Run: 10076100 was taken with single BHT3 trigger set very high at DSM value of 50, close to 12 GeV.  Events were passed to  L2W algo which did *not* abort any events, only accumulated spectra.

Fig 1 Left shows raw ADC vs. softID for all events and all towers in this run - clear band structure is visible.

Fig 1 right is projection of  ADC for all towers. The first line is drawn at approximate position of the thresholds dsmVal*16+50. Other lines are offset by 512.

The blue inset is HT energy spectrum (in GeV) accumulated by L2W algo on input shows very similar band pattern.

Fig 2. Similar pattern was observed by L2 ~2 days ago for the run 10074075, when BHT1 threshold was set to 30. Note, now separation between bands is smaller.

Fig 3. Comparison of band structure seen by L2W for different BHT3 thresholds of 20,30,39,45,50. L2W seed thresh was at 4,5,5,4,4 GeV, respectively.

Note: the right edge for every peak is the same on figs C-E, but the left edge  of all peaks seems to change with DSM thresholds for figs C-A, what implies for figs A & B they were moved more to the left and peaks merged.

Fig 4. run 10076163, no L2 algo, DMS BGHT thres=50

Fig 5. run 10076163 DSM spectra - bands are visible as well (from Pibero's plots).

Matt found explanation for this problem:

The ADCs are sent to the FEEs in 12 bits. The FEEs send 6 bits to DSM. 
The bottom 2 ADC bits are always thrown out. We configure the FEEs to 
mask the next 2 bits. Then the next 6 bits are sent to the FEEs, except 
that the most significant bit is the OR of itself and any remaining 
higher order bits (in the current configuration, there are 2).

What we figured out is that if the DSM threshold is 32 or higher, this 
logic will result in gaps. I will illustrate with an example:
If the DSM threshold is 50, that corresponds to the 6 bit threshold: 1 1 
0 0 1 0.
Now if we receive the following top 8 bits from the ADCs : 1 0 0 0 0 0 0 
0, this is above threshold, but gets "truncated" to these 6 bits: 1 0 0 
0 0 0, which are below threshold. This results in a gap between 1 0 0 0 
0 0 0 0 and 1 0 0 1 0 0 1 0. There will be gaps whenever the top 3 bits 
change.

So long as our DSM threshold is below 32, we will not have this problem. 
The current BHT3 threshold is set at 25, so we should be safe. We looked 
at some runs from yesterday with that threshold. Stephen and Jan claim 
that it is inconclusive, but there are no apparent gaps.

And Gerard can sleep well again

Doh!

So the transfer function from ADC to HT output is

    |       /   /   /   /   /...
 H  |      /   /   /   /   /
 T  |     /   /   /   /   /
    |    /   /   /   /   /
 o  |   /
 u  |  /
 t  | /
    |/
    ------------------------------
         ADC-PED

Makes perfect sense; Steve asked Chinh for the wrong thing.

What we have will work for threshold <=31 (not 32).

Fig 6a. Day 76, runs 10076136,152,153,154,161, only W-stream, 5485 L2W eve, total run time ~3700 seconds,L0:BHT3 thresh=25 & L2W thres=13 GeV are marked approximately, assuming ped=30 ADC

Fig 6b, the same data vs. soft ID.

March 19,2009, from Carl:

Eleanor has replied to my request for information about the "missing, 
dead, and noninstrumented channels".  The "missing 63's" arise from 
the HT threshold comparison logic in the FPGAs:

if (data>threshold and data<63) then
     result = 1
else
     result = 0
endif

Eleanor says there is nothing like this for the BTP 
calculations.  Therefore, we should be fine there (so long as the 
FEEs are really plateauing, not wrapping around).

Fig 7. L2W eve from days 78+79. Now 'notch' at ADC=2*512 & 3*512 is clearly visible. Now we see also threshold L2W cluster ET>13 GeV. 

 TIER1 file CHANGED on day 86, March 26 , ~9pm. "notch" is gone.

Last run w/ old tier1 file is 10085115, first run w/ new tier1 file is 10085131

Fig 8. TOP=new tier1 file : 21K L2W events from days 86, runs 10085131-10086046, using production2009_500Gev_c & trg_090325.bin

BOTTOM=old tier one file: 31K L2W events from days 82-86 using production2009_500Gev_b (or other Car's triggers with BHT3 thr=31)  & trg_090320.bin