[evlatests] Science-Engineering meeting agenda

[Barry Clark]

A couple of remarks about reducing effect of subband boundaries
by offsetting a few channels.

First, when making the final spectrum, appropriate weighting is
really important.  Example:  Three offsets.  If you do bandpass
correction and average, then in the middle of the subband, you
improve the SNR by 3^0.5 = 1.72.  For the channel including
channel 0 of the subband, which has SNR worse than the other
two by a factor of two, the SNR is improved by 3/(6)^0.5 = 1.22.
If that spectrum is weighted down by a factor of 4, the SNR
is improved by (2.25)/(2.5)^0.5 = 1.42.  So, the loss relative
to midband is 1.41 in the first, unweighted case, and only 1.22
in the second.

The proper weighting can be calculated from the well known filter
shape (it is the inverse square of one plus the ratio of aliased
power to filter response).  A very crude approximation is
weighting by the bandpass correction.

Second, there is a small (nonclosing!) loss of coherence in the end
channels due to fshift.  Standard setup is fshift-base = 25.6 kHz,
maximum fshift of 819.2 kHz, about 0.4 channel.  If you are going
to use the very end channels of the spectrum, I'd recommend cutting
this by half, to 12.8 kHz base.  (Or even further if you are using
baseline board stacking.)

On 05/08/2013 04:47 PM, Rick Perley wrote:
>      Reminder:
>
>      10AM, Rm 317.
>
>      Topics:
>
>      1) A way of eliminating the 128-MHz subband boundaries.
> Mike Revnell
>
>      2) How observers avoid damage due to these boundaries (without
> software changes)           Chris Carilli
>
>      3) Status of switched power gain correction -- and why we *must*
> regularly apply these data.   Rick Perley
>
>
> _______________________________________________
> evlatests mailing list
> evlatests at listmgr.cv.nrao.edu
> http://listmgr.cv.nrao.edu/mailman/listinfo/evlatests