Rigol DHO804 Test and Compare Thread

[2N3055]

Unless you do a piecewise FFT of partial segments of the time series, or preprocess the time series by averaging etc., of course.

Exactly like that....

[rf-loop]

Just for satisfy some peoples curiosity, not compete with any,. Here one poor sample using SDS1204XHD (at this time it is China only model)
Around same kind of signals but need run low level LSB peak instead of USB  and 10MHz for keep generator high peak level around 20Vpp (because >10MHz can not this level) (Looks like this is bit too challenging for generator what I use  and it somehow produce also USB peak etc...  )

FFT F1  normal, and Hanning
FFT F2  Max Hold, Hanning
FFT F3  Average 16, Hanning
Eta: do not care Trigger setting in image. it is Auto and other things there do not matter anything here.

[Performa01]

Average acquisition mode and/or math function are something completely different than averaging of an FFT trace.

Full featured DSOs have Average as
1.   Acquisition mode, yet with reduced max. memory depth
2.   Math function
3.   Substitute for VBW setting of an SA for the FFT traces.

But if a Math function 2. were available which averages over successive sweeps, couldn't it be be applied to the FFT to obtain 3.?

I’m not sure if I understand correctly what you mean. Of course there is a difference whether we average the records, i.e. the waveform in the time domain, which are the basis for the FFT, or the converted math result (FFT) which is now in the frequency domain.

If you mean an advanced scope with math on math, where you want to apply averaging on the FFT result, either by two math traces where one acts as input for the other, or defined in a formula in a single channel. I’ve never tried it, but would not be surprised if FFT (which delivers a frequency domain result after all) would be about the only math function that cannot act as a source for other (time domain) functions.

Since Averaging is included in the FFT package anyway, why bother.

As stated in my questions to TurboTom, a few posts above, I have been wondering that too. If the only flaw in the displayed "RBW" is that it is not actually an RBW but a frequency step, I can live with that (although Rigol should fix it, of course, and add a unit while they are at it). But I did see some screenshots or videos earlier where unexpected "RBW" values were displayed, which did not seem to make sense even as a frequency step and assuming a Hz unit. Maybe Tom or Martin can shed some light onto whether and when this happens?

Yes, you can live with it – once you know what it actually means. It’s annoying nevertheless – and up to now, you only know what I have found out by accident. Rght now you’re expressing doubts what this dimensionless number actually is. I can only tell what I see in a few screenshots, so others will have to verify it.

Is there something in the FFT which you would like to adjust but can't?

Number of points for example, I'm used to being able to manually set the number of memory points (which are also not displayed on the rigol).

I thought that's set via the acquisition memory setting (up to 1 MPts)? Is the issue that it's not in the FFT dialog, or that the available steps are too coarse? Or does that setting not actually set the number of FFT points at all?

Yes, this is one possibility, but can’t you imagine situations, where you have to capture long records, exceeding 1 Mpts, yet want a shorter FFT, maybe to get a wider RBW?

1. FFT mode : Normal, Average, Max Hold (at least)
2. Number of points/bins
3. RBW cannot be set. It is displayed in window title (which by the way is not visible if you do multiple windows, negating the benefit) and is some kind of  function of timease. You twiddle timebase left and right until you like how it looks.

1. I think we all agree that lacking Average mode is an annoying omission, as discussed separately above. I don't know about the relevance of Max Hold mode, but that's probably due to my lack of knowledge about some applications (regulatory/EMC maybe?).

2. As discussed, it's set via the acquisition memory size in my understanding. That way of doing it may be lacking something; see my questions to Martin just above.

3. Now I am confused. If you choose the window function, record length and sampling rate -- how can you set the RBW independently?

Ad 2: see above.

Ad 3: Max Hold is at least as important as Averaging. It allows the scope to do frequency response plots. Other then Bode Plot, this can use the full bandwidth up to half the sample rate and it can plot the frequency response of the scope itself. The attached screenshot shows an example of a measurement graph demonstrating the frequency response of an SDS2000X Plus in 10 bit mode, that could not be produced otherwise.


SDS2354X Plus_FR_BFull_1GSa_500MHz_10bit

[TurboTom]

RBW value that gets displayed on DHO1000 FFT screen is 3dB reolution bandwidth in units kHz. To me it appears that on the DHO800 screen, the RBW value is in units Hz but I may wrong.

[rf-loop]

RBW value that gets displayed on DHO1000 FFT screen is 3dB reolution bandwidth in units kHz. To me it appears that on the DHO800 screen, the RBW value is in units Hz but I may wrong.

