Re: [Discuss-gnuradio] Multiple USRPs

[Robert McGwier]

In addition to Matt's answer let me add a few remarks.  A few years 
ago,  while working with Tom Clark at Goddard Space Flight Center (Tom 
is an astrophysicist),  I was  asked to look at some Kalman Filtering 
routines for his VLBI work.  MANY years later,  I desperately needed to 
make widely placed, but otherwise unconnected, oscillators and receivers 
as coherent as possible.  When I realized he had solved this problem I 
asked him how.  His VLBI work has been used to do extremely accurate 
geodesy and done such things has measure the change of rotation rate of 
the earth due to El Nino current changes, etc.  I knew from the Kalman 
Filtering work that he was doing coherent signal processing on recorded 
signals.  These recorded signals were from instruments thousands of 
kilometers apart and then brought back to Goddard for processing 
sometimes months after the collection. How was this possible?  He tuned 
and transported Hy masers to his sites and then locked all oscillators 
to the masers.  Even that was not enough.  He also had to continuously 
calibrate the path/ transfer function from the antenna through the 
receive apparatus.  Using this same maser as the reference source,  he 
turned on a pulse train generator and combined it with incoming signals 
at the antenna.

Most of us here do not need this level of complexity.  We can do with 
much less.  For example,  I do coherent signal processing at HF, VHF, 
UHF to do timed difference of arrival geolocation with receivers 
covering a few hundred square kilometers by doing the same function with 
GPS receivers taming an oscillator.  The GPS tamed oscillator is used to 
cohere the systems and to generate the pulse train that aids in 
calibration of the transfer function in front of the digitization 
process.  I can geolocate  a reasonable SNR signal at (say) 10 MHz using 
this system to a few meters and carrier phase is definitely in the 
calculation.

So one of the first lessons I learned in my travels was "Who is on first 
and What is on second".  These players include:

NEVER use a PLL'd thing as the oscillator driving a system when you need 
high phase stability.  It is okay to use a temperature compensated 
voltage controlled high quality crystal oscillator.  You want the phase 
stability of any system guaranteed by a high quality oscillator 
(crystal, etc.).

If you also need ACCURACY,  you use your Hy maser or you GPS tamed 
reference oscillator to tune the high quality VCO to get on 
frequency.    Your design problem following this realization is to 
optimize which of the two different systems dominates the Allan Variance 
at what kind of time offsets.  Short term phase stability is determined 
by the high stability (crystal) oscillator.  Long term stability is 
determined by the Maser or GPS tamed beast used in a Tom Van Baak style 
mechanism to keep the VCO pulled in.  The latter is provided by (say) 
the Reflock II boards available from TAPR.

Cheers,
Bob



Matt Ettus wrote:

> > This raises a question I've always had about phase-coherence of PLL
> > oscillators that use a common
> > clock source.  Will they be coherent "enough" for things like
> > astronomical interferometry?
> >    
>
> Yes.
>
>  
>
> > The individual PLLs will still have "close in" phase noise components
> > that are unrelated to
> > on another.  I suppose that given that you typically integrate over
> > several seconds, such
> > artifacts get cancelled out.  But at very fine timescales, I imagine
> > that PLLs locked to
> > a common clock are *not* a good way to get decent coherency?
> >    
>
> The close in noise is from the COMMON reference, so there is nothing to
> cancel, it is perfect.  It is the further out phase noise components
> which are not the same between the 2 PLLs, since those components are
> from the individual VCOs.  These integrate out fairly quickly, and are
> much smaller anyway.
>
> Matt
>
>  


--
Laziness is the number one inspiration for ingenuity.  Guilty as charged!