[evla-sw-discuss] The T303 converter

[Barry Clark]

I am in the process of rewriting the LoIfSetup class.  Although there
were a few hooks, the support for the three bit samplers was essentially
nonexistent.  The code in this class is so opaque and messy that it 
is very difficult to modify.  Therefore I am rewriting ab initio.
The existing version has a VLA correlator orientation, with Widar as
an afterthought.  The new version will have a strong Widar orientation,
with a very few VLA correlator afterthoughts.

One of the things that makes the current code so very obscure is the 
possibility of changing the frequency of an object after it has been
constructed.  I plan to eliminate that capability.  (But I will still
retain the capability of trading off between the L301 and L302 settings
to get to the frequency with which the object was constructed.)

At the lower bands what I want to do is straightforward.  Running the
T303 UX converter is rather less so.

The existing version primarily uses the T303 on the "straight through" 
path, which is fine if all the bandwidth wanted lies within a single 
4 GHz block.  Ken has modified the class to allow one to directly specify 
the path desired in the constructor.  This is a reasonable path to 
follow, but it has the undesirable effect of unnecessarily exposing
the equipment's internal structure to higher level software.  (This
exposure is still needed for testing - we will likely want to see if
different paths yield equivalent results as we go along, so it will
at least always be an optional parameter to the constructor.)

An alternative approach is to have a default choice of path be made
by the software if it is not explicitly specified.  For those who care,
the rules for making the choice are given at the end of this message.

This approach has a serious flaw, in that the net sideband of the BBs
is determined by this choice.  So if we continue the approach of 
specifying signed sum LO to the class, higher level software has to be
aware of what the choice is, because astronomers are likely to give 
the frequency they want at the center of the band, not at band edge.
So at least the algorithm for choice would have to be incorporated in 
the higher level software.  This could be eliminated by providing 
constructors that operate in band centers, rather than SSLO.  (This
would be sampler band centers, so definitely a Widar concept, not a 
VLA correlator concept.  The original choice of specifying SSLO was
driven by not wanting the LO system to have to know about the filters
in the VLA backend.)

I am willing (for a while, anyway) to listen to advice on how to proceed.


Appendix:  Suggested rules for selecting path through the T303 (chosen 
for compatibility with existing practice).

For Q and Ka bands:
If all the requested bandwidth is within a 4 GHz block
   select the straight through path for both IFs, giving both IFs in USB
Else if the AC band is at higher freqency than BD
   AC uses straight through in USB, BD converts by L301-2 giving LSB
Else 
   Both use conversion path, both LSB

For K band
If all the requested bandwidth is within a 4 GHz block
   select the straight through path for both IFs, giving both IFs in USB
Else if the AC band is at higher freqency than BD
   AC uses straight through giving USB, BD converts by L301-2 giving LSB
Else
   Error - cannot be tuned

For Ku band
If all the requested bandwidth is within a 4 GHz block
   Both IFs are converted using L301-1; LSB on both
Else
   AC uses conversion by L303-1, BD by L301-2; both LSB