Choosing between entry-level 12-bit DSOs

[nctnico]

The answers over there are quite wrong.

Am I going to be blamed for that, too?

FWIW I never even read the page, I just found the image on google image search.

It's a 5Hz sine wave sampled at 11Hz intended to show where the "AM" effect comes from.

I'm not blaming you but the people who posted the graphs on Reddit and commented on it. Drawing a line through sample points as-is is never a natural representation of a sampled signal. You'll need to put the samples through an analog or digital filter in order to reconstruct the resulting signal.

[gf]

The 450MHz signal in the 450MHz screenshot definitely is a fantasy signal. The amplitude suddenly gets higher (compared to the 408MHz screenshot) while you'd expect it to be lower.

The previous one wasn't 408, but 490 MHz (the frequency counter got confused). So the amplitudes are plausible and decrease with increasing frequency.

I don't know what the signal amplitude is at the input but likely it is quite high to at least see some signal. When increasing the sensitivity, there will be more gain so likely the front-end is now distorting due to clipping / overdrive recovery / etc.

The input amplitude was said to be 1.6Vpp, which is ~13 div at 120mV/div, or 1.67x full scale. The amp certainly has some headroom, but I don't know how much. Since Performa01 explicitly mentioned the potential overdrive issue in his post, I assume that he took care not to overdrive the amp

[tggzzz]

No, it's from here:
https://dsp.stackexchange.com/questions/60859/conclusions-of-sampling-around-nyquist-rate

IMNSHO stackexchange is OK for "which button do I press to frobnitz the splurgle?".

However, stackexchange's decision to prevent multi-level quoting prevents to-and-fro discussions of subtle and interesting points. ISTR they also have downvoting, which has similar consequences. Edaboard is similar.

A derived consequence is that subtle discussions on stackexchange have a higher probablility of being wrong than discussions here.

[tggzzz]

The answers over there are quite wrong.

Am I going to be blamed for that, too?

No, but your next statement doesn't look good.

FWIW I never even read the page, I just found the image on google image search.

[nctnico]

The 450MHz signal in the 450MHz screenshot definitely is a fantasy signal. The amplitude suddenly gets higher (compared to the 408MHz screenshot) while you'd expect it to be lower.

The previous one wasn't 408, but 490 MHz (the frequency counter got confused). So the amplitudes are plausible and decrease with increasing frequency.

I don't know what the signal amplitude is at the input but likely it is quite high to at least see some signal. When increasing the sensitivity, there will be more gain so likely the front-end is now distorting due to clipping / overdrive recovery / etc.

The input amplitude was said to be 1.6Vpp, which is ~13 div at 120mV/div, or 1.67x full scale. The amp certainly has some headroom, but I don't know how much. Since Performa01 explicitly mentioned the potential overdrive issue in his post, I assume that he took care not to overdrive the amp

You are right about that. The screendumps with different time/div and v/div settings don't make the posting very clear though.

[tggzzz]

The answers over there are quite wrong.

Am I going to be blamed for that, too?

FWIW I never even read the page, I just found the image on google image search.

It's a 5Hz sine wave sampled at 11Hz intended to show where the "AM" effect comes from.

I'm not blaming you but the people who posted the graphs on Reddit and commented on it. Drawing a line through sample points as-is is never a natural representation of a sampled signal. You'll need to put the samples through an analog or digital filter in order to reconstruct the resulting signal.

Just so.

Why did they decide to use a first-order hold reconstruction (linear interpolation) rather than a zero-order hold (step)? Typically that would be an unwitting default "looks right" choice.

Personally I hate both, and much prefer to see just the samples (i.e. dots).

[Antonio90]

I'm not so sure about the tool analogies that have been made here.

Oscilloscopes being discussed in this thread are general-purpose tools. Up to around ~20K price bracket, and speciality/niche 'scopes notwithstanding, every oscilloscope you might find will do the basic stuff similarly, fit on top of a normal table, and consume less energy than most appliances.

With regards to general-purpose tools, we can generally say that better is just better. The Siglent does more things just better than the Rigol, has more features, and the hardware platform is generally more capable. The learning curve will be steeper, and you will have to RTFM, at least to an extent. I don't think that is a problem for a beginner hobbyist TBH.

