[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: preamp ASIC linearity



Hi Mitch,
  Thanks for your answers... We'll have to carry on this discussion further, I think, and agree wholeheartedly that a selection of measured waveforms from a realistic detector with realistic track angles and preferably in the realistic magnetic field, should be considered. It may be difficult to obtain this soon, though.
  I am out until middle of next week, so I thought I'd respond briefly just to some points today.
  I'm not completely clear yet, is the driver linearity plot starting from the offset baseline (-425 mV) and hence may be expected to have significant (7% ?) nonlinearity for small signals, as opposed to being very linear up to some percentage of full scale range and then only degrading to 7% nonlinearity at the top of the signal range? I completely endorse the offset baseline plan, but I think we have to be careful about nonlinearity here.
  When I mentioned double-termination, I had in mind keeping the current level the same, just double terminating and accepting the factor of 1/2 in signal amplitude. If the receiver has excellent common-mode rejection and low noise, this is probably not going to degrade the SNR but may improve the linearity considerably. Anyway, this is something we should experiment with (both with the hybrid preamp now being used for testing, and with the ASIC).
  You asked the cable length, I believe it is around 75 feet, roughly. But I don't think the cable routing has been completely specified, at least I haven't seen it.
  One other comment below-->
 
  Gerard

________________________________

From: mitch@hep.upenn.edu [mailto:mitch@hep.upenn.edu]
Sent: Mon 11/20/2006 6:40 PM
To: Visser, Gerard
Cc: halld-tracking-hw@jlab.org
Subject: Re: preamp ASIC linearity



Hi Gerard,
          there are several parts to your questions, I'll address them by topic.

1) Presentation Axes on page 18.  - I agree  they are switched. Will fix.
2) Non linearity -

   a)Keep in mind that we are presenting the response to point ionization
     pulses. If you are looking at tracks, the pulse amplitude will be
     determined by the convolution of the arrival of primaries at the wire.
     A track that results in the collection of 100fC over a 20ns period may
     have 1/2 or less of the ~200mV signal indicated by a point ionization
     calculation.  We can review the NL from 0 to XXX fc for a pt. ionization
     input. Given that you are interested in tracks, what would the interesting
     interval be for the initial charge?   XXX = 100fC?

Well, I think this is precisely the issue I'm worried about. Depending on the track parameters relative to the drift cell, the charge will be spread out over varying times, some quite extended and some short (tracks near the edge of the cell probably have drift times almost constant along the track,  right?). So, unless the cathode strip position fitting algorithm also takes into account the track parameters (to make some "correction" to the fit), the readout must be quite linear up to the level of the maximum typical charge arriving "instantly" in a "point" (I mean really, extended but isochronous) ionization. Perhaps I'm wrong though, I have no experience with cathode strip chambers really.

[For the CDC straw tubes, nonlinearity is probably not a big concern, it only means a correction to the measured dE/dx?]

 previous experience -
     We found it quite instructive to circulate a gallery of plots of typical
     track ionizations in order to understand the electronic performance
     required for the CDF central tracker and the ATLAS TRT.  The large
     variation in signal shapes we observed led us to chooose leading edge
     timing. For Gluex this may not be an option so it will be necessary to
     optimize the shaping of elements following the Preamp.
     It is also true that preamp NL can potentially degrade timing resolution
     but I suspect that the dominant uncertainty will be in the chamber response
     and that non linearities in the preamp will be negligible in determining
     the ultimate time or charge resolution.  Are there track spectra available
     with relevant shaping?  Does there exist a measurement like FWHM Q/Q for
     a point source to get a sense of the signal variation. My recollection is
     that this is something like 50% or greater for normal Argon/C02 filled \
     chambers at ~STP.

  b) Gain -  We have estimated the wiring parasitics crudely as 50fF per net,
     device parasitics are accounted for in the spice models. Since ultimately
     the gain of the shaper will depend on the final parasitics we are waiting
     till the layout is complete to finalize the gain values. We will finally
     set the offsets keeping in mind the NL implications for Device to device
     variations.  It may be a bit too large for now. Also we may want to
     revisit the target gain, based on your understanding of the minimum
     value required for reliable measurements.

  c) Driver- We can look into adding a small resistance in the sources of the
     current mirrors, this may be all that is needed to straighten out the
     reported 7% NL.  We can also look at an NMOS pair terminated into VDD
     connected termination resistors and compare the two approaches.

  d) Double termination. (31 ohms) Doubles the power(driver power) and may
     marginally improve the signal.  We had the provision on the COT but didn't
     need to use it with a 8ns peaking time and 3 meter cables to the buffer
     board. Our problem was high rate, but not analog signal quality so we might
     have chosen differently with an analog signal. What is the time constant
     (cable length) for emptying charge stored in the cable? What is the
     triggering rate?  An output current gain option can be added to the design.

   Good time to get the questions rolling. I'd much rather respond now than
     after something is submitted.

                                                                    Mitch
     Quoting Gerard Visser <gvisser@indiana.edu>:

> Hi Mitch,
>       I have a small question, wanting to make sure that I am interpreting
> your plots correctly. If I look at the linearity plot (page 18 in
>
http://www.hep.upenn.edu/HEP_INST/INST_Projects/GLUeX/Design_Status/Nov_16_TSMC_Implentation/Shaper-Line%20Driver%20Design_version1.ppt
> )
> it shows on the X-axis the test pulse amplitude and the Y-axis the line
> driver output peak-peak, correct? (Axis labels are crossed?)
>       And more importantly, the Y-axis peak-peak amplitude is starting from
> the offset baseline of about -425mV differential like on p.17, is that
> correct.? So when you have peak-peak pulse amplitude of 1V on p.18 it
> means the differential output voltage is swinging from -425 mV up to
> +575 mV. And the output linearity curve is probably symmetrical from 0
> Vpp up to 850 mVpp, correct? (I mean, an inflection point is found in
> the middle of this curve at 425 mVpp, and there is a bending over at the
> bottom end around 0 comparable to the bending over at the top end around
> 850 mVpp.) What is the nonlinearity in this portion of the curve? It
> looks by eye like it might be a few %? I think we would like to have
> more data on this if your guys have some time to run the simulations.
>       If the line is double-terminated so that each leg of the output sees a
> 31 (rather than 62) Ohm load to +1.25 V, the linearity will presumably
> be improved? I guess I am asking, are the output transistors the
> dominant nonlinearity here? Or something else?
>       Would an NMOS-only output (with common-mode termination to vdd instead
> of 1.25V) be more linear? Obviously it would allow VDS to be higher for
> the same swing on the outputs...
>
>       Gerard
>
> p.s. I would think we might well want to double terminate the lines in
> the experiment. I propose to provide the common-mode termination on the
> preamp board (since the common-mode range of the output is rather
> small), and receiver the lines differentially on the ADC board with a
> high-common-mode rejection high-common-mode impedance device like
> AD8129. So, perhaps your guys can also run some simulations with 31 Ohm
> loading on each output.
>