VEGAS Bank Discrepancy Problem
It has been observed that VEGAS measured Tsys values and line intensities can vary by ~ 10% between Banks, even when they are configured to look at the same signal. This should not be the case.
The problem
The initial evidence for the problem came from spectral line measurements of RRL lines. Both the system temperatures and the line intensities were observed to vary by peak-to-peak ~ 10 percent between identically configured bands. In an attempt to isolate the problem, we tried connecting the noise source to both polatizatiions of banks E,F,G and H. We then recorded input power snapblocks (adcpwr1,2) and output snapblocks (filtersnap). We changed the noise source gain by 1db, and divided the results from one scan by the results from the next scan. Here is the result:
As can be seen, although the input power levels change by 1 dB as expected, the "output" power levels only change by around 0.8dB. This discrepancy gets larger as the input power ratio gets larger.
Also, although statistically this may not be very significant, it seems like the output power levels vary from bank to bank.
This is the problem we are trying to solve.
Proposed Tests
- repeat noise source tests in the simulator with new BOF files, capturing filtersnap_nogain as well as filtersnap
- Look at the normality of the output filtersnap real and imaginary values
- Look at fast-dump spectra on the noise source, and see whether the passband shape varies from dump to dump
Experimental results we are putting to one side for now.
"Clumpy" ADC values
I plotted the variance of all of the measured adcpwr snapblocks (as a function of "normality" probability, but that is not relevant here). The results seemed to "clump" into two range of values.
Earlier attempts to diagnose this:
--
RichardPrestage - 2015-01-14