Magnova oscilloscope

[rf-messkopf]

AFAIK pre-ringing shouldn't exist in the real world. It likely is a problem with the sin x/x reconstruction because that can introduce pre-ringing.

This is true for an isolated edge and a causal system. However, he was showing a continuous square wave, which can show pre-ringing if it is band limited (Gibbs effect), even for causal systems. So what he is seeing can very well be a real effect.

I used isolated edges in my post above, where no pre-ringing is visible, as expected.

[tautech]

Hi tautech,

When i'am back from lunch i will test it whitout averaging.
And if there is time, also some pre-aberation pictures.

Kind regards,
Bram

Check the AC coupling too.
Your SDS2054X Plus is showing DC input coupling.

[blackdog]

Hi,

Here are a few more pictures from testing with the Magnova and the Siglent scope.
The source for the first two pictures is the Leo Bodnar BNC pulser, which is connected directly to the scope's BNC connector via a -20dB attenuator.

The -20 dB attenuator is necessary to ensure that the pulser sees the smoothest possible load and that the oscilloscope also sees the smoothest possible 50-ohm source.
I’ve conducted various tests over the past few years to determine which of the attenuators I own are best suited for these tests.
As a result, the distortions caused by the output impedance of the Leo Bodnar pulser and the oscilloscope’s input impedance are minimal at the bandwidths of the two oscilloscopes.

This is how the Magnova scoop responds to the Leo Bodnar pulse...

ntcnico: It's likely a problem with the sin x/x reconstruction because that can introduce pre-ringing.
I also think we should look into what's happening with regard to pre-ringing.

.

The image below shows the documentation that came with the Leo Bodnar pulser.

.

Same setup and now the Siglent scoop, no pre-ringing!

.

Now it's time for some exercise, because my butt was starting to get some pre-ringing, from so mucht sitting during these measurements.

Kind regards,
Bram

[rf-messkopf]

This is how the Magnova scoop responds to the Leo Bodnar pulse...

But this is again not a single isolated edge, but a square wave, isn't it? This very clean pre-ringing looks exactly like the Gibbs phenomenon, unlike the "post"-ringing following the edge, which might be a parasitic circuit effect. In other words, what you are seeing here likely is the Magnova rolling off at about 350 MHz.

Leo Bodnar's documentation was measured with a sampling scope which likely has a bandwidth in the GHz-range. And your Siglent scope also has a higher bandwidth, as you said.

You can experiment with that by limiting the bandwidth on the Siglent or the Magnova and watch the pre-ringing. You can also experiment with single pulses, e.g., one pulse every couple of seconds. I'm sure the pre-ringing will then disappear. Or you can turn the sin(x)/x-interpolation off if you suspect that being the cause.

[KungFuJosh]

The Siglent has higher bandwidth and lower vertical resolution. Both of which would reduce the effect.

[Martin72]

I wouldn't worry too much about this “natural” minor issue.
(The images show bodnar pulse signal at the 20 dB attenuator (bandwidth 2 GHz), first with the SDS3104X HD, then with the SDS7404 H12, no averaging)

[Sorama]

Hi,

Here are a few more pictures from testing with the Magnova and the Siglent scope.
The source for the first two pictures is the Leo Bodnar BNC pulser, which is connected directly to the scope's BNC connector via a -20dB attenuator.

The -20 dB attenuator is necessary to ensure that the pulser sees the smoothest possible load and that the oscilloscope also sees the smoothest possible 50-ohm source.
I’ve conducted various tests over the past few years to determine which of the attenuators I own are best suited for these tests.
As a result, the distortions caused by the output impedance of the Leo Bodnar pulser and the oscilloscope’s input impedance are minimal at the bandwidths of the two oscilloscopes.

This is how the Magnova scoop responds to the Leo Bodnar pulse...

ntcnico: It's likely a problem with the sin x/x reconstruction because that can introduce pre-ringing.
I also think we should look into what's happening with regard to pre-ringing.

.

The image below shows the documentation that came with the Leo Bodnar pulser.

.

