How much noise floor and other things matter in oscilloscope usability

[mawyatt]

The two tone IMD looks good as one would expect from a "True" 16 bit system. If you don't mind could you do this test at ~1MHz with the Picoscope 4262?

Here you go - this instrument is not sensitive enough - you can barely see the IM3 products at 990 and 1020 kHz.

Signal_IMD_40mV_1000-1010kHz

This looks good, but the tone levels are -6dBV below the level used at 20KHz so one would expect the 3rd order IMD to be significantly down from the 20KHz case.

Anyway, thanks for the test.

Best,

[G0HZU]

Note also that the other spurious terms are limiting the spurious free dynamic range on the 1MHz IMD test. A really good (old school swept) microwave spectrum analyser can achieve a typical IP3 limited SFDR of about 112dB with a very narrow RBW on the first frequency range up to a few GHz. However, 105dB is more realistic at (say) 10Hz RBW. I wouldn't expect to see those other spurious terms either when using a conventional spectrum analyser. The Pico will have a much faster refresh rate though!

On narrow spans the phase noise will slightly limit the SFDR of the swept analyser so usually stuff like this is done at a wider frequency spacing with a conventional analyser.

[Performa01]

This looks good, but the tone levels are -6dBV below the level used at 20KHz so one would expect the 3rd order IMD to be significantly down from the 20KHz case.

Sorry, this was not indended – somehow I did not pay attention to the levels.

Please find attached the correct measurement. The result is the same as at 20 kHz.

Signal_IMD_80mV_1000-1010kHz

Note also that the other spurious terms are limiting the spurious free dynamic range on the 1MHz IMD test. A really good (old school swept) microwave spectrum analyser can achieve a typical IP3 limited SFDR of about 112dB with a very narrow RBW on the first frequency range up to a few GHz. However, 105dB is more realistic at (say) 10Hz RBW. I wouldn't expect to see those other spurious terms either when using a conventional spectrum analyser. The Pico will have a much faster refresh rate though!

On narrow spans the phase noise will slightly limit the SFDR of the swept analyser so usually stuff like this is done at a wider frequency spacing with a conventional analyser.

Of course you are right – and just for others to put this into perspective, I would like to add:

Swept spectrum analyzers only “see” their resolution bandwidth at any point in time (ok, only true for the last IF), whereas the DSO always works at full bandwidth (5 MHz in this particular case). Under these conditions, some 96 dB dynamic is all you can expect from a 16 bit system – everything beyond that is just a lucky incident based on the specific conditions and the results cannot be trusted any longer.

In the previous example with the 6 dB lower level, it has been perfectly possible to measure an IMD of 109.6 dBc, since none of the spurs got in the way of this measurement. According to the textbook theory, it should have been 115 dBc though, so the measurement was flawed anyway. As mentioned before, we cannot expect great accuracy once far outside the first order dynamic range of the acquisition system.

[G0HZU]

That really is very impressive from the Picoscope in terms of SFDR.

I just turned on the RSA3408A to look at the LF noise floor up to 25kHz and I've added a plot below. This analyser is FFT only and can't do a swept measurement at any frequency. Below 40MHz it feeds direct to a 14 bit ADC and the IMD performance isn't that good. It's much worse than the Picoscope in this respect. However, the LF noise floor is quite good considering this isn't a dedicated AF analyser. I've used it to measure the noise figure of AF amplifiers a few times. As long as I provide enough gain to overcome the noise figure of the 3408A it can make fairly good noise figure measurements. I guess not many people make AF amps with 50 ohm ports but this type of amplifier is popular in direct conversion receivers.

You can see the noise floor is a fairly flat -154dBm/Hz across the AF band.

[David Hess]

I wish we had better data on available RF MOSFET noise characteristics.  What is available is intended for RF amplifier applications.

Can you measure the noise parameters yourself at audio frequencies? I've done this stuff up at RF and recently measured the s-parameters for the BF998 MOSFET at various bias points across a frequency range of a few MHz up to 3GHz and I also created some noise data for it up at VHF. This noise data gets included in the s-parameter file. I did the same for the old BF981 a few years back with good results when designing amplifiers for low noise figure. I've never tried to do this at audio frequencies though.

Up through audio frequencies would not be sufficient because RF MOSFETs can have a flicker noise corner frequency in the MHz range.

