[time-nuts] Fwd: OT: 10 MHz data capture, help

[Magnus Danielson]

Ed,

Ed Palmer skrev:
> The Agilent web site has most of the manuals for the 5372a.  The data 
> sheet and the "Condensed Specification & Reference Guide" (over 100 
> pages!) gives you most of the functionality.

It is recommended reading regardless if you have one or not.

> The 5372a's biggest deficiencies for a time nut are its resolution of 
> only 150ps and its maximum built-in timing period of 8 seconds.  This 
> means that even though it can calculate Allan Deviation, it's limited to 
> about 1.5e-10 @ 1 sec. and it can't calculate anything beyond a tau of 8 
> seconds.  You have to use external means to improve the noise floor 
> (e.g. mixers).  The 8 second limitation can be sidestepped if you use 
> GPIB to pull the data and process it externally.

As has been done for instance by Bregni. An important detail to the 
5372A architecture is the continously ticking time clock, so if you set 
up a single measure and read it out binary and then quickly setup 
another one you can run it in continous mode without problem. The 5372A 
programmers manual is about as detailed as it can get. I have not seen 
anything quite like it.

> The 13M readings per second that Magnus mentioned is limited to 
> measuring the period of the incoming signal and then incrementing a 
> count of how many measurements were in that range.

There is two modes, fast mode (13,3 MS/s) and normal mode (10 MS/s). 
Both has the same burst lengths, but for fast mode some data is not 
stored away, so counter-length is effectively reduced. It is reasnoble 
when running higher rates for shorter periods of time.

The resolution is 200 ps for all types of measurements.

>  There are 2000 'bins' that can be as small as 200ps wide.
 >  No time stamping, no
 > calculations, just how many measurements were in that range.  Useful
 > but, at the same time, limited.

That's the histogram hardware, which is unique to the 5372A compared to 
the 5371A. However, several normal measurements can also be run in fast 
mode.

> I recently used it to collect 100M 
> measurements of the 10 MHz output of a Navsync GPS receiver.  I was able 
> to infer the algorithm that they used to keep the output on frequency.  
> If I remember correctly, it took less than 20 seconds to collect and 
> plot the data!

When you come to terms with it, it can be a very efficient measurement tool!

The three most annoying things about it is memory size, time resolution 
and time-wrapping. There are a number of user interface issues and other 
limitations to annoy me, but those three annoys me mostly. Other than 
that, it is an instrument I trust and love. It still shines on some 
aspects. The Wavecrests shines on some things, the SRS 620 on others and 
the CNT-90 is to some degree much more handy. None of them fully 
replaces it however.

Cheers,
Magnus