[time-nuts] Darn you people....

[Magnus Danielson]

On 05/07/2011 02:06 AM, Bob Bownes wrote:
> Thanks to all for the discussion, but I'm still not sure why I have a
> (now) consistent 1.495 hz frequency difference between the thunderbolt
> and the VE2ZAZ FLL.
>
> On a similar note, forgive my ignorance, but is there a simple
> explanation why there are different frequency readings on my 5370B
> when selecting frequency with a 1s gate vs frequency with 100k
> samples? Not a few uHz, 1sec gate reading of the VE2ZAZ is
> 10,000,001.4946 Hz (+/- noise in the last digit) while 100k sample
> reading is 9,997,322.2 Hz +/- noise in the last four digits.

You can have time-biases and voltage-biases which comes into the 
measurement.

Recall, that the estimated frequency is

f = eventcount / time

and that an average over 100k samples will mainly average to higher 
resolution mainly in the time numbers and to a smaller amount in the 
event-count number.

Voltage-biases is due to difference in the trigger voltage of the start 
and stop comparators on the input. By use of a calibrator you can 
identify these can remove the effect, HP has patents and app-notes 
relating to their calibrator. Similar results can be achieved using 
other methods. Consider for instance that the frequency should stay the 
same regardless if the start and stop is on the rising or falling edges, 
but voltage biases can skew the time to create a time-bias.

Some counters, such as the SR-620, also has built-in time-biases in 
their frequency mode which needs to be calibrated. This is due to delay 
differences between start and stop events.

Now, with these skewed times going into the frequency formula above, it 
should not be strange if these systematic errors will cause a systematic 
skewing of the value on which no averaging is able to defeat. These 
biases needs to be handled up-front.

A classical setup to identify this type of bias is to measure the 
frequency of the counters reference output. That way you have completely 
removed any issues relating to frequency difference.

A useful method to disclose if you have a bias is to shift time-base 
length. If you have 1 second time-base and shift to 100 ms then the 
eventcount value will become one tenth and 900 ms will be removed from 
time, but the time-bias is as large as before but now has 10 times 
higher impact. If you see that, you can be quite sure that you have a 
bias and the best way to defeat it is by adjusting your trigger points.

When you have the counter under computer control, and pull out the 
event-counter and time-values separately you can do more intricate 
tricks naturally.

There are a way for a counter-designer to avoid voltage biases to 
inflict on the frequency measure, and that is to take both the start and 
stop events from the same comparator. Only very small biases of voltage 
nature would remain, but the time-difference from that comparator to the 
start and stop detectors would remain.

TIC-based comparision between two frequencies works well to measure 
relative frequencies of fairly small differences in a way that biases 
cancels on a first-degree level.

Measuring frequency has some intricacies in it.

Cheers,
Magnus