New Rigol DS1054Z oscilloscope

[Fungus]

Hi,
Probably I'm OT but I wonder if it would be possible to group buy this oscilliscope to have a bigger discount for large quantities.
I think a lot of people here would like to buy one of these...
Any ideas?
Thanks!

Sure, why don't you do it...?

Start by making some phone calls to Rigol to find out what you need to do to become a dealer, how many units you have to commit to, over what timespan, etc., to get the dealer pricing.

When you have that info, come back and we'll talk.

[Fungus]

Is 10x margin enough for sure

 

No.

(I didn't bother reading the rest of your post because it's obvious what your problem is from the first six words).

[marmad]

Is 10x margin enough for sure, or only expected to be enough? Does that 10x margin still depend on having an acceptable higher order low-pass filter in the analog front-end of the scope?

The 10x oversampling applies to linear interpolation - not sin(x)/x. If the DSO (or operator) is confident that no aliases are present (i.e. the sample rate is high enough to be beyond the transition band of the antialias filter - and if you don't know what transition band means, Google it) - the DSO should use sin(x)/x, and then 2.5 - 4x oversampling can be enough. If not, linear interpolation is/should be used. The Rigol UltraVision DSOs (as well as the Agilent X-series and many other DSOs) automatically turn on linear interpolation when the sample rate has dropped enough that it's obvious that aliasing could be present.

Really - the best thing is just to BUY A DSO and start playing around with it. You'll soon see with your own eyes why 10x oversampling is beneficial when using linear interpolation.

So although the Rigol DS2000 series goes up to Rigol DS2302A, in reality the best configuration which can be used and meets 10x margin when using 2 channels at the same time, is the Rigol DS2101A. Is that correct?

No, it depends on the characteristics of the antialias filter. From measurements that have been posted by owners, it appears that the DS2000A series attenuates frequencies above 500MHz (i.e. Nyquist for 2 channels @ 1GSa/s) enough for faithful sin(x)/x reconstruction up to the 200MHz version. The 300MHz version is a different story - and that's why I don't have that bandwidth enabled on my DS2000.

[coppice]

You seem to be unwilling (or unable) to grasp the very basics of sampling - even though they have been laid out here time and time again. The Nyquist theorem is just that: a mathematical theorem. It is not real world usage - such as trying to get an accurate image of a waveform you're trying to look at. It doesn't take much imagination to see what kind of image a waveform sampled at fs/2 is going to deliver with linear interpolation:

I think its you who are unable to grasp the basics of sampling. Your simplistic join the dots scheme massively expands the bandwidth of the signal. If you put those triangles through a bandpass filter, so remove all the out of band crud you just created, you'll get back to something similar to the original signal.

[marmad]

Your simplistic join the dots scheme...

Mine? It's just a stock Nyquist image lifted from the internet showing a sine wave sampled at 2f.

If you put those triangles through a bandpass filter, so remove all the out of band crud you just created, you'll get back to something similar to the original signal.

Oh please.. do go on! School me on how DSOs are faithfully reproducing waveforms right up to the Nyquist frequency. 

[Fungus]

...2.5 - 4x oversampling can be enough.

This only applies to signal reconstruction (ie. display), not to the actual sampling process.

[marmad]

...2.5 - 4x oversampling can be enough.

This only applies to signal reconstruction (ie. display), not to the actual sampling process.

No, it applies to the sampling rate. It's always a trade-off between the anti-alias filter and the sampling rate. From previously linked Agilent document:

"Figure 8 clearly shows that the sin(x)/x reconstruction filter perfectly reconstructs the waveform from the samples taken at a rate 2.5 times faster than the highest frequency content of the waveform."

To prevent aliasing, two fundamental items can be addressed in a digitizing system:
1) An anti-alias filter can be designed to reject all high-frequency content above fN.
2) The sample rate can be increased to an arbitrarily high frequency such that fN is above the highest significant frequency component of the signal being digitized."

In some systems, one might imagine a case where increased sample rate is inexpensive, compared to a higher-order anti-alias filter. In this case, it may be more practical to increase the sample rate to compensate for the slow roll-off of the anti-alias filter, instead of designing a more complex filter.

[coppice]

Your simplistic join the dots scheme...

Mine? It's just a stock Nyquist image lifted from the internet showing a sine wave sampled at 2f.

There are plenty of idiotic pictures on the web. You chose to quote this one to illustrate something. Maybe you didn't adequately explain what that something was.

