[mmaimcal] FW: ALMA system design issues related to the correlator

[Brian Glendenning]

>-----Original Message-----
>From: David Woody [mailto:dwoody at caltech.edu] 
>Sent: Friday, February 04, 2000 6:22 PM
>To: John Webber; Darrel Emerson; Bob Brown; bglenden at cv3.cv.nrao.edu
>Cc: Anneila Sargent
>Subject: Re: ALMA system design issues related to the correlator
>
>
>Following are some thoughts on ALMA design issues that I think need to be
>addressed.  Please forward this to people you think can contribute to the
>discussion of these issues.
>Cheers
>David
>
>______________________________________________________________________
>Comments on the present ALMA system design as it impacts
>the correlator design
>
>David Woody Feb. 4, 2000
>
>____________________________________________________________________
>The issue of single vs. double sideband receivers and phase switching
>is very complex and the implications for the system design needs
>to be very carefully worked out and discussed.
>
>The current system and correlator design assumes that the receivers
>will be single sideband or sideband separating.  Sideband separating
>receivers were initially proposed when the MMA was predominantly
>a millimeter wave (as opposed to a submillimeter wave) instrument
>to be located at a US site.  The baseline design for ALMA now extends
>to 950 GHz and possibly beyond and is slated for a site significantly
>better than any US site.  Achieving sideband
>separation for all of these bands will be more difficult and
>the potential improvement in system performance at the
>lower frequency bands will be less.
>
>In light of this, I believe it is important that
>the system at this time be able to fully utilize DSB receivers,
>i.e. not simply suppress the image sideband.
>Many observations, such as line searches,
>continuum measurements and complex regions like Orion,
>will be correlator limited.  There is a factor of two improvement
>in observing time for receiver noise dominated DSB frontends if the
>sidebands are separated in the correlator for these observations.
>
>This means that the correlator should be able to properly process
>the IF from double sideband receivers.
>Most if not all of the present millimeter interferometers utilize
>double sideband receivers with sideband separation
>accomplished via phase switching
>of the 1st LO and binning of the correlator output.  This system works
>extremely well and essentially processes twice the on-the-sky bandwidth
>for the same amount of correlator hardware at the cost of adding
>four-state binning of the correlator output.
>
>There are issues associated with sideband
>and spurious response suppression that may need to be
>looked at in more detail.
>The result of CW frequency offsets, 180 deg 1st LO modulation with
>bit inversion of the digitizer output along with the effects of
>.5 msec time slices for the correlator processing can be complex.
>There could be subtle artifacts that show up in the correlator
>output.  Some of the questions that come to mind  for the present
>baseline design or any other phase switching scheme are:
>1) If the 180 deg Walsh sequence is to complete in .5 msec, what
>phase lock relock times are required for the 1st LO?
>2) If the 180 deg Walsh period is .5 msec then the meaningful
>minimum integration time becomes ~50 msec).  Is this acceptable?
>3) In the highest resolution mode, the 90 deg sideband suppression
>offset frequencies and possibly the 180 deg phase switch periods
>will be within the frequency resolution of the correlator, i.e.
>lag length longer than the clock period of the Walsh steps.
>What artifacts does this produce?
>4) Are the 180 deg Walsh states synchronous with the 90 deg frequency
>offsets, and if so what effects does this have on the output?
>5) What is the effect of incomplete cross correlation at
>the ends of the .5 msec time slices for the 90 deg frequency offsets
>and the 180 deg Walsh sequence averaging?
>6) Similarly, if there is any dead time between the .5 msec slice
>or 16 msec "frames", then the averaging over an integral
>number of sequence cycles is compromised.
>7) Even the type of phase switch sequence to use is not obvious and
>is an active area of research.  Eric Keto (SMA), Steve Scott (OVRO)
>and Lynn Urry (BIMA) are even now discussing how to do this for
>much smaller sparse arrays.  All of the desirable properties
>you want from phase switching are hard to achieve in a limited
>sequence time.
>
>My guess is that ALMA will have to support two phase switch modes:
>1) fast complete integration times of ~16 msec with imperfect
>image and artifact suppression to be used for solar and on-the-fly
>mosaicing.
>2) slower complete dump times >50 msec with full
>phase switch state binning at the correlator output for the
>best possible artifact suppression and with image recovery
>(i.e. DSB receivers).  This mode will be used for most observations
>and will be required for difficult detections
>requiring long integration times.
>
>The phase switch sequence for 64 antenna will require Walsh sequences
>that are ~100 cycles long.  If binning can be accomplished at the
>.5 msec time scale with blanking for phase lock recovery,
>then a full sequence and hence the minimum integration
>time would be ~50 msec.  The LTA could accomplish this for the widest
>band mode where each plane contains .5 msec of data, but it may be
>impractical for the highest resolution modes where lags are distributed
>between planes and possibly between quadrants.
>These modes may require longer cycle times.
>
>_____________________________________________________________________
>There are several future upgrades that should be kept in mind for the
>ALMA correlator.
>
>1) more bandwidth:
>Although 16 GHz seems like a tremendous amount of
>bandwidth, many continuum and line search observations could use
>even more bandwidth.  Dual polarization 3mm HEMT amplifiers
>and sideband separating mixers actually have ~32 GHz of IF bandwidth
>available.
>
>2) more bits:
>There is also an efficiency loss of ~12% for 2-bit
>quantization which can be improved by going to 3 or even 4-bit
>quantization.  Note that a 1% improvement in efficiency is
>worth 2-4M$ to the project.
>The current design already calls for 3-bit
>digitizers to improve the FIR filter performance.
>These bits should be available at the correlator for future upgrades.
>
>3) more independently tunable bands:
>It may also be advantageous to have more than four independently tunable
>IF-band dual polarization pairs to optimally cover many lines
>simultaneously.
>(The current millimeter interferometers have observed more than
>five lines simultaneously using double sideband mixers and four
>tunable bands.)
>
>These improvements will require roughly double or quadruple the
>correlator computations but Moore's law will make this feasible
>at the same cost as the current correlator in only a few years.
>The system design should keep these issues in mind
>to allow implementation of a next generation correlator with
>minimum changes to the rest of the system.
>*****************************************************
>| David Woody
>| Assistant Director of Instrumentation
>| Owens Valley Radio Observatory
>| California Institute of Technology
>| 100 Leighton Lane (PO Box 968)
>| Big Pine, CA 93513
>| 760-938-2075ext111, FAX 760-938-2075
>| dwoody at caltech.edu
>*****************************************************
>
>