In DHO800 image by @Martin72 (Reply #956)  it looks that it is simply sample frequency / FFT length.  (62500000/625000=100)
And then there read just plain RBW:100   and also then same RBW even for Hanning and Flattop..  They did not use  window factor but print only Bin interval as RBW. RBW we normally think as Resolution Band Width.
Of course it can say it is RBW_Bins  but more commonly with spectrum analyzers etc we use RBW_3dB  and for some purposes RBW_6dB (example EMI measurements). But tightly speaking it is not absolute error to tell RBW_Bin but why then not tell it using example  Δf as example Siglent use what do not leave anything unclear. Also then can tell "normal" RBW also and its width depending used window. As can see example in my last image.

Well, if someone names it RBW, shame on him (Rigol) and even without a unit... well, what about the units of measurement... this is just a measuring device where guessing is part of the game. (sarc.)

[gf]

Of course there is a difference whether we average the records, i.e. the waveform in the time domain, which are the basis for the FFT, or the converted math result (FFT) which is now in the frequency domain.

I'm just wondering: When does it make sense to average in the time domain instead of averaging the spectral power of the FFT bins?
I would always prefer the latter.
The former suffers from trigger jitter due to noise, which slightly misaligns the traces to be averaged.
OTOH, the sampling phase (or IOW the trigger point of the traces) does not matter at all for spectral power averaging.

The attached screenshot shows an example of a measurement graph demonstrating the frequency response of an SDS2000X Plus in 10 bit mode, that could not be produced otherwise.


SDS2354X Plus_FR_BFull_1GSa_500MHz_10bit

What is the blue "Ref" trace?

In DHO800 image by @Martin72 (Reply #956)  it looks that it is simply sample frequency / FFT length.  (62500000/625000=100)

I agree with that. It is obviously just the FFT bin spacing, not the -3dB bandwidth of the corresponding spectrum analysis filter.

[ebastler]

If you mean an advanced scope with math on math, where you want to apply averaging on the FFT result, either by two math traces where one acts as input for the other, or defined in a formula in a single channel. I’ve never tried it, but would not be surprised if FFT (which delivers a frequency domain result after all) would be about the only math function that cannot act as a source for other (time domain) functions.

Since Averaging is included in the FFT package anyway, why bother.

Yes, that's what I meant. Agree, if there is a dedicated averaging mode in the FFT, nothing else is needed. I just wanted to confirm my understanding what Averaging FFT mode does, and whether a secondary math averaging function is equivalent (if it can be applied to the FFT result, of course). Thanks!

[...] can’t you imagine situations, where you have to capture long records, exceeding 1 Mpts, yet want a shorter FFT, maybe to get a wider RBW?

But what would the FFT operation do then? Split the long record into multiple segments, Fourier-transform them individually, and combine them via Averaging or Max Hold? If so, isn't that equivalent to setting a shorter record length in the first place and keeping the operation running?

(Honest question; I might be missing something.)

Ad 3: Max Hold is at least as important as Averaging. It allows the scope to do frequency response plots. Other then Bode Plot, this can use the full bandwidth up to half the sample rate and it can plot the frequency response of the scope itself. The attached screenshot shows an example of a measurement graph demonstrating the frequency response of an SDS2000X Plus in 10 bit mode, that could not be produced otherwise.

OK, thanks. So Max Hold would be used in scenarios where you sweep the input signal, as 2N3055 also suggested? It's a bit counter-intuitive to me that you would always do a full FFT during such a scan, rather than a Bode-plot style detection of the known fundamental frequency. If a system has strong distortion, say due to clipping, at a low input frequency, that might appear as an apparent response at a higher harmonic?

Anyway -- if it's a feature that is commonly available, it seems like an omission on Rigol's part not to implement it. Like averaging mode, it should be relatively low-hanging fruit.

[Performa01]

I'm just wondering: When does it make sense to average in the time domain instead of averaging the spectral power of the FFT bins?
I would always prefer the latter.
The former suffers from trigger jitter due to noise, which slightly misaligns the traces to be averaged.
OTOH, the sampling phase (or IOW the trigger point of the traces) does not matter at all for spectral power averaging.

Of course it makes no sense to average the acquisitions before an FFT is applied. It kills all modulation and other dynamic effects.

Yet I think I need not stress how important Average acquisition mode and/or math function can be when we do not intend to do the FFT processing. This mode even allows the scope to act as an autocorrelator to clearly display weakest signals deeply buried in noise, if only we have a synchronous reference signal for the trigger available.

