[evlatests] VLA image -- and an interesting clue ...

[Ken Sowinski]

On Tue, 14 Jul 2009, Michael Rupen wrote:

>> 5. Both the bandpass and gain calibration were applied to all of the data
>> (the gain calibration differently for the two fields as noted below).  I had
>> no trouble calibrating the bulk delay (cf Ed), but I note that the constant
>> delay calibration performed here leaves behind a residual of a few 10ths
>> of a nsec that fluctuates on timescales of a few seconds to minutes.  This
>> residual is nominally too small to explain the freq-dep effects discerned
>> from the images and described below.
>
> This is related to a point Ken has often made: the residual delay clunking
> we see with WIDAR is much less than the residual delay clunking with the
> VLA correlator, where we don't notice them because the dump times are long
> and the frequency resolution is poor.  If delay clunking were a killer we
> would have much worse problems with the VLA correlator.  Ken, could you
> send out a brief summary of this for the VLA compared to the WIDAR correlator?
> I think at this point that would be useful to have written down for all of us
> to ponder.

Not much less, just comparable.  For the VLA the data is sampled at
100 MHz and the delay lines run at 100 MHz so there is a 10 ns
resolution in digital delay.  Delay is further refined by shifting
the phase of the sample clock to effectively produce delay steps
of 10/16 = 0.625 ns.  Thus the phase change introduced by a delay 
step, if all the phase shifts in the sample clock are correct, is
(.625/20)*360 = 11.25 degrees.  In practice getting the samplers
adjusted so that the fine delay steps are uniform is tricky and
it was often imperfect.


For Widar the eight bit sampler runs at 2 GHz and so are the delay
modules, thanks to parallelism.  The 1 GHz baseband is then reduced
to eight 128 MHz subbands with a maximum delay error in each subband
based on the 2 GHZ initial sampling rate.  Thus the phase change
induced by a delay step is (.5/8)*360 = 22.5 degrees, a factor of
two larger than with the old correlater.  If we choose to divide
the 1 GHZ baseband into 16 64 MHz subbands then the result would be 
the same as for the VLA correlator, (.5/16)*360 = 11.25 degrees.