New Rigol DS1054Z oscilloscope

[David Hess]

Only the 120 MHz results looks weird to me but I think I know what causes it in this case.  I have seen something similar which appeared to be related to the aperture time of the digitizer implying a non-linear frequency response but I do not think that is it.

The 120MHz result is what I would expect as the sine wave frequency starts to approach the border of the Nyquist frequency (125MHz) - the appearance of amplitude modulation due to 'leakage' (perfectly reproducing a frequency exactly half that of the sampling rate only works in theory). As the frequency reaches Nyquist, the AM will become more extreme. At some point past Nyquist, there is the reappearance of an alias that looks like a normally amplified sine wave again.

I agree in principle although I would not call it leakage.  I took another look at that test and did some tests of my own to replicate it.  What I see does look like an alias folded back below the Nyquist frequency and relatively close to the fundamental.  What I was saying in my later post is that we cannot know if this was produced in the DSO or if it originated from the test signal because the later has much higher distortion than the DSO has.

It is tedious to do but looking at the 120 MHz example carefully, it looks like a 120 MHz fundamental which is expected with a large spur at 95 MHz and offhand I do not see a way to produce that with aliasing in this case so I think it is just distortion in the source.

[pa3bca]

Only the 120 MHz results looks weird to me but I think I know what causes it in this case.  I have seen something similar which appeared to be related to the aperture time of the digitizer implying a non-linear frequency response but I do not think that is it.

The 120MHz result is what I would expect as the sine wave frequency starts to approach the border of the Nyquist frequency (125MHz) - the appearance of amplitude modulation due to 'leakage' (perfectly reproducing a frequency exactly half that of the sampling rate only works in theory). As the frequency reaches Nyquist, the AM will become more extreme. At some point past Nyquist, there is the reappearance of an alias that looks like a normally amplified sine wave again.

I agree in principle although I would not call it leakage.  I took another look at that test and did some tests of my own to replicate it.  What I see does look like an alias folded back below the Nyquist frequency and relatively close to the fundamental.  What I was saying in my later post is that we cannot know if this was produced in the DSO or if it originated from the test signal because the later has much higher distortion than the DSO has.

It is tedious to do but looking at the 120 MHz example carefully, it looks like a 120 MHz fundamental which is expected with a large spur at 95 MHz and offhand I do not see a way to produce that with aliasing in this case so I think it is just distortion in the source.

Based on the deep AM I think it is very unlikely that this is produced by harmonics of the TG at 120 MHz. Those are at least 20 dB down wrt the fundamental and the "modulation depth" of the signal as shown on the screen is almost 50%.

Did a quick re-test with FTT on. TG at 120 MHz. THis is the best (resolution) I can get at these settings.
The 120 MHz component can be clearly seen, and it looks like at 130 MHz there is another one (confabulated by the scope  ). Exactly mirrored around the 125 MHz nyquist.
hmmm..

[leppie]

Using seronday's chart, I've revamped my earlier figure showing the (even bigger) risk of aliasing if not limiting the BW to 20MHz when running with 3/4 channels on. The roll-off of the DS1000Z is so slow, that frequency content above Nyquist could be aliasing as high as almost -2db when using 3/4 channels @ 250MSa/s.

The reference dB was 0.98, so 125Mhz should be at around -3.2dB (given the -2.2 reading), no?

[Fungus]

Analog ICs may get binned based on linearity, noise, bandwidth, offsets, current draw, etc.

Yep, but he was talking about speed.

The binning if they did this would be for distortion produced anywhere from the BNC to digitizer.

Based on what we know about the amplifiers and ADCs used, it could vary by about 12 dB or so.

As much as that? That's surprising.

I assume this is what the self calibration is for.

Does that affect bandwidth or just voltage measurements?

[marmad]

The reference dB was 0.98, so 125Mhz should be at around -3.2dB (given the -2.2 reading), no?

Thanks for pointing out the error. I screwed-up in a few different ways when I edited together the charts: I didn't move from log scale to linear for dB; I didn't fix seronday's odd BW scaling; and most importantly - I forgot that his chart was made on an unaltered DS1074Z - so it reflects a "70MHz" BW limit.

So I've redone and re-posted the chart back at the original message - adding dots and staight lines reflecting Stan Perkins' posted measurements for his 100MHz modified DS1054Z - but if anyone has a more detailed BW frequency response chart for a stock DS1104Z or modified DS1054Z, I'd be happy to add the data.

[pa3bca]

Did a quick re-test with FTT on. TG at 120 MHz. THis is the best (resolution) I can get at these settings.
The 120 MHz component can be clearly seen, and it looks like at 130 MHz there is another one (confabulated by the scope  ). Exactly mirrored around the 125 MHz nyquist.
hmmm..

Well better resolution is possible. See attached for FFT's with the TG at 120, 110 and 100 MHz.
the mirror image goes from a few dB's under the fundamental @ 120 Mhz to more than 20 dB with an 100 MHz input signal.
Definitely a DSO artifact, and not caused by spurious/harmonics of the TG..
The AM you saw in the earlier screenshot of the 120 MHz signal is definitely caused by mixing this 120 MHz and the confabulated 130 MHz by the DSO.

[i4004]

Me? I'm gonna relax and enjoy my 4-channel, top-build-quality, sub-100Mhz-with-four-channels-enabled, 300 Euro oscilloscope...  (thanks, Rigol!)

that is probably price without vat...on batronix....and that's not the best price in europe it seems to me (marmad already mentioned another option
https://www.eevblog.com/forum/testgear/rigol-ds1074z-oscillosope/msg514730/#msg514730
see marmad, i do know how to use search    i can also read. a lot. esp. if it saves me money.  )

come to think i almost bought 1052 from china. 

and some leftovers from back pages:

rolycat (about the rigol and hacking)..that's a gray area but also an area where it would be real easy to prove (in the court of law) that rigol actually likes things to remain just like this. (just by showing sales figures for 1052) 

marmad (about search capabilities)...just like i wrote in added edit "it's over, google found us just now, now everybody knows!" which was written prior to your post.
so yeah, now you can, but then? no.
what surprised me was the forum search had the simillar lag as google....

it essentially boils down to what miguelvp said few posts later

They tried to prevent it but they didn't use the algorithm properly and the private key kind of leaked, don't recall the details, they are buried in the monster thread somewhere.

or what you said

(part of the problem with spread-out information on this blog)

now, i know what that thread is, but 99% of the planet doesn't, but in essence you're right...who cares if they don't? 1% (or less) of populaton knows what scope is anyway.... 

overall a good post from miguel there, and just about the only thing rigol could do (bare in mind that they have rather lil incentive to do anything if sales go like this, and they will go) is to connect scope to web and then update every so often to check if it was hacked (if so, then you..huhm...pay to u nlock it again? hehe).
but nobody would buy such a scope. 


somebody mentioned tek analog scopes of high bandwidth: these flopped for me just based on dimensions....1054z is about the size of 10mhz portable scopes that are now 30 years old.
that's pretty amazing.
(not going into performance specs at all, tek just won't fit on my table (and is not protable at all), and i won't be getting bigger table because of analog scope...)

[marmad]

Definitely a DSO artifact, and not caused by spurious/harmonics of the TG..
The AM you saw in the earlier screenshot of the 120 MHz signal is definitely caused by mixing this 120 MHz and the confabulated 130 MHz by the DSO.

When the upper and lower sidebands created by sampling start overlapping (when approaching Nyquist), first you can get false AM (taught as "leakage" by an old EE teacher of mine) and then aliases.

[Bert Camper]

For people in Belgium, the Netherlands or Luxembourg: The importer arBenelux has only 1 DS1054Z oscilloscope in stock, they hope to receive more before November 1. There is a big electronics fair in the Netherlands on November 1 "Dag voor de Radio Amateur"
http://www.veron.nl/activiteiten/details/activiteiten_dvdra.html

--Bert

[marmad]

It is tedious to do but looking at the 120 MHz example carefully, it looks like a 120 MHz fundamental which is expected with a large spur at 95 MHz and offhand I do not see a way to produce that with aliasing in this case so I think it is just distortion in the source.

No, it's a byproduct of sampling a frequency very close to Nyquist: a false appearance of amplitude modulation. The "modulating signal" is caused by leakage of the fundamental frequency. Here's an image I altered demonstrating the process in action (the red dots are the sample points connected via linear interpolation):

[i4004]

did you just steal my book?
 

yes, it's a good one...

as for quasi am, got that recently by connecting square wave gen (2.4khz) to tank circuit...on analog scope...

[David Hess]

The binning if they did this would be for distortion produced anywhere from the BNC to digitizer.

Based on what we know about the amplifiers and ADCs used, it could vary by about 12 dB or so.

As much as that? That's surprising.

I assume this is what the self calibration is for.

Does that affect bandwidth or just voltage measurements?

I am going by the datasheet specifications for subranging ADCs and amplifiers typically used in these DSOs.  The distortion of the ADC and signal conditioning could vary over that much range between good and not quite as good parts.  Self calibration only applies to the ADC and the numbers include it.  The ADCs are actually surprisingly good.

That number is only an estimate.  It could be worse in poor designs if for instance decoupling is poor or if the sample clock has high jitter which is a common problem.

Note that I do not think the ICs are binned.  I am just suggesting that Rigol might bin the completed boards during self testing because distortion could vary over that wide a range and it gets worse at higher frequencies.

[David Hess]

Do we know what ADC the DS1054Z uses?

Just posted the datasheet (Hittite website: HMCAD1511) right above your comment two minutes before you posted this. 

Doh!

It was early and I had too much blood in my caffeine stream and the Tardis is in the shop.

Reading the ADC datasheet shows how Rigol likely did gain correction.

I will trade you links though.  A friend forwarded this link to me.  His daughter has to study this video for some class she is doing.  Check out the explanations for Gibbs phenomenon and how reconstruction of a perfect edge works toward the end.

D/A and A/D | Digital Show and Tell (Monty Montgomery @ xiph.org) - YouTube

[David Hess]

What I meant when I refereed to this is that the distortion from the TG is going to conceal the distortion in the DSO unlike that other test with the video I linked to where an RF synthesizer intended for receiver testing was used.  The DSO will show aliasing from the TG distortion products.

Ah yes I now see what you mean.  Unfortunately I have nothing here (readily available) that can produce a 100MHz-ish signal where spurious is down more than 50 dB. Could build it (say a 5 pole low-pass after a 100 MHz generator) but not now.
But then again: are we going to see -50dB spurious signals back into the "passband" and into the display? with 8 bit A/D? what am I missing here...

I was thinking the same thing.  Hack together a simple tuned LC bandpass filter or actually, the frequency range is high enough that maybe a helical resonator could be used.

The 8 bit subranging  ADCs typically used have a spurious free dynamic range which could be as low as 47 dB and and effective number of bits which could be as low as 7.1 bits.  The signal conditioning before that probably will not make it any worse if it is at that level but if it was better which will be typical then it might.

-50 dB spurs may or may not be visible.

While I was out today I had a better idea for a test or at least a more relevant one.  Use a fast transition 1 to 4 MHz flat level crystal controlled square wave and measure the jitter.  Aliasing will occur but how well the DSO handles that counts.  To me that is a more relevant measurement of performance anyway.

I have a couple signal generators as described specifically intended for testing horizontal sweeps and transient response.

[Fungus]

Me? I'm gonna relax and enjoy my 4-channel, top-build-quality, sub-100Mhz-with-four-channels-enabled, 300 Euro oscilloscope...  (thanks, Rigol!)

that is probably price without vat...on batronix....

Yep. I got mine from Batronix ... and I didn't pay VAT  (because I'm European VAT registered and I don't live in Germany).

Check out the explanations for Gibbs phenomenon and how reconstruction of a perfect edge works toward the end.

D/A and A/D | Digital Show and Tell (Monty Montgomery @ xiph.org) - YouTube

Nice video. Those last few minutes certainly need some mulling over to adjust my worldview.

[coppice]

I will trade you links though.  A friend forwarded this link to me.  His daughter has to study this video for some class she is doing.  Check out the explanations for Gibbs phenomenon and how reconstruction of a perfect edge works toward the end.

D/A and A/D | Digital Show and Tell (Monty Montgomery @ xiph.org) - YouTube

I have found that video, and a couple of related ones by the same guy, to be the best things to refer someone to when they start some brain dead discourse about a sampled signal being a staircase or triangles or other bizarre notion.

[Fungus]

I have found that video, and a couple of related ones by the same guy, to be the best things to refer someone to when they start some brain dead discourse about a sampled signal being a staircase or triangles or other bizarre notion.

One thing I don't understand about sampling.... how are the samples actually taken?

