Answers to STT review panel questions
======================================
(questions dated 4 April 1998, answers sent 10 April 1998)


    Members of the STT group have provided answers to the questions on the
    design and  performance of the STTpp that have been asked by the review
    panel. The questions and answers are appended to this message.
    It should be noted that while we did answer the technical questions,
    this may be of limited relevance since the design is still under discussion
    and may turn out to be  quite different. On the other hand, even if details
    of implementation change, we do have all  intention to keep the
    functionality of the device as outlined in the draft-TDR.
    
                 the STT group

------------------------------------------------------------------------------

Question 1.:
   Is a threshold applied to the pulse height of the seed strip before a
   cluster is generated by the STTpp? Or is every unsuppressed strip
   that's read out used even if it forms a cluster from only the one
   strip? (i.e., is the SVXII threshold sufficient?)

Answer 1.:
  we foresee that we will have the capability to:
    * mask out known noisy or bad channels
    * correct for pedestals and gains
    * apply a threshold
    These corrections could be applied on a channel-by-channel basis if needed,
    but some probably will only be needed on a chip-by-chip basis. The exact
    procedure can only be determined once we know the noise level and
    uniformity across readout chips. However the capability will be built into
    the design.

Question 2.
   We are not sure we understand some of the numbers and plots in
   Meenakshi's note "Dataflow Requirements for the Silicon Track Trigger".
   In particular, in tables 1 and 2, we do not understand why the average
   number of clusters in a road for sublayers 7 and 8 goes down as the
   number of interactions in a collision rises. This comes from the
   "Clusters in Road" plots, where there are a very large number of zero
   clusters (not) found, but these missing clusters do not even lie just
   outside the road, as shown by the large peaks at 0 in the plots of
   numbers of clusters in 30 degree sectors in the various barrel modules.
   Why is the SMT apparently so inefficient? Or has something anomalous
   happened during the studies?

Answer 2.:
    The "CLusters in Road" go down for sublayers 7 and 8 as the
    road purity of the CFT gets low for high interaction events.
    The problem is that if the CFT road gives a fake track, then
    most probably there are still alot of wrong hits being picked
    up in the inner layers, while the outer layer does not have contribution
    from real tracks. As can be seen in table 2 (Zbb events), though the
    "Clusters in road" go down for these sublayers, the numbers
    of clusters in 30 degree sectors goes up with increasing number
    of interactions for all layers/sublayers as expected.

    In Table 1 (minbias events) one would have expected the same behaviour,
    but the occupancy in the outerlayers also goes down for 6 and 8
    interactions. For one/two and four interaction files the behaviour
    is as expected. I strongly suspect that there is a problem with Geant
    here, I had noticed this weird feature before writing the dataflow note and
    concluded that it was not the STT/CFT algorithms in this case.
    The 6 and 8 interaction point should not be believed too much.

    For the occupancy calculation for the TDR we actualy took the rates
    for the 4th layer same as for the 3rd layer to circumvent this problem.
    I do not know how to do a better job at this point given these Geant
    files.
    
Question 3.:
   For the timing studies, the "TDR" uses 2% occupancy whereas some other
   studies have used 5%. What is the expected noise? What happends to your
   design if the occupancy is actually as bad as 5%? Is there any noise
   from the previous crossing?

Answer 3.:
    The number 2% was derived from the "physics" occupancy seen in the
    simulations plus the  "noise" occupancy seen in the testbeam. The largest
    effect on the experiment  of a 5% occupancy will be that the SVX2 readout
    time will be 2.5 times  longer. The maximum times given in the TDR would
    easily encompass such an  increase in occupancy. If the extra hits are due
    to noise they will be successively eliminated in the pattern recognition
    steps within the STT (ie,  cluster -> roads -> tracks)

Question 4.:
   Has any thought gone into how this STTpp works if the luminosity rises
   to 5 x 10^32? 2 x 10^33? (132 ns Run III) for the same SMT geometry (if
   we get no tracking upgrade) or for an enlarged SMT (more barrel
   modules, more layers per barrel?) - is the STTpp upgradable for one or
   both of these scenarios?

Answer 4.:
    we have not simulated a higher luminosity. Presumably the main effect
    will be  an increasing occupancy due to soft tracks from additional minbias
    interactions.

    The processor can easily be extended to cover more barrels or more layers
    by adding more trigger cards. If we use the VRB backplane the number of
    cards with VME buffer control (for L3 readout) can be at most 12.
    12 trigger cards can accommodate 8 barrels and 4 layers (the only possible
    short term expansion) or 6 barrels and 5 layers (if the 5th layer has as
    many HDIs as layers 3 and 4). For further expansion we would have to double
    the number of crates from 6 to 12.

    It should also be noted that at present the accelerator has no approved
    plan to  increase the luminosity beyond  2x10^32 cm^-2sec^-1. Even if TeV33
    happens one will try to keep the peak luminosity to 5x10^32 cm^-2sec^-1 for
    example. This will keep the interaction per crossing low.
    The goal is to have higher integrated luminosity, not peak luminosity.