<html>
<head>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
Sho,<br>
<br>
I think the data speaks for itself, the 3-pole is much better! In
particular the edge of the distribution will be much better, that is
important to reproduce the time precision of the experiment in the
monte-carlo! That might impact trigger algorithm, so it is critical
not to miss this part of the distribution!<br>
<br>
Best regards,<br>
<br>
Le 04/10/2014 11:18 PM, Sho Uemura a écrit :
<blockquote
cite="mid:alpine.LRH.2.00.1404101414580.25883@noric40.slac.stanford.edu"
type="cite">Realized this discussion didn't make it to the ECal
list.
<br>
<br>
Would like more ECal experts to weigh in. I think we need to
decide by the end of the month for a parametrization to use in the
mock data.
<br>
<br>
On Fri, 4 Apr 2014, Sho Uemura wrote:
<br>
<br>
<blockquote type="cite">Mean chisq/DOF is 0.56 for 3-pole, 0.68
for double-Gaussian.
<br>
<br>
<a class="moz-txt-link-freetext" href="http://www.slac.stanford.edu/~meeg/ecalpulsefit/3pole/rchisq.png">http://www.slac.stanford.edu/~meeg/ecalpulsefit/3pole/rchisq.png</a>
<br>
<br>
<a class="moz-txt-link-freetext" href="http://www.slac.stanford.edu/~meeg/ecalpulsefit/2gaus/rchisq.png">http://www.slac.stanford.edu/~meeg/ecalpulsefit/2gaus/rchisq.png</a>
<br>
<br>
I think that in every fit the 2-Gaussian shape overestimates the
rising edge; 3-pole looks better. As Andrea says, the pulse peak
is definitely better fit by 3-pole.
<br>
<br>
On Thu, 3 Apr 2014, Sho Uemura wrote:
<br>
<br>
<blockquote type="cite">chisq/dof plot attached - I'm using
TGraph to fit, which assumes errors of 1 mV per data point. I
use the time range (-200, 60).
<br>
<br>
On Thu, 3 Apr 2014, Gabriel CHARLES wrote:
<br>
<br>
<blockquote type="cite">Could you both provide an average
value of chi square that the different parametrization can
be compared easily, please ?
<br>
<br>
Also, from the simulation it appears that the rising edge
could be present. In attachment you will find a picture with
two plots. The top one corresponds to the signal after the
crystal and the APD, that is the input of the preamplifier.
<br>
It is obtained by the convolution of the signal of the
crystal and the APD. The crystal response is composed of the
sum of two decreasing exponential governed by different time
constants. The APD transfert function is given by the bottom
plot (sorry for the wrong Y axis units).
<br>
<br>
There is no reason for the preamplifier to reduce the tail.
<br>
<br>
I think that if there is no huge difference between the chi
square it would be better to keep the two gaussian function.
<br>
<br>
---
<br>
Gabriel CHARLES
<br>
Institut de Physique Nucléaire d'Orsay
<br>
<br>
On Thu, 3 Apr 2014 13:15:00 -0700 (PDT), Sho Uemura wrote:
<br>
<blockquote type="cite">I tried two more parametrizations.
These are parametrizations
<br>
commonly used for the APV25 preamp that we use in the SVT.
<br>
<br>
CR-RC: t*exp(-t/tp)
<br>
3-pole, or CR-RC-RC: t^2*exp(-t/tp)
<br>
<br>
3-pole seems to fit well, I think better than the
asymmetric
<br>
Gaussian. CR-RC seems no better than the Gaussian. Other
<br>
parametrizations I tried (variations on CR-RC or 3-pole
using more
<br>
than one time constant) were degenerate with CR-RC or
3-pole, so I
<br>
didn't include those plots.
<br>
<br>
Plots attached are for 3-pole function. All plots for
3-pole and
<br>
CR-RC, and the pyroot scripts I used, are online:
<br>
<br>
<a class="moz-txt-link-freetext" href="http://www.slac.stanford.edu/~meeg/ecalpulsefit/">http://www.slac.stanford.edu/~meeg/ecalpulsefit/</a>
<br>
<br>
I also see what you see, where there are 2 clusters in the
<br>
distribution of shape parameters. I chose the center of
the larger
<br>
cluster (with the faster time constant) and refit all the
events with
<br>
this time constant fixed; those plots are named "fit2" and
as expected
<br>
they fit the faster pulses well and the slower pulses
poorly.
