Frequency measurements with oscilloscope - what accuracy to expect?

[John_ITIC]

I'm validating an oscillator and clock generator, which should output a 250 MHz clock from a 40 MHz crystal. Internally, a PLL is used (black box IC). My 13 GHz infiniuum scope shows that one oscillator runs at 39.985971 MHz with output frequency 249.40278 MHz. The other oscillator is 40.00921 MHz for an output (after PLL) 249.42577 MHz. The multiplier is 6.23725 and 6.23421 times, respectively. The correct multiplier (250/40) should be 6.25 so something is wrong. However, I'm not sure I can trust the accuracy of the oscilloscope although a high-performance unit. How would one go about verifying the scopes functionality with a "frequency reference standard" of some kind? Which one to buy? How would one get best accuracy when measuring frequency with oscilloscope? Currently, I just set the averaging to 16. The scope additionally, calculates "mean". It is an Infiniuum DSO81304B. http://www.keysight.com/en/pd-734419/infiniium-high-performance-oscilloscope-13-ghz?cc=US&lc=eng

Also, would a frequency counter be more accurate? It would need to be high-impedance to not disturb the oscillator, I suppose. How to connect to DUT?

Thanks.

[Psi]

Most scopes calculate the frequency in software from a subset of the sample data. (not very accurate)
Some scopes do it from the pixel data on screen (terrible)
Others have a hardware frequency counter built in (good)

In general the frequency measurements on a scope are not that accurate.

[AndyC_772]

How exactly are you measuring frequency with the scope?

If you use the scope's memory to store a single complete cycle, and use the cursors to measure the period as accurately as possible (by zooming in to position each one precisely), then you should get a much more accurate result than if you just rely on built-in measurements.

[borjam]

Rigol DS1000Z scopes have a buit in frequency meter, it seems. I have checked it against known sources and the accuracy is quite good.

[John_ITIC]

My scope has a 13 GHz analog bandwidth and a 40 GS/s sample rate so I would assume that the time-base is sufficiently accurate to measure correct frequency of a 40 MHz and 250 MHz sine wave. I simply use my 2.5 GHz active probe and press the "quick meas" button on the front of the scope. It shows various statistics about the signal (period, frequency, including mean period and frequency). I essentially want to know that what I see on the scope is the true frequency. Assuming that it is, there is lots of jitter in the clocks and the oscillator is way off. Perhaps bad crystals? They are both right next to ICs that have been reworked many times. Perhaps the crystals have been damaged during multiple heating cycles (260 degrees C).

I have attached some scope pictures.

[alsetalokin4017]

Perhaps.
You should check your scope's accuracy against a known accurate standard at the frequency range of concern, for absolute measurement accuracy. But since you are aslo seeing a relative difference between the two oscillators under test, you could be dealing with both scope inaccuracy and DUT variation.

In general a dedicated frequency counter with an oven controlled crystal oscillator or rubidium internal reference will be more accurate than an oscilloscope at measuring frequency. Your scope is giving you eight digits of precision which is pretty good for a scope, but you can get nine or ten digits or even more from an actual frequency counter.

[Alex Nikitin]

1) If you do several sequential measurements with your scope - are results the same every time?

2) A good frequency counter will be much more accurate. For frequencies above few Mhz, it would be wise to use a matched impedance cable into 50 Ohm input on the meter and a resistive divider (say, 1K/51 Ohm) directly on the output of the circuit you are trying to measure.

Cheers

Alex

[Wuerstchenhund]

My scope has a 13 GHz analog bandwidth and a 40 GS/s sample rate so I would assume that the time-base is sufficiently accurate to measure correct frequency of a 40 MHz and 250 MHz sine wave. I simply use my 2.5 GHz active probe and press the "quick meas" button on the front of the scope. It shows various statistics about the signal (period, frequency, including mean period and frequency). I essentially want to know that what I see on the scope is the true frequency.