The Rigol has a more welcoming UI, I don't have it but it seems to me more intuitive. It also has less functionality. The VESA mount, HDMI, better rendering and much better probes are relevant QOL improvements though.

Just my two cents, but if you don't have any other oscilloscope, I would get the Siglent. It will work perfectly well for basic tasks, it can do Bode plots (with a Siglent AWG) which I find invaluable for learning, and will let you quite a bit more space to grow into. The growing part I only find relevant because they cost roughly the same, otherwise it's a never-ending escalation.

I would certainly have preferred the Siglent to have VESA and HDMI, but I wouldn't base the purchasing decission on that. If you are sure you will be just probing around, decoding the odd I2C/SPI around an MCU etc., then the Rigol is competitive. Or as a second 'scope to complement an older, slower, but more featured device.

[Mechatrommer]

It's a 5Hz sine wave sampled at 11Hz intended to show where the "AM" effect comes from.

lets ignore wild guess such as pulse nitpick BS by others. its clearly a simple illustration (small set from infinitely long repetitive signal) of 5Hz sampled at 11Sps. yes its good example of how AM modulated like shape is derived without interpolation. but poor interpolation still shows that AM shape, but a good Sinc implementation WILL NOT. a good practical Sinc implementation will get back the gray signal. if its not clear enough for anybody this is the second time i state it.

[Performa01]

Perhaps a better question would be whether the Rigol's interpolation implementation is flawed and starts to suffer from aliasing too early.

Reconstruction failure and aliasing are different matters. We can never see aliasing with a low distortion sine wave, as long as we stay below Nyquist.

There were some decent indications that the Siglent is better, but to properly compare apples to apples we would need to capture signals that have frequencies which are equal not absolutely, but as fractions of the sampling rates of both, and that, ideally, stay within the spec'd bandwidth, for which we would need to reduce the sampling rate artificially. Those could be say 10 frequency values from 1/5 to 1/2.5 the respective sampling rates. If someone's willing to do such a test with SDS800X HD (which goes to min. 500 MSa/s, right?), then I can do it with DHO800.

That’s a sensible approach. Trying to analyze signals that are heavily attenuated after the PGA might not always lead to meaningful results.

It doesn’t make sense to lower the sample rate on the SDS824X HD by activating more channels, because additional AA-filtering (analog + digital) gets enabled as soon as there are more than two channels active and the sample rate drops down to 500 MSa/s. Thus, anything above 200 MHz gets heavily attenuated in this configuration.

But that’s not a problem, since newer Siglent DSOs have the ability to freely choose the sample rate for all those cases, where it matters – and you’ve just created a use case for this. Of course, for this particular case, you can get away without that on a scope that lacks AA-filtering – and/or is generally undersampling with more than 2 channels active.

I’ve restricted the SDS824X HD to 500 MSa/s, hence a Nyquist frequency of 250 MHz even in single channel mode. The frequency response is the same as shown in an earlier post. The amplitude is 800 mVpp, so the signal would fill the entire screen height at 10 MHz. There is certainly no need for 10 different frequency values – I’ll show a few strategic ones only.

100 MHz = SR/5


SDS824X HD_Sine_100MHz_500MSa


200 MHz = SR/2.5


SDS824X HD_Sine_200MHz_500MSa


210 MHz = SR/2.38


SDS824X HD_Sine_210MHz_500MSa


220 MHz = SR/2.273


SDS824X HD_Sine_220MHz_500MSa

[2N3055]

The answers over there are quite wrong.

Am I going to be blamed for that, too?

FWIW I never even read the page, I just found the image on google image search.

It's a 5Hz sine wave sampled at 11Hz intended to show where the "AM" effect comes from.

I'm not blaming you but the people who posted the graphs on Reddit and commented on it. Drawing a line through sample points as-is is never a natural representation of a sampled signal. You'll need to put the samples through an analog or digital filter in order to reconstruct the resulting signal.

Just so.

Why did they decide to use a first-order hold reconstruction (linear interpolation) rather than a zero-order hold (step)? Typically that would be an unwitting default "looks right" choice.

