[evlatests] Aliasing in narrow-BW observations

[Jim Jackson]

Rick,

I'll put on both systems engineering and project management hats to 
answer this.  I spoke with Mike Revnell about this as soon as we saw 
your message.  We are both convinced the problem is almost certainly 
with the FIR filters implemented in the D351 deformatters. The 
solution would likely be a more complex filter in a larger FPGA. A 
larger FPGA means redesigning the deformatter which would obsolete 
the roughly half project worth of them we've already built as well as 
the expensive FPGA's we already bought for the rest of them.  It 
would likely also involve about a year of Mike's time to build it and 
possibly mean descoping a receiver from the project to pay for it.  I 
think we will need to find a way to live with this until the WIDAR 
comes along or find a way to correct for it in post processing (if 
that's possible).

Jim


At 04:43 PM 10/9/2007, Rick Perley wrote:
>     Miller Goss ran a spectral line test last Thursday which has
>revealed an heretofore unsuspected problem which has significant
>consequences for 'transition' observations.
>
>     In his test, Miller observed in a very narrow bandwidth -- 780 kHz,
>in mode '2AC', producing 256 spectral channels for each of the two
>correlations.  After generating the bandpasses (which looked entirely
>normal),  he soon found that all, and only, the EVLA-EVLA baselines had
>large offsets in their bandpass-corrected visibility amplitudes.
>'Large' here means of order 20% or more.  VLA to VLA baselines were
>fine, as were EVLA to VLA baselines.
>
>     There are only two ways that I could imagine this happening --
>either there is RFI in EVLA antennas, which only correlates between
>them, or there is some lower level broadband signal which is present in
>EVLA antennas only.
>
>     The first hypothesis was easily dismissed, as all the spectra are
>clean.  There is no RFI.
>
>     For the 2nd hypothesis to be viable, it has to only affect narrow
>bandwidth observations, as numerous (and nearly innumerable) tests and
>observations have been made at wider bandwidths, without any hint of an
>EVLA-EVLA extraneous correlation.  From this consideration, it was
>suggested that perhaps we are seeing some aliasing around the baseband
>end, which is present only in EVLA antennas, and which is significant
>only in the bottom ~500 kHz of so of the passband.
>
>     A test for this was executed this afternoon, and has clearly shown
>that this is the origin of the large offsets seen in Miller's data.  In
>this test, I observed a strong calibrator in mode '1A', at BW = 3 MHz.
>(Other BW were observed also -- and some other problems were found -- to
>be reported on later).  There are then 256 channels across the spectrum,
>for  12 kHz resolution.  Good-looking bandpass solutions were found.
>Bandpass-calibrated correlator spectra were then generated, which
>clearly shows the extent of the problem:
>        Extending up from baseband, for about 700 kHz, there is a large
>(up to a factor of two!) additional correlation in the signal.  Both
>amplitude and phase are affected, but the effect is much stronger in
>amplitude.  Both 'positive' and 'negative' additions are seen.  Beyond
>about 1 MHz from the baseband end, the effect is no longer detectable in
>any of the individual spectra.
>
>     This is a serious 'transition' issue.  The effect does not originate
>in the EVLA's antenna or DTS electronics, nor in the VLA's baseband
>filters.  It must be due to aliasing in the digital-to-analog conversion
>system we use to enable EVLA signals to be used for science observing
>(and tests!).   Unlike most 'transition' issues, this one will not
>lessen in time -- it grows with the number of EVLA to EVLA pairs we have
>-- and will not be eliminated until the new correlator arrives.
>     I believe the effect is time-variable (based on a small amount of
>data), so a simple 'closure' correction will not be easy, and perhaps
>not viable at all.  It is perhaps important to measure this variability,
>to see if it exists, and what its characteristics are.
>     This aliasing effect is small for wide-band continuum observations,
>and for relatively wideband spectral line observations, one can always
>choose to remove the affected channels on EVLA-EVLA baselines.
>     The effect is quite serious for narrow BW spectral observations --
>anything taken at BW less than ~3 MHz will be notably degraded.
>
>
>
>_______________________________________________
>evlatests mailing list
>evlatests at listmgr.cv.nrao.edu
>http://listmgr.cv.nrao.edu/mailman/listinfo/evlatests