EEVblog Electronics Community Forum
Products => Test Equipment => Topic started by: besauk on August 15, 2018, 05:52:20 pm
-
Having recently acquired a DSO, I was curious to see how accurate the timebase was. The built-in frequency counter tied to the triggering seemed decent enough - although not enough digits to really get a solid determination. While fiddling with a long time scale on a fast signal (was looking to get a high resolution FFT at the time), I went well past the Nyquist limit, and realized I could use that to compare timebases of my signal source and scope. Below is a screenshot taken with a 10MHz sine wave from my signal generator into the scope (used two channels out from the signal generator into channels one and two of the scope, just to convince myself there was no timebase variation between the channels). Note that the sample rate is 2.5MSa/s - so we are aliasing big time. The frequency of the aliased signal gives the delta from the the scope's timebase compared to the signal generator's. By adjusting the signal generator up or down that amount, it was trivial to determine which direction it was off - in this case the scope was on the low side. I had previously done an analysis of my signal generator compared to a WWV disciplined clock and determined it was low by 0.4 ppm. The scope is an additional 1.6 ppm lower than the signal generator (16 Hz / 10MHz) for an absolute error of -2ppm. Of course, if I had a solid reference like a Rubidium source, or GPS disciplined receiver, I could have gotten to the errors even faster, with more confidence.
One side note, there is some amount of wiggle on the aliased signal - you can see a std dev of about 0.5 Hz - so the relative jitter between the two devices is not too bad. Can't tell which one is contributing more to the jitter.
Just wanted to share a fun use of aliasing to quickly gauge how accurate (or more correctly how closely matched) two timebases are. All in all, I'm pretty pleased with the absolute accuracy of both units - both well within spec.
-
That is one of the most impressive applications of understanding to a problem I have ever seen. Very cool.
Thank you!
Reg
-
The old fashioned way to compare frequencies, which also works for low-order harmonics, is to display lissajous figures.
And they are prettier :)
-
Old DSOs use the deliberate aliasing trick as part of their calibration procedure.
-
The old fashioned way to compare frequencies, which also works for low-order harmonics, is to display lissajous figures.
And they are prettier :)
If I had access to the internal clock (well really, a suitably divided down signal), I would have done a lissajous figure - agreed on their utility (and aesthetics!) One of my first projects was a freq counter built on a design published in Pop Elec some 40 years ago - to calibrate it, I went over to a friend's house to use his dad's scope, and we used line frequency lissajous figures to work our way up to many multiples of line. Looking back now, two things about that were notably bad - first, short term line freq is really pretty awful as a timebase (although it was only a 5 digit counter), and second, I used monostables to drive all sorts of state in the design. Horrible! Eh, live and learn.
-
Won't scopes with automatic anti-aliasing prevent this?
-
Won't scopes with automatic anti-aliasing prevent this?
No such feature on my Siglent... But if a higher-end model BW limits the input based on the sample rate, and there is no way to turn it off, well then you are right, no dice on this approach. Not a lot of effort to try in any case.
-
Tried on my Keysight, no matter the time base setting it wouldn't do it and the frequency counter was always bang on.
-
The DSO equivalent to the Lissajous pattern.
-
Tried on my Keysight, no matter the time base setting it wouldn't do it and the frequency counter was always bang on.
And it have real digital side trigger system? (and if freq counter is trigger counter)
-
Tried on my Keysight, no matter the time base setting it wouldn't do it and the frequency counter was always bang on.
I remember reading that Keysight has an anti-alias algorithm, so it's harder to get it to alias. The algorithm dithers the capture data while it's decimating it for display, so the displayed capture points are not evenly spaced in time and this prevents aliasing.
However, on some functions you don't want dithering. They turn it off when either the math FFT or differentiation functions are on (and perhaps others?).
I think this was mentioned in a paper or app note but I can't find it at the moment.
Here's a 10MHz input signal at 200ms/div without differentiate and then with differentiate turned on, using a X3104A.
-
It's interesting to set up your scope for this measurement and then turn it off and let it cool down. When you turn it on later, watch how the frequency error shifts as it warms up. My LeCroy 9384L starts out about 10 Hz off. The frequency drops, goes through zero, keeps on going, and within 10 minutes settles to its final value of around 6.5 Hz. So the total shift is about 16 - 17 Hz.
Ed
-
Tried on my Keysight, no matter the time base setting it wouldn't do it and the frequency counter was always bang on.
I remember reading that Keysight has an anti-alias algorithm, so it's harder to get it to alias. The algorithm dithers the capture data while it's decimating it for display, so the displayed capture points are not evenly spaced in time and this prevents aliasing.
However, on some functions you don't want dithering. They turn it off when either the math FFT or differentiation functions are on (and perhaps others?).
I think this was mentioned in a paper or app note but I can't find it at the moment.
Here's a 10MHz input signal at 200ms/div without differentiate and then with differentiate turned on, using a X3104A.
Very cool, I'll give it a shot tonight.
-
Tried on my Keysight, no matter the time base setting it wouldn't do it and the frequency counter was always bang on.
I remember reading that Keysight has an anti-alias algorithm, so it's harder to get it to alias. The algorithm dithers the capture data while it's decimating it for display, so the displayed capture points are not evenly spaced in time and this prevents aliasing.
However, on some functions you don't want dithering. They turn it off when either the math FFT or differentiation functions are on (and perhaps others?).
I think this was mentioned in a paper or app note but I can't find it at the moment.
Here's a 10MHz input signal at 200ms/div without differentiate and then with differentiate turned on, using a X3104A.
Oh neat trick there. I gotta remember that one as most of my scopes are Agilent.
-
The MSO/DSO6000 series will also disable the randomizing feature when you select hi-res acquisition mode, if I recall correctly.