Personally I hate both, and much prefer to see just the samples (i.e. dots).

And on Siglent you can do just that. It has Dot mode.

[ebastler]

If you look carefully at the 450MHz screenshot Performa01 posted, you'll see that the displayed signal can't be right. Just look how the histogram for the frequency measurement shows a wide spread instead of a single frequency! And also look at the amplitude compared to the 408MHz signal. IOW, the screenshot doesn't show what is actually happening. Just stick to testing fs / 2.5 + 2% or so. DSOs can show fantasy signals in such circumstances!

The mini histogram icons don't show a scale, so they mainly give qualitative information, e.g. to point out bimodal distributions. You could open the full histogram with axis scales by tapping the icon.

In the screenshot, you can look at the standard deviation of the amplitude and frequency to get an idea of the scale. Both are not concerning in my view. Hence I would stick with my conclusion that we still get a stable interpolation at sampling freq / 2.22.

[Performa01]

But I'm still wondering how 490 MHz @1GSa/s and 10ns/div look like?

490 MHz is a bit nasty – but I’m sure you are aware of that and just want to be mean 😉

Thank you! Yes, of course I wanted to be mean

If you display the same with a single-shot, then I think we would see the same kind of "AM envelope" which was observed on the Rigol.
I think it was just unable to trigger on the envelope, therefore the trace looks so weird.

Yes, we should not forget what I’m demonstrating here: torturing a dirt-cheap entry-level instrument operating far outside its specifications, with a signal that is ~2.5 times the specified bandwidth. A heavily attenuated signal close to Nyquist, with a remaining amplitude barely high enough to get a solid trigger, sitting right on the edge of a steep AA-filter, is naturally not going to be a benchmark for the scope performance.

Here you go:


SDS824X HD_Sine_490MHz_1GSa_Single

If we could do an FFT on the interpolated trace, then I'd expect to see peaks at 490 and 510 MHz (image).
[ If I think about it, maybe it is even possible with the Interpolate math function. ]

SDS824X HD_Sine_490MHz_1GSa_FFT

A less demanding configuration would be 450 MHz – this time at 2 ns/div, to show the sine wave better. There are no cyclic effects to see anymore. Reconstruction works as it should, but the amplitude is so low that I needed a little help from the vertical gain again – at 120 mV/div the trigger was fully working again, as can be seen from the trigger frequency counter.

The trace is still a bit thick. Is this really just noise? Could also be the same effect (but with much lower amplitude and 5x higher frequency of the envelope). Even if it is hardly visible in the time domain, an FFT of the interpolated trace should still reveal the amount of attenuation for the 550 MHz image (ideally it drowns in the noise floor).

Of course it cannot drown in the noise floor. There are no special measures for dual channel mode, hence the frequency response for 2 GSa/s shown earlier still applies. That means “only” some 28 dB difference between 450 and 550 MHz. But it means almost 50 dB attenuation with regard to 200 MHz, hence a respectable amount of aliasing protection. 450 MHz signal frequency on the other hand is just not a valid use case for a 200 MHz instrument.


SDS824X HD_Sine_450MHz_1GSa_FFT

Just for comparison purposes: the actual bandwidth of the SDS824X HD in dual channel mode is 245 MHz, so we need to relate the mirrored signal to that:


SDS824X HD_Sine_240MHz_1GSa_FFT

Here we see a whooping 91 dB suppression for the mirror signal.


For those thinking and asking themselves how such a good AA-protection can be possible in a serious DSO which has still good pulse response and -fidelity, here’s a tiny hint: the ADCs in this DSO can still work at 4 x 1 GSa/s…

Bottom line: even though we cannot use the high sample rate for acquisition, it’s still not wasted in this instrument (and the SDS1000X HD) 😉

EDIT: legend corrected for the first screenshot.

[shapirus]

I’ve restricted the SDS824X HD to 500 MSa/s, hence a Nyquist frequency of 250 MHz even in single channel mode. The frequency response is the same as shown in an earlier post. The amplitude is 800 mVpp, so the signal would fill the entire screen height at 10 MHz. There is certainly no need for 10 different frequency values – I’ll show a few strategic ones only.

