[evlatests] Troubles with WIDAR-0

[Ken Sowinski]

On Fri, 5 Jun 2009, Rick Perley wrote:

>    I made images of both the noise field, and the two calibrators, from
> the WIDAR-0 5-hour database taken last night.  The issues found
> yesterday in the northern source are evident in the maps of the two
> calibrators from last night.  However, the images of the noise field
> look fine.
>
>    To better clarify things, I extracted six databases for each of the
> three sources (two calibrators, one noise field).  The 6 databases  have
> differing bandwidth, spaced by factors of four:  0.125, 0.5, 2.0, 8.0,
> 32.0 and 128 MHz.
>
>    The histograms of nearly all databases look normal, in both real and
> imaginary parts, with the expected relationship between noise and BW.
> However, there are excess counts in the tails of the histograms for the
> calibrators (to be explained in a moment).
>
>     The images tell a very different story.  Images of the noise field
> were correct, and declined as expected until the widest bandwidths,
> where edge effects can be expected.
>    Images of the calibrator are uniformly terrible, at all bandwidths,
> with the background dominated by 'lumps and bumps' which are certainly
> not real.  The rms noise has no relationship with BW, until the widest
> bandwidth, where it is twice the value of any other image!
>    These effects are due to large 'closure' errors, which are quite
> evident in the visibility plots of the strong calibrator.
>
>    So clearly something is wrong.  But whatever it is, it affects only
> the correlated signal, and was not present in the PTC.
>
>    Ken reports that Martin has a theory, which can be tested next
> week.  I'll be in Pasadena, so somebody will have to step forwards to
> work with this.

We are now using multiple backend processes to write the disk
files and suppose that interactions in the file system may cause
the data to be wrongly represented.  As a test of this hypothesis
we plan to take an hour or so of data Monday morning with the
backend restricted to use only one process.  Since Rick will be
away next week we need a volunteer to examine the result of this
experiment.