Frequency response with a TEK MDO3000

[onesixright]

Hi All,

I was watching this tutorial from TEKl "How do I use the MDO3000 to generate a frequency response curve?"

).

According to TEK, you can create a FR up to 50 Mhz like this. Enough for audio filters i would think.

I always understood that for a Frequency Response, you would need a Spectrum Analyzer with a Tracking Generator.

My question:

Does this method actually differs (by results) from a SA+TG approach? Or could you state that, along your with in the 50Mhz bandwidth,  its exactly the same?

Many thanks!

[Wuerstchenhund]

I always understood that for a Frequency Response, you would need a Spectrum Analyzer with a Tracking Generator.

No, you don't. A swept SA + TG is just one way of doing it.

My question:

Does this method actually differs (by results) from a SA+TG approach? Or could you state that, along your with in the 50Mhz bandwidth,  its exactly the same?

The result is pretty much the same. Using a wideband noise source instead of a TG is a valid alternative for frequency response measurements, especially with FFT-based SAs like the one in the MDO3000 (which is pretty much just FFT as on any other scope) which don't sweep and therefore can't have a TG. You can't get phase information out of it but  then most TGs aren't phase coherent so they can't be used to measure phase response.

It also works very well with a swept SA, i.e. I use a (borrowed) 18Ghz wideband noise source together with a 13Ghz swept SA for frequency response measurements.

[tekzilla]

Tektronix Application Engineer and author of this video here.

Wuerstchenhund is correct in saying that a wideband noise source with an FFT spectrum analyzer is a valid way of performing a frequency response measurement. After all, when you think about the fundamental nature of a frequency response curve, you're getting the same information from both methods (amplitude response of the device under test across a range of given frequencies). The main difference is that with a TG/Swept SA you're taking one frequency measurement at a time and plotting it on screen, whereas with a wideband noise source/FFT spectrum analyzer, you're measuring all the frequencies produced by your noise source simultaneously. In both cases, the accuracy of the measurement comes down to both the quality of the source and the receiver. If you have a noise source with poor frequency response or a TG with a frequency offset, you'll have to take that into consideration when making the measurements.

One thing to note, the spectrum analyzer in the MDO3000 is not the same as an FFT on a scope. The difference here is that a normal FFT on a scope channel is just using the scope's standard analog input, whereas the MDO3000's spectrum analyzer uses a separate RF signal path and provides a much more user-friendly interface than a scope FFT.

[w2aew]

<snipped>
One thing to note, the spectrum analyzer in the MDO3000 is not the same as an FFT on a scope. The difference here is that a normal FFT on a scope channel is just using the scope's standard analog input, whereas the MDO3000's spectrum analyzer uses a separate RF signal path and provides a much more user-friendly interface than a scope FFT.

To add to this - the MDO's spectrum analyzer features some additional differences compared to a simple FFT on a scope channel.  The SA input does not have to support DC operation, and does not have to include DC offset adjustment capability. These omissions make it easier to optimize the RF path in terms of noise and distortion.  Also, the use of DDC (digital down conversion) helps with the process gain afforded by the discrete transform.  The transform used is a chirp-z transform, which removes the 'power of 2' vector size limitations of an FFT, thus giving the user more flexibility in setting a desired RBW.  Additional processing is employed (using dither) to reduce the distortion products that result from the DNL (differential non linearity) of the DAC.  These contribute to giving the SA similar performance (DANL, SFDR, etc.) to many entry level conventional spectrum analyzers.  So, it really is a bit more that a simple scope channel FFT.

[Wuerstchenhund]

Tektronix Application Engineer and author of this video here.

Nice! Thanks for participating!

After all, when you think about the fundamental nature of a frequency response curve, you're getting the same information from both methods (amplitude response of the device under test across a range of given frequencies). The main difference is that with a TG/Swept SA you're taking one frequency measurement at a time and plotting it on screen, whereas with a wideband noise source/FFT spectrum analyzer, you're measuring all the frequencies produced by your noise source simultaneously. In both cases, the accuracy of the measurement comes down to both the quality of the source and the receiver. If you have a noise source with poor frequency response or a TG with a frequency offset, you'll have to take that into consideration when making the measurements.

You have to leave it to Tek that they made use of the MDO's AWG in such a way. It's just 50Mhz but still useful for many areas.