<br>
<br>
More data will help.
<br>
<br>
I plotted the three parametrizations we have, see
plot4.pdf attached.
<br>
If we agree that the Gaussian has an unphysical rising
edge, I think
<br>
we should use 3-pole.
<br>
<br>
On Tue, 1 Apr 2014, Andrea Celentano wrote:
<br>
<br>
<blockquote type="cite">Dear all,
<br>
here are some results about HPS Ecal signals
parametrization.
<br>
I took data with the crystal placed vertically, APD gain
150, room temperature. I put a threshold ~ 20 mV to keep
only big enough signals, out of the noise.
<br>
I acquired data with a 2.5Gs/s oscilloscope, 1 GHz
bandwidth, 50 Ohm input impedance.
<br>
<br>
I used the same* configuration employed at JLab for
cabling: 8m 3M cable ---> passive splitter ---> 3m
lemo cable.
<br>
<br>
*actually I employed an 8 meters 3M cable instead of 7m
because the latter is not available here in Genova.
<br>
<br>
Attached you find a postcript file with the results.
(outGood.ps shows the fit results covering some parts of
the signal, outGood1.ps no)
<br>
<br>
- Neglect first blank page
<br>
- Pages from 2 to 32 are the 31 signals I got, with
superimposed the fit performed with the two-gaussians
parametrization. Each chi2 fit is performed
independently.
<br>
Signals are in mV and ns.
<br>
Note that near ~ 100 ns there is probably a reflection
due to some impedance mismatch in the cables chain.
<br>
However, I am not using those points to fit. I am
fitting the data in between -200 ns and +80 ns. The
function is then plotted in the full time range.
<br>
<br>
- Last page is a summary of the fits performed. Two
1d-histograms are the distributions of the two time
constants used in the parametrization. Then I am
plotting also their correlation, as well as the
correlation of the rise-time (par[1]) with the signal
amplitude (from the fit).
<br>
<br>
I noted that the fit parameters Trise, Tfall are not
distributed as two gaussians. In particular, for Trise
there is an accumulation of events at ~ 5 ns and ~ 7 ns,
correlated with corresponding Tfall at ~ 15 and ~20 ns.
Actually, I see that, other than the amplitude, signals
do not have always the same shape: look, for example, at
signals n.5 and n.6 (ps pages n.5 and n.6).
<br>
<br>
Attached you find also the C implementation of the
signal parametrization, in form of a "double fun(double
*x,double *par)" used by ROOT when fitting trough TF1.
<br>
Finally, I am attaching also the raw data for the 31
signals I got, so if you're interested you can play with
different signal parametrizations.
<br>
<br>
I am planning to take more data these days.
<br>
<br>
<br>
Bests,
<br>
<br>
Andrea
<br>
<br>
</blockquote>
</blockquote>
<br>
########################################################################
<br>
Use REPLY-ALL to reply to list
<br>
<br>
To unsubscribe from the HPS-SOFTWARE list, click the
following link:
<br>
<a class="moz-txt-link-freetext" href="https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=HPS-SOFTWARE&A=1">https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=HPS-SOFTWARE&A=1</a>
<br>
</blockquote>
</blockquote>
<br>
</blockquote>
<br>
<fieldset class="mimeAttachmentHeader"></fieldset>
<br>
<pre wrap="">_______________________________________________
Hps-ecal mailing list
<a class="moz-txt-link-abbreviated" href="mailto:Hps-ecal@jlab.org">Hps-ecal@jlab.org</a>
<a class="moz-txt-link-freetext" href="https://mailman.jlab.org/mailman/listinfo/hps-ecal">https://mailman.jlab.org/mailman/listinfo/hps-ecal</a>
</pre>
</blockquote>
<br>
<pre class="moz-signature" cols="72">--
Raphaël Dupré
Institut de Physique Nucléaire d'Orsay
Université Paris-Sud
</pre>
</body>
</html>