I am just not setup to make that kind of measurement easily.  I can make spot noise measurements up to 1 MHz but even that would not be high enough.  I would have to build something custom and I would prefer a more general solution.

[Performa01]

You can see the noise floor is a fairly flat -154dBm/Hz across the AF band.

Yeah - bipolar technology, using rf-transistors (with very low intrinsic base resistance) makes for a good noise matching at low impedances like 50 ohms – and a low 1/f corner frequency.

High impedance FET input stages are noisy under such conditions. On the other hand, high impedance inputs are much more versatile. We can adapt them to any impedance we like by means of a pass through terminator (at least at low frequencies).

Meanwhile I’ve experimented a bit further and detected at least two flaws in my previous noise measurement:

1.   The input was AC coupled by accident, which of course increases LF-noise significantly.
2.   The input had a 50 ohm through terminator fitted, but since this scope is sensitive to the source impedance, an additional 50 ohm end terminator should be used to complete the 50 ohms setup.

Now look at the screenshot attached.

Pico4262_Noise_5MHz_D50kHz

I’ve tried to resemble your settings as close as possible but still kept the total FFT bandwidth at 5 MHz in order to keep the high frequency noise out of the LF region. Display units are dBm now for better comparability. Frequency step is 38.15 Hz, which is equivalent to a RBW of 112 Hz with the Flat-Top window – so noise levels will read slightly higher than in your setup.

With a noise level of -134.7 dBm this is very comparable to your RSA 3408A above some 30 kHz.
At 1 kHz, the FET input goes up by 17.8 dB to -116.9 dBm, but obviously stops at -114 dBm with this RBW.

So I’m confident to claim that the Pico 4262 has the same low noise in a 50 ohm system, as long as you keep the input DC-coupled and stay above 30 kHz.

EDIT: I have updated my original posting, where you can also see the updated noise density plot.

[Fungus]

I would like to thank you everybody for the many replies.
You have been very important and educative to convince me in the decision that in my work It would be better an oscilloscope with a low noise front end than one with a very fast ADC like the Rigol.
I am receiving an sds2104x plus in the next two days so I will do a limited comparison with the Rigol mso5000 that I still have.

Still waiting for the screenshots of your ripple with averaging turned on... 

[gf]

I would like to thank you everybody for the many replies.
You have been very important and educative to convince me in the decision that in my work It would be better an oscilloscope with a low noise front end than one with a very fast ADC like the Rigol.
I am receiving an sds2104x plus in the next two days so I will do a limited comparison with the Rigol mso5000 that I still have.

Still waiting for the screenshots of your ripple with averaging turned on... 

Looking at this post again, I can believe that the first image shows noise from the scope (although it is quite a lot). But I rather cannot believe that the 40mV_pp "noise band" on top of the sawtooth in the second image is scope noise as well (apparently scope settings are the same as in the first image). I guess the latter is already present in the input signal. I don't feel able to assess whether it is random noise, or rather a high-frequency oscillation. FFT should help to reveal it.

[Fungus]

Still waiting for the screenshots of your ripple with averaging turned on... 

Looking at this post again, I can believe that the first image shows noise from the scope (although it is quite a lot). But I rather cannot believe that the 40mV_pp "noise band" on top of the sawtooth in the second image is scope noise as well (apparently scope settings are the same as in the first image). I guess the latter is already present in the input signal. I don't feel able to assess whether it is random noise, or rather a high-frequency oscillation. FFT should help to reveal it.

I'm just interested in what an MSO5000 can do with a signal like that when a user really uses all the provided features.

[Fungus]

Also... the color gradient mode as mentioned on the first page. How would the ripple appear if you enable that?

[nctnico]

For color grading to be usefull you need a perfectly repetitive signal. After all color grading is a form of averaging. Power supply ripple isn't perfectly repetitive so color grading won't help at all.

Just face it: you will want to use an oscilloscope with the least amount of internal noise to look at any signal. After all the purpose of an oscilloscope is to look at the shape of a signal and the less an oscilloscope distorts that signal, the better. It seems Rigol dropped the ball where it comes to noise reduction and at some point cheap doesn't make up for poor performance.

[Fungus]

For color grading to be usefull you need a perfectly repetitive signal. After all color grading is a form of averaging. Power supply ripple isn't perfectly repetitive so color grading won't help at all.