It would be nice if scopes (OK, not necessarily entry-level) came with a wideband noise source option covering the whole BW of the scope, which would allow to make such simple measurement with the scope. I can't believe the costs would be prohibitive.

One thing to note, the spectrum analyzer in the MDO3000 is not the same as an FFT on a scope. The difference here is that a normal FFT on a scope channel is just using the scope's standard analog input, whereas the MDO3000's spectrum analyzer uses a separate RF signal path and provides a much more user-friendly interface than a scope FFT.

You're right, it's a separate high BW input but at the end of the day that input goes to the same 8bit ADC, using pretty much just plain standard 1Mpts FFT as the FFT math function (until recently I had a MDO3054, and within the scope's 500MHz BW I got exactly the same RF performance in FFT math mode on one of the analog inputs as in SA mode via the N input). And unless you pay for the MDO3SA option the high BW input is limited to the scope's analog BW anyways.

I agree that on the MDO the SA UI is more user-friendly than the FFT math mode, but other scopes have had "SA"-like controls for its FFT mode for ages.

So yes, it really is like FFT on other scopes, unless you pay for the MDO3SA option which only increases the BW

[G0HZU]

It also works very well with a swept SA, i.e. I use a (borrowed) 18Ghz wideband noise source together with a 13Ghz swept SA for frequency response measurements.

Wow, you would need a very powerful 18GHz noise source to be able to make even basic measurements like the measurement in the Tek video.

What make/model noise source is it?

[Jwalling]

Tektronix Application Engineer and author of this video here.

Excellent!  Hope that you contribute on a regular basis. Keysight (Daniel) has been doing so for quite awhile.

[Wuerstchenhund]

It also works very well with a swept SA, i.e. I use a (borrowed) 18Ghz wideband noise source together with a 13Ghz swept SA for frequency response measurements.

Wow, you would need a very powerful 18GHz noise source to be able to make even basic measurements like the measurement in the Tek video.

What make/model noise source is it?

A Noisecom NC3609 with wideband amplifier.

[jjoonathan]

You're right, it's a separate high BW input but at the end of the day that input goes to the same 8bit ADC, using pretty much just plain standard 1Mpts FFT as the FFT math function

You have now had a Rohde&Schwarz engineer, a Tek engineer, and an independent forum-goer who spent time designing wideband SAs in a radio astronomy lab personally explain to you why this  is not the case, yet you keep repeating it.

(until recently I had a MDO3054, and within the scope's 500MHz BW I got exactly the same RF performance in FFT math mode on one of the analog inputs as in SA mode via the N input). And unless you pay for the MDO3SA option the high BW input is limited to the scope's analog BW anyways.

On the other hand *that* is a completely fair criticism.

[Wuerstchenhund]

You're right, it's a separate high BW input but at the end of the day that input goes to the same 8bit ADC, using pretty much just plain standard 1Mpts FFT as the FFT math function

You have now had a Rohde&Schwarz engineer, a Tek engineer, and an independent forum-goer who spent time designing wideband SAs in a radio astronomy lab personally explain to you why this  is not the case, yet you keep repeating it.

   I really have no idea what you're talking about. I'm sure you can provide references where this should have happened, right?

Why I would even ask a R&S engineer how a Tek scope works, I don't know, that would be a pretty idiotic thing to do. Maybe you ask your plumber when there's something wrong with your car, and discuss health issues with your carpenter instead of a healthcare professional, but I doubt a representative of a competitor is the right person to ask about Tek scopes. Oh btw, I think it was actually Dave who mentioned that he got an "explanation" of the MDO3k by a R&S guy, not me.

I also can't remember that a an "independent forum-goer who spent time designing wideband SAs in a radio astronomy lab" ever explained to me how my MDO3054 works (again, please provide some reference). But foremost it's up to you to explain why being an "independent forum-goer who spent time designing wideband SAs in a radio astronomy lab", as admirable as that work is, makes someone an expert in scopes.

What I do remember though is that a certain individual who clearly loves to watch videos about scopes he doesn't own or know goes to quite some length trailing me because for some reason he can't grasp the fact simple that the "SA" part isn't as sophisticated as marketing and wishful thinking suggests.

