Tiger Team Week - 13 - 17 April 2015


The goal of the Tiger Team Week is to brainstorm possible causes for the VEGAS bank-to-bank variation problem, and define tests we can do to isolate the origin of the problem. We already have a number of tests defined which we can perform starting on Tuesday morning, but it is still worthwhile discussing other possibilities.

The problem: Tsys, the ratio of off/(on-off) times a constant Tcal, as calculated from VEGAS HPC spectral line data is wrong, and can vary (sometimes wildly, depending on setup) from bank to bank and poln to poln.

What I think we know (others, feel free to clarify / add to this list)

  1. The problem is not present in total power in Mode 1, as measured by the ADCs, or the overall (mean central 80%) Tsys as calculated by GBTIDL. It is present if Tsys is calculated for a narrow portion of the overall Mode 1 passband. It is obvious in Mode 4.
  2. The problem does not appear to be as obvious in total power, as measured by the VEGAS ADCs.
    1. Dave MacMahon has evidence it is present (which I have forgotten...)
    2. They are clearly not as bad as the spectra, when VEGAS is "detuned".
  3. The problem does not appear to be present when the VEGAS IF modules (or ADCs) are fed with a "well-behaved" lab noise source:
    1. The ratio appears to be approximately correct (using different input power levels for "ons" and offs");
    2. The ratio appears to be approximately identical from bank to bank / poln to poln for "identical" lab noise source inputs.
  4. The problem can be made significantly worse by applying a bad lbw_gain value.
  5. The problem can be made significantly worse by not applying ADC calibration (we only did this as an all or nothing test, we did not try the individual components.)
  6. The problem does not appear to be present in the GBT spectrometer. Differences from VEGAS are:
    1. The Spectrometer is fed by the "Analog Filter Rack" after the Converter Rack, as opposed to the VEGAS IF modules
    2. The Spectrometer (at least in wide-band mode) has 3-level samplers
    3. The Spectrometer samples in the second Nyquist zone (I think)
    4. Anything else?
  7. The problem does not appear to be GBT receiver specific, implying it is somewhere in the common IF system.
  8. We have quite thoroughly (but not recently) investigated and ruled out "housekeeping" errors (such as miscalculating exposure times).
  9. We have not recently looked closely at the behavior of spectra produced by off-line FFTs of the ADC or filtersnap snap blocks.

Hypotheses (speculations?) as to cause

  1. These is something "nasty" in the GBT IF, which is not seen by the GBT spectrometer, and is not grossly obvious in ADC total power samples.
  2. This effect either
    1. varies wildly from IF to IF, or
    2. the interaction between this effect and VEGAS varies wildly from IF to IF.
  3. (4) and (5) above might indicate that hypothesis (2b) is more likely than hypothesis (2a).
  4. Possible causes of this might be:
    1. Some interaction with the ADC "DC" spikes
    2. some processing error (e.g. truncation or overflow or something) in the FPGA / HPC code

Potential tests

  1. Look very carefully at the VEGAS IF Module output and GBT Spectrometer IF inputs, over at least DC to a few GHz, and look for discrepancies.
  2. split the signal going into the GBT Spectrometer, and feed one copy into:
    1. VEGAS IF module
    2. directly into ADCs.
      Above may require lab components such as amplifiers / attenuators / filters 3. Look much more closely at spectra derived manually offline from the ADC / filtersnap snapblocks, for existing or new data.
  3. Do manual (tp_nocal) off/(on-off) observations with VEGAS detuned. Look at results as calculated from the ADCs
  4. Isolate different aspects of ADC calibration (e.g. just MMCM calibration, just OGP calibration).
  5. What else?

Daily Logs

Monday 13th April 2015

Tiger Team Meeting 2:30pm

  1. Agree on list of "knowns", add other facts if possible
  2. Brainstorm above and additional hypotheses
  3. Define additional tests / data reduction steps which may help further isolate the problem, or distinguish between hypotheses
  4. Define needed test equipment: GBT Spectrometer working, cables, filters...
  5. Fully specify individual tests.
  6. Agree order and approximate time requirements for planned tests
  7. Identify personnel to perform tests, and process data
  8. Decide whether we need Wednesday / Thursday
  9. Any other business

-- RichardPrestage - 2015-04-13
Topic revision: r3 - 2015-04-13, RichardPrestage
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding NRAO Public Wiki? Send feedback