Same setup and now the Siglent scoop, no pre-ringing!

.

Now it's time for some exercise, because my butt was starting to get some pre-ringing, from so mucht sitting during these measurements.

Kind regards,
Bram

what about switching probes between scopes?
Maybe it is probe related?

[Rubo]

what about switching probes between scopes?
Maybe it is probe related?

You might want to re-read his post.

The source for the first two pictures is the Leo Bodnar BNC pulser, which is connected directly to the scope's BNC connector via a -20dB attenuator.

[nctnico]

This is how the Magnova scoop responds to the Leo Bodnar pulse...

But this is again not a single isolated edge, but a square wave, isn't it? This very clean pre-ringing looks exactly like the Gibbs phenomenon, unlike the "post"-ringing following the edge, which might be a parasitic circuit effect. In other words, what you are seeing here likely is the Magnova rolling off at about 350 MHz.

Leo Bodnar's documentation was measured with a sampling scope which likely has a bandwidth in the GHz-range. And your Siglent scope also has a higher bandwidth, as you said.

You can experiment with that by limiting the bandwidth on the Siglent or the Magnova and watch the pre-ringing. You can also experiment with single pulses, e.g., one pulse every couple of seconds. I'm sure the pre-ringing will then disappear. Or you can turn the sin(x)/x-interpolation off if you suspect that being the cause.

Turning sin x/x off is a good suggestion. However, I don't think you can call the pre-ringing a Gibbs effect. The Gibbs effect are residuals when trying to construct a square wave from fourier series. It is a math problem, not something that happens in the real world. If blackdog is testing using a square wave, hus result could be the effect of bandwidth limiting a square wave indeed.

[rf-messkopf]

AFAIK pre-ringing shouldn't exist in the real world. It likely is a problem with the sin x/x reconstruction because that can introduce pre-ringing.

This is true for an isolated edge and a causal system. However, he was showing a continuous square wave, which can show pre-ringing if it is band limited (Gibbs effect), even for causal systems. So what he is seeing can very well be a real effect.

I used isolated edges in my post above, where no pre-ringing is visible, as expected.

Thinking about this again, I have to correct myself. A causal system or a causal filter should not show pre-ringing, even when it is subjected to a square wave that started at minus infinity. This follows simply by looking at the convolution integral of the square wave and the causal impulse response of the system. Now, the Gibbs effect, i.e., the oscillatory behavior of the N-th partial Fourier series, which does show pre- and post-ringing, corresponds to a brick-wall filter in the frequency domain, which is non-causal. A real physical system is, of course, causal and therefore should not show this behavior.

Now I don't think that the Magnova applies any FIR or other filtering unless this is selected in the channel menu. Also, the bandwidth limit filters should be first order and not show any ringing, similarly the HiRes filter (a moving average does not have overshoot and ringing as well). Moreover, I cannot get my Magnova to how pre-ringing. I suspect that the pre-ringing @blackdog shows is real and present at the scope input.

What I would do next to investigate this phenomenon is not to apply a continuous square wave, but single pulses. Can the Leo Boadnar pulser be externally triggered? This would allow to look at single pulses and exclude that the pre-ringing is anything that is coming from the previous cycle.

[rf-messkopf]

However, I don't think you can call the pre-ringing a Gibbs effect. The Gibbs effect are residuals when trying to construct a square wave from fourier series. It is a math problem, not something that happens in the real world. If blackdog is testing using a square wave, hus result could be the effect of bandwidth limiting a square wave indeed.

You are right, see my previous post. But if the bandwidth limiting is happening in the analog domain by a causal filter, no pre-ringing would be expected. Maybe it is coming out of the pulser? As I said, I would try to generate single pulses, and single-shot capture them.

Edit: The sin(x)/x reconstruction filter is non-causal. It can certainly exhibit pre-ringing. So it is indeed a good idea to turn it off for this experiment-

[Sorama]

what about switching probes between scopes?
Maybe it is probe related?

You might want to re-read his post.