If you put those triangles through a bandpass filter, so remove all the out of band crud you just created, you'll get back to something similar to the original signal.

Oh please.. do go on! School me on how DSOs are faithfully reproducing waveforms right up to the Nyquist frequency. 

They don't faithfully display right up to Shannon, for reasons many people have explained. However the practical limitations of not being able to impose a brick wall filter without nasty phase effects don't make linear interpolation, with its consequent massive increase in bandwidth, any more meaningful. After digitising a signal you don't get back to the original signal just  by feeding it through a DAC. You have to filter back to the Shannon bandwidth. If your DAC is a ZOH type, you also need to compensate for its funky frequency response. If you are to display the digitised signal in any meaningful form, you need to do the same kind of filtering digitally.

[marmad]

There are plenty of idiotic pictures on the web. You chose to quote this one to illustrate something. Maybe you didn't adequately explain what that something was.

Maybe you didn't adequately follow the conversation that has been taking place over several pages of this thread.

They don't faithfully display right up to Shannon, for reasons many people have explained.

Yes, me being one those people that have been attempting to explain it to the person I was responding to.

However the practical limitations of not being able to impose a brick wall filter without nasty phase effects don't make linear interpolation, with its consequent massive increase in bandwidth, any more meaningful.

Wow, I really don't have the energy to go through this again. Congratulations, you win!

[Orange]

Wow, I really don't have the energy to go through this again. Congratulations, you win!

Does this mean we stop talking about it, and start using our scope ? 
Oh and Pascal, when do you buy an Rigol scope ??

[Fungus]

Wow, I really don't have the energy to go through this again. Congratulations, you win!

Does this mean we stop talking about it, and start using our scope ? 

Yes. I say we leave him in his little world of doubt. The only hope is that he buys a 'scope and sees it for himself.

[marmad]

Does this mean we stop talking about it, and start using our scope ? 

Yes.... and pray we don't have another Al Pacino-Godfather 3-"Just when I thought..." moment   

[xnaron]

Apologies if this has been pointed out before...

The resistor jumper configuration for "Hardware Version" is different on DS1054Z compared to DS1104Z.

I was watching Dave's DS1054Z teardown and I took note of the part where he mentioned the resistor configuration blocks.  I started thinking about whether the resistor jumpers would be the same on the DS1104Z.  I searched youtube and found a DS1104Z teardown video.  I watched the video and found a spot where the resistor jumpers were shown.  I took a screenshot of them.  I also took a screenshot of the jumpers from Dave's Flickr teardown.  I am including them in this post.  If you examine the photos you can see there is one resistor jumper in a different place.  Anyone have any info on how these resistor jumpers effect functionality?

This is the DS1104Z teardown video I took the screenshot from. 

Here is the link to Daves teardown image on Flickr I cropped my image from https://www.flickr.com/photos/eevblog/15354535170/in/set-72157646442125864/lightbox/

[tom66]

This argument over sampling rates and bandwidth is really quite silly.

What Rigol have done is given people a solid and low cost 4 channel digital scope with good sampling rate and intensity graded display. This thing absolutely destroys the competition in features and performance. What other scope offers 24Mpt in the same price range? The only other one I can think of is the DS2000... Even Tek's high end scopes only offer 10Mpt... Agilent's 2000X is 4Mpt/4ch...
 
I've owned a DS1074Z for about 3 months now. How many times have I used the full bandwidth on all four channels? Never. I can't even think of an application where I would do so.

The primary applications for 4 channels are:
 - SPI+result analysis (DAC etc)
 - I2C+result analysis (DAC etc)
 - 3ph motor drives (max 1MHz)

.... that is, when you can only afford a $400 scope... Of course you can analyse DDR buses but the DS1xxxZ probably isn't for you then....

You volt-nuts may say the scope only has real 25MHz bandwidth on four channels. If that really bothers you, don't buy it. Or turn on the 20MHz filter. (It'd be really nice if Rigol let users enable the 50 and 70MHz filters too, for the 100MHz scopes. I can hope.)

[marmad]

This argument over sampling rates and bandwidth is really quite silly....

You volt-nuts may say the scope only has real 25MHz bandwidth on four channels. If that really bothers you, don't buy it. Or turn on the 20MHz filter. (It'd be really nice if Rigol let users enable the 50 and 70MHz filters too, for the 100MHz scopes. I can hope.)

