In aperture synthesis, the angular resolution scales with the frequency: in fact the whole Fourier plane scales with
this frequency, so that the angular scaled defined by the baselines varies across the frequency coverage. Accordingly,
when using sufficiently wide bandwidths (and/or imaging sufficiently large areas), it becomes important to account
correctly for this effect. As a result, beams (whether primary or synthesized)
can be 4-D arrays, as they may depend on Frequency and Field (for Mosaics).
The deconvolution implementation of
IMAGER in the
UV_MAP and
CLEAN commands is designed such that ensembles of contiguous channels
(“chunks”) are treated at once and share the same synthesized beam.
Deconvolution with
CLEAN then proceeds by using the synthesized
beam with the appropriate frequency for each channel. The user can
control the “chunk” size, and hence the precision of the process
given the desired field of view.
The user control is done through
BEAM_STEP.
-
BEAM_STEP = 0 instructs
IMAGER to try to produce a single synthesized beam for all channels. This is
appropriate for narrow bands and spectral channels with very similar noise (ideally identical). Unavoidably, the
synthesized beam is only an approximation, exact only at the reference frequency.
-
BEAM_STEP = N instructs
IMAGER to ensure that N consecutive channels share the
same synthesized beam. This can be used when channels have the same weights (same effective UV coverage) but
the overall bandwidth is too large: the number of image channels being M, the number of different beams
will be M/N.
-
BEAM_STEP = -1 instructs
IMAGER to derive the number N of consecutive channels
that can share the same beam to within a certain precision in the imaged field of view. This precision
is given by
MAP_PRECIS. The derived N will depend of this number, on the relative bandwidth and on
the size of the synthesized field (hence
MAP_FIELD or
MAP_CELL
\(\times\)
MAP_SIZE).
The above works in an optimal way in all cases where all channels have the same weight distribution.
This is in general true for NOEMA data coming from
CLIC, but unfortunately,
this may not always be the case. Common cases where this does not occur are:
- Band edges, where the noise may increase
- Data resulting from concatenation in frequency of different spectral windows
- Cases where the system temperature is highly frequency dependent (and carried along in the weights)
UV_CHECK
BEAMS will verify the channel ranges that share the same weights
and return in
BEAM_RANGES the range boundaries.
UV_MAP uses this
information in an “intelligent” way: band edges will be dropped if needed, and small regions
(up to
BEAM_GAP, default 3) with different weights are ignored in this process.
In most cases, this behaviour allows a common beam for all channels, or an automatic guess of N if
BEAM_STEP = -1.
If this fails,
UV_MAP will complain and stop with an error. The user must then adjust
BEAM_STEP to either
-
BEAM_STEP = 1, one beam per channel, mimicking to some extent the behaviour of CASA,
at the expense of speed and memory/disk use (twice more needed)
-
BEAM_STEP = -2, forcing
IMAGER to produce a single synthesized beam,
but at the expense of some sensitivity loss in channels whose weight do not match the reference one,
the so-called weight channel
WCOL.
Subsections
