Transcript iTOP_carrier_amplifier_studies_at_IUx

iTOP carrier amplifier test at Indiana University
G. Visser 4/27/2015 (latest update)
I use the recently (here) characterized JT0947 ch3 as input source for this work.
At 3110V, mean gain is measured 3.64×105.
3010 V
3010 V
As before we use this (800MHz analog BWL)
for the gain measurement work.
Pedestal σ = 0.02332 Melectrons
Charge histogram method as before.
Fit pedestal first, then fit whole curve (from 0.072 to 0.70 Melectrons)
to a phenomenological form (red curve) for signal ignoring electronics
noise + the fixed pedestal fit.
Signal in fit
g(x)=b*x**p*exp(-x**q/c)
b = 12372.9 +/- 2999
p = 0.94528 +/- 0.07329
q = 1.24050 +/- 0.04934
c = 0.0901407 +/- 0.00088
 137640 events in signal ( 6.3% )
[We expected 2200000 −12612(cut)
−2062040(pedestal fit) = 125348
events in signal. Reasonable.]
 mean signal size 0.197 Melectrons
JT0947/ch3
3010 V
direct, no amp
Same plot except on linear scale
JT0947/ch3
3010 V
direct, no amp
Pedestal σ = 0.025083 Melectrons
Charge histogram method as before.
Fit pedestal first, then fit whole curve (from 0.082 to 1.47 Melectrons)
to a phenomenological form (red curve) for signal ignoring electronics
noise + the fixed pedestal fit.
Signal in fit
g(x)=b*x**p*exp(-x**q/c)
b = 1220.41 +/- 44.58
p = 0.36669 +/- 0.01891
q = 1.70531 +/- 0.03849
c = 0.78737 +/- 0.02436
 139592 events in signal ( 6.3% )
[We expected 2200000 −18190(cut)
−2049780(pedestal fit) = 132030
events in signal. Reasonable.]
 mean signal size 0.364 Melectrons
JT0947/ch3
3110 V
direct, no amp
Same plot except on linear scale
JT0947/ch3
3110 V
direct, no amp
Pedestal σ = 0.02805 Melectrons
Charge histogram method as before.
Fit pedestal first, then fit whole curve (from 0.11 to 2.36 Melectrons)
to a phenomenological form (red curve) for signal ignoring electronics
noise + the fixed pedestal fit.
Signal in fit
g(x)=b*x**p*exp(-x**q/c)
b = 1220.41 +/- 44.58
p = 0.36669 +/- 0.01891
q = 1.70531 +/- 0.03849
c = 0.78737 +/- 0.02436
 139034 events in signal ( 6.3% )
[We expected 2200000 −22086(cut)
−2042780(pedestal fit) = 135134
events in signal. Good.]
 mean signal size 0.664 Melectrons
JT0947/ch3
3210 V
direct, no amp
Same plot except on linear scale
JT0947/ch3
3210 V
direct, no amp
Pedestal rms in terms of input referred current noise :
Assume perfectly white noise.
One-sided noise current density s, units of A / Hz .
Integration gate length  .
Then the rms noise charge, i.e. observed pedestal rms, should be


2
s.
For example, and to check the constant here is correct :
We know that shot noise of current I has current noise 2qI .
From above, integrating a current with shot noise should give
  qI . And that is correct because the charge I came with
I
I
carriers and so has   q
 qI .
q
q
Of course, the result is general, for any white noise, not only
for shot noise.
For example, the JT0947 ch3 3010V data above, was taken with the scope set for
1.2mV/div. At that setting, input-referred current noise is 53 pA/sqrt(Hz). So, with
a 9 ns integration gate we expect to have pedestal rms of 22.2 k electrons. This
compares well with the fitted pedestal rms 23.3 k electrons.
We want to use same method to check the amplifier input-referred noise in
different configurations.
From SPICE for the amplifier in so-called “1x” configuration, the input-referred
noise is 48 pA/sqrt(Hz), and in “4x” configuration (baseline for carrier rev E2/3)
the input-referred noise is 26 pA/sqrt(Hz). So perhaps optimistically we expect
pedestal rms of 20.1 k electrons and 10.9 k electrons, respectively.
20 events, with “1x” amplifier configuration
Other than signal polarity flip, the min & max voltage
versus charge histogram looks about the same as
before. We make cuts similar to before, to eliminate
saturated events and some two-photon events.
It is necessary to be careful that cuts (or saturation, if
not cut!) don’t bias the upper end of the charge
spectrum too much. Judgement is needed, inspect
charge histogram w/ & w/out cuts.
JT0947/ch3
3010 V
Amp “1x”
Red: raw charge histogram
Blue: charge historgam after cuts
JT0947/ch3
3010 V
Amp “1x”
We assume the amplifier gain
from SPICE: 3067 Ω.
Pedestal σ = 0.02130 Melectrons
SPICE: Expected
0.0201 Melectrons.
Charge histogram method as before.
Fit pedestal first, then fit whole curve (from 0.064 to 0.64 Melectrons)
to a phenomenological form (red curve) for signal ignoring electronics
noise + the fixed pedestal fit.
Signal in fit
g(x)=b*x**p*exp(-x**q/c)
b = 14460.1 +/- 3374
p = 0.90774 +/- 0.06894
q = 1.27845 +/- 0.04929
c = 0.07701 +/- 0.0008367
 135619 events in signal ( 6.2% )
[We expected 2200000 −2899(cut)
−2073110(pedestal fit) = 123991
events in signal. Reasonable.]
 mean signal size 0.175 Melectrons
11% low from expected
result. Reason understood
(see next slides).
