Siglent SDS800X HD 12 bit DSO's

[tautech]

I have a probably stupid FFT question, but I can't figure out what I am seeing here.

Pulse trigger set vs Edge for Martin's screenshots.

[Martin72]

NTP server not yet set ? 

Looked one up for you:
TU Berlin   times.tubit.tu-berlin.de   130.149.7.7

Sync failed, tried several ones..

[tautech]

NTP server not yet set ? 

Looked one up for you:
TU Berlin   times.tubit.tu-berlin.de   130.149.7.7

Sync failed, tried several ones..

Gateway correct in LAN settings ?

[egonotto]

Hello,

Maybe you can transfer the data to a PC and then do an FFT with Octave or Matlab. Then you can see whether the problem is the data or the FFT

Best regards
egonotto

[Martin72]

Tell me about your generator settings, my signal is a bipolar square wave.

[gf]

I already wanted to point out the 1.000010 kHz versus 1.000012 kHz difference as well, but you were faster.
What is the risetime of the square wave? I guess the edges are significantly undersampled at 5 MSa/s?

[tautech]

My efforts using the 1kHz probe comp output providing ~10Hz RBW

[gf]

My efforts using the 1kHz probe comp output providing ~10Hz RBW

Thanks for looking into this as well. To try whether you get the surprising "sidebands" on the individual peaks, zoom in on one of the peaks in the spectral domain, say from 4.5 to 5.5 kHz. Then gradually switch to slower time bases, 50 to 500 ms/div.

And a decibel scale is stringently required, since your spurs are in the order of -60dBc.

If you increase the risetime of the SDG2000X squarewave more and more, does the effect vanish at some point?

[tautech]

No sidebands displayed on the fundamental with ~4Hz RBW

[tautech]

What ghosts are we chasing @-70dBm ? 
https://www.eevblog.com/forum/testgear/sds800x-hd-review-demonstration-thread/msg5293765/#msg5293765
https://www.eevblog.com/forum/testgear/sds800x-hd-review-demonstration-thread/msg5384846/#msg5384846

[rf-loop]

I have a probably stupid FFT question, but I can't figure out what I am seeing here. I was looking at the FFT of a 1 kHz square wave. As expected, the odd harmonics show up nicely. But when I go to slower time bases to increase the spectral resolution, each harmonic peak splits up, as if there was a modulation on top. In addition, the "modulation fequency" changes with the time base chosen.

Please see two example screenshots attached, at 50 ms/div and 100 ms/div, showing side peaks at 30 and 60 Hz increments, respectively. If I can provide any additional information, of course please let me know. I appreciate any hints what is causing this -- thanks!

Sometimes Oscilloscope + HP67 or slide rule is good combination, or just pencil and grid paper (= oscilloscope + thinking)

One could also say that if it didn't work like that - it would be broken. 



5 point hint : Try to adjust your input fundamental freq. as exactly as possible to sync with oscilloscope sampling clock. 


Here 3 images.
If think oscilloscope sampling clock is perfectly right (so adjust first generator so that oscilloscope trigger counter display is between 999.9999Hz and 1.000000kHz for rough "sync" ).


1kHz square is here 10Hz up (ref oscilloscope)
Because this is 5th harmonic you see sides 50Hz raster (and remember our signal is here square, sidebands looks like bit same as sine modulated with square)


1kHz square is here 10Hz down (ref oscilloscope)
Because this is 5th harmonic you see sides 50Hz raster


But here input signal is  very close sync with oscilloscope sampling clock... no sidebands

It is fun exercise if go to 2499 kHz (2499th harmonic) and if look it ... you can see peaks under 2499 and over 2499 (depending how close 1kHz Square signal is sync with sampling clodck. Adjust more and more close and when perfect you see only one peak exatly 2499kHz... (1)  remember fNyquist and fold back... and remember there (it need be really close 1kHz because error is multiplied with 2499 (ref is oscilloscope clock - and because we can not lock generator and scope to same reference...we may see some continuous fluctuation/drifting there )
(1) This is now quite exactly 1kHz as oscilloscope see it.