And no, trigger jitter absolutely isn’t a concern on these machines, thanks to the incorporation of a fully digital trigger system with proper trigger point interpolator. It is measured in the picoseconds...

What is the blue "Ref" trace?

That is the frequency response in the 2nd Nyquist zone, i.e. from 1 GHz – 500 MHz 😉

[dmulligan]

The discussion about FFT has been very interesting and made me fire up my DHO804 to see what it can do.  I was disappointed when I couldn't select any of the math operations as a source.  I checked and according to the manual FFT on the DHO800 only works with one of the 4 channels as the source.  It doesn't work from either a math or reference source.

[Performa01]

[...] can’t you imagine situations, where you have to capture long records, exceeding 1 Mpts, yet want a shorter FFT, maybe to get a wider RBW?

But what would the FFT operation do then? Split the long record into multiple segments, Fourier-transform them individually, and combine them via Averaging or Max Hold? If so, isn't that equivalent to setting a shorter record length in the first place and keeping the operation running?

(Honest question; I might be missing something.)

Don’t make it more complicated than it is. We can of course just select the record length so that it makes for the desired FFT length.
We can also take the freedom to select the record length using one criterion and a different criterion for the FFT length. What’s wrong with that?

Of course the record length cannot be shorter than the FFT length. But it could be way longer.
As to the question what happens then – the same what happens whenever we capture e.g. 10 Mpts but have only 1 Mpts FFT available. The data gets decimated before FFT processing.

OK, thanks. So Max Hold would be used in scenarios where you sweep the input signal, as 2N3055 also suggested? It's a bit counter-intuitive to me that you would always do a full FFT during such a scan, rather than a Bode-plot style detection of the known fundamental frequency. If a system has strong distortion, say due to clipping, at a low input frequency, that might appear as an apparent response at a higher harmonic?

Just because we can always find a scenario where something might not work well, this doesn’t make a certain feature less valuable. In my example it worked just perfect.

Of course we don’t have a frequency synchronous measurement and yes, because of the very high dynamic range of the FFT we can see the harmonics as false signals even from sources with low distortion – but these spurious signals get overwritten in most wideband measurements anyway – except for the cases where we want to measure the stop band of a LP filter of all things.

If we stick to a decent signal source (AWG or signal generator) the harmonic level shouldn’t be much of a problem.  And if it still is, we can do a separate scan for the stop band alone - there are so many ways to solve a problem. Finding solutions is the main job of engineers, by the way. But for this, they need reliable tools and the knowledge how to use them properly – which includes a clear judgement what is going to work and what is not.

[Martin72]

Hi,
I just wanted to go through various time bases and capture the displayed values in the FFT window via screenshot, once with automemory, once with max memory.
But there are already inconsistencies at the beginning...
I started with 2µs/div., was over at 10µs, went back to 10µs/div.
And I have 2 different resolutions (RBW)
How does that work...
Oh yes, the unit Hz is not displayed at any time.
Instead, there are decimal numbers such as 19.99k, and a sample rate of 1.26GSa/s was also displayed...
I'll continue, but here are two pictures with the different RBW with the same time base

[Performa01]

The discussion about FFT has been very interesting and made me fire up my DHO804 to see what it can do.  I was disappointed when I couldn't select any of the math operations as a source.  I checked and according to the manual FFT on the DHO800 only works with one of the 4 channels as the source.  It doesn't work from either a math or reference source.

It cannot do an FFT on reference traces, because these are heavily decimated (maybe even screen) data. You would need so called Memory Traces for this, which are similar to Reference traces, but with the the original data, hence original sample rate.

It's similar for math traces. Math channels as a source for math (aka math on math) is an advanced feature, and makes mostly sense in scopes that work on undecimated data for the math.

[gf]

Wouldn't worry too much about this since the most relevant window functions von Hann, Blackman-Harris and rectangular are available and appear to work correctly.

I would. A flattop window is important to get good amplitude accuracy for frequencies that are not integer multiples of the bin spacing. The maximum amplitude error of the flattop window ¹⁾ is only 0.016 dB. Hann, Blackman and various others still have significant scalopping loss for frequencies that fall in the middle between two bins. Blackman-Harris has also ~0.83 dB. And a rectangular window leaks as hell if the frequency is not an integer multiple of the bin spacing.

¹⁾ I consider the Matlab variant here, but there exist several other variants too

[ebastler]

Don’t make it more complicated than it is. We can of course just select the record length so that it makes for the desired FFT length.
We can also take the freedom to select the record length using one criterion and a different criterion for the FFT length. What’s wrong with that?

