attenuated version of the minutes: the postscript files have been
deleted - with a reference to their location.
--------------------------------------------------------------------
ca-forum #8 (6 April 1995) -- MINUTES
1. Ghost images. (R. Sault). see appendix A for background on the problem.
The symptom is a weak replica (~0.5%) of the true image, reflected through the
phase centre. In principle, then, our dynamic range will have been
limited to this order - say 200:1; we gain some relief if the observations
require phase calibration - this sharpens the image and dilutes the ghost.
Bob has developed a lag-spectrum weighting scheme which defeats the
ghosting. It has the effect of further down-weighting the amplitude
at large delays. Thus there may be a problem with wide field
imaging - and it may affect mosaicing at L-band.
Bob's weighting scheme is defined operationally - it produces the
lowest sidelobe content within the central channels of the frequency
spectrum. - see appendix A2. It results in an increase of the
effective channel width - ~ 10MHz.
recommendations.
a. Correlator configuration files.
Warwick points out:
- one could concatenate 2 modules at 128 MHz bandwidth to
obtain a lag window which is flat over from zero to ~channel 16, then
tapered out to channel 48.
- one could reconfigure a single module at 128 MHz to have a flat
central region, then a steeper taper.
- wide-field imagers should consider using a 64 MHz bandwidth.
b. Bob will examine the consequences of the anti-ghost weighting scheme
on wide-field imaging.
2. Mosaicing.
Some further refinements are (still) required before we can mosaic the
Cen A region. The problem is that the Tsys changes enough that the
sampler levels need to be reset for many fields - ie, separate environments
are needed for each field. In addition, the attenuator settings also
differ between fields.
Neil outlined a scheme which will overcome most of these difficulties
without requiring a large software investment.
- Every pointing centre of the mosaic will be assigned a different
environment number by CAOBS.
- At the end of every cycle the ACC will read the sampler statistics,
compute the new sampler levels. These levels will be assigned to the
environment of the cycle just completed.
- Before a cycle begins the sampler settings will be read from the
(ACC) database and loaded into the samplers. That is, the environment
number of the upcoming cycle will define the sampler levels to be
used.
- In normal observing where the pointing centre doesn't change
the environment number doesn't change either; there will be
a slight increase in the traffic to the database, but otherwise no
observable change.
[**** Lister points out that this strategy may be undesirable:
if the temporal variations are greater than the spatial variations
then the auto-sequencing of the environment number should be disabled. ***]
A rapid AGC in the samplers will obviate these stratagems, but there
are no concrete plans for such right now.
- CAOBS will associate an attenuator setting with each environment
and request the appropriate setting before the cycle starts. For the
next year or so the data at the end of a cycle may be compromised
by the attenuator step: the present hardware does not allow the
change to occur in response to an event, but rather it happens when
the message arrives at the dataset. The hardware revision is in
George Graves' charge, and will not arrive for some while.
Recommendations.
MJK will modify CAOBS as required, to be tested towards the end
of the april shutdown. Some ACC mods. may be needed, but it is
hoped that these can be kept to a minimum.
3. Monitor capabilities and strategies.
Derek has provided a summary of the current state of affairs in
the narrabri home page of the WWW - The narrabri implementation
is a good example of the kind (far better than the epping attempt),
and should be consulted. Its URL:
Little software effort will be available in the coming year to
make serious changes, given the impending upgrade to the parkes
control system. It is important to highlight now any gross deficiencies.
Russl Gough argued that the maintenance of the cryogenic systems
benefited greatly from long term (1-2 year) displays .. thus he
asked that the ability to produce plots covering long time spans
be given some consideration - not be lost (with changes to the
archive formats), and made easier.
----------------------------------------------------------------
Appendix A - the ghost problem.
Three items: A1 - Bob's description of the problem.
A2 - plots of several taper functions (postscript file)
A3 - example of "before" and "after" (postscript file)
-------
A1 -- The problem
ATNF ATCA DATA INFORMATION REPORT #17
Description: When observing in continuum mode, each real source produces a ghost
source at the 0.1% -- 0.5% level positioned diametrically opposite the phase
centre.
Date of affected data: all 33-channel/128-MHz data loaded before xxxxx
Problem in Detail: XF correlators, such as the ATCA's, inherently suffer
from a problem known as the Gibbs phenomena (see Albert Bos, in "Image
Formation from Coherence Functions in Astronomy"). This is caused by
a discontinuity in the visibility spectra at zero frequency in the baseband
signal. As this discontinuity is different for the real and imaginary parts of
the signal, the resultant spectral sidelobes in the real and imaginary parts
of a spectrum are also different. This results in an effective difference in
the gain between the real and imaginary parts of a correlation. The triangular
lag weighting function used for the continuum system has the undesirable
property that this error does not average out across the visibility spectrum
(for spectral line correlator configurations it will tend to). In the
image plane, the error manifests itself as ghost sources, which are located
on the opposite side of the phase centre to true sources.
The real/imaginary gain different is small (~1%) away from the band edges,
and so the ghost sources are also weak. Like the true source,
antenna phase errors will cause ghost sources to decorrelate. Unlike
true sources, ghost sources are decorrelated further by the calibration
process (primary calibration or self-calibration). The net result is that
the ghost sources are only apparent in high dynamic range images, where
observation conditions have been moderately good (i.e. no antenna phase to
cause decorrelation). Typically the ghost sources will not be apparent
in images with dynamic ranges less than a few hundred to a thousand. The
problem is likely to be more pronounced in more recent data, because of
general improvements in telescope stability (e.g. LO phase stabilisation).
Solution: Although solvable off-line, the correction must be performed at
the very start of the reduction process (whereas the problem will not
normally become apparent until the end). Using the option
`reweight' with Miriad's atlod task reweights the visibility spectrum in the
lag domain in a fashion which eliminates the problem. This option also
prevents a given narrowband interferencing signal corrupting more than one
channel on each side. For data already loaded, Miriad's atxy task also has a
reweight option. This correction must be applied before any calibration
(including XY phase calibration), frequency averaging or channel selection
in performed.
Signed: Bob Sault (24-Mar-95)
--------
A2 - some examples of taper functions.
see ~rsault/gibbs/taper.ps
A3 - Before and After
see ~rsault/gibbs/ghosts.ps