VEGAS Bank Discrepancy Problem - 24 January 2015
More attempts to diagnose the cause of this problem, after the requantization balancing fix.
Measurements were made using Astrid to configure the system, so all configurations should have been self-consistent. Could not use L-band, since staff were working on the MUSTANG rotator, so had to use KFPA. The antenna was tipped in elevation a couple of times during the day, during which time the input power was slowly varying during a measurement.
KFPA was configured to send "identical" signals to both polarizations of Banks E,F,G,H (same banks as used for the noise tests).
The Astrid "Balance()" utility was used to step the converter module attenuators across their whole usable range. This resulted in nominal input levels at the VEGAS ADCs ranging from -36dBm to -6dBm, in steps of 2dB (normal target is -20dBm). Note: the balancing didn't always work correctly, so some data were taken with the input powers not at the nominal value.
This should be obvious in the input adcpwr histograms, however, and I don't think it was a major problem.
I used pyfits and MATPLOTLIB to read the first snapblock (16k samples) of pol 1 of Bank E for each scan (different mode / input power levels).
- In all cases, I read "adcpwr1"
- For the L1 modes, I read the "filtersnap_nogain" snapblocks, i.e. before the rebalancing
- For the L8 modes, I read the sb1snap snapblocks, i.e. after the rebalancing - there does not seem to be a corresponding "nogain" snapblock for these modes.
- all snapblock values were written as unsigned bytes in the FITS file. I converted them to signed values before performing the histograms.
All of the plots are here
- For the L1 modes, regardless of input power level, the filtersnap_nogain values have no dynamic range; essentially all of the counts are 0, with only a few outliers.
- For quite a few modes with a reasonable spread of input power ADC values, there seem to be "drop-outs", or specific count values which have either no, or a low number of counts (I suppose this could be a MATPLOTLIB issue, although that seems highly unlikely.
- For all modes, the real and imaginary histograms look quite non-Guassian; should they be Gaussian?
- The BOF files are broken...