I'm not sure it's actually been an argument (although perhaps there's been 1 or 2 people fighting against the idea). Personally, I talked about it just so prospective owners would understand the limitations - but I wholeheartedly agree with you. I think the DSO is a great bargain - and I'd be pleased as punch if my DS2000 grew a couple of extra channels, even if they just ran at 1/4 the maximum sample rate - because as you say, there are plenty of applications where that works just fine.

[DanielS]

Apologies if this has been pointed out before...

The resistor jumper configuration for "Hardware Version" is different on DS1054Z compared to DS1104Z.

That number does not necessarily mean anything: many months have gone by since the DS1104Z launched so it is highly probable that Rigol has tweaked the PCB or BoM since then and DS1104Z manufactured more recently might be revision 01 as well.

You would need to find someone who cracked open a post-DS1054Z DS1104Z to confirm that the 01 revision is specific to the DS1054Z.

[Mark_O]

Oh and Pascal, when do you buy an Rigol scope ??

Now there's an interesting question!  Let's look back at the last year...

January 01, 2014, 02:40:30 PM »
I am going to buy a digital oscilloscope, but am doubting between the DS1074Z (4 channels) or the DS2072 (2 channels).

January 01, 2014, 05:38:35 PM »
I am interested in DS2072A but want to wait until the hack is confirmed with new FW and all options enabled, including 300MHz and CAN decoding.

January 02, 2014, 04:20:22 PM »
is the DS2072A hackable to 300 MHz with all options or not?
How long does it typically take to come up with a new hack? =) I need to buy my scope latest next week

January 21, 2014, 09:49:04 AM »
Are there any rumours or confirmations about upcoming DS2000 series with built-in LA?
Will the LA have 8 channels or 16 channels? When is it expected? Details on launch date and model numbers? Pictures?

~~ Purchase decisions got deferred, more questions were asked, the LA option awaited.  Then he focused on the MSO1074z (after flirting for a bit with the Siglent SDS2000).

June 30, 2014, 03:56:18 AM »
I am considering to buy a new scope now, and am thinking about MSO1074Z, as it seems to have everything I need.  4 Channels is very handy.

July 11, 2014, 07:57:32 AM »
Anybody with more feedback about the MSO2072A or the MSO1074Z?
If you are doing digital design 2 analog channels are more than enough given that you have 16 digital channels.
So that makes me conclude to go for the MSO2072A. Now just need more feedback on it

August 04, 2014, 09:21:00 AM »
Myself am going to order very soon an MSO1074Z, or an MSO2072A. Still doubting about which one is the best, but most likely it will be the MSO1074Z, as 100 MHz (after possible upgrade) would be enough for my current applications.  But I really want confirmation that feature wise both scopes are as good.

August 07, 2014, 12:33:46 PM »
From that perspective I think I have finally made up my mind. It is simply going to be a MSO2072A

~~ Finally!     But not so fast...

August 09, 2014, 04:49:16 PM »
It actually seems that intensity grading is better on 1074Z series than on 2000A series.
Also the screen contrast is better on the cheaper 1074Z series.
Now I am really confused. Although the screen is smaller on the 1074Z series, the screen quality is better than the 2000A series.
Honestly I am clueless again, and still can not decide then on MSO1074Z or MSO2072A, after seeing this video =)

~~ Marmad finally inquired, 2 months ago:  "seriously, man, you've been posting these same questions for over 7 months now... perhaps it's time to bite the bullet and buy a DSO?"

August 13, 2014, 05:43:40 AM »
Yes, I agree. I will buy the scope in September =)   [MSO2000, after debating viewing angles vs. 1000z]

~~ But hold on a minute...

September 10, 2014, 05:21:47 AM »
Considering to buy MSO1074Z or MSO2072A.  Most likely I will go for MSO2072A.

October 09, 2014, 10:51:54 AM »
I haven't bought the scope yet, as I was waiting on some videos for the LA functionality in the MSO series.  But scope will be ordered soon (other priorities now).

~~

My personal suspicion is he won't ever order any DSO.  Because: a) he wants it to be perfect, and the "best" in every possible way, b) he wants it to be trivially easy to hack, c) he wants guarantees that it will have no limitations that he might some day run into, and the major #1 reason...  (drum roll please)

d) he really has no need for a DSO!   

[Orange]

Oh and Pascal, when do you buy an Rigol scope ??

Now there's an interesting question!  Let's look back at the last year...

January 01, 2014, 02:40:30 PM »
I am going to buy a digital oscilloscope, but am doubting between the DS1074Z (4 channels) or the DS2072 (2 channels).

