[Halld-cal] Input to the next iteration on electronics

[Fernando J Barbosa]

Hi Andrei,

There have been a number of questions on item 3 and I thought I had 
presented the reasons quite some time ago. In a previous email you 
suggested a threshold of 50 mV and a gain of X2 for the T channel.  I am 
not sure where you got the 50 mV but here is the rationale for the gain 
of the T channel:

a) set the gain of the A channel to cover the full dynamic range of the 
ADC, i.e. 2V.
b) given the dynamic range of yours and Elton's simulations (400 for 
round numbers), the minimum signal of interest is 5 mV. Since we want to 
set the threshold at half of this value (at least), then 2.5 mV.
c) set the gain of the T channel to get the threshold at a reasonable 
level, above the noise level, or the resolution, of the discriminator. 
For a T channel gain of X10 (original), the threshold would then be set 
at 25 mV; for a gain of X5 (now), 12.5 mV.

The amplifier works well under clipping and does not impact the slew 
rate of the following pulses; there is no additional power dissipation 
to speak of during clipping; the pulse is not stretched (maybe a couple 
of ns max for the recovery time) when clipping occurs that would affect 
subsequent pulses. So, you can expect excellent timing performance and 
high reliability from the readout.

Another point of interest, which required a big effort during the design 
stage, is that the noise performance is excellent. The A output was 
measured on the BCAL sector with everything on at 2-3 mVpp. For purposes 
of not triggering the discriminator on noise, we are interested in the 
peak value or 1.5 mV and at a gain of X10 the noise peak level would be 
at 15 mV (compare to Vth=25 mV); at X5, the level would be 7.5 mV 
(compare to 12.5 mV).

I hope this clarifies the justification for the T channel gain.

Best regards,
Fernando

On 5/23/2012 12:43 AM, Fernando J Barbosa wrote:
> Hi Andrei,
>
> I agree with your comments on the bias distribution non-symmetry, i.e. 
> we have one board used both U and D. That implies that you will need 
> additional runs during calibration or have two boards for U and D with 
> mirrored bias distributions. The reasons for what we have presently 
> have been summarized by Zisis and Elton. I should note that it is 
> reasonably easy, for an additional NRE, to have two PCB versions (U 
> and D) where the only difference is the bias distribution on the inner 
> 8 SiPMs. Taking the bias distribution table defined for U, the bias 
> distribution defined for D would have Bias 1 on SiPMs 3 and 4; Bias 2 
> on SiPMs 1 and 2; Bias 3 on SiPMs 7 and 8; Bias 4 on SiPMs 5 and 6. 
> This is , just a mirror of the bias on the inner 2 layers. The other 
> layers would be the same. This will, of course, complicate assembly 
> and testing a bit - just a matter of logistics.
>
> Regarding the gain for the output going to the discriminator (T 
> outputs), maximum amplitudes are clipped at ~3.2V, independent of the 
> gain. The discriminator will never have inputs above this value. The 
> amplifier behaves well when clipping occurs, much like a discriminator 
> (I can send you some scope pictures if you want). Alex, Elton and 
> Vitaly have looked at the timing performance with a gain of X10 about 
> a year ago. Higher gain, of course within reason, allows for better 
> resolution in setting the threshold at the discriminator.
>
> It seems to me that we have now three items to decide on:
>
> 1. One or two board versions (U & D) due to bias distribution symmetry 
> needs
> 2. Gain for dynamic range conformance (% of present)
> 3. Gain for outputs to discriminators. (X5 now)
>
> Best regards,
> Fernando
>
> On 5/22/2012 8:39 PM, semenov at jlab.org wrote:
>>>> 2. I am not sure I understand the muon/cosmics requirement for
>>>> powering in layers.  If we want the whole inner layer on, we power
>>>> both lines 1 and 2.  We can then leave 3 and 4 off.  Bias 1 powers
>>>> cells 1 and 2, and the second row in the double sum (9-12).  Bias 2
>>>> powers cells 3 and 4, and the first row in the triple sum (13-16). So
>>>> there is no conflict in operating individual rows in sums. For muons I
>>>> would expect nominally all units to be on, with the exception of runs
>>>> were we want to sample the individual ones.  Layer biasing looks nicer
>>>> but I don't see what we lose as it is.  Obviously, the two
>>>> arrangements lead to different patterns (e.g the current system cannot
>>>> power the inner layer and the 1st row in the double sum at the same
>>>> time), but the current one respects load on line Fernando can probably
>>>> provide numbers on load. Am I missing something that cannot be done?
>> Zisis:
>>
>> The cosmics tests are pretty good opportunity to balance the gains of
>> SiPMs (because of well-predicted and almost uniform energy depositions
>> from muons); but for these runs we need to provide the following
>> conditions:
>> 1. It should be 1 active SiPM per readout cell,
>> 2. we want to be sure that the cell to the left and to the right of the
>> cell of interest have no energy depositions =>  we want whole raw 
>> active,
>> 3. We want to have many of these rows preferably distributed over whole
>> thickness of the BCAL to made a track.
>>
>> All this is achievable with 4 runs with level-wise distribution of 
>> biases
>> (viz., U1-D1, U2-D2, U3-D3, U4-D4).
>>
>> If we go for the present bias distribution scheme (with mirrored
>> distribution on the opposite side), we will need more runs:
>> U1-D1, U2-D2, U3-D3, U4-D4 to figure out the gains in the outer part of
>> the calorimeter, and U1-U2-D1-D2 and U3-U4-D3-D4 for inner layers 
>> (so, 50%
>> more time on cosmics that might be non-negligible).
>>
>> But this is not the biggest problem. With U1-D1, U2-D2, U3-D3, U4-D4 
>> runs
>> to calibrate 3rd and 4th readout layers, we will have no good "tracking"
>> of muon in the first 2 readout layers (only half of each inner layer 
>> will
>> be active) =>  we can not guarantee that we will work with muons but not
>> with protons producing the showers...
>>
>> Elton: Discussing gain in the timing channels, I can not understand 
>> these
>> arguments in favor of working with saturated pulses. First of all, the
>> saturation is bad because your amplifiers are overloaded =>  extra heat
>> problems, failure problems etc. Secondly, the saturated pulses have
>> distorted shape and (that is more important) this shape is unstable (for
>> example, the subject of the local load - viz., frequency and type and
>> energy and trajectory of the particles that came before and after the
>> particle of interest). Also I'm not sure how stable will be delay 
>> between
>> the input pulse and output pulse for overloaded amplifier. All this 
>> means
>> that we probably should forget about reliable and precise timing for the
>> biggest-energy-deposited particles in BCAL if we go for saturated 
>> pulses.
>> (Even time-walk correction will be a nightmare.)
>> And finally, I would like to see the faces of people who will read the
>> chapters in our papers where we will declare that we intentionally 
>> set our
>> detectors to work with saturated pulses when we had perfect 
>> opportunity to
>> work with normal pulses from non-overloaded amplifiers. I don't think 
>> that
>> any "flexibility" arguments will help us then...
>>
>> Andrei
>>
>>
>> _______________________________________________
>> Halld-cal mailing list
>> Halld-cal at jlab.org
>> https://mailman.jlab.org/mailman/listinfo/halld-cal
>