The data referred to in the last report were analyzed for its
bandpass stability characteristics.
Summary:
The ~4 MHz ripple which is the dominant source of bandpass
instability for VLA antennas is completely absent in EVLA antennas (as
expected). The stability of EVLA bandpasses, over frequency scales of a
few MHz, on time scales of tens of minutes, is about 0.1% in amplitude,
and 0.1 degrees in phase (pk - pk) -- a factor ~20 larger than the PB
requirements. However, when the 4 MHz ripple present in the VLA
antennas is removed through careful calibration, the residual stability
for VLA antennas is found to be the same as the EVLA antennas, leaving
open the possibility the instability origin is within electronics common
to the two systems. The WIDAR prototype correlator will be needed to
make significant further progress on this requirement.
Details:
The PB requirements call for a bandpass stability of 1e-4 in power,
and 0.006 degrees, over timescales of an hour, and on frequency scales
less than the RF frequency/1000. For X-band, this frequency scale is ~8
MHz. To explore our current stability, I observed a strong source with
a BW of 25 MHz, and channel resolution of 0.8 MHz, for 4.5 hours.
The variations in the bandpass are what interest us, so I removed
the fundamental bandpass shape by finding the mean bandpass shape for
the entire 4.5 hour run. The AIPS program BPASS was then applied to
find the time variation of the bandpass w.r.t. this mean.
The results clearly show the dominant VLA 4 MHz ripple. Its
characteristics are:
1) Frequency scale is close to 4.0 MHz.
2) The amplitude of the ripple varies widely between VLA antennas --
pk-pk ripple maximum is about 1.5% on antennas 1 and 12, but nearly
invisible on others, where the amplitude is at least a factor of 5 less.
3) The phase scale is maximum of about 1 degree on antennas 1 and
12, less on others.
4) The pattern changes amplitude in time -- inverting its amplitude
and phase with a period of about 2 hours. The frequency of max and min
did not change in this experiment, but in a previous one (done with 12
MHz BW), the peaks/troughs clearly moved in time.
The amplitude and phase changes for EVLA antennas are clearly much
better, with no sign of the ripple. The bandpasses computed once every
30 minutes show variations with a maximum amplitude of 0.1%, and phase
excursions of about 0.1 degree -- about 20 times too high (if the PB
requirements are interpreted as peak-peak, rather than some sort of rms
over a good part of the spectrum).
The structure of the EVLA Bandpass variations is quite different
than the 4 MHz ripple. They are typically on a much longer frequency
scale -- ~15 MHz seems dominant, but scales down to a few MHz are also
visible.
The PB requirements state that the variations are to be determined
over an hourly timescale -- this presumes that the observer who wants
super-accurate determinations of weak absorption or emission will
calibrate on a strong bandpass source on that timescale. To mimic the
effect of this, I formed a bandpass solution on an hourly interval, then
determined the differential bandpass after interpolating the hourly
solution.
The result of this procedure was that the VLA ripple nearly
disappeared, and the resultant spectral shapes have the same shape and
variations as the EVLA bandpasses (either with or without the hourly
solutions).
The similarity of the bandpass variations between the EVLA and VLA,
after hourly calibration has removed the dominant VLA ripple, suggests
a common origin. This could be the receiver itself, or in the backend
analog electronics. We'll have to await the WIDAR prototype
correlator to be able to make further progress (unless somebody can
think of a better test).