Some times it is really annoying that still in this millenum with modern digital oscilloscopes we do not have freq reference input in most oscilloscopes and internal reference is very cheap simple xtal osc.

[gf]

Thanks, rf-loop!

We were getting close yesterday night -- had noticed that the effect disappears when the input signal was tuned to exactly 1.0000 kHz (#948), and gf must have been thinking of high harmonics beyond the Nyquist frequency (#949). I just didn't think aliasing could be an issue, since I also saw this at 100 MSa/s and with the 20 MHz bandpass engaged. But I guess that bandpass is not perfect...

It is still counter-intuitive to me that the calibration square wave, with its risetime of nearly 1µs, produces aliasing even at 50 MSa/s. But of course it's way down in level.

Thanks for the hints! This is along the lines I expected from the start: some artefact due to the principle of operation which I should be aware of so I don't misinterpret real-world signals going forward. Having a low-noise 12-bit scope lets me see things, inlcuding such artefacts, which I never came across with the old scope!

The 20 MHz limit is first order, i.e. only 6dB/octave or 20dB/decade.
So even at 1GHz (for example) you get only 34dB (additional) attenuation.

Look below what you get if you sample an ideal (not bandwidth limited) 1000.012 Hz square wave at 5 MSa/s.
Does the plot remind you of something?
Your noise floor is higher, though, so you can't see the bottom of this plot any more on the scope.

Code: [Select]

x = [-1 1](1+floor(mod([0:2*1024*1024-1]*1000.012/5e6*2,2))); % sample square wave
win = blackman(2*1024*1024,"periodic")';
win /= mean(win);
f = [0:2*1024*1024-1]/(2*1024*1024)*5e6;
semilogy(f,abs(fft(x.*win)/length(x))); grid on
xlim([0 6000])
ylim([1e-6 1])


[tszaboo]

I'm liking this scope so far. The software seems a lot better than the SDS1000x that I tried, and it's seems a lot more capable. The most important functions are avaliable on buttons and knobs, all the rest is on the touch. The UI is built up around the touch interface, and it's not an afterthought.
I also found the first annoying part, the overload recovery of the front end. This is with the built in probe compensation output. The signal doesn't look good at all, voltage level is incorrect, shape is incorrect.

[2N3055]

I'm liking this scope so far. The software seems a lot better than the SDS1000x that I tried, and it's seems a lot more capable. The most important functions are avaliable on buttons and knobs, all the rest is on the touch. The UI is built up around the touch interface, and it's not an afterthought.
I also found the first annoying part, the overload recovery of the front end. This is with the built in probe compensation output. The signal doesn't look good at all, voltage level is incorrect, shape is incorrect.

As was demonstrated before, 12 bit dynamic range and vertical zoom (not overdriving inputs, but actually entering zoom mode) renders overdriving inputs mostly unnecessary.

[tszaboo]

I'm liking this scope so far. The software seems a lot better than the SDS1000x that I tried, and it's seems a lot more capable. The most important functions are avaliable on buttons and knobs, all the rest is on the touch. The UI is built up around the touch interface, and it's not an afterthought.
I also found the first annoying part, the overload recovery of the front end. This is with the built in probe compensation output. The signal doesn't look good at all, voltage level is incorrect, shape is incorrect.

As was demonstrated before, 12 bit dynamic range and vertical zoom (not overdriving inputs, but actually entering zoom mode) renders overdriving inputs mostly unnecessary.

Unnecessary maybe, but you might end up with the wrong settings on your scope when probing an unknown signal.
I don't think it should be a dealbreaker for anyone.

Trying it a little bit more. Here is a stepdown power supply, which is not properly bypassed, and compensated, driving a motor. This is in DC coupling, and zoomed in 10x on the horizontal scale  . You might need a 2000 EUR power rail probe to probe this not that long ago.

[2N3055]

I'm liking this scope so far. The software seems a lot better than the SDS1000x that I tried, and it's seems a lot more capable. The most important functions are avaliable on buttons and knobs, all the rest is on the touch. The UI is built up around the touch interface, and it's not an afterthought.
I also found the first annoying part, the overload recovery of the front end. This is with the built in probe compensation output. The signal doesn't look good at all, voltage level is incorrect, shape is incorrect.