So instead of you spreading nonsense, how about you prove me wrong for a change? I'll eagerly await your proof showing that the MDO3000 "SA" doesn't use the same 8bit acquisition system as the analog channels and that the "SA" is more than just FFT (even though that's also confirmed by the manual).

 

[David Hess]

It would be nice if scopes (OK, not necessarily entry-level) came with a wideband noise source option covering the whole BW of the scope, which would allow to make such simple measurement with the scope. I can't believe the costs would be prohibitive.

For oscilloscopes which already have an AWG, I would settle for an impulse or even swept output allowing low frequency vector network analysis.

EDN had an article last year contrasting the three methods.

[w2aew]

Tektronix Application Engineer <snip> here.

That makes two of us ;-)  Nice to to see you here, M.A.

[onesixright]

@tekzilla thank you for joining and explaining!

To the others, thank you for going in such technical detail! TBH not sure if i full understand it all. Keep learning as we go :-)

My main question was (there numerous question about a TG with the MD3000 series) if the video (provided by Tek) is a alternative for measuring Frequency response.  I understand you can do a FR by hand (lots of work), you could automate it (takes some time), the TG seems the preferred method.

That said, maybe it would be handy that Tek explains (for the less experienced), that this method is a valid alternative for a TG (but with BW limitations)?

Once again, thank you all!

[G0HZU]

It would be nice if scopes (OK, not necessarily entry-level) came with a wideband noise source option covering the whole BW of the scope, which would allow to make such simple measurement with the scope. I can't believe the costs would be prohibitive.

Maybe, but I think that would rely on the user being very aware of the limitations of this approach. The source itself would need a fairly beefy output stage to cope with the task of generating a useful spectral density over such a wide bandwidth. It would also be easy to fall foul of overload or alias issues in the scope and this could cause some very confusing/false results depending of the filter type being tested. Things get even riskier if the device being tested is active like an amplifier. I have a few high power noise sources here but I rarely use them for measuring filters with a scope. I usually control the bandwidth of the noise source very carefully if I do.

[snoopy]

You're right, it's a separate high BW input but at the end of the day that input goes to the same 8bit ADC, using pretty much just plain standard 1Mpts FFT as the FFT math function

You have now had a Rohde&Schwarz engineer, a Tek engineer, and an independent forum-goer who spent time designing wideband SAs in a radio astronomy lab personally explain to you why this  is not the case, yet you keep repeating it.

   I really have no idea what you're talking about. I'm sure you can provide references where this should have happened, right?

Why I would even ask a R&S engineer how a Tek scope works, I don't know, that would be a pretty idiotic thing to do. Maybe you ask your plumber when there's something wrong with your car, and discuss health issues with your carpenter instead of a healthcare professional, but I doubt a representative of a competitor is the right person to ask about Tek scopes. Oh btw, I think it was actually Dave who mentioned that he got an "explanation" of the MDO3k by a R&S guy, not me.

Youtube is littered with examples of product comparisons where all of a sudden the competing vendor knows all about how the oppositions test equipment works when it suits them !! Funny about that

[Wuerstchenhund]

It would be nice if scopes (OK, not necessarily entry-level) came with a wideband noise source option covering the whole BW of the scope, which would allow to make such simple measurement with the scope. I can't believe the costs would be prohibitive.

Maybe, but I think that would rely on the user being very aware of the limitations of this approach. The source itself would need a fairly beefy output stage to cope with the task of generating a useful spectral density over such a wide bandwidth. It would also be easy to fall foul of overload or alias issues in the scope and this could cause some very confusing/false results depending of the filter type being tested. Things get even riskier if the device being tested is active like an amplifier. I have a few high power noise sources here but I rarely use them for measuring filters with a scope. I usually control the bandwidth of the noise source very carefully if I do.

I agree that using a scope for determining the frequency response of a device is far from perfect, and usually a scope wouldn't be the first instrument one would use for something like that, but for situations where one could live with the inherent limitations the scope approach may well be good enough.

I guess a wideband (with scope BW) noise source could be offered as option for a reasonable price for scopes with low (<1Ghz) and moderate (<5GHz) BW scopes. A better option would be a sweeping RF generator but I assume that would be cost-prohibitive, especially in the low BW class.

[Wuerstchenhund]

Youtube is littered with examples of product comparisons where all of a sudden the competing vendor knows all about how the oppositions test equipment works when it suits them !! Funny about that

Indeed. The first thing that comes to mind is why are they talking so much about their competitors instead about their own product, which suggests a certain fear.

'Ours-vs-others' comparisons need to stop. Manufacturers probably don't realize how pathetic they come across as.

[snoopy]

Youtube is littered with examples of product comparisons where all of a sudden the competing vendor knows all about how the oppositions test equipment works when it suits them !! Funny about that

Indeed. The first thing that comes to mind is why are they talking so much about their competitors instead about their own product, which suggests a certain fear.

'Ours-vs-others' comparisons need to stop. Manufacturers probably don't realize how pathetic they come across as.

It's like the US election at the moment. Getting dirty

cheers

[jjoonathan]

You're right, it's a separate high BW input but at the end of the day that input goes to the same 8bit ADC, using pretty much just plain standard 1Mpts FFT as the FFT math function

You have now had a Rohde&Schwarz engineer, a Tek engineer, and an independent forum-goer who spent time designing wideband SAs in a radio astronomy lab personally explain to you why this  is not the case, yet you keep repeating it.

   I really have no idea what you're talking about. I'm sure you can provide references where this should have happened, right?

The Tek engineer was w2aew, just now. As for R&S, you're right, it was Dave's anecdote, my bad. Times two, because it looks like I made the misattribution in the other thread as well  . In any case, the claim that R&S made that I expanded on was that Mpts should not be used as a figure of merit on modern FFT implementations. Using it to describe a limitation is like looking at a racecar and saying "pff, it's a four-wheel car." The categorization misses the point even if three wheels is a dumb idea and some applications demand six. My real objection was that every time I hear the drivers talking about acceleration and braking distance I shortly thereafter hear "pff, four wheels" and that's a too.

It's also possible that I'm simultaneously missing your point. As you delicately note I am not currently a consumer of mid to high end scopes. I'd like to ask a question about these standard FFT math functions with SA controls. On a non-MDO 5GSa/s scope with 42MPt of memory and SA-like FFT configured to a baseband 1 GHz span (or 1MHz span starting at 999MHz) with RBW=SPAN/1000, how many independent "frames" of spectrum (gate intervals do not overlap) fit into the memory? 42000? Fewer? None at all? And what does the narrower span buy you in comparison to the wider span?

But foremost it's up to you to explain why being an "independent forum-goer who spent time designing wideband SAs in a radio astronomy lab", as admirable as that work is, makes someone an expert in scopes.

It shouldn't come as a surprise that wideband SA architectures have similar implementations in both radio telescopes and MDO oscilloscopes: 8 bit N GS/s ADCs going into FPGA/ASICs implementing a "FFT+" with extra attention paid to spurs, linearity, and phase noise. I'm far from an oscilloscope expert, but I spent enough time playing with the very type of SA implementation that Tek is using -- in matlab, simulink, paper, and FPGAs -- that when I hear "it's just an 8-bit 1MPt FFT" I get a highly unfortunate case of someone-is-wrong-on-the-internet disease

[Wuerstchenhund]

In any case, the claim that R&S made that I expanded on was that Mpts should not be used as a figure of merit on modern FFT implementations. Using it to describe a limitation is like looking at a racecar and saying "pff, it's a four-wheel car." The categorization misses the point even if three wheels is a dumb idea and some applications demand six. My real objection was that every time I hear the drivers talking about acceleration and braking distance I shortly thereafter hear "pff, four wheels" and that's a too.

I get your point, however what you ignore is that on a DSO, the memory depth in FFT mode is actually relevant because, together with the sample rate, it determines the frequency resolution we get. 

For FFTs, resolution bandwidth is equal to the sample rate divided by the number of points in the FFT. That means to get a more narrow RBW the scope must acquire and process more samples.

I'm away from home for the next two weeks otherwise I'd have posted some demo screenshots.

It's also possible that I'm simultaneously missing your point.

I guess so. You're arguing from a mathematical POV which is not wrong, and I am well aware that you can perform a DFT in different ways, some of where memory requirements don't play a large role as FFT as the way it's done on a scope, but since scopes don't work that way this isn't very relevant to the topic at hand.

It shouldn't come as a surprise that wideband SA architectures have similar implementations in both radio telescopes and MDO oscilloscopes: 8 bit N GS/s ADCs going into FPGA/ASICs implementing a "FFT+" with extra attention paid to spurs, linearity, and phase noise.