The source for the first two pictures is the Leo Bodnar BNC pulser, which is connected directly to the scope's BNC connector via a -20dB attenuator.

just noticed it myself.
tnx and sorry...

[blackdog]

Hi,

I’ve done some more measurements and tinkered with the scoops using Sinus-x and other settings, as well as trying out different generators.
Sometimes there are slight changes, but I need to investigate these further, first to make sure I’m not fooling myself—or you guys.
Some of the minor errors that occur could be due to common-mode errors, even though I’ve already been careful to address them, so I need to get some certainty about the common-mode issues first.

Even during my walk through beautiful Amsterdam, which lasted over two hours, I was confronted with "pre-ripple", see the photo in the foreground, which shows a lot of ripples... 

.

Now I’m too tired to take any more measurements; tomorrow is another day!

Best regards,
Bram

[gf]

This is how the Magnova scoop responds to the Leo Bodnar pulse...

But this is again not a single isolated edge, but a square wave, isn't it? This very clean pre-ringing looks exactly like the Gibbs phenomenon, unlike the "post"-ringing following the edge, which might be a parasitic circuit effect. In other words, what you are seeing here likely is the Magnova rolling off at about 350 MHz.

Leo Bodnar's documentation was measured with a sampling scope which likely has a bandwidth in the GHz-range. And your Siglent scope also has a higher bandwidth, as you said.

You can experiment with that by limiting the bandwidth on the Siglent or the Magnova and watch the pre-ringing. You can also experiment with single pulses, e.g., one pulse every couple of seconds. I'm sure the pre-ringing will then disappear. Or you can turn the sin(x)/x-interpolation off if you suspect that being the cause.

Turning sin x/x off is a good suggestion. However, I don't think you can call the pre-ringing a Gibbs effect. The Gibbs effect are residuals when trying to construct a square wave from fourier series. It is a math problem, not something that happens in the real world. If blackdog is testing using a square wave, hus result could be the effect of bandwidth limiting a square wave indeed.

Although Gibbs ringing/overshoot is traditionally explained via the truncation of square wave harmonics, the concept applies analogously to Heaviside steps (the underlying change being merely the transition from a discrete spectrum to a continuous one). Furthermore, complete elimination of higher frequencies is not required; sufficient attenuation is enough to produce the effect. All that is necessary is a sufficiently steep high-frequency cutoff, whereas a soft roll-off—such as a Gaussian profile—will not produce it.

However, the temporal location of the ringing (i.e., whether it manifests as pre-ringing, post-ringing, or a combination of both that is either symmetric or asymmetric) is primarily determined by the phase response. A linear-phase system exhibits symmetric pre- and post-ringing, whereas a minimum-phase system shows exclusively post-ringing. Any intermediate phase characteristic will distribute the ringing accordingly between these two extremes.

[ From a more specialized mathematical perspective, the Gibbs Phenomenon states that a square wave cannot be synthesized exactly, not even with an infinite number of harmonics—at least when the error metric is the maximum absolute deviation. However, when people refer to "Gibbs" in practical engineering contexts, they usually mean the overshoot/ringing introduced by a steep high-frequency cutoff. ]

[woody]

Even during my walk through beautiful Amsterdam, which lasted over two hours, I was confronted with "pre-ripple", see the photo in the foreground, which shows a lot of ripples...

And also with a lot of potentially hot air, coming from the "stoomhelling" on the Kromhout wharf on the right.

[Martin72]

I recreated this again using the Magnova.
Thanks to the numerous options in the display menu, you can configure the signal display accordingly; most scopes I know of only offer two options to choose from.
A nice little feature.
Here are two more pictures taken in high-resolution mode; of course, they're not very useful due to the reduced bandwidth—but I did it anyway.
Either way, I wouldn't want to get hung up on this slight fluctuation in the slope of the curve, let alone expect an “improvement.”

Martin

[Rubo]

just noticed it myself.
tnx and sorry...

No worries 
Just wanted everyone to be on the same page.

[rf-messkopf]

I recreated this again using the Magnova.
Thanks to the numerous options in the display menu, you can configure the signal display accordingly; most scopes I know of only offer two options to choose from.

