[evla-sw-discuss] OTF Mosaicing

[Barry Clark]

ALMA memo 331 discusses ways of doing on-the-fly mosaicing.  It
considers three possible techniques:
a.  The phase pointing center is changed for each integration, whereas
the antenna pointing changes smoothly.  The antenna is pointed at the
phase tracking center in the middle of the integration.
I shall refer to this as the saltating phase center method.
b.  The phase tracking center moves smoothly on the sky, along with the
antenna beam.
I shall refer to this as the gliding phase center method.
c.  The phase tracking center remains fixed on the sky while the antenna 
moves by many beams.
I shall refer to this as the fixed phase center method.

The fixed phase center method has a number of variants, such as keeping
the phase center fixed, but stepping the delay tracking center so that 
wider channels may be used, and implementing some processing in the 
correlator backend, to take out gross fringe rate changes and enable
integrating to longer times.  I shall ignore the variants.

Memo 331 quite rightly rejects the gliding phase center method; the 
required correlator output data rates are simply absurd.

Memo 331 complains about the difficulty of implementing the saltating
phase center method, because phase rotation is done in analog at a site
distant both physically and conceptually distant from where correlator
integration timing is determined.  This does not apply to the EVLA -
both phase rotation and integration dump times are determined on the 
station board, within inches, and very close conceptually.

The fixed phase center method requires coding of new data reduction
algorithms, whereas the saltating method can use current mosaicing code
with little change.

The requirements for integration times are different for the fix phase
center method and the saltating phase center method.  Consider a OFT
scan line of length N half power beamwidths, and a scanning speed of 
S beams per second (roughly 5*S*lambda times sidereal, lambda in cm for the
VLA).  For the EVLA case, saltating phase center method, the required 
integration time is the lessor of 0.5/S (for half beamwidth spacing), and
t0, the configuration dependent time for full beam mapping, about 3 seconds
for A configuration.  The fixed phase center method requires an integration 
time of the lessor of 0.5/S or t0/N. This integration time consideration 
is important because the limitation on output data rate is a critical 
constriction which forces us to make integration times as long as we can.

Since the integration time is never shorter for the fixed phase center
method, and the technical difficulties cited for ALMA do not apply to
the EVLA, it seems clear that the saltating phase center method is the
one to use.  The implication is that new delay models have to be calculated
once per integration, for the new phase tracking center.

How far the phase tracking center should jump, in the saltating case,
depends to some extent on the dynamic range desired.  The dynamic
range may be limited by non-linear effects due to the antenna beam
moving during the integration.  For a high dynamic range, I would 
expect that a jump substantially shorter than the "Nyquist" half beamwidth 
would be advantageous.  If a high dynamic range is not needed, it may
be possible to go a bit beyond the half beamwidth spacing without 
serious loss of SNR.

At the higher frequencies, the speed of mosaicing is likely to be set
by the permitted data output rates.  The project book speaks of 
mosaicing at ten times sidereal rate, 2.5 arcminutes/sec.  With
this mosaicing speed, at Q band, with half beamwidth spacing, this
translates to 0.2 second integrations, which, with a 25 Mbit/sec output
data rate and 10 bytes per channel, limits you to about 1000 channels.  
It is likely that in many cases one would choose a slower speed and 
larger number of channels.