Nothing wrong with that, of course. But on the other hand, I can't see where the serious omissions in adjustable parameters are in Rigol's FFT implementation. That's what I was trying to understand, after hearing that complaint many times from the Siglent crew.

It seems to me that you can adjust all relevant parameters on the Rigol: Record length and sampling interval (and hence FFT frequency step) via the scope's regular acquisition and horizontal controls; center frequency and span via the FFT dialog (and the touch screen too).

I get the impression that there is mainly a difference in approach here: Rigol sees this as a regular time series acquisition, with its standard controls, followed by a specific math operation -- while Siglent pulls all parameters together in an FFT-specific menu, to make FFT mode behave similar to a spectrum analyzer.

Having said that -- yes, Rigol's displayed "RBW" is not an RBW and lacks a unit, and Rigol really should implement Averaging and Max Hold modes! I hope they do that sooner rather than later...

[Martin72]

I hope they do that sooner rather than later...

Just had a look, even the MSO8000A does not offer more than what the DHO800 offers in terms of FFT.
But at least the unit is given with RBW, although I know this from the MSO5000 as well.
Apart from that, if you can't set the number of points, it would be great if the rigol could at least show how many points are currently available.
"Max. 1Mpt" (from the manual)doesn't say anything at first.

Apropos, It's been a long week, it's evening, you can have a bit of a mental block...
Like me right now, operating the rigol from the living room.
Span= half of the current samplerate
RBW= 1/capturetime
If the timebase ist set to say 50µs/Div, capturetime should be 500µs, so RBW should be 1/500µs  = 2000Hz.
With the Rigol, I have an RBW of 5000(Hz) precisely under these parameters.
What else does the RBW depend on?
And:
Span is currently 625Mhz, which makes sense if the sample rate is 1.25GSa/s.
I know from siglentscopes that the FFT sample rate is equal to the sample rate of the time domain if the number of FFT points is greater than the number of points of the current memory points.
Since the FFT sample rate is the same, I assume that rigol "has" more than 625kpts for the FFT (unfortunately it is not displayed).
If I now set the memory from auto to manual and let's say 10Mpts, the FFT sample rate should go down.
But it doesn't, Instead, the RBW is now reduced to 2000Hz - i.e. to the value actually to be expected at 50µs/div.
And that's where my mental block is, I must be missing something.

Edit: Forget it, the rigol drives me crazy...
While typing this post, the rigol shows 5K RBW until I´ve set the memory 10 Mpt as mentioned above.
So I make a pic from it, then I want to make a pic from the state before, switch back to Automemory and 625kpts - RBW remains 2K...
And it´s getting better:
OK, let´s change the timebase and go back to 50µs/Div.
Now I have 1K RBW....

[Martin72]

OK, it´s a bug...
When changing the timebase, the scope "forget" (often but not everytime) to refresh the RBW value.
If you then move the position (vertical offset) of the FFT, the RBW value is updated.

[Mechatrommer]

Apart from that, if you can't set the number of points, it would be great if the rigol could at least show how many points are currently available.

istr when i ordered my dho804 few weeks ago, someone keep complaining about dho800/900, how 900 (same HW as 800) can easily aliased (by struggling hard enough to show gibbs effect phenomenon as aliasing etc)  how bad things including its fft, so i let him find the formula by himself. usually if people think he is clever i'll let him find the formula himself... not sure if he already figure that out... the formula is very simple..

number of FFT point (bins you saw plotted on screen) = FFT frequency range / rbw
FFT frequency range = DC - sampling rate / 2.

both frequency range and rbw can be seen on screen..
working example: sampling rate: 1.25GSps, hence FFT range = DC - 625MHz (sampling rate / 2), RBW (shown) = 2K (Hz).... so number of FFT bins = 625MHz / 2KHz = 312500 points (bins).

since FFT only use/show half of calculated points (another half is only mirror image), number of sampled points used to generate FFT is twice bins number. you are right what you have figured, dho800 has some bug on showing RBW, i reported in bug thread but not sure if people understood when i reported it.. i figure when you increase time/div and then back down/ reduce it again, it will show correct rbw, but not when you increase time/div.

so in conclusion, we can indirectly change how many FFT or sampled points used by dho800 by changing time/div and use formula above to figure how many points currently used for FFT, ymmv.
and btw, if you have traditional real/sweeping SA, you'll know you only have freq range and RBW figure, you wont have sampled/fft points count anywhere there is no such thing, so rigol is being traditious ymmv.

[Performa01]

When trying to sound smart, people should at least post correct information.

That includes not using points and bins as synonym first, and making contradicting statements later by (accidentally?) stating something that sounds right, namely that FFT bins are half the FFT points.

One should also not confuse frequency step (df) with resolution bandwidth (RBW).

One should also understand the vast difference between a traditional swept SA and the FFT on a DSO – or even a combination of both to form a powerful realtime SA…

When in doubt, one can always look up the fundamentals in this old post (reply #23):

https://www.eevblog.com/forum/testgear/rohde-schwarz-rtb2002-vs-siglent-sds2104x-plus/msg3239832/#msg3239832

[gf]

number of FFT point (bins you saw plotted on screen)

In common terminology, the "number of FFT points" usually refers to the number of points in the time domain from which the FFT is computed, not to the subset of frequency bins which happen to be displayed.

[gf]

When I see screenshots showing 62.5MSa/s, 625000 points, and RBW 100 (assuming that 100Hz is not really the -3dB RBW, but actually the frequency bin spacing), then I get the impression that the FFT size is the full record length of 625000 points, which would also imply that this FFT implementation can handle sizes that are not a power of two. Of course, there has to be a limit to the maximum FFT size, so the question remains what happens if the acquired records are longer than the limit. What subset of samples is selected? Start of the record? Center? Or are long records even split into multiple chunks of the maximum FFT size?

If we distrust the numbers displayed on the screen, ist it possible to export the FFT math trace to a text file? If yes, how many points does it export if full span (0...fs/2) is selected? The number of saved points and the difference between adjacent frequencies should give an indication for the actual FFT size and frequency bin spacing (unless only decimated screen data are saved - that's useless, of course).

[Martin72]

You wouldn't be so much in the dark if there was something like this - and it can't be difficult to implement.
But so far I haven't seen a Rigol that does it, neither expensive nor cheap.

[Mechatrommer]

When trying to sound smart, people should at least post correct information.

That includes not using points and bins as synonym first, and making contradicting statements later by (accidentally?) stating something that sounds right, namely that FFT bins are half the FFT points.

One should also not confuse frequency step (df) with resolution bandwidth (RBW).

One should also understand the vast difference between a traditional swept SA and the FFT on a DSO – or even a combination of both to form a powerful realtime SA…

When in doubt, one can always look up the fundamentals in this old post (reply #23):

https://www.eevblog.com/forum/testgear/rohde-schwarz-rtb2002-vs-siglent-sds2104x-plus/msg3239832/#msg3239832

i may mixed up / confusing what a specific DSO brand refering to "FFT points" (sampled points? or bin points? i dont own "THAT brand" DSO) but... what a waste lengthy talk about -3dB points of FFT Windowing function that looks like smart but in fact a "more than necessary" a complicated misinformation... better refer to "known brand"'s definition, than one guys' own definition from an unknown location cheers...

https://www.analog.com/en/design-center/glossary/frequency_bin.html
https://dsp.stackexchange.com/questions/26927/what-is-a-frequency-bin
https://www.ap.com/technical-library/more-about-ffts/
https://www.ni.com/docs/en-US/bundle/ni-rfsa/page/nirfsa/resolution-bandwidth.html

there is NONE talking about WINDOWING function relating to this...

https://googlethatforyou.com?q=fft%20bins
https://googlethatforyou.com?q=resolution%20bandwidth%20fft

[gf]

i may mixed up / confusing what a specific DSO brand refering to "FFT points" (sampled points? or bin points? i dont own "THAT brand" DSO) but... what a waste lengthy talk about -3dB points of FFT Windowing function that looks like smart but in fact a "more than necessary" a complicated misinformation... better refer to "known brand"'s definition, than one guys' own definition from an unknown location

FFT and window functions are pure math - nothing brand spacific. The term "RBW" comes from spectrum analyzers, and there it means the -3dB bandwidth of the spectrum analysis filter. Anyone familar with spectrum analyzers associates RBW with this bandwidth.

[Martin72]

If we distrust the numbers displayed on the screen, ist it possible to export the FFT math trace to a text file?

Directly from the FFT menu not, only the peak/marker table.
But I'll try another thing, maybe that works.

[Mechatrommer]

You wouldn't be so much in the dark if there was something like this - and it can't be difficult to implement.
But so far I haven't seen a Rigol that does it, neither expensive nor cheap.

if you are depending on that figure, then good for you, you dont have to know formula... (my program also provides such figure) i know rigol is a bit behind in this, but now its catching up... "the traditional way", which somehow bring back nostalgic emotion.. but please dont argue smart like some people.. only providing own's definition without being backed up by industrial name's definition. cheers.