Are the samples taken at an instant in time (or in a narrow window) as the theories seem to suggest? In that case, what happens if you sample a sine wave at exactly twice the frequency (eg. 1kHz sine wave, 2kHz sampler)?

a) If I sample at the peaks/troughs, everything is OK.

b) If I sample at the zero-crossing points then the wave will completely disappear (all my samples will be zero).

c) I can also sample at any point between (a) and (b), getting a 1KHz wave with different amplitudes.

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?

Widening the sample window and taking an average doesn't seem to help, so... how do samplers work?


PS: Sorry for the off-topic, but the last few pages have been full of signal theory.


Edit:

This image shows what I mean: https://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_sampling_theorem#mediaviewer/File:CriticalFrequencyAliasing.svg

At the "critical frequency" (I just found out it has a name) the original wave could have any amplitude at all...even infinite amplitude!

[marmad]

Are the samples taken at an instant in time (or in a narrow window) as the theories seem to suggest? In that case, what happens if you sample a sine wave at exactly twice the frequency (eg. 1kHz sine wave, 2kHz sampler)?

a) If I sample at the peaks/troughs, everything is OK.

b) If I sample at the zero-crossing points then the wave will completely disappear (all my samples will be zero).

c) I can also sample at any point between (a) and (b), getting a 1KHz wave with different amplitudes.

In DSOs, most sampling normally occurs at uniformly-spaced, instances of time - although there are exceptions. For example, Agilent uses random decimation for slower sample rates to produce nonuniform sampling in order to eliminate aliasing, but at the expense of introducing noise.

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.

As seen in pa3bca's images, fs/2.5 (250MSa/2.5 = 100MHz) is working correctly - at least for the simple sine wave .

[pascal_sweden]

It would be nice if there was a website e.g. scopeperformance.com, where experts have measured the actual bandwidth for commercial scopes using the proper equipment.

They could sort the scopes per brand, and by advertised bandwidth, and add a confirmed bandwidth next to it.

[marmad]

They could sort the scopes per brand, and by advertised bandwidth, and add a confirmed bandwidth next to it.

Just the BW alone would be pointless information by itself. As David Hess noted in this story before:

There is a guy on Ebay who rebuilds 150 MHz Tektronix 2445 oscilloscopes by removing the hardware bandwidth filter, setting a jumper to make the firmware think it is a 2465, and changing the faceplate to that of a 300 MHz 2465 oscilloscope.  These faux 2465s do indeed have 300 MHz bandwidth or higher but because the original 2445 lacks the rather ingenious frequency and phase compensation network included in a true 2465, the transient response is severely compromised.  Nobody noticed this on these Ebay specials for a long time because they just checked the bandwidth.

As he mentions elsewhere in his post, owners reporting that they've changed their DS1000Zs, DS2000As, DS4000s, etc. into the highest BW models by just measuring the BW doesn't actually prove it conclusively. Rigol may be storing filter coefficients determined by calibration at the time of manufacture to implement frequency and phase compensation - and even though you can enter a keycode to change the low-pass filter, it doesn't necessarily mean that the DSO is perfectly calibrated for that BW.

Remember, Rigol doesn't actually SELL bandwidth upgrades. If they did, it would be more likely that those BW-unlocking codes were doing the job completely - since money would be riding on it.

[pascal_sweden]

Can these calibration coefficients be read-out?
This way it would be possible to verify if they differ a lot between e.g. an original MDO2302A and an MSO2072A.

Or can the errors be measured with more extensive measurements besides BW measurement?
Is there a systematic way to do this, according to a unified, standardized test framework, where different test users could contribute to complement the database on the Internet?

[marmad]

Can these calibration coefficients be read-out?
This way it would be possible to verify if they differ a lot between e.g. an original MDO2302A and an MSO2072A.

Or can the errors be measured with more extensive measurements besides BW measurement?
Is there a systematic way to do this, according to a unified, standardized test framework, where different test users could contribute to complement the database on the Internet?

To what end? By using unauthorized codes to unlock options, you are getting some extra bandwidth (whether it's perfectly compensated or not) for free. If your livelihood depends on that extra bandwidth, you really should be paying for it - rather than risking some hack.

[David Hess]

I have found that video, and a couple of related ones by the same guy, to be the best things to refer someone to when they start some brain dead discourse about a sampled signal being a staircase or triangles or other bizarre notion.

One thing I don't understand about sampling.... how are the samples actually taken?