All right, let's see. My signal is 75 mV p-p at 10 MHz (into a 1 MOhm input after a 50 Ohm feed-through terminator), which fits entire screen height as well.

Sample rate was set to 625 MSa/s.

100 MHz = SR/5

125 MHz:

200 MHz = SR/2.5

250 MHz:

210 MHz = SR/2.38

262.6 MHz:

220 MHz = SR/2.273

275 MHz:





I guess this comparison can now be considered complete. Rigol's interpolation implementation is way worse than that of Siglent.

It looks like a software flaw indeed. There is no other reason why the two scopes would show such strikingly different results given equal conditions and the fact that the processing responsible for this takes place, as far as I understand it, entirely after the ADC stage.

[shapirus]

Yes, we should not forget what I’m demonstrating here: torturing a dirt-cheap entry-level instrument operating far outside its specifications, with a signal that is ~2.5 times the specified bandwidth. A heavily attenuated signal close to Nyquist, with a remaining amplitude barely high enough to get a solid trigger, sitting right on the edge of a steep AA-filter, is naturally not going to be a benchmark for the scope performance.

At the same time, it doesn't mean that such test is useless. Quite the opposite: it's good to know the actual limits of the instrument's usability. Knowing that helps to use your tools more efficiently.
Besides -- remember what thread we're in! -- it helps potential new users to make informed choices (or no choices at all, lol).

[Mechatrommer]

I guess this comparison can now be considered complete. Rigol's interpolation implementation is way worse than that of Siglent.

imho screenshots with persistent mode ON is less helpful... other than, "thicker" is worse than "thinner". single line trace display (whether using maximum memory or single trigger mode) can show clearly how Sinc implementation works. but i think we are going round and round, first capture from you and 2n3055 already showed sds800x is better to my eye. ymmv.

[Mechatrommer]

The answers over there are quite wrong.

Am I going to be blamed for that, too?

FWIW I never even read the page, I just found the image on google image search.

It's a 5Hz sine wave sampled at 11Hz intended to show where the "AM" effect comes from.

I'm not blaming you but the people who posted the graphs on Reddit and commented on it. Drawing a line through sample points as-is is never a natural representation of a sampled signal. You'll need to put the samples through an analog or digital filter in order to reconstruct the resulting signal.

Just so.

Why did they decide to use a first-order hold reconstruction (linear interpolation) rather than a zero-order hold (step)? Typically that would be an unwitting default "looks right" choice.

Personally I hate both, and much prefer to see just the samples (i.e. dots).

zero-order hold you mean like stair steps plot? that will be awful in any circumstances in any scope. how do you come up with such an idea?

[nctnico]

The answers over there are quite wrong.

Am I going to be blamed for that, too?

FWIW I never even read the page, I just found the image on google image search.

It's a 5Hz sine wave sampled at 11Hz intended to show where the "AM" effect comes from.

I'm not blaming you but the people who posted the graphs on Reddit and commented on it. Drawing a line through sample points as-is is never a natural representation of a sampled signal. You'll need to put the samples through an analog or digital filter in order to reconstruct the resulting signal.

Just so.

Why did they decide to use a first-order hold reconstruction (linear interpolation) rather than a zero-order hold (step)? Typically that would be an unwitting default "looks right" choice.

Personally I hate both, and much prefer to see just the samples (i.e. dots).

zero-order hold you mean like stair steps plot? that will be awful in any circumstances in any scope. how do you come up with such an idea?

A step (staircase) plot style is easier to see compared to a single dot and it doesn't pretend to plot a waveform shape that is completely wrong. So there is merit in showing sample values this way. However I have not come across a practical situation where showing samples that way is usefull. But that could be different next week or never.

[Mechatrommer]

Here you go:


SDS824X HD_Sine_450MHz_1GSa_Single

here you demo that at some setup Sr / 2.22, sds800x is unable to reconstruct signal properly? from the look of it, yes it is.

