John Kelley, September 2006
This page documents the results of various test runs with experimental FPGA builds to try and eliminate the baseline shift in the ATWD. For background information on the problem, please see this page.
The data for these tests were recorded with TestDAQ running on two DOMs in a chest freezer in Chamberlin at UW (triggering on dark noise). The DOMs were set for local coincidence mode, with only DOM A (Platypus) recording waveform data. The discriminator for DOM B (Titanium) was tuned to provide random LC hits for Platypus for approximately 1/2 of the triggers.
The data are calibrated with DOM-Cal runs taken with the same FPGA build. DOM-Cal calibrates the ATWDs (pedestal pattern + baseline) using CPU triggers and does not ever enable LC.
FPGA builds tested:
Jim Braun has also investigated the ATWD ghost pulse effect in these and later builds. See his page here.
For Thorsten Stezelberger's descriptions of the various FPGA builds, see his page here.
In the released test FPGA, the baselines show a distinct bi- (and maybe even tri-) modality.
Build 413 was the initial hack which essentially removes the LC abort from the ATWD control. The baseline bimodality is eliminated.
Build 420 does not eliminate the bimodality and introduces a slope to the waveform which is clearly visible in ch1 and ch2 data. This slope is also present in the raw data. Since the calibration runs were also taken with build 420, this indicates the slope is also related to the LC aborts.
Build 421 has the bimodality present, but at a reduced level in ch1 and ch2. The slope from build 420 is not present. However, one can see that both modes are forms of the "shifted" baselines, since the hook is still present at the end of the waveforms.
Build 423 exhibits both bimodality and waveform slope.
Histogram of baselines (in ATWD counts) as determined by the first sample (channel 0, sample 0):
Histogram of raw data from channel 2 (baseline average in ATWD counts vs. time between hits), courtesy of Jim Braun:
Build 423 is qualitatively similar to build 423 and exhibits both bimodality and waveform slope. As one can see in the final histogram of raw baseline average, the split between the modes is reduced a bit.
Also, the offset of the modes is different as is most evident in channel 0, with one of the modes being below the calibrated baseline. The frequency of each of the modes has shifted as well, with the lower mode being less frequent than the higher mode.
Histogram of baselines (in ATWD counts) as determined by the first sample (channel 0, sample 0):
Histogram of raw data from channel 2 (baseline average in
ATWD counts vs. time between hits), courtesy of Jim Braun:
FPGA build 425 eliminates the slope seen in some of the previous builds, but the bimodality is back to levels seen in the released builds. It appears that the two modes are shifted vs. unshifted baselines, as opposed to the split shifted modes seen in, for example, build 424.
Build 426 shows subtle bimodality, but this time around the "unshifted" raw baseline level. There is no waveform slope.
The bimodality in channel 2 is invisible in the waveform view, but it shows up clearly on a scatter plot of the raw waveforms:
The delta_T distributions for the different baselines are standard for an effect of LC_abort:
FPGA build 427, which has a longer reset time than build 426, looks almost good -- the baselines are at "unshifted" levels, bimodality is eliminated in channel 1 and 2 (but not channel 0), and there is no apparent waveform slope.
The bimodality in channel 0 only shows up in the scatter plot of the raw baseline versus hit delta_t:
Scatter plots of raw baseline vs. hit delta_t for channel 1 and channel 2 show no hint of bimodality:
