New Rigol DS1054Z oscilloscope

[marmad]

If the sampler is very slightly out of sync with the incoming waveform (eg. if I sample at 2.0001 kHz) then I ought to see a 1kHz sine wave with pulsating amplitude, right?

If everything is perfect then yes...

Absolutely not. As already shown in both my earlier posted drawing and pa3bca's 120MHz image, both the amplitude and frequency are shifted by leakage when close to Nyquist. This is not much of an issue in the time domain, since no one normally attempts to sample and reconstruct signals close to fs/2 - but in the frequency domain it's a different story because of the nature of the FFT process, and has led to various adaptive sampling-frequency algorithms.

...and this is where I disagree with some here.

Sorry David, you may disagree - but you're just plain wrong.   This has nothing to do with digitizers - it's just math. I dug around online and found the following info for you from a computational physics course:

"To see the problem of frequency leakage, one can simply look at a plot of the function f(t)=cos(3t) and the sampled data points with a sampling rate of one per second ( red circle)."



"In this case, the sampling rate is one per second and its corresponding Nyquist frequency is 0.5 Hz. The actual frequency of the function is 0.477 Hz (3/2pi) and under-sampling is not a problem. However, the mismatch of the the fundamental frequency and the sampling frequency introduces spurious low frequency components in the sampled data. Note the slow oscillation in the sampled data. The result of this mismatch is that the power of the fundamental frequency leaks into other frequencies. This can be seen from the following power spectra."



"The peaks at the fundamental frequency are broadened by this effect. As the total length of the data increases, the resolution in frequencies increases and the peak at the fundamental frequency gets sharper."

[pascal_sweden]

The Rigol R&D department consists of many highly skilled electrical engineers, with master degrees, and probably many with PhD degree as well. They know all these mathematics and physics very well.
So from that perspective their equipment should be designed to meet all the constraints imposed by physical laws and sampling theorems.

Therefore I believe that their low-pass filter, ADC specs, and software algorithms are most likely well engineered and configured in an optimal way.

[pa3bca]

The Rigol R&D department consists of many highly skilled electrical engineers, with master degrees, and probably many with PhD degree as well. They know all these mathematics and physics very well.
So from that perspective their equipment should be designed to meet all the constraints imposed by physical laws and sampling theorems.

Therefore I believe that their low-pass filter, ADC specs, and software algorithms are most likely well engineered and configured in an optimal way.

The first sentence: yes, obviously. "You canna break the laws of physics captain"
Second sentence: No.
You left out at least two other important dimensions. Money and time. Where money is probably the most important one..
For instance very good/sharp roll off filters are expensive and introduce other issues. So no, at $399 this is not the best that is technically possible given physics laws.. But for the money it comes d*mned close and it is certainly well engineered..

[pa3bca]

Absolutely not. As already shown in both my earlier posted drawing and pa3bca's 120MHz image, both the amplitude and frequency are shifted by leakage when close to Nyquist. This is not much of an issue in the time domain, since no one normally attempts to sample and reconstruct signals close to fs/2 - but in the frequency domain it's a different story because of the nature of the FFT process, and has led to various adaptive sampling-frequency algorithms.

...and this is where I disagree with some here.

Sorry David, you may disagree - but you're plain wrong.   This has nothing to do with digitizers - it's just math. I dug around online and found the following info for you from a computational physics course:

Agree with Marmad.
With the fundamental near fs/2 the mirror frequency (i.e. fs-f) has almost the same amplitude shown by the scope (see the FFT's)
My 120 MHz has a "mirror frequency" of 250(fs)-120 = 130 MHz. Add these to together and you get an amplitude modulated signal with an modulation frequency of 130-120 = 10 MHz. See attached screenshot with the TG at 120 MHz and 250 Ms/s.

No way this can be caused by harmonics/spurious of the 120 MHz TG signal. (I can find no other mathematical relation, and the amplitudes would be way off)

[Fungus]

And the quandary you bring up about sampling a frequency at fs/2, or even frequencies very close to the Nyquist frequency (like pa3bca's 120MHz - which was fs/2.08) - i.e. that the frequency locations are unknown - is one of the reasons that, although sampling theory states that a sample-rate of 2*BW or larger is sufficient to reproduce frequencies < BW, most papers on DSO sampling and interpolation speak about a fs/2.5 ratio as the absolute minimum for reconstruction.

...and presumably the"bandwidth times ten" grows upwards from there.