January 01, 2014, 05:38:35 PM »
I am interested in DS2072A but want to wait until the hack is confirmed with new FW and all options enabled, including 300MHz and CAN decoding.

January 02, 2014, 04:20:22 PM »
is the DS2072A hackable to 300 MHz with all options or not?
How long does it typically take to come up with a new hack? =) I need to buy my scope latest next week

January 21, 2014, 09:49:04 AM »
Are there any rumours or confirmations about upcoming DS2000 series with built-in LA?
Will the LA have 8 channels or 16 channels? When is it expected? Details on launch date and model numbers? Pictures?

~~ Purchase decisions got deferred, more questions were asked, the LA option awaited.  Then he focused on the MSO1074z (after flirting for a bit with the Siglent SDS2000).

June 30, 2014, 03:56:18 AM »
I am considering to buy a new scope now, and am thinking about MSO1074Z, as it seems to have everything I need.  4 Channels is very handy.

July 11, 2014, 07:57:32 AM »
Anybody with more feedback about the MSO2072A or the MSO1074Z?
If you are doing digital design 2 analog channels are more than enough given that you have 16 digital channels.
So that makes me conclude to go for the MSO2072A. Now just need more feedback on it

August 04, 2014, 09:21:00 AM »
Myself am going to order very soon an MSO1074Z, or an MSO2072A. Still doubting about which one is the best, but most likely it will be the MSO1074Z, as 100 MHz (after possible upgrade) would be enough for my current applications.  But I really want confirmation that feature wise both scopes are as good.

August 07, 2014, 12:33:46 PM »
From that perspective I think I have finally made up my mind. It is simply going to be a MSO2072A

~~ Finally!     But not so fast...

August 09, 2014, 04:49:16 PM »
It actually seems that intensity grading is better on 1074Z series than on 2000A series.
Also the screen contrast is better on the cheaper 1074Z series.
Now I am really confused. Although the screen is smaller on the 1074Z series, the screen quality is better than the 2000A series.
Honestly I am clueless again, and still can not decide then on MSO1074Z or MSO2072A, after seeing this video =)

~~ Marmad finally inquired, 2 months ago:  "seriously, man, you've been posting these same questions for over 7 months now... perhaps it's time to bite the bullet and buy a DSO?"

August 13, 2014, 05:43:40 AM »
Yes, I agree. I will buy the scope in September =)   [MSO2000, after debating viewing angles vs. 1000z]

~~ But hold on a minute...

September 10, 2014, 05:21:47 AM »
Considering to buy MSO1074Z or MSO2072A.  Most likely I will go for MSO2072A.

October 09, 2014, 10:51:54 AM »
I haven't bought the scope yet, as I was waiting on some videos for the LA functionality in the MSO series.  But scope will be ordered soon (other priorities now).

~~

My personal suspicion is he won't ever order any DSO.  Because: a) he wants it to be perfect, and the "best" in every possible way, b) he wants it to be trivially easy to hack, c) he wants guarantees that it will have no limitations that he might some day run into, and the major #1 reason...  (drum roll please)

d) he really has no need for a DSO!   

For sure he keeps us busy 

[pascal_sweden]

Good summary! Yes, keeps everybody busy!
I will come with the big announcement at the right time

[Bert Camper]

Good summary! Yes, keeps everybody busy!
I will come with the big announcement at the right time

Pascal,

Er is alleen maar tijd, geen "goede" tijd. Maak de beslissing en word blij

--Bert

[David Hess]

Now I am confused.  Are you referring to the analog antialiasing filter before the digitizer or the reconstruction filter after the digitizer?  If aliasing occurs then the later cannot do anything about it.

Both are low-pass filters. Leakage is a phenomenon that is not associated with aliasing (since it happens below Nyquist) and is not preventable by antialiasing or reconstruction filters.

Yes, and we established that in the screenshots I made there were no frequency components above Nyquist, so aliasing was not occurring.

This is the same screenshot of the 120 MHz signal where I counted wrong and you made the same counting mistake:

https://www.eevblog.com/forum/testgear/new-rigol-ds1054z-oscilloscope/msg532229/#msg532229

The signal is a 120 MHz sine wave.  The image above the 125 MHz Nyquist frequency is at 130 MHz.  After reconstruction with a low pass filter, any remaining part of the 130 MHz image should result in constructive and destructive interference with a beat frequency of 10 MHz which is what the blue line I have added shows.

