any beam current?
>
> +++ CM offset increases up to -35? What does that mean? è baseline shift
> is only relevant w.r.t different time beam of the same APV and the same
> event, It is not with respect to any “true baseline” level, the CM offset
> parameter is precisely the magnitude of the correction you apply to the
> baseline shift it does not matter if it is -15 or -35 as long as you can
> calculate it properly and it is.
>
> +++ The point I was making to Ben at the meeting last week is that if the
> shift is a baseline shift, then it does not matter what cause it or even
> understand its source as you can correct it. If it is shift for individuals
> channels in the APV because of high rate and large signal (which I observe
> with APV data) then it also does not matter because these channels are not
> used for the CM calculation
>
> +++ Do you have a plot of average CM (new calculation) as a function of
> the beam current
>
> 1. I selected one MPD from the front most UV layers, which naturally
> has the most background. One MPD on a UV layer covers half of the active
> area, so this does cover the hottest region of the detectors.
> 2. You are right, I have not thoroughly investigated the possibility
> of the entire hit being flipped. I have only looked at a hit flipping
> within the 6 time samples. I am still working on using tracking to check
> for negative hits in "expected" areas of the detectors. I will just say
> that from my plots of "APV Negative Signal Peak Time" we do not see a
> normal pulse shape. But of course this could be due to saturation. or
> something else.
> 3. My apologies, my wording was not correct here. My conclusion should
> be more like "There is no evidence of positive signals flipping into
> negative signals inside of the trigger window". As you mentioned there is
> also the scary possibility that the negative strips are not hits, but they
> "occupy" the strip and do not let positive hits in the same location be
> recorded. This can be investigated by checking if positive clusters ever
> have a "break" in the middle from negative strips. Or perhaps anyone else
> has a suggestion.
>
> +++ Yes and this was my actual initial worry in October when we observe
> the negative pulses for the first time, l whether it is polarity flip or
> not, the strip is occupied by a wrong signal. I became aware of the second
> problem i.e. CM correction later, but I am less worried about this one
> because it can be dealt with “easily”
>
> 1. Wouldn't the negative pulses saturate at 0? Or am I
> misunderstanding something?
>
> +++ The “0” here is your baseline that level is set through some register
> that you configure. So you optimize this level to increase the dynamic
> range for the polarity of the signal you want to measure, then you
> automatically limit the range for the opposite signal (which is negative on
> the plots here), so no matter how large is the pulses (whether it is real
> signal or not) it will be limited by the 200 – 400 ADCs unlike the positive
> pulses that can go up to 2000 ADCs … è that is why the analysis between
> positive and negative pulse i.e your 2D correlation plots are biased. At
> least that is my understanding, maybe I am wrong on some details but Ben
> can correct me
>
> Any way the most important thing is to have a strong CM algorithm that
> deals with negative pulses and it looks like you and Andrew are making
> progress so I just raise points to understand the phenomenon we are seeing.
> It is not criticism
>
> Best regards
>
> Kondo
>
> Best,
>
> Sean
>
>
>
> On Fri, Feb 18, 2022 at 7:04 PM Gnanvo, Kondo (kg6cq) <kg6cq at virginia.edu>
> wrote:
>
> Hi Sean,
>
> Thanks for the studies. It is pretty extensive and a lot of information to
> digest so probably I am still missing a few things, so maybe l can ask a
> couple of few points that might help clarify a little bit things.
>
>
>
> 1. You use only 20 middles channels to calculate CM using the sorting
> method, meaning that you remove the 54 highest ADC channels and 54 lowest
> ADC channels from the calculation è Is that correct? If so, yes this
> will probably remove the negative pulse effect in almost all cases even
> though the precision of CM calculation will be strongly affected by
> definition
>
>
>
> 1. What is the difference between the 2 plots of slide 2 in attached
> pdf. I see the same title in the y axis è is it typo where you meant
> Danning CM – baseline CM (on left) and sorting CM – baseline CM (on right),
> baseline meaning the CM calculated from pedestal data (ideal case?)
>
>
>
> 1. Do you actually have any evidence from the data of the voltage
> shift downward due to high occupancy and why is it an issue w.r.t to CM
> correction?
>
>
>
> 1. Even for the 7uA on LD2 data that you are showing, I don’t really
> see high occupancy data, and since you said that you analyzed data from
> only 1 MPD, did you select MPD reading out data from the hottest area of
> the chamber?
>
>
>
> 1. In my view, there are several aspect of what I mean by polarity
> flip.
>
>
> 1. It looks to me like you are looking for a flip for a same strip
> across the 6 time sample. This is just one way, but very likely the most
> unlikely
> 2. On slide 1 of the attached pdf, you can clearly see an event
> with have both positive ADC strips and negative ADC strips on the same APV.
> The question is are the negative signal flip or not? Unless these strips
> are always below the baseline (i.e. their pedestal level, which we al agree
> it is not the case), something makes them flip their ADC level w.r.t the
> baseline while there was good hit in other strips of the same APV
>
>
>
> 1. Even if there is never an actual signal flip, with your correct
> conclusion that the negative pulse occupancy increases in similar fashion
> than the positive one. I think regardless whether we convince ourselves
> that signal can flip or not:
>
>
> 1. it is not correct to say that negative pulse does not cause the
> loss of positive pulse
> 2. there is a high probability that a good event hitting these
> strips might be lost (because of negative occupancy) and I think that is
> the scariest scenario.
>
>
>
> 1. Most negative pulses have the same amplitude across time sample can
> be just due to saturation, there are actually some case where they have the
> typical APV signal waveform (see slide 3)
>
>
>
> Any way these are just a few questions I have after looking at the results
> you present.
>
>
>
> Best regards
>
> Kondo
>
>
>
> *From:* Sbs_gems <sbs_gems-bounces at jlab.org> *On Behalf Of *Sean Jeffas
> *Sent:* Friday, February 18, 2022 4:31 PM
> *To:* Sbs_gems at jlab.org
> *Subject:* [Sbs_gems] [EXTERNAL] Negative Pulse Analysis Update
>
>
>
> Hi All,
>
>
>
> I touched up my negative pulse analysis results that I showed to everyone
> last week. I realized I was making a serious mistake by simply cutting on
> signal > 80 ADC, because the first few strips in the APV always have large
> signals on them from the event header. Therefore this was significantly
> skewing my data. I am now using the correct method of cutting using 5*sigma
> of the pedestal noise, as is done in the tracking analysis.
>
>
>
> Please see my new results in the following slides
> https://logbooks.jlab.org/files/2022/02/3984417/Neg_pulse_study_2_18_22.pdf.
> Also please take the time to look over all the plots in my log entry,
> https://logbooks.jlab.org/entry/3984417. I try my best to pick some event
> examples to show, but I encourage everyone to scroll through the event
> displays I have posted, to get an idea of what the negative pulses look
> like. Please let me know your thoughts and questions.
>
>
>
> Best,
>
> Sean
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/sbs_gems/attachments/20220221/eef5e459/attachment-0001.html>
More information about the Sbs_gems
mailing list