[mmaimcal] Followup on sideband separation

[Al Wootten]

Folks:

For discussion today.  Apologies for the late distribution; discussion has
been developing on these topics.

A question arose--how does ALMA produce singledish single sideband spectral 
line data from a double sideband receiver.

I thought that the Science Software Requirements cover that.  See
http://almasw.hq.eso.org/almasw/bin/view/SSR/SsrDocuments

The Use Cases state:
"Measuring the sideband gain ratios could be done in the preamble if both sidebands can be measured."
and what is I think the same thing:
"Measuring the sideband gain ratios could be done in an initialization step if both sidebands can be measured."
"User specifies whether to use software sideband suppression or sideband suppression by LO offset."
"The system will allow the user to examine which transitions
could be expected from image sidebands and as a result tune the receiver
appropriately."

The actual requirements are:
2.3-R12 When a band is equipped with a double sideband receiver, it shall be possible to process and
store data from both sidebands using software sideband separation or to store data from a single
sideband using sideband suppression by local oscillator offset.
Priority: 0

3.6.3.3 Astronomical Calibration: Single Dish Data
6.3.3-R1 The Pipeline shall reduce the temperature scale calibration, using sideband ratios determined
from the most recent interferometric calibration. In on-line mode, the results must be made
accessible to the dynamic scheduling system; they must also be made available to later convert
the raw data into temperature scale whenever required.
Priority: 1

------------what does this mean?---------------
I think that software sideband suppression or sideband suppression by LO offset
must both be available.
What is 'software sideband suppression'?  I think that this may be the sort of
deconvolution done by the CSO folks when they produce a spectral survey.  
Herschel plans to implement this.  This option is available to both single
dish and interferometric observers.  Single dish observers benefit in that
during initialization, the sideband gain ratio of the receiver using the
observing setup may be determined.  See C) below

What is LO offset suppression? 
A) Suppression of the Other sideband in DSB receivers. 
Darrel described it:
'Peter's memo on LO switching (draft of 1/18/2007)
http://www.cv.nrao.edu/~awootten/mmaimcal/LO-Offsetting.pdf
included a section (#7, p.10) on "Total Power 
Considerations", where he states:
"... None of the phase switching or LO offsetting techniques 
discussed ... remove spurious signals, improve undesired 
sideband rejection or enable sideband separation for 
autocorrelation spectra. ...".
Peter then envisages an LO offsetting scheme where each total 
power antenna has a different LO offset, such that when total 
power from all antennas is combined, the unwanted sidebands will 
occur at different frequencies in the IF and so smear out.  This 
is exactly equivalent to the "sideband smear" operation, where 
we stepped the LOs at the 12 M as a function of time.  So, some 
version of "sideband smear", whether as a function of time or as 
a function of antenna, would probably help in some cases.  Peter 
ends up suggesting that science simulations might help decide if 
this would be good enough.'

  So the smearing only works when n antennas signals are
combined, where n is left for us to determine, through simulation or math.
 Clearly n=1 does not work well.
Darrel commented: 'I don't think the "N antenna" version on its own its going 
to be adequate.  I think you need more different frequency 
offsets even than the number ~50 of antennas.'

A1) A version of LO offset smearing was contemplated with the NRAO 12m
antenna once.  This works with n=1.  We could think of using this in
combination with A) above.
Darrel explained the NRAO 12m concept: 'With that, you step 1st 
LO and 2nd LO in synchronism, which means that the unwanted sideband 
smears out.  For example, you might step the LOs once a second, which 
means that after 60 seconds you've smeared the unwanted sideband into 60 
spots over the spectrum.  That would be essentially identical to Peter's 
concept of averaging 60 antennas each with a different but constant-in- 
time LO offset.'
Q: Can the computing and electronics work with this?  We think the answer
is yes (LO changes at ~1Hz rate).

Proto-recommendation:
* Maybe the optimum will be a combination of 12 M 
frequency-with-time stepping and the N-antenna version, making 
sure different frequency offsets are used for each antenna. 
That would enable the stepping-with-time to get through a 
complete set of steps N times more quickly, which is probably a 
good thing.
* Simulations are needed! One value of 'n' which we should consider is n=4 
as that is the number of ACA total power antennas.

ATF testing:  This should be tried with a DSB receiver.  There will be none
which can be used at the ATF under the present plan.  This can only be
sky-tested on an ALMA system when antennas are installed at the AOS (when
there are plenty of other things to keep us busy).    It may be possible
to test the scheme at the ATF using the one-polzn DSB evaluation receiver
at 1.3mm.  As we do not plan to use this receiver for any other tests, this
also would entail effort.

B) Improving the suppression in 2SB receivers.
Darrel noted:
'One good thing is that for most bands we do at least have 10 
dB of suppression between the sidebands, which is a starting 
point.  The better thing is that I think most of the sideband 
separation mixers have a good deal better than 10 dB 
suppression, with 10 dB being the worst case spec that _has_ to 
be met over most of the band.

    'One sideband separation option has been pointed out by Tony 
Kerr in a draft memo that he has at: 
http://www.cv.nrao.edu/~akerr/EnhancedImageRejection01.pdf .
Tony's scheme can only be used with sideband separating mixers, 
not with true double sideband mixers, and relies on very careful 
calibration measurements that enable you to separate the 2 
sidebands by a process of linear combinations in the post 
processing.  Tony's scheme might work quite well, but so far as 
I know hasn't actually been tried on a mm-wave telescope.'

This should be tried with a 2SB ALMA receiver.  There will be none
which can be used at the ATF under the present plan.  This can only be
sky-tested on an ALMA system when antennas are installed at the AOS (when
there are plenty of other things to keep us busy). 
It may be possible to test the scheme at the SMT, probably at large expense.

C) What is software suppression?
This would be a version of the 'sideband clean' algorithm which Schilke
and collaborators used on the CSO spectral surveys.  ALMA has the advantage
that the sideband gain ratio is measured.  However, in all implementations of
this algorithm I am aware of there is some residual 'ghosting'.

Clear skies,
Al