This post covers the same problem and mentions the Agilent MSO6034 manual which includes a statement about their reconstruction filter having a bandwidth of half the Nyquist frequency:

https://www.eevblog.com/forum/chat/rigol-ds1000e-series-possible-errorfail-in-sin%28x%29x-interpolation/msg2545/#msg2545

note: what results is not really an AM signal, but a DSB signal (or AM with suppressed carrier, the sine does not ride the wave) -equivalent to the summation of the original "real" 120 MHz and the leaked "mirror" at 130 MHz. In the FFT's a few pages ago and attached here you can see that the amplitude of the mirror grows as the sampled frequency approaches Fnyquist. The result is shown as a double-sideband signal, not really an AM signal as there is no power in the central frequency of 125 MHz.

And the higher sideband grows stronger as Nyquist is approached because the reconstruction filter has lower attenuation.  The waveform shown in the image does not completely cancel because the 130 MHz image is already smaller than the 120 MHz signal.  With an input of 124.995 MHz, the peak to trough frequency would be 10 kHz and cancellation would be almost complete.

All this is after the digitizer and no aliasing occurring.
So as Marmad noticed this leakage is no issue till say Fsample/2.5 (my 100 MHz). There the leakage is so small is does not show any more.

The leakage is still there.  If you look really carefully at the lower frequency examples, you can see the amplitude still varying but it happens at a higher frequency because the signal and image are further apart.

As I understand it (now) up until Fnyquist you can reconstruct the frequency, but the amplitude information gets lost above Fsample/2.5. Just too few samples, and these samples "shift" along the wave resulting in the AM like waveform
Looking at the samples near Fnyquist themselves I am not even sure that a reliable reconstruction of the amplitude is even possible, and that it has nothing to do with errors in the reconstruction algorithms used by Rigol? At least that is what the documentation Marmad supplied is suggesting.
Interesting learning experience. This goes to show that one should be really really careful when interpreting displayed waveforms even long before FNyquist.  So the DS1000Z's are 100 MHz scopes? Yes, but beware..

If the filter was long enough, then the 130 MHz image would have been removed.  The filter however is shorter than this because of both performance reasons and because the filter length subtracts from the usable record length at the ends where a full set of samples is not available to calculate it.

This post was late not because I had given up but because the EEVBlog forum was returning database errors for me most of yesterday.   I am not giving up because I want to understand exactly what is going on here.

[marmad]

This post covers the same problem and mentions the Agilent MSO6034 manual which includes a statement about their reconstruction filter having a bandwidth of half the Nyquist frequency:

No, this doesn't cover the same problem; this is you attempting to connect this with a problem the DS1000E had - just as you tried to do before earlier in this thread. Rf-loop and jahonen were discussing the problem of the sin(x)/x interpolation on the DS1000E not matching the actual sample points. That is not the issue at all here - the interpolation matches the sample points just fine (see images).

The leakage is still there.  If you look really carefully at the lower frequency examples, you can see the amplitude still varying but it happens at a higher frequency because the signal and image are further apart.

Huh? No it doesn't. I just examined his 100MHz image in Photoshop and the amplitude only varies +/- 1 pixel - certainly within the DSO's margin of error, especially since the display memory is 2x scaled from the intensity buffer .

If the filter was long enough, then the 130 MHz image would have been removed.  The filter however is shorter than this because of both performance reasons and because the filter length subtracts from the usable record length at the ends where a full set of samples is not available to calculate it.

By long enough, do you mean infinitely or impossibly longer? If not, please produce a SINGLE real-world example of a DSO sampling a frequency between fs/2.1 and fs/2.05 and reconstructing the waveform correctly. I would love to see it. 

[David Hess]

This post covers the same problem and mentions the Agilent MSO6034 manual which includes a statement about their reconstruction filter having a bandwidth of half the Nyquist frequency:

No, this doesn't cover the same problem; this is you attempting to connect this with a problem the DS1000E had - just as you tried to do before earlier in this thread. Rf-loop and jahonen were discussing the problem of the sin(x)/x interpolation on the DS1000E not matching the actual sample points. That is not the issue at all here - the interpolation matches the sample points just fine (see images).

Not having the sin(x)/x reconstruction match the sample points is a different problem.  It just means that the filter did not use enough terms or used windowing or it was not a sin(x)/x filter which amounts to the same thing.  Rigol might have used a more aggressive filter to prevent the issue being discussed and raise the real time bandwidth in the way Agilent used the term even though it would mean that the actual sample points would not be on the reconstructed waveform.