JT0947/ch3
3010 V
Amp “1x”
Same plot except on linear scale
JT0947/ch3
3010 V
Amp “1x”
Charge histogram method as before.
Fit pedestal first, then fit whole curve (from 0.07 to 1.3 Melectrons) to
a phenomenological form (red curve) for signal ignoring electronics
noise + the fixed pedestal fit.
We assume the amplifier gain
from SPICE: 3067 Ω.
Pedestal σ = 0.02145 Melectrons
SPICE: Expected
0.0201 Melectrons.
Signal in fit
g(x)=b*x**p*exp(-x**q/c)
b = 3762.15 +/- 325.9
p = 0.61477 +/- 0.03078
q = 1.40095 +/- 0.03595
c = 0.20195 +/- 0.00266
 140482 events in signal ( 6.4% )
[We expected 2200000 −4320(cut)
−2060310(pedestal fit) = 135370
events in signal. Reasonable.]
 mean signal size 0.326 Melectrons
10% low from expected
result. Reason understood
(see next slides).
JT0947/ch3
3110 V
Amp “1x”
Same plot except on linear scale
JT0947/ch3
3110 V
Amp “1x”
Conclusions so far:
• Signals with amplifier, read through scope at IU, make sense in
same analysis framework. (Of course, this is already known from
IRSX work.)
• Pedestal rms corresponds very well with expected value from
measured scope noise or from amplifier SPICE noise.
• “1x” amplifier gain looks ~10% lower than SPICE. In part this may be
resistor tolerance, in part perhaps an effect of bandwidth limitation
or AC coupling, or some charge lost due to nonlinear effects? Not
clear.
• Next up: “4x” case. Hopefully it will be similarly 10% lower than
SPICE, although IRSX results suggest it will be a further 22% lower.
 The 10% discrepancy is explained: prototype circuit didn’t exactly
match carrier rev E / SPICE circuit. Was effectively a 100 Ω / 1 kΩ
voltage divider at the input, owing to bias resistor connection. Changed
this now to match carrier rev E circuit, and replaced the R-C network
capacitor w/ 5 pF (just in case that was not it’s value before) and the R
with 49.9 Ω (it was for some reason 100 Ω).
Following work uses this corrected amplifier circuit.
Minor notes (ignore this page):
run 8: corrected amplifier (should have 3067 Ohm gain)
run 9: changed to “4x” plan (34.8->69.8, 40.2->20)
run 10: same but at 3010 V
run 11: Wanted to understand pulse shape out of
amplifier more certainly. Removed LMH6559, replaced it
with 909 Ohm resistor. So output is now 20.18×
attenuation not 2 × attenuation. (50 Ohm output R is also
still in there, not touched!) Look at pulse shape, as well as
comparing gain results run 11 – run 9.
in all of these cases, scope is set with the proper
attenuation factor. (Except, in LMH6559 case, we assume
unity gain not the actual gain which may be ~1% lower.)
Charge histogram method as before.
Fit pedestal first, then fit whole curve (from 0.07 to 1.0 Melectrons) to
a phenomenological form (red curve) for signal ignoring electronics
noise + the fixed pedestal fit.
We assume the amplifier gain
from SPICE: 3067 Ω.
Pedestal σ = 0.019328 Melectrons
SPICE: Expected
0.0201 Melectrons.
Signal in fit
g(x)=b*x**p*exp(-x**q/c)
b = 1010.72 +/- 90.14
p = 0.40558 +/- 0.03489
q = 1.68143 +/- 0.06435
c = 0.24595 +/- 0.003926
 74342 events in signal ( 3.4% )
[We expected 2200000 −2749(cut)
−2125900(pedestal fit) = 71351 events
in signal. Good.]
 mean signal size 0.342 Melectrons
6% low from expected
result.
JT0947/ch3
3110 V
Amp “1x”
CORRECTED
In this run, unfortunately
a small problem with
negative saturation above
about 1 Me. Ignore that
data for fit.
Same plot except on linear scale
JT0947/ch3
3110 V
Amp “1x”
CORRECTED
20 events, with “4x” amplifier configuration, 3110 V
Charge histogram method as before.
Fit pedestal first, then fit whole curve (from 0.039 to 1.15 Melectrons)
to a phenomenological form (red curve) for signal ignoring electronics
noise + the fixed pedestal fit.
We assume the amplifier gain
from SPICE: 11.32 kΩ.
Pedestal σ = 0.010583 Melectrons
SPICE: Expected
0.0109 Melectrons.
Signal in fit
g(x)=b*x**p*exp(-x**q/c)
b = 360.753 +/- 13.23
p = 0.24404 +/- 0.01528
q = 2.20030 +/- 0.05604
c = 0.25298 +/- 0.003507
 72489 events in signal ( 3.3% )
[We expected 2200000 −1470(cut)
−2127158(pedestal fit) = 71372 events
in signal. Good.]
 mean signal size 0.336 Melectrons
8% low from expected
result.
JT0947/ch3
3110 V
Amp “4x”
Same plot except on linear scale
JT0947/ch3
3110 V
Amp “4x”
Conclusions part deux:
• Pedestal rms still corresponds very well with expected value from
measured scope noise or from amplifier SPICE noise.
• “1x” amplifier gain looks slightly lower than SPICE predicted (6%).
But, this may include a few percent for HV repeatability, resistor
tolerance, AC coupling effects, nonlinearity effects, etc. I think it is
close enough for our purposes.
• “4x” / “1x” amplifier gain matches very well to the predicted value
from SPICE (3.69). I think the IRSX measurements should agree on
this point. (Unless for instance the nonlinearities there are playing a
role?)