[marmad]

So no, at $399 this is not the best that is technically possible given physics laws.. But for the money it comes d*mned close and it is certainly well engineered..

The price is pretty astounding. The ADC alone is listed at ~$62 in quantities of 500 from Digikey.

[Fungus]

So no, at $399 this is not the best that is technically possible given physics laws.. But for the money it comes d*mned close and it is certainly well engineered..

The price is pretty astounding. The ADC alone is listed at ~$62 in quantities of 500 from Digikey.

Rigol can probably get it cheaper but yeah, it would be interesting to see the BOM. It's hard to see how they're making any money on these. As noted, they don't seem to be skimping on components or materials.

I guess it's because they've been selling them more expensive for a while (DS1074Z/DS1104Z) and have got enough R&D investment back to make the $400 DS1054Z possible.

[coppice]

The price is pretty astounding. The ADC alone is listed at ~$62 in quantities of 500 from Digikey.

Pricing up the BOM of a product from Digikey prices is like pricing up the BOM of a car from the cost of spares at your local dealer.

I suspect Rigol buy at least 501 of these devices and get a somewhat better price. 

[marmad]

Pricing up the BOM of a product from Digikey prices is like pricing up the BOM of a car from the cost of spares at your local dealer.

Not quite. It's more like getting the price of 500 catalytic converters from a wholesaler of automobile parts.

I suspect Rigol buy at least 501 of these devices and get a somewhat better price. 

Obviously. But this is an oscilloscope, not an iPhone - Rigol's not selling 12 million (or even 12 thousand) a month. I never suspected they paid close to that amount, but that (and the $73 single-unit price) gives me a ballpark idea of what they might be paying for it. 

[Fungus]

Just watched the teardown video...

The analog input part is a lot more complicated than I would have guessed.

[coppice]

But this is an oscilloscope, not an iPhone - Rigol's not selling 12 million (or even 12 thousand) a month. I never suspected they paid close to that amount, but that (and the $73 single-unit price) gives me a ballpark idea of what they might be paying for it. 

Oscilloscope volumes can be surprisingly high, and Rigol is one of the more successful players. I wouldn't be surprised if they made 12k a month of this entry level model. I would be very surprised if they paid even $10 for that ADC chip.

[marmad]

Oscilloscope volumes can be surprisingly high, and Rigol is one of the more successful players. I wouldn't be surprised if they made 12k a month of this entry level model. I would be very surprised if they paid even $10 for that ADC chip.

Would like to post some actual data to support any of this?

[coppice]

Oscilloscope volumes can be surprisingly high, and Rigol is one of the more successful players. I wouldn't be surprised if they made 12k a month of this entry level model. I would be very surprised if they paid even $10 for that ADC chip.

Would like to post some actual data to support any of this?

I was amazed to find a very obscure maker of simple scopes was making a couple of thousand a month some years ago, so 12k a month for a world class producer doesn't sound odd to me.

As for the IC price, just look at the pricing of most stuff on Digikey, compared to what real negotiated prices look like. Some of them are 50 times a reasonable price. You go to Digikey to get a few bits quickly to make experimental setups and prototypes. Its not where you go for serious production. I'm not knocking Digikey. Convenience is expensive to provide. The die for this chip isn't going to be huge, although the test costs might be significant - high speed mixed signal testing costs.

[David Hess]

"To see the problem of frequency leakage, one can simply look at a plot of the function f(t)=cos(3t) and the sampled data points with a sampling rate of one per second ( red circle)."

"In this case, the sampling rate is one per second and its corresponding Nyquist frequency is 0.5 Hz. The actual frequency of the function is 0.477 Hz (3/2pi) and under-sampling is not a problem. However, the mismatch of the the fundamental frequency and the sampling frequency introduces spurious low frequency components in the sampled data. Note the slow oscillation in the sampled data. The result of this mismatch is that the power of the fundamental frequency leaks into other frequencies. This can be seen from the following power spectra."

"The peaks at the fundamental frequency are broadened by this effect. As the total length of the data increases, the resolution in frequencies increases and the peak at the fundamental frequency gets sharper."

I do not disagree with any of the above.  I think it is being applied incorrectly here.

The sampled points shown alone display exactly what is described but the examples I am familiar with include sin(x)/x reconstruction after sampling which reproduces the original waveform including the peaks correctly.

On DSOs that I have used under the proper conditions (I would not normally be measuring pure sine waves close to Nyquist) I can generate exactly what is shown and described with reconstruction disabled and then restore the original waveform, which is close to Nyquist but not exceeding it, with reconstruction turned on.  If aliasing is present for whatever reason, and it will always be present because of imperfections in the digitizer no matter how pure the input signal is, then that fails and I get the result described anyway in a lessor or greater form.  That is a distinction without a difference on the practical side but is not what is being described above.

[marmad]

I was amazed to find a very obscure maker of simple scopes was making a couple of thousand a month some years ago, so 12k a month for a world class producer doesn't sound odd to me.

As for the IC price, just look at the pricing of most stuff on Digikey, compared to what real negotiated prices look like. Some of them are 50 times a reasonable price. You go to Digikey to get a few bits quickly to make experimental setups and prototypes. Its not where you go for serious production. I'm not knocking Digikey. Convenience is expensive to provide. The die for this chip isn't going to be huge, although the test costs might be significant - high speed mixed signal testing costs.

I love your enthusiasm and righteousness, but once again - I posted a fact: HMCAD5211 at $62 for 500 units - you've posted stories and speculation, but no actual data. Most people know good (and even not so good) ADCs are expensive - even in quantities of 10k.

[coppice]

I love your enthusiasm and righteousness, but once again - I posted a fact: HMCAD5211 at $62 for 500 units - you've posted stories and speculation, but no actual data. Most people know good (and even not so good) ADCs are expensive - even in quantities of 10k.

Get real. You quoted a Digikey price. For a quantity of 500 the vendor won't talk to you, but their main distributors will. If you'd quoted a price from them you'd have more credibility.

Let's say Rigol only make 2k a month. They've made the previous model for about 7 or 8 years, haven't they? If this model runs as long that would be 170k or so. You can start talking about serious contracted prices with the vendor for a production run like that and a fairly high value part.

[David Hess]

I love your enthusiasm and righteousness, but once again - I posted a fact: HMCAD5211 at $62 for 500 units - you've posted stories and speculation, but no actual data. Most people know good (and even not so good) ADCs are expensive - even in quantities of 10k.

I found the same price and did not think much of it other than what an amazing deal even though Digi-Key and similar distributors will be at the high end of the price scale.  I would hope Rigol is buying directly from Hittite with their volume.

I remember similar parts in the past costing at least an order of magnitude more for less performance and functionality.

[marmad]

Get real. You quoted a Digikey price.

Yes, I quoted a price. OTOH, you've invented sales volumes and prices. 

I would hope Rigol is buying directly from Hittite with their volume.

I would think so. Not so sure about the < $10 a piece price though.

[coppice]

I love your enthusiasm and righteousness, but once again - I posted a fact: HMCAD5211 at $62 for 500 units - you've posted stories and speculation, but no actual data. Most people know good (and even not so good) ADCs are expensive - even in quantities of 10k.

I found the same price and did not think much of it other than what an amazing deal even though Digi-Key and similar distributors will be at the high end of the price scale.  I would hope Rigol is buying directly from Hittite with their volume.

I remember similar parts in the past costing at least an order of magnitude more for less performance and functionality.

Its not that long ago you couldn't buy such functionality as a chip, even at exotic niche product prices. Times change, and devices like this do have volume applications now. Hittite describes (or is it described now they are part of ADI) itself as a microwave company. That tells you what their core market is.

[marmad]

The sampled points shown alone display exactly what is described but the examples I am familiar with include sin(x)/x reconstruction after sampling which reproduces the original waveform including the peaks correctly.

I'm not sure I understand you. It's obvious that sin(x)/x being run on the displayed sampled points will not reproduce the original waveform correctly. In fact, not only does sin(x)/x interpolation reproduce just what we saw in pa3bca's 120MHz image, but I can reproduce the exact same results by setting my DSO at any sampling rate and sending it a sine wave with the frequency of ~fs/2.08 (and other frequencies close to - but less than - the Nyquist frequency).

On DSOs that I have used under the proper conditions (I would not normally be measuring pure sine waves close to Nyquist) I can generate exactly what is shown and described with reconstruction disabled and then restore the original waveform, which is close to Nyquist but not exceeding it, with reconstruction turned on.

I don't know what you mean by 'the proper conditions', but the math above says that it would be extremely difficult to do that over much time. The Nyquist theorem simply states that if a signal contains no frequencies higher than B, it is mathematically possible to reproduce it by a series of points spaced 1/(2B). It says nothing about the length of the sampling. Leakage occurs over time; it's window dependent - e.g. if you just sample a cycle or two of a 120MHz sine wave at 250MSa/s, you can get a reproduction of the frequency of the waveform - thus satisfying the Nyquist theorem. But if you begin to sample more cycles than that, leakage WILL occur.

