[time-nuts] TI to Freq. conversion

[Magnus Danielson]

From: Glenn <glenn at net127.com>
Subject: [time-nuts] TI to Freq. conversion
Date: Wed, 02 Aug 2006 14:59:04 -0400
Message-ID: <44D0F5F8.5050904 at net127.com>

Glenn,

A few words from my experience, I'm sure others have deeper knowledge and knows
these things better.

> Is computing the frequency of an oscillator using TI more or less 
> accurate than the measuring the frequency directly? I think, 
> theorically, they should have the same accuarcy, but I doubt that's 
> actually the case.

It really depends on the details of the counter mode, mix that with numeric
precission issues and it is a bit more mudded waters than one think at first.
Regardless of what you call the measurement (Time Interval or Frequency),
measure a start position in a local clock and then a stop position in the same
local clock will usually be better than making a number of shorter
measurements, calculate frequency and then average those. One reason is that
each measurement add new +/- 1 cycle of measurement error in the local clock.
However, if the measurements run back-to-back with zero death-time the
difference disappears. Numeric precission issues will be an issue if the
time-gaps are very small, but averaging compensate it. The linearity of
interpolators is another issue, and this would improve by averaging many
measurements.

Some counters (HP5335A for instance) make calibrations of their interpolators
through a short and a long pulse. The difference in interpolator reading is
being used to correct the scale of the reading, while the short pulse reading
is being used to remove the offset. This sounds all nice and dandy, until you
realize that the division causes a form of quantization noise. This is one
example where averaging over many measurements would be beneficial.
Interpolators isn't linear, so there is another source. Averaging will smooth
some of that error out.

Like any form of measurement, detailed knowledge about the inner workings of
any particular measurement mode and processing is the only really good source
to make the most to come out of any instrument. You not only have to mitigate
the setup behaviour, you also need to mitigate the counters behaviour. They are
not black boxes which will magically make the optimum measurement whatever you
toss into it.

I've found documentations being all from muddy to brilliant for counters.

Cheers,
Magnus