It isn't. Your DSO81304A (which I assume is what you have) has a timebase accuracy of +1ppm, which isn't great for frequency measurements. You can connect a 10MHz reference source (i.e. GPSDO) to increase the accuracy, usually by several magnitudes.

[Alex Nikitin]

My scope has a 13 GHz analog bandwidth and a 40 GS/s sample rate so I would assume that the time-base is sufficiently accurate to measure correct frequency of a 40 MHz and 250 MHz sine wave. I simply use my 2.5 GHz active probe and press the "quick meas" button on the front of the scope. It shows various statistics about the signal (period, frequency, including mean period and frequency). I essentially want to know that what I see on the scope is the true frequency.

It isn't. Your DSO81304A (which I assume is what you have) has a timebase accuracy of +1ppm, which isn't great for frequency measurements. You can connect a 10MHz reference source (i.e. GPSDO) to increase the accuracy, usually by several magnitudes.

The timebase accuracy of the scope tells you very little about the frequency measurement accuracy, especially if the frequency is measured from the acquired waveform (meaning that the amplitude errors are converted to frequency errors and the results are barely good to something like 1000ppm  ). look at the frequency min/max results on the screen shots - i.e. 39.898/40.058 MHz. Even an RC oscillator should have a better frequency stability! Unless the scope has a hardware frequency counter it is pretty useless for accurate frequency measurements IMHO.


Cheers

Alex

[rf-loop]

I'm validating an oscillator and clock generator, which should output a 250 MHz clock from a 40 MHz crystal. Internally, a PLL is used (black box IC). My 13 GHz infiniuum scope shows that one oscillator runs at 39.985971 MHz with output frequency 249.40278 MHz. The other oscillator is 40.00921 MHz for an output (after PLL) 249.42577 MHz. The multiplier is 6.23725 and 6.23421 times, respectively. The correct multiplier (250/40) should be 6.25 so something is wrong. However, I'm not sure I can trust the accuracy of the oscilloscope although a high-performance unit. How would one go about verifying the scopes functionality with a "frequency reference standard" of some kind? Which one to buy? How would one get best accuracy when measuring frequency with oscilloscope? Currently, I just set the averaging to 16. The scope additionally, calculates "mean". It is an Infiniuum DSO81304B. http://www.keysight.com/en/pd-734419/infiniium-high-performance-oscilloscope-13-ghz?cc=US&lc=eng

Also, would a frequency counter be more accurate? It would need to be high-impedance to not disturb the oscillator, I suppose. How to connect to DUT?

Thanks.

If think just numbers in your message it can clearly see that your scope is "far" away from accuracy what you need. And in this scope is quite good freq reference +/-1ppm.   But for your frequency measurements it is not at all enough.
There is external reference input in your scope.  You can use external reference and get much more accuracy and also stability of course.  For someone who may think some Rigol cheap bottom entry level scope have "good" accuracy. Well... +/25ppm  is  +/- 2500 Hz at 100MHz.  Is it good accuracy? 
OP scope have +/- 1ppm class reference. Is it enough? IMHO not. It is still +/- 100Hz @ 100MHz.

From scope screen (or acquisition memory) automatic measurements from trace. It do not give anything acceptable for this purpose, just nothing, how ever it do.

Also many oscilloscopes HW counters may have severe things what affect badly for accuracy, even if scope reference is good.

For middle level accurate frequency measurements  I will recommend real frequency counter what use known accuracy reference.
Example something like HP53131A or something in this class of meters with known accuracy reference.

[nctnico]

I agree. For frequency measurements nothing beats a frequency counter. Use the scope to see if the signal is suitable and use the frequency counter to measure the frequency.

[wraper]

Isn't there hardware frequency counter available in the scope (many scopes have)? Measurements based on the waveform may be not very accurate.