The leakage is still there.  If you look really carefully at the lower frequency examples, you can see the amplitude still varying but it happens at a higher frequency because the signal and image are further apart.

Huh? No it doesn't. I just examined his 100MHz image in Photoshop and the amplitude only varies +/- 1 pixel - certainly within the DSO's margin of error, especially since the display memory is 2x scaled from the intensity buffer .

It looks like a difference to me because it repeats over the entire displayed waveform and I was specifically looking for it.  At some point in the past, I would have missed or dismissed it.

The amount of leakage is continuous with the input frequency.  At some frequency it will be low enough not to be seen visually but it will still be present which an FFT will show.  At a lower frequency yet it will become indistinguishable from noise.  Pick the right higher frequency and it will be +/- 2 pixels.  Pick a higher one yet and it will be +/- 3 pixels. 

If the filter was long enough, then the 130 MHz image would have been removed.  The filter however is shorter than this because of both performance reasons and because the filter length subtracts from the usable record length at the ends where a full set of samples is not available to calculate it.

By long enough, do you mean infinitely or impossibly longer? If not, please produce a SINGLE real-world example of a DSO sampling a frequency between fs/2.1 and fs/2.05 and reconstructing the waveform correctly. I would love to see it. 

I mean significantly longer then the filter Rigol is using now and gave the reasons why it is not longer.  I do not expect the Rigol to have perfect sin(x)/x reconstruction up to Nyquist.  I have yet to see a DSO which does.  It might be nice though to get an idea of the actual real time bandwidth available before distortion from the image becomes significant.  It looks like they did barely meet their 100 MHz specification at a sample rate of 250 MS/s.

I cannot give an example of a DSO doing this for the reasons discussed above but textbooks give graphic examples of ideal sin(x)/x reconstruction all the time when they discuss representing all input frequencies up to but not including the Nyquist frequency.  How else could a sampled data stream represent all frequencies up to but not included the Nyquist frequency accurately otherwise?

DSOs with longer and sharper reconstruction filters will display more accurate results up to their Nyquist frequency.  If a sin(x)/x filter is used, then the reconstructed waveform will also pass through the actual sample points which is a desirable feature.

The FFT of the 120 MHz input that pa3bca posted is magnified so any low frequency aliasing is left out but look carefully at the 120 MHz and 130 MHz signals.  The later one is about one minor division less or 4 dB.  If that FFT is of the reconstructed signal (otherwise how could it display the 130 MHz sine wave?), then that represents the attenuation difference of Rigol's reconstruction filter over that 10 MHz span.

[marmad]

At a lower frequency yet it will become indistinguishable from noise.  Pick the right higher frequency and it will be +/- 2 pixels.  Pick a higher one yet and it will be +/- 3 pixels.

Of course. But I think it's indistinguishable from noise in the 100MHz image - and I don't think a low-cost DSO has to do any better than fs/2.5 using sin(x)/x.

I cannot give an example of a DSO doing this for the reasons discussed above but textbooks give graphic examples of ideal sin(x)/x reconstruction all the time when they discuss representing all input frequencies up to but not including the Nyquist frequency.  How else could a sampled data stream represent all frequencies up to but not included the Nyquist frequency accurately otherwise?

Yes, textbooks - but who does this in reality? Who is capturing and reproducing frequencies right up to Nyquist? Everybody oversamples. And every paper on sin(x)/x interpolation on DSOs that I've read - although they might mention that it's mathematically possible to represent all frequencies up to Nyquist - ALWAYS mention fs/2.5, fs/3 or fs/4 as the minimum that should be used.

[pa3bca]

This is the same screenshot of the 120 MHz signal where I counted wrong and you made the same counting mistake:

https://www.eevblog.com/forum/testgear/new-rigol-ds1054z-oscilloscope/msg532229/#msg532229

The signal is a 120 MHz sine wave.  The image above the 125 MHz Nyquist frequency is at 130 MHz.  After reconstruction with a low pass filter, any remaining part of the 130 MHz image should result in constructive and destructive interference with a beat frequency of 10 MHz which is what the blue line I have added shows.

Oops. My bad. The 10 MHz is indeed your blue line riding the wavecrests, not my (5 MHz) red line. Was not paying attention (and counting) as I was trying to reconcile the two signals (120 and 130) into a DSB waveform.