I'd still like to see a screenshot of it.

[Fiorenzo]

I am going to do the photos you asked, maybe this evening or tomorrow

[David Hess]

Frequency step is 38.15 Hz, which is equivalent to a RBW of 112 Hz with the Flat-Top window – so noise levels will read slightly higher than in your setup.

With a noise level of -134.7 dBm this is very comparable to your RSA 3408A above some 30 kHz.
At 1 kHz, the FET input goes up by 17.8 dB to -116.9 dBm, but obviously stops at -114 dBm with this RBW.

If I did my math right, that comes out to  3.9 nV/SqrtHz so similar to a well designed 100 MHz JFET input, and consistent with the specified 15 picofarad input capacitance.  (1) For a lower frequency singled ended JFET input instrument, 1 nV/SqrtHz is possible (LSK170) but the input capacitance would be 2 or 3 times higher.  So why is the input capacitance low and noise high for such low bandwidth?

Given their dynamic range and distortion requirements for 16-bits, a simple FET source follower would have too much distortion; feedback is required to lower the distortion.  So they probably used a JFET operational amplifier, and that would be consistent with higher noise, 5 MHz bandwidth, and a 15 picofarad input capacitance.

That also places this instrument into a different class than an oscilloscope, although similar to the old Tektronix oscilloscopes which used the 5A22 or 7A22 differential amplifier.

(1) There is a close relationship with input capacitance, bandwidth, and input noise.  Lower bandwidth FETs have lower noise and higher input capacitance.

[David Hess]

I would like to thank you everybody for the many replies.

Hopefully the discussion was of some help.

You have been very important and educative to convince me in the decision that in my work It would be better an oscilloscope with a low noise front end than one with a very fast ADC like the Rigol.

I do not think low noise and high sample rate are mutually exclusive because the digitizer's input noise should be insignificant compared to the noise from earlier stages and especially from the amplified noise of the high impedance input buffer.  The instruments in question seem to suffer from higher noise in general rather than because of sample rate.

[Fiorenzo]

So i did a lot of photos.
First two comparison: no probe, 1mv/div, 20mhz BW limit. There is a huge difference

[Fiorenzo]

Same no probe

[Fiorenzo]

4x avarage no probe

[Fiorenzo]

Eres 3db on siglent and hi-res on rigol
no probe

[Fiorenzo]

Normal mode
Probe connected 1x and grounded

[Fiorenzo]

Ripple same settings on both oscilloscope
AC, normal mode, 1x probe

[Fiorenzo]

There is a huge difference between this two scopes.
The Rigol has a faster update of the image, almost double... and the quality of the rappresentation, i mean the graphics, of the signal is a lot better than the Siglent. It seem like the Siglent had a lower resolution. There is not a huge difference in the speed of the user interface and also the Rigol appear to have a better quality of the material of the scope. But the front end noise is a lot different.
Judge by your self from the photos attached.

[David Hess]

Normal mode
Probe connected 1x and grounded

Shorting the probe tip can be tricky.  For lowest noise it is not sufficient to simply clip the probe's ground lead to the probe's tip because the loop will pick up ambient noise.  Best is to short the tip out with a coaxial probe tip to BNC adapter plugged into a BNC short or 50 ohm termination, but winding wire around the probe tip also works.

[nctnico]

This is even worse than I expected ( ). I don't care about the open / shorted inputs at the most sensitive V/div (Rigol does digital zoom there so it is not an apples for apples comparison) but I do care about the display of an actual signal. On the Siglent you can clearly see spikes on the signal which are completely obscured on the Rigol.

[Fungus]

So i did a lot of photos.
First two comparison: no probe, 1mv/div, 20mhz BW limit.

Irrelevant. Only real signals count.

4x avarage no probe

Averaging only works when there's a periodic signal.

Ripple same settings on both oscilloscope
AC, normal mode, 1x probe

No averaging?

That's the only thing that counts - if averaging mode can better show the underlying signal or not.

Normal mode
Probe connected 1x and grounded

Shorting the probe tip can be tricky.  For lowest noise it is not sufficient to simply clip the probe's ground lead to the probe's tip because the loop will pick up ambient noise.

Yep. Connecting the ground clip to the probe actually creates an antenna.