--Erik
================================================================================
Ciao a tutti,
Following Doug's e-mail, suggesting further investigation of the noise
present at the modified fanouts (with the 1mF electrolytic cap), we
(mostly Chris Orth and myself) looked into it and we found the following:
(1) The noise that is present has its primary component @ 50Hz, and the
next couple harmonics of it (100, 150, 200). This is very clear if
you look at the output of the especially modified for this purpose fanout
with DC instead of AC coupling. This output does not only have a DC bias
but also a very clear noise component at the above mentioned frequencies.
(2) The arrival of Chuck Lane at Gran Sasso provided us with a SPICE
simulation of the fanout-daughtercard discriminator circuit that is *very*
close to reality. In this way, we could "test" a few things before
soldering components on the test WFD daughtercards. Thus, we concluded
(both from the simulation and at the test bench) that a significantly
smaller cap on the fanout can do essentially the same job as the 1mF
electrolytic one. We have to keep in mind that the reason why we chose
this big cap, was to eliminate the positive overshoot, by making its
amplitude never to exceed the ~3mV discriminator level. But, we have
changed approach by utilizing a small cap on the WFD-daughtecard
discriminator circuit. Thus, the positive overshoot is no longer a
problem, in the sense that it might trigger the POSITIVE discriminator.
But, the problem of the negative undershoot still remains: the positive
overshoot of the original 2.2uF cap, combined with the diode clipping,
can built enough positive charge to create this side-effect that can
blind the WFDs for more than 100usec following a big pulse.
So, our criterion in choosing a (bigger than 2.2uF) cap for the fanout
is to eliminate the NEGATIVE overshoot.
Unfortunately, analytic expression of this negative overshoot is not
as simple as that of the simple high-pass RC circuit (even though it
may be worth spending the extra time to do the Laplace-domain analysis
of this circuit, thus obtaining a model of this undershoot).
But, the test-bench has shown to us that we can go down to 47uF (tantalum)
capacitor, without creating any negative undershoot; not even for a
10V-30usec pulse, as long as the capacitor on the WFD daughtercard is
of the order of 0.1-0.2uF. This result coincides with Chuck's SPICE
simulation.
So, someone may ask: what values shall we use ?? I proceeded as follows.
As known, the 3-db cutoff frequency of a high-pass RC circuit is given
by
f_{3dB} = 1/(2*pi*R*C) (in Hz)
If we set f_{3dB}=50 and solve the above equation, we get
C = 64uF,
and the closest commercially available tantalum of C=47uF gives
f_{3dB} = 67Hz, which should attenuate the 50Hz noise-component by
roughly 6dB.
We proceeded and tested the 47uF tantalum on the fanout with a
0.2uF on the daughtercard. This took place last Tuesday.
At first, we looked for the two side-effects we had noticed before
(noise and 2-5% attenuation) and this time it was not present.
We postponed the rest of the tests for the following day, since that
day was also maintainance day.
The rest of the testing targeted the sensitivity to big and small
pulses. Keep in mind that we are still loosing sensitivity in these
regions for the following reasons:
(1) For big pulses (especially for long ones), the positive overshoot
of the internal R-C circuit (the one after the 3x1 amplifier, right
before the negative and positive discriminators) can last for more
than 100usec.
(2) For small pulses (again, the situation becomes worse for long ones),
the decay of the same internal R-C circuit differentiates them; in this
way we rely on the spe statistics to trigger the discriminators.
But due to the slow transition of the pulse and the reduced sensitivity
to spes (see Erik's e-mail on this subject), large pulses are not
digitized continously. Instead, there are "dents" on the waveform.
Someone may argue that such "dented" pulses should not be considered
lost, but at least for the time being we chose this as a measure of
the "small-pulse sensitivity" for any proposed solution.
Chris Orth kept a full log of our measurements - he can mail it
to everybody if needed. In a nutshell, the sensitivity we measured for
the 47uF - 0.2uF solution is quite good. My personal opinion is that
we need to fine-tune the cap on the daughtercard to ~0.15uF, depending
on what is commercially available. It may also be that we can use a 33uF
cap on the fanout (which is 3 times cheaper) instead of the 47uF, without
any measureable difference.
But, what kept us from declaring it as the "final solution" is that
the noise problem is still there. A very strong 50Hz + harmonics, 3mV p-p
was present on the output of the modified fanout (both the one with the 47uF
and the one with the 33uF cap) the second time we checked for it, even
though it was not present the first time. The DC-coupled output showed it
as a 6mv (!!!) p-p noise.
What changed from the first time we tested it ?
Not too many things: we moved the crate a foot or so, we plugged and
unplugged some cards, we turned it on and off a number of times, but
(in my opinion the most important of all) the electric power guys at G.S.
did some work in the meantime...
Also in my opinion, it is not really important *what* caused the noise
to be present on the output of the fanouts, when driven with no input.
It is most important whether something like that *can* happen to the 12
fanout-crates working now on MACRO at any time, and whether it is a problem
for us or not. It is quite clear, by the way, that this noise was NOT
present on the un-modified channels (with the 2.2uF the f_3dB is at 1.5KHz).
For this reason, I believe that we
(i) either find the source of the noise and eliminate it there; not with
a high pass RC filter at the output of the signal fanout
(ii) or decide that this noise does not bother us (remember: these
50,100,150 oscillations will never reach our discriminators because of
the 0.2uF in-series cap)
(iii) abandon all solutions that increase the capacitor on the fanout
and look for something else.
Ioannis