So it clearly is the sinc-filter in this case, as one can see when looking at the data with no interpolation. And this is a linear phase system, which is non-causal, and therefore has every right in the world to show pre-ringing.

[nctnico]

Even during my walk through beautiful Amsterdam, which lasted over two hours, I was confronted with "pre-ripple", see the photo in the foreground, which shows a lot of ripples...

And also with a lot of potentially hot air, coming from the "stoomhelling" on the Kromhout wharf on the right.

That is one of the nicer bits of Amsterdam indeed. Today's cycling trip to a customer only got me to the outskirts of Amsterdam.

[Andre77]

Then there's another counter issue: I've tried a variety of combinations, and while not all of them result in an error, the two measurements below do.
[...]

Hi Bram and everyone else who contributed,

Thanks for the detailed testing. First, I'd like to address the counter deviation you observed at 200 MHz with a specific channel configuration.

The counter can only measure accurately up to a certain frequency. Above this limit, not every signal edge is detected reliably, which leads to the incorrect readings you observed.

The maximum frequency that can be measured accurately depends on the sample rate and whether hysteresis is enabled. Of course, it also depends on factors such as the signal shape, signal-to-noise ratio, how much of the vertical range is used, and other conditions, but that's not what we're focusing on here.

When hysteresis is enabled, the counter can, under typical conditions, measure accurately up to approximately the following frequencies:
- 155 MHz at 1.0 GSa/s
- 250 MHz at 1.6 GSa/s

Without hysteresis:
- 450 MHz at 1.0 GSa/s
- 720 MHz at 1.6 GSa/s

At very high frequencies, it can therefore be useful to disable hysteresis to extend the counter's usable frequency range. At lower frequencies, however, hysteresis helps prevent the same edge from being detected more than once due to noise and should normally remain enabled.

This also explains the different results with your channel combinations. When using channels within the CH1/CH2 or CH3/CH4 groups, the sample rate is 1.6 GSa/s. With combinations across these groups, such as CH1 + CH4 in your second setup, the sample rate is 1.0 GSa/s.

So with your 200 MHz signal and hysteresis enabled, the 1.6 GSa/s configurations are still within the reliable measurement range, while the 1.0 GSa/s configuration is already above it.

We will also check whether we can reduce the number of sample points the counters require per signal period when hysteresis is enabled. This could allow higher frequencies to be measured accurately with hysteresis in the future.

Here are a few more pictures from testing with the Magnova and the Siglent scope.
The source for the first two pictures is the Leo Bodnar BNC pulser, which is connected directly to the scope's BNC connector via a -20dB attenuator.
[...]

Regarding the pre-ringing associated with the very fast edges of the Leo Bodnar pulser: these edges are extremely fast. According to the manufacturer, the rise/fall time is around 30 ps. The corresponding frequency components therefore extend well into the GHz range.

The very fast edges contain frequency components far beyond the oscilloscope's Nyquist frequency. These components cannot be represented directly by the sampled data and are therefore strongly limited by the oscilloscope's analog and digital bandwidth. A sharp bandwidth limitation, such as the one associated with sinc-type reconstruction/interpolation filtering, can produce ringing before and after a fast edge. The analog front end removes most of the out-of-band energy, but some very high-frequency content still reaches the sampling system and contributes to the observed pre-ringing. At a higher sample rate, the Nyquist frequency is higher, so more of the edge spectrum can be represented before this limitation takes effect, which reduces the visible ringing.

As some of you have already pointed out here, this effect in the oscilloscope display can be avoided by disabling sinc interpolation and setting Interpolation = None. In that case, the sampled points are displayed without sinc interpolation, so the associated pre-ringing is not introduced by the display interpolation.

One additional note for anyone who wants to reproduce this effect using a fast function generator instead of the Leo Bodnar pulser: many function generators with very high sample rates, such as the excellent SDG6000X series, use DACs with built-in sinc interpolation. This is necessary because the FPGA cannot directly provide new sample data at the DAC's full sample rate. This interpolation itself can produce pre-ringing that is physically present in the output signal. Since this pre-ringing can be several times larger than the effect caused by the oscilloscope, in practice you may end up seeing mainly the generator's pre-ringing rather than the scope's.

