[evla-sw-discuss] Tuning for high frequency bands

[Ken Sowinski]

The note included here describes in some detail changes I have made to the
executor with the goal of having better agreement in tuning decisions
between OPT and executor.  The summary so that most of you will not have
to read very much is that there are three changes.

1.  Change the use of the 'hint' parameter in the constructor to agree
     with what the OPT thinks it means.
2.  Improve the algorithm used to choose the first LO.
3.  Only reject the LoIfSetup if the requested center frequencies
     cannot be placed at the center of a baseband.  In particular,
     slopping over the edge of an If passband is allowed.

The changes described here have been made and are being tested.  I will
commit the source code after I am satisfied that it works.  It will not
surprise me if after that we uncover a new bug or two in the executor
or OPT/m2s.

The note includes a list of general guidelines, no doubt incomplete
or wrong in part, to what tuning cobinations are allowed.

Any comments or corections are welcome, especially before I commit these
changes.

Ken



Over the years there has been much confusion arising from the fact that
the OPT and executor do not always agree on how to set the LOs to achieve
a requested sky frequency.  I have decided to reorganize how the executor
deals with the wide bandwidth, high frequency bands to try to address 
these problems.  I plan to discuss what the OPT does, how the executor
ought to respond, and end with some general guidelines.  Familiarity with
the T303 module is assumed.

To begin with I review what model2script creates LoIfSetups.  It always 
uses the full blown constructor described below in a snippet from the 
executor.

  /**
    * Constructor for Widar with SSLOs or centerfrequencies,  and either
    * 8bit or 3bit samplers.
    * @param cenfreq false if SSLOs are specified
    * @param band receiver name
    * @param ac1  Cenfreq for the AC T304s, output 1 in MHz
    * @param ac2  Cenfreq for the AC T304s, output 2 in MHz (see note)
    * @param bd1  Cenfreq for the BD T304s, output 1 in MHz
    * @param bd2  Cenfreq for the BD T304s, output 2 in MHz (see note).
    * @param t303 Path for UX converter (0=default; 1=AC straight througe,
    * BD converted; 2=both converted)
    * NOTE:  if ac2 is supplied as zero, the LoIFSetup will be configured
    *   to have an eight bit sampler at the cenfreq specified by parameter ac1
    * NOTE:  if bd2 is supplied as zero, the LoIFSetup will be configured
    *   to have an eight bit sampler at the cenfreq specified by parameter bd1
    */
    public LoIfSetup(boolean cenfreq, String band, double ac1, double ac2,
        double bd1, double bd2, int t303)

On Tue, 6 Jan 2009 David Harland described what the OPT does.  The param
he refers to is the one called 't303' above.

> As it stands now, we're duplicating the test in LoIfSetup that sees if the
> IFs are close enough to both go through w/out conversion.  If they are, we
> use the "0" param; if they are not, we adjust the B/D SSLO, ensure that A/C
> is the higher sky freq, and use the "1" param.  Having decided conversion
> was necessary for calling LoIfSetup, it seemed prudent to tell it explicitly
> to use that path.  Otherwise, if the logic outside and inside LoIfSetup had
> different ideas about when it was necessary to convert, using the "0" might
> lead to a suprise, and a B/D SSLO that was not calculated consistently w/
> the chosen pathway.
>
> As an aside, i've taken the easy way out w/ Ku and am always using "2"
> there.

I have examined a few months worth of observing scripts and find that what
Dave wrote then remains true except for the statement about Ku band.  The 
t303 parameter is always zero in that case rather than two.  This observation
is supported by an examination of the m2s code.

I propose two changes in executor behavior.  First if the t303 parameter is
provided it will be taken to have the same meaning as in the OPT:
t303 == 0:  Use the direct copy path in the T303,
t303 == 1:  Use the conversion path for BD in the T303,
t303 == 2:  Use both mixers in the T303 (only works for Ku band).

Internally this parameter is initialized to -1, and if not supplied in
the constructor, the current 'default' behavior will be elicited: the
direct path will be used if the total frequency spanned is small enough,
else the conversion path.  Thus old functionality, probably of use only for
testing, is not lost by this change.  Explicitly choosing T303 behavior by 
use of this parameter means that the executor and OPT will never have
contrary notions about the net sideband.  Ku band is a special case, perhaps
outside the scope of this note.  The OPT always sets the t303 parameter to
zero for Ku band and that is not good because the direct path is never
appropriate for Ku band.  To address this there is a kludge in the executor
to set this parameter to -1, only for Ku band, allowing an appropriate choice 
of path through the T303.  When m2s is changed, the kludge may be removed. 
For Ku band, all the baseband pairs always have the same net sideband, so 
there is no penalty for disagreeing with the OPT.  Note that the value of 
this parameter is irrelevant for all the low frequency bands and the 
constructor which omits this parameter may be used.

The second change is the algorithm used to choose LO settings.  The algorithm
we have used and tinkered with is premised on the false assumption that we 
want to maximize total bandwidth by setting basebands next to each other. 
Instead it has become common to overlap basebands, sometimes almost 
entirely in an attempt to provide uniform sensitivity over the subband gaps.
This has lead to much confusion and incompatilities between OPT and executor.
The key is to recognize that when there is overlap, or only a very small gap,
between the AC and BD basebands then the overlap region should be centered
as close to 12 GHz as possible.  If the gap is large then the AC basebands
should snuggle up against the 8 GHz lower limit leaving as much room as 
possible for the BD basebands in the conversion space above 12 GHz. 
Implementing this change will have no effect on the tuning strategies for
Ku band and all the lower frequencies.


There are some fundamental constraints to tuning set by filter bandwidths
in the IF paths and synthesizer tuning ranges.  The executor and OPT each
represent these constraints in their own way, but our intent is that the
executor should accept any tuning which the OPT considers valid.  Only the 
OPT knows what part of the baseband presented to the sampler will actually 
be correlated so the executor should not be concerned when signals are 
apparantly out of band.  Adopting the philosophy that anything tunable is
acceptable means that we will have to occasionally test how effective some 
of the extreme tunings really are.

While the discussion in this note is limited to the three high frquency 
bands the limitations here apply equally to all bands except where noted.

1.  When using three bit samplers the two basebands pairs sharing a path
(AC1 and AC2 for example) must not span more than 5048 MHz.  There might be
examples where spans less than this will be rejected because of L302 tuning
restrictions, but the expectation is that the OPT would warn about these.

2.  If the AC and BD basebands overlap by more than 1 GHz, then the full
baseband BW will not be available.  Nonetheless the the tuning will not be
rejected as long as LO tunings can be found to achieve the request.  It
is expected that the OPT knows that only subbands with valid signal will
be correlated.

3.  If the T303 conversion path is to be used the AC basebands must be tuned
higher in sky frequency than the BD basebands.  This constraint is enforced 
in the OPT.

4.  The total bandwidth spanned must not exceed 11 GHz.  The same caveats
as for number 2. above apply here.  This arises only for Ka and Q bands.

5.  There are special rules for Ka band because the allowable first LO
range is very limited and the reciver bandwidth so large.  The AC If pair 
may not extend below 31.408 GHz.  This limit is set by LO tuning range but
allows the baseband to extend well beyond the advertized IF path bandwidth. 
Much thought is needed if the intent is to allow the AC IF pair below 32 GHz. 
This constraint means that the full 8 GHz from 26 to 34 GHz cannot be covered 
instantaneously.  In particular the AC IF pair may be tuned to cover 32 to 
36 GHz and the BD IF pair to cover any 4 GHz span between 26 and 32 GHz.