As was demonstrated before, 12 bit dynamic range and vertical zoom (not overdriving inputs, but actually entering zoom mode) renders overdriving inputs mostly unnecessary.

Unnecessary maybe, but you might end up with the wrong settings on your scope when probing an unknown signal.
I don't think it should be a dealbreaker for anyone.

You are correct, no dispute there. But I simply try to fit whole signal to screen first (vertically), and then go from there to what is the best way..
Like you have shown, if you have large DC offset, that works well, as long as dynamic part of waveform stays in screen bounds...

For example, a 100mv P-P signal riding on top of 4,95V DC offset. On top of that again, zoom, looking at detail with 10mV/div :

[rf-loop]

I'm liking this scope so far. The software seems a lot better than the SDS1000x that I tried, and it's seems a lot more capable. The most important functions are avaliable on buttons and knobs, all the rest is on the touch. The UI is built up around the touch interface, and it's not an afterthought.
I also found the first annoying part, the overload recovery of the front end. This is with the built in probe compensation output. The signal doesn't look good at all, voltage level is incorrect, shape is incorrect.

Not personal, not so seriously...

An oscilloscope is a relatively complex measuring device. I have often seen it misused. A hammer is a fairly simple tool, and even that can sometimes be seen being misused.
If one have a multimeter with manual range selection. He set the range to 1V and then put 3V into it and then not happy when it measures wrong.
Should the oscilloscope have made with the blonde protection so that it sets the voltage range automatically. 


With this V/div offset max is 8V. And full range is tiny bit over 800mVpp.
With this kind of signal, offset 8V and signal full scale, it works perfectly as it is designed. (There is still bit more room before front end overdrive but not much)

And if want more room before overload...  102mV/div instead of 100mV/div and it change lot.

[gf]

It is still counter-intuitive to me that the calibration square wave, with its risetime of nearly 1µs, produces aliasing even at 50 MSa/s. But of course it's way down in level.

If we talk about the same kind of aliasing spurs that have been discussed, I would not expect them to be more than about -100dBc (relative to the amplitude of the 1000.012 Hz fundamental) if the rise/fall time is 800ns, and at a sample rate of 5 MSa/s. See attached plots from a simulation.

If you nevertheless see larger spurs, then it might be worth to investigate if they have a different root cause. Also try to compare the calibration signal to a square wave with the same frequency (as measured by the scope) and 800ns rise/fall time from the SDG200X. Is there a significant difference?

Also beware of window function's selectivity. With the Siglent flattop window, disregard anything in the neighborhood of a large peak which is more than 93dB below the peak. [ For my plots I eventually switched to a Kaiser window with beta=15 to get side lobes below -120 dBc. ]

EDIT: figure2 assumes trapezoidal edges with sharp corners, and figure3 assumes smooth, S-shaped edges.

[pdenisowski]

But I simply try to fit whole signal to screen first (vertically), and then go from there to what is the best way.

Based on my experience, this is one of the first things someone should learn about using an oscilloscope.  No point in having a 12 bit ADC if you're not even using half the bits

[tszaboo]

But I simply try to fit whole signal to screen first (vertically), and then go from there to what is the best way.

Based on my experience, this is one of the first things someone should learn about using an oscilloscope.  No point in having a 12 bit ADC if you're not even using half the bits

Sure all this theory talk about how I don't know how to use an oscilloscope because pointing out a flaw...
Ok, to give you an example where you would intentionally overload the frontend:
Ultrasonic ranging sensors, single trasducer. The TX pulse is in the tens of volt range, the RX part is several orders of magnitude less. I opened a recent project of mine, and the difference was x100. And you want to place the trigger on the transmit, see any ringing after the last pulse, and see the received signal. Even with the 12 bit ADC you wouldn't have any reasonable resolution on the RX if you put everything on the screen. I'm sure there are other applications where this is regularly done.

