[time-nuts] Measuring frequency

Magnus Danielson cfmd at bredband.net
Sat Feb 24 06:51:06 EST 2007

From: Hal Murray <hmurray at megapathdsl.net>
Subject: [time-nuts] Measuring frequency
Date: Sat, 24 Feb 2007 00:02:04 -0800
Message-ID: <20070224080205.AD880BE04 at ip-64-139-1-69.sjc.megapath.net>

> If I understand things correctly, when a box like the 5334 takes a sequence 
> of frequency measurements, each measurement has a start time and a stop time. 
>  For each measurement, you get out the number of ticks (including fraction) 
> on the input signal between those times converted into reasonable units.
> What if a sequence of measurements combined the stop from one measurement 
> with the start of the next measurement?  Then any error in one measurement 
> has the same magnitude but opposite sign in the next measurement.  How much 
> does that help in the downstream data processing?

Yes. The HP 5372 does this (as does HP 5371). You end up having the
interpolator errors of the first and last measures since these will not be
canceled. This only works for a measuring block. When doing this to the
time-base you end up on the same interpolator values (within a small range) so
the difference between the start and stop interpolator for that measure will
affect the time. It's all clear to me now. For a non-synchronous signal is it
much more likely that things would smooth out over time. The start-to-stop
time delay difference will also come into play. Fortunatly they would have the
same polarity and trigger point value, which simplifies things alot.

> That would be easy to implement that with a FPGA.

It is. They are in modern counters. FPGA implementations for the backend
counter and memory stuff is trivial. There are some issues with FPGA for the
front-end stuff.

> This seems like an obvious idea.  What's it called and/or what is a clump of 
> non-linked measurements called?

The block measurement technique has been commercially available for quite some
time, the first instrument to have it is HP 5371A as far as I know. For lower
rate signals you could do it by just pulling the data out of your counter
quick enought anyway.

One of the big differences is that you don't clear the counters but rather only
let the time and event counters to free-run and then sample their values for
each measurement. The start and stop interpolators is sampled and provide the
finer-grained time than the time counter. You do want to read the HP 5372A
programmers manual, it's really a inside-out description.

There is also a heap of HP patents which is being used in the HP 5372A.
The main thing which makes the HP 5372A excell over the HP 5371A is really its
hardware accelerated histogram engine. I rarely use that thought.


More information about the time-nuts mailing list