VEGAS Bank Discrepancy Problem - 2015 March 21

For earlier discussions, see VegasBankDiscrepancyProblem2015Jan30, and subsequent links

Expected VEGAS Astronomical Operation

When connected to the same receiver, feed, and polarization, with the same center frequency and bandwidth, and operated at "reasonable" input power levels, we expect that the system temperatures (Tsys) and line antenna temperatures (Ta) as determined from each individual bank of VEGAS should produced identical results,to within ~ 1%. These results should also be consistent with our previous backends (specifically the GBT Spectrometer). The consistency between banks specification is based upon the equivalent bank-to-bank results obtained by the GBT Spectrometer (which shares the same IF path up to and including the GBT Converter Modules), while the consistency with previous backends is required since these have been extensively tested and demonstrated to be correct via detailed astronomical commissioning and science observations.

Currently it is possible for the Tsys and Ta calculations performed by GBTIDL on the spectra produced by the VEGAS HPC system to vary by more than 10% from bank to bank.

Internal VEGAS Consistency Properties

  • For the same lbw_gain, and reasonable input power levels, the ratios of ADC inband total power, filtersnap total power, and HPC spectrum total power should be identical for every bank.
  • Above should also be true for bandpass shapes within a single bank.
  • Bandpass shapes may differ somewhat from bank to bank, due to differences in the VEGAS analog IF modules. But, the ratios of different input powers should be flat for all banks, and for each bank the ratio of two successive scans should exactly match the change in the input power levels.
  • For the same frequency range, spectra produced by different VEGAS modes (and in particular in comparison with Modes 1,2,3, which use a simpler chain of digital signal processing firmware / software) should be identical, apart perhaps from digital filter shapes used in modes 4 and up.

Status of recent concerns

  • The storage of the ADC, filtersnap_nogain and filtersnap data samples in the Archivist FITS files must be correct. This can be determined by looking at the statistics of the data in the Archivist FITS files for inputs of random Gaussian noise, and/or by comparing the values stored in the FITS files with values independently captured directly from the Roach boards.
    • This was found to be a bug in the FITS Archivist when writing arrays of ints (it was writing them as longs). Will be patched during maintenance Tuesday 10th March.
  • The total power results calculated by the Manager for both real and complex input data samples must be correct (although these values are not strictly needed for commissioning tests, they have been used in the past).
    • This will be corrected by Ray during maintenance on Tuesday 10th March.
  • We need to confirm that the correct polarization (XX or YY) products being produced by GBTIDL are being compared with the same polarization snapblock values from the Roach.
    • Not clear whether this is in SDFITS, or within the GBTIDL program. Within GBTIDL, can check the actual polarizations corresponding to the PLNUMs. Should be self-consistent for data taken straight from the Manager.

Potential (or repeated) Commissioning Tests

Test equipment:
  • BOF 123
  • New wideband noise source and computer controlled attenuator connected via an eight-way splitter into 4 banks x two pols
  • Can we completely configure VEGAS without Astrid? Can we use a second invocation of the config_tool? How do we know what parameters to set?
    • The "real" VEGAS hardware can be made part of the simulation environment, allowing use of the config tool.
  • C-band receiver with high Tcal noise diode (previously, we were using K-band, which only has a low Tcal noise diode).
  • even when testing with the noise source, VEGAS should be configured as for astronomical observations, to ensure all parameters are set correctly.
  • __remember to set lbw_gain to be the same between all banks.
    • Ray can add the storage of this parameter to the VEGAS FITS file (by adding it to shared memory).

  • We would still like the Manager to be modified to allow simultaneous triggering of all three varieties of snapblock.

  • We need to check that we understand how the GBTIDL Tcal values are derived, and that these change correctly from mode to mode.

1. Manager / Archivist snapblocks / total power values

  • With nominal inputs, use the Manager to capture some snapblocks (all three types). Capture some corresponding snapblocks with a python script. Compare the statistics of the two data-sets.
  • For "good" archived snapblocks, calculate the total power offline (using a known correct algorithm), and compare to the Manager-calculated values. These should be identical to within numerical precision.
  • We believe these problems should be resolved now, but still prodent to check.
  • We can capture the identical snaps with the Manager and a python script, by setting the Manager snapblock update rate to be very slow.
    • Richard to configure VEGAS
    • Ray to capture snapblocks
    • Ray / Dave to compare statistics, and calculated total power values.

2. Digital Gain in L1 BOF - Resolved

  • Randy expects this to need to be around 2**25.
  • Current value (Manager parameter lbw_gain) needs to be around 2**27 to have filtersnap measpwr equal to ADC measpwr - but filtersnap measpwr values may be being calculated incorrectly?
  • Randy can account for the two extra powers of two observed; one from some digital processing performed before the gain block, and one since the gain is applied to both real and imaginary values.

3. Total power and spectrum ratios ADC / filtersnap / HPC spectra

  • For reasonable inputs, the ratios of these should be indentical from bank to bank.
  • Glenn has produced some plots which appear to confirm this, but Adam cannot replicate these results.
  • Once (1) and (2) above are complete, capture snapblocks and total-power no-cal spectra for each bank on noise source
  • Compare ratios for consistency
  • Remember to make sure like pols are being compared with like
    • Richard to run test
    • Glenn and Adam to independently process the data; should get identical results.

4. Identical ratios for varying input power.

  • Ratios of successive scans done with different input powers should correspond to the change in input (total power and spectrum at all three stages).
  • Use noise source through eight-way splitter as test source.
  • capture snapblocks and total-power no-cal spectra with nominal input level.
  • Obtain successive scans with +6dB, +3dB, 0, -3dB, -6dB input powers (using computer controlled attenuator).
  • Form ratios and confirm consistency
    • Richard to run test
    • Adam to process data

5. Above test for different inputs

  • Perform test (4) with noise source connected directly to ADCs; noise source connected to input of VEGAS IF module, C-band receiver
  • Should get identical results in all cases.
    • Richard to perform the test
    • Adam to process the data.

6. Identical Tsys values from bank to bank

  • Use X-band receiver with nominal input levels, since C-band Tcal values are suspect.
  • execute total-power with cal observations; perform a series of observations in total-power no-cal mode, switching cal on and off manually.
  • bracket this with identical observations performed in mode 1.
  • Use GBTIDL to calculate Tsys, and perform the same calculation manually for separate scans.
  • Can also run standard total power with cal observations, and "slow switched cals" driven by the DCR, and compare all three approaches.
  • Compare consistency of results
    • Richard to run test
    • Adam to process data

7. Spectral lines should have identical intensities from bank to bank.

  • Configure for X-band RRL lines; with identical inputs going into each bank.
  • Perform position-switched observations, on a known bright RRL source
  • Compare consistency of results
    • Richard to run test
    • Adam to process results.

-- RichardPrestage - 2015-03-07
Topic revision: r1 - 2015-03-21, 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