[pdenisowski]

Sure all this theory talk about how I don't know how to use an oscilloscope because pointing out a flaw...
Ok, to give you an example where you would intentionally overload the frontend:
Ultrasonic ranging sensors, single trasducer. The TX pulse is in the tens of volt range, the RX part is several orders of magnitude less. I opened a recent project of mine, and the difference was x100. And you want to place the trigger on the transmit, see any ringing after the last pulse, and see the received signal. Even with the 12 bit ADC you wouldn't have any reasonable resolution on the RX if you put everything on the screen. I'm sure there are other applications where this is regularly done.

Actually, that's an excellent example of when NOT to make sure the entire signal fits on the screen    Thanks!

Edit:  Seriously, that is a really good example - I hope you wouldn't mind if I use it in a presentation someday

[gf]

When I switch to 50 ms/div, resulting in 5 MSa/s for the FFT as you modelled it, the spurs are about 95 dB below the fundamental.

-95 dBc is indeed completely in the ballpark, with 800ns edges from a 1st order filter
I'm overwhelmed by the good match between theory and practice

[2N3055]

But I simply try to fit whole signal to screen first (vertically), and then go from there to what is the best way.

Based on my experience, this is one of the first things someone should learn about using an oscilloscope.  No point in having a 12 bit ADC if you're not even using half the bits

Sure all this theory talk about how I don't know how to use an oscilloscope because pointing out a flaw...
Ok, to give you an example where you would intentionally overload the frontend:
Ultrasonic ranging sensors, single trasducer. The TX pulse is in the tens of volt range, the RX part is several orders of magnitude less. I opened a recent project of mine, and the difference was x100. And you want to place the trigger on the transmit, see any ringing after the last pulse, and see the received signal. Even with the 12 bit ADC you wouldn't have any reasonable resolution on the RX if you put everything on the screen. I'm sure there are other applications where this is regularly done.

Well, you missed the point that I said you are right, overload recovery distortion IS there. It is because of split path front end architecture. All modern scopes have it in some form.
You seem to misunderstood my example where I managed to achieve decent resolution from 20x magnification  from already small signal.
Software magnification of 100x is entirely usable.
Nobody said you don't know how to use scopes you have.
We are pointing out that that new capabilities introduce new methods with much enhanced results.

I don't know of ANY modern digital scope that will have no overdrive recovery distortions if you go 100x or 1000X over range..??

If you know one please let me know. I will buy one immediately..

Once you asked about comparison of this (very inexpensive, cheapest of new Siglent 12bit scopes gamut) to very expensive Keysight MSOX3000T "classic".
Since I also have MSOX3104T we can do exactly that.

Example:

4MHz damped oscillation.
15V P-P, 8V top peak.
1ms repetition.
10MHz superimposed 20mV P-P signal on top.

MSOX3104T 2V/div for full screen and 200mV/div "magnified" with front end amplification. Pay attention that DC level after the pulse is wrong. That is from recovery distortion.
Then same signal at 10mV/div. It is completely destroyed. That was 200x overdrive.

As comparison little Siglent:

SD800xHD same signal, 2V/div for full screen and 200mV/div "magnified" with front end amplification.
SDS800xHD  at 200mV/div with additional 20x zoom factor,
That is effective 200x magnification.... effective 10mV/div.
If you want to save screen space and not use zoom, you can use math channel: F1 magnified to 10mv/div and F2 ERES 10mV/div demonstrated.
10 Mhz signal is clearly visible. With cursors and little eyeballing we estimate 11mV P-P (it should be 10mV). Not bad for 200x magnified portion. If signal is repetitive we can also use averaging.

And that is a 700€+VAT scope (list price) compared to US$ 22,587+VAT list price for Keysight MSOX3104T fully loaded (right from Keysight today). I know nobody pays full price but if you get for half the price, that is still 11000€ +VAT...

SDS2000xHD has much larger input offset range while still being fraction of the price of said Keysight.
SDS3000xHD also has much larger input offset, and is also fraction of the price of Keysight while being a true powerhouse in comparison.

Hope this answers some of your questions...
Best,

[Martin72]

Earlier at work, during the break....
1kHz from the SDG1062X fed into the lecroy HDO6034A scope.
It was just a gimmick, there was no time for more settings.

[Martin72]

The vertical axis of the CH1 trace is quite a mess

We almost wanted to give it back, but then we realized that the rest is pretty cool.