[Wuerstchenhund]

The timebase accuracy of the scope tells you very little about the frequency measurement accuracy

The timebase accuracy tells you that you what basic uncertainty you'll have for your measurements, i.e. it won't be any better than 1ppm.

especially if the frequency is measured from the acquired waveform (meaning that the amplitude errors are converted to frequency errors and the results are barely good to something like 1000ppm  )

That's another problem, but to what extend depends on the signal and how the actual measurements are performed.

I agree that a frequency counter would be a better choice but you can do pretty accurate frequency measurements with a scope which is connected to an accurate reference source.

[rf-loop]

Isn't there hardware frequency counter available in the scope (many scopes have)? Measurements based on the waveform may be not very accurate.

Just short time ago:
https://www.eevblog.com/forum/testgear/frequency-measurements-with-oscilloscope-what-accuracy-to-expect/msg937581/#msg937581

And also some other members comments.

Most today oscilloscope have HW counter. Just this accuracy is question. And accuracy is what here is needed. (related to OP and numbers there.

Measuring from captured waveform, who even try 1ppm or better what here is needed.

[Someone]

Also, would a frequency counter be more accurate? It would need to be high-impedance to not disturb the oscillator

The timebase accuracy of your scope is already quite good and similar to a dedicated frequency counter, as discussed above its how you're resolving the measurement that is the problem. One useful method since you are looking for small deviations from a known frequency is to set the pre trigger far off by say 1000 cycles (25us) and measure the phase offset, you can keep multiplying this pre trigger out for finer and finer measurement as limited by the timebase accuracy.

[Someone]

Most scopes calculate the frequency in software from a subset of the sample data. (not very accurate)
Some scopes do it from the pixel data on screen (terrible)
Others have a hardware frequency counter built in (good)

In general the frequency measurements on a scope are not that accurate.

I'm wondering what the memory depth was for the measurements shown in this thread, its possible to make high accuracy period and/or frequency measurements with scopes and I've had no problem getting jitter data below 0.001UI from the realtime measurement on an Agilent MSOX.

[alsetalokin4017]

Accuracy and precision are not the same thing, and both are required in this case.

 

[VK5RC]

Isn't the answer in the data sheet, about 1/3 the way down, in "Delta time measurement accuracy"?   http://literature.cdn.keysight.com/litweb/pdf/5989-4604EN.pdf?id=2051558
0.9pS averaging disabled is pretty darn hot for any counter over 5 cycles  but the accuracy for frequency (esp at 40-250MHz)  is getting back to the limit of the crystal from a  few of my back of the envelope calculations.  e.g. 5 cycles at 2.5 e-8 / 1 e-12  approx 1 part in 10e5. Averaging helps a lot but then the time reference would really be the limit of accuracy, resolution would be better I presume.
KE5FX - the time nut -who often visits eevblog -  could provide some useful comment I am sure.

[nctnico]

Accuracy and precision are not the same thing, and both are required in this case.

Actually absolute accuracy isn't required here since John is comparing two frequencies to determine whether he programmed the PLL correctly. A simple crystal referenced frequency counter will do just fine if the measurements are taken within a couple of minutes. It also depends on the programming steps of the PLL. If it can do 6.0, 6.25 or 6.5 then 6.23 is close enough to say it is OK. If it can do much finer steps then the measurement needs more significant digits. The question boils down to: how many significant digits does the measurement need?
Another thing to look for is whether the PLL has a stable lock. For that purpuse it is a good idea to use FFT or a spectrum analyser to look at the spectrum of the output signal.

[MT]

Simple frequency counters may be based on a simple crystal, quality frequency counters are based on a aged ovenized crystals
and some people use NIST http://www.nist.gov/pml/div688/grp40/wwvb.cfm

I would assume that the time-base is sufficiently accurate to measure correct frequency of a 40 MHz and 250 MHz sine wave.
I essentially want to know that what I see on the scope is the true frequency

To do this he needs a "known" accurate time base.