I was trying to figure out why the frequency counter wasn't working; it turned out to be the "Level" was set to "0."
I think it would be handy to set this to (auto) by default, so you can always see that the counter for the channel you’re setting is working—then you can adjust it to the desired setting later if needed.

Good point. In the current firmware versions, the counter level is already set to Auto by default.

So with the default settings, the counter should start with automatic level detection enabled, and you can switch to a manual level afterward if needed.

Best regards,
Andre

[blackdog]

Hi Andre77,

Thank you for your awnsers!

I’d still like to show you something and explain it...
First of all, I’ve made sure that no common-mode interference can occur due to shared ground connections.
The Leo Bodnar unit is now powered by a PowerBank and is not connected in any other way to the ground plane of the scope being used.
By doing this, minor distortions in the signal being measured on the scopes disappeared.

Now I’d like to show you some differences between the three oscilloscopes I tested: the Magnova, the Siglent 12-bit 500 MHz, and the Hameg-R&S 8-bit 500 MHz.
Here’s a comparison between the Vector/Sine-x and Dot modes of the oscilloscopes.
I tried to keep the conditions as consistent as possible across the oscilloscopes; I displayed the waveform being measured as large as possible on the screen, which is especially important in Dot mode.
I didn't notice any significant change in the waveform when I used the amplitude fine-tuning on the scoops, so this isn't a problem.

Picture Time!
First, I took measurements using the Hameg HMO3054.
Every time I use this scope for a series of measurements, it always works very well.
All the buttons respond almost instantly, and the averaging function works very well when needed.
The variable bandwidth setting also works well (digital filtering).
In use, it feels more like a 10-bit model than the 8-bit model this scope actually is, likely due to the very good AD converter.

Hameg Sine-X modus

.

Hameg Dot modus, and yeah, I noticed the typo—I'm still looking for an 8 Hz, -20 dB SMA attenuator *grin*

.

Siglent Vector Modus, and there is a small amount of pre-ringing visible.

.

Siglent Dot Modus, no ringing.

.

Magnova Sine-X modus, Again, I didn't resize the photo here for clarity.

.

Magnova Dot Modus, Even in Dot Mode, Pri-Ringing is clearly visible here,
so it's possible that the frequencies caused by the edges of the fast pulse are entering the AD converter and interfering with the sample clock.
This happens much more noticeably with the Magnova than with the other two oscilloscopes.

.

I'm still having some trouble understanding pre-ringing when no processing—such as Sinus-x or others—is applied.
Perhaps pre-ringing doesn't occur if I follow what rf-messkopf says: use a single pulse.
I tested with my function generator, which produces the fastest edges, but with an edge steepness of just over 3 nsec, it isn’t fast enough.
I tested with 10 kHz and a 0.05% duty cycle; this test did not produce any significant pre-ringing  in Dot or Sine-X mode.

Counter problems
IIf the user exceeds the maximum frequency or selects functions that cause the displayed value to be incorrect,
display a message on the screen that refers to the selected functions currently causing the error condition.
And yes, that could involve multiple steps.

When testing the Dot/Sine-X functions on the Siglent, there were also two settings in a row that had to be reset to their default values.
If the user exceeds the maximum frequency or selects certain functions.
The first feature was a trigger window that wasn’t visible on the screen; I had to turn it off. Then, when I tried to use Dot Mode again,
I was told that averaging had to be turned off—otherwise, I couldn’t use Dot Mode.
That was mentally challenging :-) I was focused on completely different things (taking accurate measurements), and my measuring equipment kept correcting me!

The Siglent only displayed one error at a time, and it wasn't very clear—especially the trigger window, which was difficult to deal with because you couldn't see it.
But I was already glad that a message appeared that pointed me in the right direction.