[pa3bca]

The sampled points shown alone display exactly what is described but the examples I am familiar with include sin(x)/x reconstruction after sampling which reproduces the original waveform including the peaks correctly.

I'm not sure I understand you. It's obvious that sin(x)/x being run on the displayed sampled points will not reproduce the original waveform correctly. In fact, not only does sin(x)/x interpolation reproduce just what we saw in pa3bca's 120MHz image, but I can reproduce the exact same results by setting my DSO at any sampling rate and sending it a sine wave with the frequency of ~fs/2.08 (and other frequencies close to - but less than - the Nyquist frequency).

On DSOs that I have used under the proper conditions (I would not normally be measuring pure sine waves close to Nyquist) I can generate exactly what is shown and described with reconstruction disabled and then restore the original waveform, which is close to Nyquist but not exceeding it, with reconstruction turned on.

I don't know what you mean by 'the proper conditions', but the math above says that it would be extremely difficult to do that over much time. The Nyquist theorem simply states that if a signal contains no frequencies higher than B, it is mathematically possible to reproduce it by a series of points spaced 1/(2B). It says nothing about the length of the sampling. Leakage occurs over time; it's window dependent - e.g. if you just sample a cycle or two of a 120MHz sine wave at 250MSa/s, you can get a reproduction of the frequency of the waveform - thus satisfying the Nyquist theorem. But if you begin to sample more cycles than that, leakage WILL occur.

Yes and what we have seen has nothing to do with the impurities of the applied signal. Yes impurities above Nyquist will fold back and show up, but even with my DSA815's 'crappy' TG this would be far down in the noise (i.e invisible).

As further proof: 24 MHz generated by my DG1032Z, DS2072A at 50 MSa/s.
Spurious of the DG1032 is at -60dB(2nd harmonic) to -70dB for others. Not stellar but quite sufficient thank you. So now we might all agree that spurious signals cannot have any visible influence on the displayed signals.
And as expected the scope shows an 'Amplitude modulated" signal. Modulation frequency of (25-24) - (25-26) = 2 MHz, 26 MHz being the "mirror" of the 24 MHz signal wrt to Fnyquist. All as explained by Marmad.
And (but maybe I misunderstood David) reconstruction will not restore the original 24 MHz waveform.

[Electro Fan]

As further proof: 24 MHz generated by my DG1032Z, DS2072A at 50 MSa/s.
Spurious of the DG1032 is at -60dB(2nd harmonic) to -70dB for others. Not stellar but quite sufficient thank you. So now we might all agree that spurious signals cannot have any visible influence on the displayed signals.
And as expected the scope shows an 'Amplitude modulated" signal. Modulation frequency of (25-24) - (25-26) = 2 MHz, 26 MHz being the "mirror" of the 24 MHz signal wrt to Fnyquist/2. All as explained by Marmad.
And (but maybe I misunderstood David) reconstruction will not restore the original 24 MHz waveform.

pa3bca, what Rigol product or user interface is shown in the image below? Maybe PC software or something else associated with the DG1032Z?

[pa3bca]

pa3bca, what Rigol product or user interface is shown in the image below?

Rigol DSA815 TG Spectrum Analyzer.
I wanted to have an SA since I was 14 years old or so (looong time ago). Too expensive (then) or to bulky&expensive for hobby use until this came out.
Bought one immediately, very happy with it!

[Electro Fan]

Very nice - please explain the cabling setup you used with the SA, DG1032Z, and scope.  How do you like the DG1032Z?

[pa3bca]

Very nice - please explain the setup you used with the SA, DG1032Z, and scope.  How do you like the DG1032Z?

Oh I used the DG1032Z to output a 24 MHz sine at 0dBm. Into the spectrum analyzer to check the output/spurious. (as shown above).
Notice how the SA agrees exactly on the output level of 0.00 dBm  I find both the SA and the DG performing very nice (well, all my Rigol stuff performs nice, especially if you look at the price points).
Then the the output of the DG to my DS2072 ("upgraded").
The DG1032Z performs nicely, in spec, use it a lot. The fan however is was rather noisy, easily the noisiest of all my Rigol devices. And it has I think the lowest power consumption of them all so go figure. I replaced the 50mm fan with a quite one, now it is almost inaudible and all stays cool enough.