[evla-sw-discuss] Delay models to station boards

[Bruce Rowen]

The delay model schema needs to be updated (I'll do that shortly) so  
what you are seeing is not 100%

On May 21, 2008, at 4:45 PM, Barry Clark wrote:
> OK, Kevin's remark about location seems compelling, and Rich has  
> already
> implemented it (thanks Rich), so I'm ready to comment about content.
>
> I am still dubious about the use of ISO 8601, because it looks like
> encoding it is a significant impact on the Executor.


The correlator is a very-discrete time system and everything is in  
10ms steps. The 'epoch' or time the model becomes active really needs  
to be in some time that represents this. The CMIB seems to be able to  
do the conversion from ISO 8601 to (int)secs:(int)usecs in a few  
microseconds so it seems to have an advantage over mchost.

't0' certainly can be double, I'll make it this way in the updated  
schema.

>
> float is inadequate for t0, it must be double.  On the other hand, for
> all of the coefficients except index=0, float is fine.  But it  
> probably
> isn't worth the effort to take advantage of that.

Agreed, I think just making them all double makes everything cleaner.

>
> The coefficient index and subband ID should be type xs:int - jaxb does
> very annoying things with xs:integer.

One of the changes is to make subband ID (and the subband models) a  
sub element of the baseband. The correlator has no capacity for  
subband IDs greater that 17 (0-17) and the subbands are a part of the  
baseband, not independent. I'd think this would have no great impact  
on the model writer but if I am missing something....

I also want to eliminate the 'numCff' (number of coefficients) since  
it is redundant. The number of coefficients would then be unlimited  
(as far as the schema cares) and probably of more use, the number  
could be independent from bb to bb.

>
> The units of the coefficients should be stated in the documentation.
> I'm not sure what units you are expecting but something like  
> microseconds,
> microseconds per day, microseconds per day per day make nice readable
> numbers.  Clocks instead of microseconds is still pretty reasonable,
> but clocks per clock is annoyingly small.
>
> Similarly a documentatin matter - are the sbDelayModel coefficients
> ab initio or as an offset from the bbDelayModel?  (if the latter,
> floats are adequate.)
>
> I think for testing purposes, we need a few fields to make it easy to
> check if we think something might be going to the wrong place.  These
> would be antenna ID (xs:string) and a sequence number (xs:int) as
> attributes of vciStbDelayModel and an IF ID (xs:string) as an  
> attribute
> of the bbDelayModel.

These are inconsistent with the rest of the VCI identifiers.  
Currently there is an element 'boardId' which the station board  
ignores since it is not used anywhere. I'll leave it in, but make it  
optional so we don't need to send it when we are done testing the  
mapping.

>
> Do we want to retain the 'vci' epithet in vciStbDelayModel?  It  
> seems to
> me that we've gotten rather far from the rest of the VCI.

It doesn't really matter, but currently there is a lot of code that  
looks for 'vciStbDelayModel'. I'd leave this decision for a post- 
testing code cleanup period (which needs to be done elsewhere as well)

>
> It is likely that all records for some years will use three  
> coefficients.
> The nasty OTF case requires three, and three is adequate for  
> everything
> we currently are thinking about.  The only very obvious cases where  
> more
> are needed are VLBI and tracking LEOS.  Currently not a lot of  
> constituency
> for either.
>
> Similarly, all initial testing will be without sbDelayModels.  I've  
> done
> no infrastructure, expecting to eventually import code from the OPT  
> when
> it learns how to deal with them.

Ken suggests that the sb delays be processed regardless, but  
initialized to 'zero'.
Internally I have everything in place to support the subband models  
and Dave Fort has added them to his module test file format so their  
performance can be at least verified within the driver code.


Hopefully Rich will be able to jaxB-ify the updated delay schema soon  
so that all the changes can be verfied to work. Also note that you  
will not need to wrap the vciStbDelayModel within a vciDelayModel  
element since the models by their nature are station board based and  
each XML document describes one and only one station board.

>
> Not sure when I can support 3bit samplers, either.  I don't really  
> want
> to do anything about it until I at least have a design for supporting
> the baseband slope correctors, and I haven't a clue.
> _______________________________________________
> evla-sw-discuss mailing list
> evla-sw-discuss at listmgr.cv.nrao.edu
> http://listmgr.cv.nrao.edu/mailman/listinfo/evla-sw-discuss