How would one go about verifying the scopes functionality with a "frequency reference standard" of some kind? Which one to buy?
How would one get best accuracy when measuring frequency with oscilloscope?

http://www.thinksrs.com/products/SR625.htm
http://www.thinksrs.com/products/FS725.htm

[John_ITIC]

Thanks all for the input. So, I could (for instance) buy an Agilent 53131A/53132A with GHz option and possibly a higher precision time base built-in. And then use the 1K/51 resistive divider to connect to the 50 ohm input on the counter?

I'm currently reading in on the Agilent 53131A/53132A to get familiar with its feature. It looks like the 10 MHz reference on the rear panel can be output for use by my  DSO81304A's "ext reference" input so I don't have to use a dedicated external 10 MHz reference?

What external 10 MHz reference would you recommend, if I need/want to buy a separate one?

I also have a stack of my books on the subject to read in on the theoretical details...

[AndyC_772]

What frequencies do you get if you measure carefully and manually using the cursors, rather than relying on built-in measurements?

[John_ITIC]

What frequencies do you get if you measure carefully and manually using the cursors, rather than relying on built-in measurements?

Based on the above first scope screen shot, there is exactly 5 x 5ns between the x-axis transitions, which works out to exactly 40 MHz. But that is only for a single cycle. Variation in the measured period is large. Possibly due to spread-frequency clocking, which I'm not sure how to measure. I still have yet to read in on if/how a time-base improvement would improve the max/min/mean frequency reported by the scope.

I'm quite close to buying a HP 53131A with the 2.5E-9 (0.0025 PPM or 2.5 PPB) OCXO built-in. I can then also use this oscillator to drive the scope's time base to see if it changes the result. I then have two ways to validate the frequency. I doubt I will need a rubidium reference. The specs for the system that will use the 40 MHz oscillator states +/- 50 PPM. The one thing that it a mystery is why my scope, which is supposed to handle 1PPM accuracy, measured some 2000 PPM error in the oscillator frequency. It seems to indicate a flakey oscillator, possibly due to bad crystals after multiple reflow cycles. The HP 53131A should give me more data points...

[tggzzz]

Based on the above first scope screen shot, there is exactly 5 x 5ns between the x-axis transitions, which works out to exactly 40 MHz. But that is only for a single cycle. Variation in the measured period is large. Possibly due to spread-frequency clocking, which I'm not sure how to measure.

You can use a hammer to put screws in, but a screwdriver is usually preferable.

In this case, use a spectrum analyser, which is the traditional tool for measuring clock quality. It will, when properly used, show both jitter (a.k.a. phase noise) for a nominally single-frequency clock, and also the deviation for a modulated (e.g. spread spectrum) clock.

[AndyC_772]

Sorry, John, I think you're in danger of throwing additional equipment at a problem which your scope is perfectly capable of solving.

Set your scope to trigger on the rising edge of the clock, and set its time base such that you get one complete period on the screen, with the trigger point near the left edge of the graticule. Set the vertical scale and position such that the mid-point of the rising edge (ie. the trigger threshold) is in the middle of the screen, and such that the signal occupies most of the height of the display.

Set one X cursor precisely at t = 0, and the other cursor on the next rising edge, which should be near the right edge of the display.

Now, adjust the horizontal position and time base settings so you can see that second rising edge in detail, ie. so it completely fills the display. It should look like a diagonal line from bottom left to top right. Using this zoomed-in view, position the second X cursor as precisely as you can on the point where the trace crosses the trigger threshold.

This should give you a much more accurate indication of the period of one cycle.

If the trace isn't steady, then this indicates either noise or the deliberate use of spread-spectrum techniques. In this case, try turning on display persistence to see the extent of the variability, and take measurements to each extremity.

Another technique you can use is to measure over multiple periods. So, rather than placing the second X cursor on a point which is one cycle after the trigger, put it on, say, the 10th rising edge. As before, zoom right in to that edge, place the second cursor as precisely as you can on the point where the rising edge crosses the trigger point, then divide the measured X distance by 10.