Are the samples taken at an instant in time (or in a narrow window) as the theories seem to suggest? In that case, what happens if you sample a sine wave at exactly twice the frequency (eg. 1kHz sine wave, 2kHz sampler)?

With a sampling converter, they are taken in a narrow window which usually has nothing to do with the sample rate.  With an integrating converter, the sampling duration is intimately related to the sample rate.  Delta-sigma converters are integrating converters and DSOs act as if they have integrating converters when operating in high resolution mode.

a) If I sample at the peaks/troughs, everything is OK.

b) If I sample at the zero-crossing points then the wave will completely disappear (all my samples will be zero).

c) I can also sample at any point between (a) and (b), getting a 1KHz wave with different amplitudes.

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 and this is where I disagree with some here.  The behavior of practical instruments with perfect inputs close to the Nyquist frequency is a reflection of imperfections in the digitizer including sampling error and non-linearity which produce distortion not present in the original signal.  If the distortion produced in the digitizer is above the Nyquist frequency, it causes aliasing.  With a better digitizer, you can get more accurate reconstructions closer to the Nyquist frequency.  Usually the simplest way to get a better digitizer is to increase the sampling rate.

Widening the sample window and taking an average doesn't seem to help, so... how do samplers work?

As a practical manner they do average or sort of average the input over some duration of time but usually this is much smaller than the length of time between samples.  The sampling duration sets an upper limit on the input bandwidth and creates a non-linear frequency response limiting the 3 dB bandwidth to 0.442/t.  If the sampling duration was 1ns for instance, there would be a 3 dB input bandwidth of 442 MHz and null at 1 GHz and every integer multiple of 1 GHz which is not something you find in a linear system.  This also changes the bandwidth to rise time relationship so it is no longer the bandwidth = 0.35/tr found with a single pole rolloff but more like 0.47/tr.  This is different than using a sharp anti-aliasing filter which is linear but which also affects the rise time similarly.

Old style sampling oscilloscopes have tiny sampling durations measured in picoseconds giving them very high bandwidths but their frequency response is decidedly non-linear and the bandwidth = 0.35/tr relationship does not usually apply to them.

This image shows what I mean: https://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_sampling_theorem#mediaviewer/File:CriticalFrequencyAliasing.svg

At the "critical frequency" (I just found out it has a name) the original wave could have any amplitude at all...even infinite amplitude!

If you sample synchronously so the input is at the Nyquist frequency, then you get an in indeterminate output which changes based on the phase relationship.  Sampling phase detectors take advantage of this.

[David Hess]

Can these calibration coefficients be read-out?
This way it would be possible to verify if they differ a lot between e.g. an original MDO2302A and an MSO2072A.

If they exist then maybe but it is not required.  See below.

Or can the errors be measured with more extensive measurements besides BW measurement?

Is there a systematic way to do this, according to a unified, standardized test framework, where different test users could contribute to complement the database on the Internet?

Sure, I do it all the time.  Use a reference flat top pulse generator and measure the transient response.  I suspect there is also a way to do this using vector network analyser techniques but a VNA guru should answer that.

Almost as good and more useful would be to compare the transient response of the upgraded DSO to the true one.  If there is a calibration or grading difference, it should be revealed.

[David Hess]

Can these calibration coefficients be read-out?
This way it would be possible to verify if they differ a lot between e.g. an original MDO2302A and an MSO2072A.

Or can the errors be measured with more extensive measurements besides BW measurement?
Is there a systematic way to do this, according to a unified, standardized test framework, where different test users could contribute to complement the database on the Internet?

To what end? By using unauthorized codes to unlock options, you are getting some extra bandwidth (whether it's perfectly compensated or not) for free. If your livelihood depends on that extra bandwidth, you really should be paying for it - rather than risking some hack.

The guy who was converting those analog Tektronix oscilloscopes I mentioned made the same argument and I largely agree.  What we disapproved of was him selling the modified oscilloscopes and representing them as the originals as far as performance with no notification of the difference which would have lowered their selling price.

If you used one of the modified oscilloscopes expecting the clean transient response of the true ones, your measurements would be compromised.  People who are using 300+ MHz oscilloscopes with passive x10 probes are unlikely to see the difference except under ideal conditions.  The same is likely to apply here.  How many people are looking at 3.5 nanosecond and faster transition time signals with their Rigol DS1054Z which has been hacked to 100 MHz and would notice?