4
\$\begingroup\$

I am a bit puzzled. I want an oscilloscope which allows me to watch two sinus albeit with a very small phase shift.

I have considered the usual parameters of scopes, namely: bandwidth, resolution, sampling rate and rise time. But I don't manage to put my head around it.

The sine have a frequency on the 100-200 Hz and amplitudes of about 2 Vpp. So typically small bandwidth will do. But I want to see phase shifts of 5ns -- 5 µs. I guess I need at least 18 bits resolution, but then? I don't want to go to the motto, I'll take a scope that overkills all, and be sure, because, those tend to cost slightly more.

Just to try and be clearer, I am not asking for a scope recommendation (brand and all), but which specification of a scope should I check to see such small shifts?

\$\endgroup\$
11
  • \$\begingroup\$ A scope is the wrong tool. At those frequencies, use a PC sound card. Write your own analysis s/w, or you may be able to find audio analyser software that does something suitable. And yes, 5nS is resolvable, with a long enough record. \$\endgroup\$
    – Neil_UK
    Jun 17, 2017 at 20:23
  • 4
    \$\begingroup\$ Oscilloscopes don't have resolutions of 18-bits. Most are 8-bit! Amplitude resolution may not be important for measuring phase anyway. \$\endgroup\$ Jun 17, 2017 at 20:42
  • 2
    \$\begingroup\$ Your problem is that 5ns intervals occur 200 million times a second, ie. 200 MHz! You might need a sample rate of 2GSa/s to see this clearly and that's getting pricey. \$\endgroup\$
    – Paul Uszak
    Jun 17, 2017 at 20:58
  • \$\begingroup\$ This is a fairly dubious requirement, but the best attempts at reaching it would probably involve acquiring a lot of data (both sample rate and amplitude resolution) from each, performing a very long fourier transform on each, and comparing the calculated phases to one another. \$\endgroup\$ Jun 17, 2017 at 23:24
  • \$\begingroup\$ @user287001 so are you telling me that when I resolved a 0.01 degree phase change between two 10kHz waveforms, a time difference of 2.8nS, back in 2013, using a PC card at 48kHz, PortAudio, python and numpy, I was somehow hallucinating? When I hear an assertion I don't understand, I ask 'how?'. You on the other hand say 'you can't', like you know better or something. It's rude to the person, unscientific, and deprives you of an education. It required long averaging times, and all of my DSP and measurement skill, to separate the measurement from confounding factors. \$\endgroup\$
    – Neil_UK
    Jun 18, 2017 at 4:16

2 Answers 2

3
\$\begingroup\$

If you want to see 5ns timing difference between your two 100Hz 2Vpp sinewaves from a single cycle observation, you have taken a hopeless task. At zero crossing the voltage difference of those sinewaves is about 3uV. You really cannot expect to see reliably that low voltage differences in ordinary oscilloscopes. The signals aren't sharp pulses but slooooow ramps.

24 bit voltage resolution (1V full scale) would in theory allow about 2% accuracy for that 3uV if the timing resolution is in accordance and the noise is filtered to low enough with band limiting and long averaging. The needed timing resolution is about 5ns/50 = 0,1ns. That means 10GHz sampling rate.

Forget it. Or specify some lower needed accuracy "how much error is allowed to the 5ns timing difference". Until this you have specified nothing. 18bit resolution is given, but not said what range must be quantized to 18 bits.

ADDENDUM:

Commentator @Neil_UK proposes to get smart instead using brute force. So, in theory you for example could generate sharp pulses from the zero crossings, extract a high harmonic and get more easily measurable phase difference from that 5ns time shift. That's possible in analog domain. But to get it properly extracted from the noise needs a signal processing mathematician to help. That is well beyond my zone of comfort.

\$\endgroup\$
3
\$\begingroup\$

It depends what you mean by 'see'. As you mention an oscilloscope, you might mean, take a single shot picture of a two-channel capture.

Unfortunately, you are up against the fundamental limit of phase noise on your signals. A 100Hz signal has a period of 10mS. If you want to 'see' 5nS, that's one part in 2 million of a cycle. Even if your oscilloscope had a 2Mpixel wide display, or even if you zoomed in 20,000:1 to concentrate on the zero-crossing part of the waveform, the noise present on the signal would exceed the signal by orders of magnitude, making such a measurement impossible.

Your only hope to resolve such a small difference is averaging to reduce the noise, and a lot of averaging. I discovered that to see a 3nS difference start to emerge from the noise, between two 10kHz signals using a cheap PC sound card, took about 10 seconds of reading, that's half a million samples at 48kHz, and to be confident took another factor of 10 above that.

Oscilloscopes don't tend to have the facility to do that. Because they are usually (well, obviously) geared to seeing waveforms in a single shot, the vertical resolution tends to be very limited, 8 bits is usually pretty good going for a 'scope, only specialist slow ones have more.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.