[Mechatrommer]

A step (staircase) plot style is easier to see compared to a single dot and it doesn't pretend to plot a waveform shape that is completely wrong.

as sample rate is increased, say 10-50 points per cycle, stair step plot will be annoyingly disturbing imho, its not suitable to represent continuous analog signal, for statistics that maybe all right. the purpose of graph plot on dso is try to mimick as close as possible actual signal. highly undersampling such case above Sr / 2.2 Fungus demo'ed surely create distortion when plotted with so called first order (line) but thats not an excuse to use zero order. you can sim sampling at 10-20 points per cycle and try to plot it using stair step and look how bad it is.. i cant imagine how bad it is if lecroy is using such an idea for example in the attached plot

[Fungus]

262.6 MHz:

That's just the "AM" signal with persistence.

It looks like a software flaw indeed.

...except that other software does the same thing, and I've seen the exact same effect using hardware reconstruction filters (from a DAC feeding into an RC filter).

What it means is that Siglent is doing something weird. I'd like to see an actual AM signal fed into the Siglent to see what happens.

nb. It doesn't have to be at the outer limits of bandwidth. You can drop the sample rate down by fiddling with the horizontal timebase and figure out a suitable frequency from that.

(I can get aliasing on the 1kHz probe compensation signal by making the sample rate drop low enough)

[ebastler]

...except that other software does the same thing, and I've seen the exact same effect using hardware reconstruction filters (from a DAC feeding into an RC filter).

What it means is that Siglent is doing something weird.
I'd like to see an actual AM signal fed into the Siglent to see what happens.

I am afraid it's too late, you have already lost all credibility. Comparing Rigol's slightly flawed implementation with Mechatrommer's admittedly wrong first homebrew attempt -- and reaching the conclusion that Siglent's implementation must be "weird"?

 

[nctnico]

A step (staircase) plot style is easier to see compared to a single dot and it doesn't pretend to plot a waveform shape that is completely wrong.

as sample rate is increased, say 10-50 points per cycle, stair step plot will be annoyingly disturbing imho, its not suitable to represent continuous analog signal, for statistics that maybe all right. the purpose of graph plot on dso is try to mimick as close as possible actual signal. highly undersampling such case above Sr / 2.2 Fungus demo'ed surely create distortion when plotted with so called first order (line) but thats not an excuse to use zero order. you can sim sampling at 10-20 points per cycle and try to plot it using stair step and look how bad it is.. i cant imagine how bad it is if lecroy is using such an idea for example in the attached plot

Almost all DSOs support linear interpolation to connect the dots but you have to be really careful to have enough sample points to make a visual representation which is close to the actual signal. However, In most situations this is not the case. A step-wise signal at least doesn't pretend to be more than it is. I never use linear interpolation mode as it is very easy to get tricked into mis-interpreting the signal.

[nctnico]

That's just the "AM" signal with persistence.

What it means is that Siglent is doing something weird. I'd like to see an actual AM signal fed into the Siglent to see what happens.

No, the sin x/x interpolation is broken on the Rigol; the screenshots made by shapirus show that clearly. Siglent has implemented sin x/xt correctly. It is simple as that.

[Mechatrommer]

...except that other software does the same thing, and I've seen the exact same effect using hardware reconstruction filters (from a DAC feeding into an RC filter).

What it means is that Siglent is doing something weird.
I'd like to see an actual AM signal fed into the Siglent to see what happens.

I am afraid it's too late, you have already lost all credibility. Comparing Rigol's slightly flawed implementation with Mechatrommer's admittedly wrong first homebrew attempt -- and reaching the conclusion that Siglent's implementation must be "weird"?

 

i didnt aware he's using my post as example... so i redo sim based on below... 446MHz @ 1GSps...

Here you go:


SDS824X HD_Sine_450MHz_1GSa_Single

my SW shows AM for each 3-4 cycles. but siglent plotted it like for each 24 cycles, so yeah it weird and why? its kind of aliasing to me.

[shapirus]

I'd like to see an actual AM signal fed into the Siglent to see what happens.

This is a very good point. It would be interesting to see how the Siglent copes with an AM signal with the carrier frequency same as those used for the screenshots above and the modulation frequency say somewhere around 1/20..1/5 the carrier frequency. It would allow to see if the Siglent applies any special "cheating" for better interpolation. I personally doubt that, but doubts can only serve as a motivation to conduct an experiment.