So clear error messages on the screen are always helpful for the user; they’ll always appreciate it when this is handled properly. This applies, of course, to all equipment manufacturers.
And when creating error messages, keep in mind that the user is often doing something completely different from what the message indicates.
The user certainly wasn’t present at the meetings where the functions and error messages were discussed. 

Quicksave function
It almost always works fine when I use the Quicksave touch function at the top of the screen.
Since I have 71-year-old fingers, this often doesn’t work well on many touchscreen devices, but with the Magnova scope it almost always works fine, so kudos to you in that regard.

But, yeah, yeah, there’s always a “but”...
Move the notification that Quicksave has been used to a position just below the Quicksave button, to make it much clearer.
I also recommend a clear background and border for this text, as well as for error messages, for example.

Sorry for the poor-quality photo, but it’s clear enough—the red box indicates the current setting, hidden among other text at the bottom of the screen.
Your eyes go to where you tapped the function, and it’s more convenient if the notification appears nearby as well.
Like my green text and box.
Error messages, for example, in the red box, and others in, say, an orange box—just my two cents.

.

And Andre77, wat about this picture witch a showd before, is it possible to correct this.
is it possible to adjust this behavior so that it matches the behavior of other scoops—that is, not to display amplitude modulation when it isn't present?

.

Thank you for your time.

Kind regards,
Bram

PS
My texts with dyslexia go through an AI translator, which isn't always a good combination, that's why the sentences sometimes sound a bit off. 

[tautech]

Hi Andre77,

Thank you for your awnsers!

I’d still like to show you something and explain it...
First of all, I’ve made sure that no common-mode interference can occur due to shared ground connections.
The Leo Bodnar unit is now powered by a PowerBank and is not connected in any other way to the ground plane of the scope being used.
By doing this, minor distortions in the signal being measured on the scopes disappeared.

Now I’d like to show you some differences between the three oscilloscopes I tested: the Magnova, the Siglent 10-bit 500 MHz, and the Hameg-R&S 8-bit 500 MHz.
Here’s a comparison between the Vector/Sine-x and Dot modes of the oscilloscopes.
.................

FYI
SDS2000X Plus models in 10bit mode are BW limited to 100 MHz.

[blackdog]

Hi tautech,

My Siglent scoop is as show here is: SDS2104X HD and is 12Bit en 500MHz.

Kind regards,
Bram

[tautech]

Hi tautech,

My Siglent scoop is as show here is: SDS2104X HD and is 10Bit en 500MHz.

Kind regards,
Bram

So you might think.

SDS2000X Plus used in 10bit mode will give measurement errors over 100 MHz.....in all models !
The oversampling acts as a BW limiter.
Dave Jones (EEVblog) discovered this fact when reviewing this range at release some years ago.

While SDS2000X Plus are a 500+ MHz 8bit design the BW for each model is limited by SW in the same IC that performs the 20 MHz BW limit.
SW sets BW and even the 500 MHz SDS2054X Plus is limited to 350 MHz when 3 or more channels are active.
You can see the Full (BW) change to 350 MHz in the channel tab when 3 channels are active.

Carry on.... 

[rf-messkopf]

I'm still having some trouble understanding pre-ringing when no processing—such as Sinus-x or others—is applied.
Perhaps pre-ringing doesn't occur if I follow what rf-messkopf says: use a single pulse.

As already speculated, the pre-shoot/pre-ringing might be there for real, and just not be visible on the 8 or 10 bit scopes due to their limited resolution (count the number of pixels of its actual height in dots mode, 12 bit vs. 10 bit is an enhancement by a factor of four). For example, avalanche pulsers tend to show pre-shoots due to the coupling of the trigger pulse at the base of the transistor through junction capacitances before the avalanche breakdown takes place. I understand (correct me if I'm wrong) that the Bodnar pulser uses a fast SiGe comparator, but there still may be an internal coupling mechanism from the input/trigger circuit, which will be much slower than the comparator and in the range of the pre-ringing that you see.

As Andre mentioned, DDS generators often use pre-distortion and digital filtering, which may lead to pre-ringing, and thus cannot be fully trusted for such a test.