And still, neither is any better as with FFT on a decent scope (and phase noise of the "SA" in my MDO3k was even slightly worse than my 3GHz scope at similar settings).

I'm far from an oscilloscope expert, but I spent enough time playing with the very type of SA implementation that Tek is using -- in matlab, simulink, paper, and FPGAs -- that when I hear "it's just an 8-bit 1MPt FFT" I get a highly unfortunate case of someone-is-wrong-on-the-internet disease

You said you spent enough time playing with the very type of SA implementation that Tek is using, but how do you know what principle Tek uses in the MDO3000 SA?

You continue to claim that the MDO3k's "SA" mode works differently than scope's FFT, but so far you have not supported that claim with any facts or evidence.

So please, prove me wrong. Show me how Tek has managed to cram two completely different FFT implementations into the MDO3k which still perform exactly the same (BW aside, as this is limited by the inputs)?

[w2aew]

The Tek engineer was w2aew

w2aew *is* still a Tektronix Apps engineer.  tekzilla is one of my colleagues.

You said you spent enough time playing with the very type of SA implementation that Tek is using, but how do you know what principle Tek uses in the MDO3000 SA?

You continue to claim that the MDO3k's "SA" mode works differently than scope's FFT, but so far you have not supported that claim with any facts or evidence.

So please, prove me wrong. Show me how Tek has managed to cram two completely different FFT implementations into the MDO3k which still perform exactly the same (BW aside, as this is limited by the inputs)?

There are a couple of differences between a scope channel FFT, and the implementation in the MDO3000 and MDO4000 SA:
 
1) The SA doesn't have to pass DC, and doesn't have to provide an adjustable DC offset like the scope channels do, so the design is optimized for RF signals (lower distortion, lower noise).

2) The RF input is always sampled at 10GS/s.  The scope channel FFT uses the sample rate of the waveform (determined by record length and horizontal timebase setting).  The oversampling on the SA input improves the process gain.  The samples from the 10GS/s ADC go through a digital down-converter (DDC) to baseband, including filtering and decimation determined by the SA settings (primarily Span).  The baseband is in the form of quadrature (IQ) data.  The oversampling, DDC and filtering provide process gain which lowers the noise floor.

3) The actual transform used is not an FFT.  It uses a Chirp-Z transform. The CZT removes the "power of 2" limitation on the FFT vector size.  This gives the flexibility of a fully user-selectable RBW, not bound by the ^2 array size.

4) Out of band dithering is employed to reduce the distortion from the DNL (differential non-linearity) of the ADC.  This improves the input-related harmonic distortion performance.   

Yes, there are definitely similarities to a scope channel FFT, but there are definite and deliberate differences implemented to improved the RF performance beyond a basic scope channel FFT.

[Keysight DanielBogdanoff]

A scope can definitely capture frequency response in a valid manner. The measurements may not be exactly the same as an FRA or VNA due to differences in calibrations, components, accuracy of the generator, etc., but you could theoretically get identical results.

As a side note, we actually do a sweep on the InfiniiVision 3000T and 4000 X-Series scopes very similar to what a standard FRA or VNA would do. We built it for power supply testing - namely control loops response and PSRR. We plot both gain and phase up to 20 MHz (the bw limit of our "WaveGen"). Johnnie Hancock did a video on it recently, too: http://bit.ly/2dfGaLY

[Someone]

A scope can definitely capture frequency response in a valid manner. The measurements may not be exactly the same as an FRA or VNA due to differences in calibrations, components, accuracy of the generator, etc., but you could theoretically get identical results.

As a side note, we actually do a sweep on the InfiniiVision 3000T and 4000 X-Series scopes very similar to what a standard FRA or VNA would do. We built it for power supply testing - namely control loops response and PSRR. We plot both gain and phase up to 20 MHz (the bw limit of our "WaveGen"). Johnnie Hancock did a video on it recently, too: http://bit.ly/2dfGaLY

Is the phase measurement going to be implemented on the other scopes in the line? The 3000A only has amplitude at the moment.

[Keysight DanielBogdanoff]

There unfortunately aren't plans to do more Bode enhancements on the "A" model at this point.

[TheSteve]

There unfortunately aren't plans to do more Bode enhancements on the "A" model at this point.

BOO