Filterbank Sample Times vs Wobbler Chopping Frequency

    Poor choices of wobbler frequency can result in a significant reduction
in the duty cycle over what one would expect from the wobbler blanking time.
The time required for the wobbler to move from one position to another is
about 40 ms.  If one is chopping at 2 Hz, then the non-blanked time is
210 ms per 250 ms half-cycle, corresponding to a duty cycle of 84%.  The AOS's
have two advantages over the filterbanks:  their sample time is 10 ms and the
controller synchronizes itself with the falling edge of the blanking pulse.  
So with the above wobbler configuration, the AOS's achieve the theoretical duty
cycle since the non-blanked time is a multiple of 10 ms. 

    The filterbanks have a sample time of 50 ms and the controller samples
asynchronously with respect to the subreflector.  The blanking pulse is checked
at the beginning and at the end of a 50 ms sample, and the 50 ms sample is
discarded if blank is found to be high at either time.  At 2.0 Hz, the 
subreflector moves every 5th sample, but the offset between the sample time
end and the chop time is random.  (It is determined by the time elapsed since
the chopper was started and the antenna reached the position of the source.)
For roughly 80% of the subscans (40/50), the time between samples will fall
during the blank time causing two consecutive samples to be discarded.  So
there will only be 150 ms integration during each 250 ms half-cycle, and the
duty cycle will be only 60%.  Since the above chop time happens to be a
multiple of the sample time, if two samples are discarded at one chop, two 
samples will be discarded at every chop.  For the lucky 20% of the subscans,
only one sample is discarded per chop and the duty cycle rises to 80%. 

    The easiest way to increase the duty cycle is to simply slow down the chop
frequency.  Obviously at 1.0 Hz, the duty cycle would be up to 80% (vs the
maximum possible of 92%).  I don't know at what chop frequency the sky/receiver
variations begin to degrade the spectra.  Certainly the Position Switch
frequency of 0.02-0.03 Hz is much worse than the usual Wobble Switch frequency.

    But one can also get better duty cycles by simply changing the chop
frequency slightly.  If the half-cycle time was 270 ms, then the filterbank
discard pattern would be discard 2, 1, 1, 2, 1, 1, ... yielding a duty cycle of
75% (vs 85% maximum).  270 ms corresponds to 1/.54 = 1.85185 Hz.  (Maybe
there's an even better choice?) 

    The filterbank dump period is currently limited by the read out time of the
filters: about 40 ms.  There are three A/D converters, but the software reads
them out serially not in parallel.  John Hughes was going to modify the
software to fix this, and actually did began to to this, but he never debugged
it.  He doesn't even work for Steward Observatory anymore, so someone else
would have to debug it if this is to be fixed.  With the current software, the
dump time could easily be reduced somewhat from the nominal 50 ms, but how much
has never been tested.  There needs to be some safety margin, but we don't know
how much is required.