Frequency counter - gate time and digits below 10^0

[tkamiya]

I thought I understood this, but apparently I don't.

With frequency counters, I thought gate time determined the resolution.  Say I have 10MHz clock.  I apply 10MHz to input.  I set gate time to 1 second, 10,000,000 cycles pass through while the gate is open.  So I can read down to 1Hz.  I thought it would be impossible to resolve 0.1Hz or below, for example.  I thought to have 0.1Hz resolution, I need 10 second gate time.

I have a HP53131a.  If I set gate time to 1Hz, it will happily display down to millli-hertz.  How is this possible?  Shuldn't I need 1000 second to do that?

For this discussion, please disregard accuracy/precision of clock.  Let's say it's accurate enough.

[Tom45]

A reciprocal frequency counter can measure the period of the input signal using (for example) a local 10 MHz reference.

Measuring a 1 Hz signal this way it would gate the clock for 1 cycle of the input signal. That 1 cycle would count up 10,000,000 counts. Giving a resolution of 1 clock cycle (1 / 10,000,000 seconds) for the period of the input signal. The input frequency is obviously the reciprocal of the input period.

One cycle of a 1 MHz signal would gate 10 counts of the local clock giving a period of 10/10,000,000 seconds. Gating for more cycles of the input, say 1000 cycles, would give better resolution for the period while still being much faster than a direct measurement as you would have thought.

See: https://www.best-microcontroller-projects.com/article-frequency-counter.html for more detailed examples of direct and reciprocal frequency counting.

[tkamiya]

So basically, it's measuring period and take an inverse of that to arrive at frequency?  I never thought of that....  I guess it would be easily possible with micro processors.  (in every counter)

[Tom45]

There are ways to quickly compute a number's reciprocal without dividing. But that is a different topic.

[oPossum]

It measures the time between the first edge in the gate period and the last edge in the gate period instead of just using the duration of the gate period. The resolution of this time period is determined by the master clock frequency, and that (in combination with the gate period) determines the resolution of the measurement.

Frequency = edge count / time

[beanflying]

I was actually having a read of this ysterday. Goes into some detail about period counting http://www.hpl.hp.com/hpjournal/pdfs/IssuePDFs/1969-05.pdf

[tkamiya]

So this isn't exactly new.....  but now it's affordable.

[DaJMasta]

Definitely not new, it's been a standard for a while, especially in very high resolution counters.  You set your trigger up and then time the interval between triggers and do a bit of math (better implementations do fitting and interpolation), giving you your period and thus frequency.  This is also is required for burst frequency counting, with the originally described method, a 1ms long pulse of 10MHz once a second would register as 10kHz, as that would be the total pulse count in the 1s gate time.


Basically once they became frequency counters instead of just pulse counters, this functionality was important for usability and accuracy, so they've all done some form of that for a while now.

[Dr. Frank]

I strongly recommend these great Application Notes AN 200, 200-1, 200-2, 200-3 and 200-4 from hp.

These explain in a condensed way the different counter techniques for frequency, period, time interval, and the different RF conversion techniques, and other aspects of time keeping.

These are still hosted on the hp Memory Project:

http://hpmemoryproject.org/ressources/resrc_an_03.htm


So your 54131A and the 5335A are reciprocal and interpolating frequency counters.

Frank

[tkamiya]

Thank you Dr. Frank.