CA Problem Forum #1
May 18, 1994
Mosaicing.
What follows is a loose paraphrase/summary of the discussion.
This mode of observing was invented to circumvent the 2-cycle
start-up associated with each scan. In Mosaic's original manifestation
the antenna was expected to slew a small distance between pointing
centres - 2 seconds was the typical figure mooted. The antenna
was expected to pause for just a few cycles at each centre.
The present form of "mosaicing" is a complicated kluge which evolved
from that simple model. A mosaic file contains a list of pointing
centres: the coordinates of each centre and the number of integration
cycles (which may be different for each centre).
This scheme works, but poorly - a number of defects have surfaced.
It was early recognised that some mechanism was required to cope
with the fact that the antenna may not have reached its target
when the integration cycle starts. This lead to the HOLD, a
blanking signal used by the correlator on the first cycle of each
centre. This is the same for all centres specified in the mosaic
file, and has to be a pessimistic estimate. If it is too small
then the entire cycle will be flagged as bad since the antennas will
not be ON-SOURCE when the correlator data-taking starts.
The problems:
a. HOLD is not applied out at the antenna. This means that Tsys
is wrong for the first cycle when slewing to a strong source.
It may also mean that the sampler statistics are wrong.
b. Indeed, given that several cycles are required by the
samplers to adjust to a new power level, they could conceivably
be incorrect for the entire mosaic if the field is complex,
and the number of cycles/centre is small.
c. A single HOLD time for all pointing centres, for all time
is not appropriate - greater flexibility is needed.
d. Scheduling is difficult, in that SCHED knows nothing about
mosaics (so the user gets no help in setting up his schedule),
and the drive-time algorithm in sched (and CAOBS) is inadequate.
Bottom line:
The hardware problems mean that the some visibilities will be incorrectly
scaled. A software palliative is possible.
The scheduling inflexibility leads to observing inefficiency and
(one presumes) observer irritation and frustration.
-----------------------------------------
Some solutions:
A. Flexible HOLD.
The correlator group propose to implement a dynamic HOLD option.
This will be quantised in units of (probably) 1/2 second. A different HOLD
can be requested for every cycle. The request could be sent along
with DELBAT, for example.
This will likely be available after the August shut-down.
B. Dave McConnell will attempt to offer a similar quantised dynamic
HOLD at the antenna.
C. There are available spare event lines between the ACC and the
vertex junction box. Additional wires will be needed between the
junction box and the conversion rack.
D. A digital GTP/SDO unit has been built by WEW's group. The receiver
group have already allocated space in the conversion rack, along with
a dataset to collect the digital data. G.Graves will coordinate with
the correlator group to match the space available with the new boards.
DMcC sees no problem in adding 8 new monitor points.
E. This is a suitable place to raise the problem of the attenuators:
their transitions should be tied to cycle boundaries. G.Graves assures
us that the digital conversion rack interface will resolve this problem.
Perhaps its priority could be raised.
F. The sampler question is much less tractable. The current design
does not lend itself to any easy implementation of a HOLD. The new
design used in the VLBI samplers are probably what is needed, but an
AT design is way in the future.
In the meantime:
NEBK will run some tests to quantify the problem.
It is possible that with careful calibration of the samplers one
could make a reasonable estimate of the required sampler levels
directly from the Tsys. MJK will retrieve some earlier experimental
data bearing on this matter.
G. A useful dynamic HOLD requires a reliable drive-time algorithm. The
recent revision of the ACC servo loop makes this possible. DMcC
will look into this.
H. It does not seem to be a big problem to get CAOBS to compute
the dynamic HOLD once the drive algorithm is verified.
I. SCHED. This could be revised to understand mosaic files and to
assist the user with realistic schedule timing. But perhaps a major
rethink is required. Ie, cure the problem (the 2-cycle overhead),
not the symptom (the escalating complexity).
For example: perhaps we could have just one scan/project, or
one scan/correlator configuration/observer.
A scan would have a preamble (header) listing the sources to be observed
as well as the frequencies. Thereafter the scan would be "mosaic" in form
listing source/frequency/#cycles.
Bottom line:
a. The dynamic HOLD for the correlator is in the pipeline. Hardware and
software effort required to make use of it:
b. Hardware required:
- digital Tsys (WEW/GG)
- strobed attenuators (GG)
c. Software:
- ACC closed-loop slewing. (DMcC)
- ACC drive-time algorithm. (DMcC)
- CAOBS to compute the dynamic HOLD. (MJK/DMcC)
- ACC .. quantised dynamic HOLD. (DMcC)
d. Interim measures
- NEBK/MJK to investigate cost of sampler errors
and possibility of assisting the setting of the
sampler levels.