LeCroy 3034 - has anyone got the FFT to do anything?

[Mechatrommer]

I also could not see why you need so many cycles on the screen. I wrote an FFT program at univ, 1970s, Fortran of course and it needed just ONE cycle, and you got the spectrum from that. That is how an FFT is supposed to work, and the ability to do that enables a spectrum of a single shot event to be obtained, too.

Can any of these do an FFT on just one cycle of the waveform?

with such query, backed up by proof that you tried to observe 500Hz's FFT @ 2GS/s, added with missed formulation recap and important hints given earlier by members, i have a high confidence that you've lost your grasp totally on FFT algorithm...

Can any of these do an FFT on just one cycle of the waveform?

That is not how FFT works..... 

lol, thats exactly how FFT works.

yes thats how it works, but it will produce garbage of confusing readings...

[gf]

Can any of these do an FFT on just one cycle of the waveform?

This is possible if your (periodic) signal happens to have the "right" freqency. Or more specifically, if your signal's frequency is fs / N, where fs is the selected¹⁾ sampling rate and N is the number of selected FFT points (usually a power of 2).
Example: If fs=100 MSa/s and N=524288, and your signal freqency happens to be 190.73486328125 Hz, then the FFT will be calculated from exactly one cycle of the waveform.

What you are asking may be nice for a school example. But when are these border conditions ever granted in practice?
Another point is: You do not really want to get a frequency resolution (bin spacing) as coarse as the signal frequency, but you likely want a (significantly) higher resolution. But the former is what you get when you calculate the FFT from a single cycle only.

And what the time domain display shows on the screen is still yet another issue, since the selected FFT size does not necessarily match the selected memory depth, but the FFT size is possibly just a small subset of the captured samples, while the displayed time domain trace is can be a different subset of the captured samples.

Ediit: ¹⁾ On many scopes you cannot select the sampling rate directly, but only indirectly via memory depth, #channels, timebase, and maybe other parameters.

[gf]

FFT also uses windowing

Can/may/might/could/possibly use windowing depending on the application. Windowing is not inherent to FFT.

Well this may be a matter of the point of view²⁾. If we consider that FT and DTFT are actually defined for an infinte time line, then the process of picking-out a finite number of adjacent samples from the infinite time line, in order to apply the DFT only to the samples in this time window, is already "windowing". The relative weights of the points in the window is then defined by a window function. If the weights are equal, then we are dealing with a rectangular window, but even a rectangular window is still yet another window function, still leading to a very particular¹⁾ frequency response of the spectrum analysis filter that corresponds to this window function. It just has been adopted to use "no window" as synonym for a rectangular window function.

¹⁾ Btw, the frequency response of a rectangular window is of course lousy and leads to much spectral leakage. For a spectrum analyzer like display, rather a window functions with very low side lobes (in the frequency domain) is desired, which concentrates most of the energy in the main lobe, and particularly this is desired in conjunction with a large number of FFT points which lead to a high processing gain and a low noise floor.

Edit: ²⁾ It is of course not inherent to the FFT per se (which is just a mathematical algorithm), but it basically applies to the context in which the FFT is used here.

[peter-h]

Obviously I don't know how to use this scope properly, which is why I posted here (it's not because I wanted people to tell me to read the manual, which I have done ) but the discussion shows that it is not trivial, even for those who do understand it.

What I don't get is why somebody in LC Marketing didn't suggest copying e.g. Tek and initialising the Math/FFT function with some useful defaults, so it actually does something useful. It is clearly possible to do, but despite my Q further back nobody has said how to work out the optimum settings.

[gf]

Obviously I don't know how to use this scope properly

My feeling is rather that you are possibly struggling to determine your desired FFT parameters (samplig rate, number of FFT points, desired window function,...), which lead to the result you expect to see.  If you don't feel able to determine these parameters, because you don't understand their impact on the result and the relation between them, then you rather need to read-up a bit about FFT/DSP topics (one can actually spend weeks or months on these topics, when diving deeply into them). But this is rather not lack of knowledge "how to use this scope properly" then.

Once you know which FFT paramers you really want, I'm pretty sure you will find a way to enter them into the scope somehow, even if the sampling rate can only be selected indirectly, as a function of the selected timebase, memory depth, etc.
And if your needs exceed the FFT capabilities of the scope (say you would need 10M points), then you still had the option to download the captured samples, and to do any desired FFT processing yourself offline on the PC, e.g. in Matlab, Octave, etc.

but despite my Q further back nobody has said how to work out the optimum settings.

Please define your criteria for "the optimum". A global optimum likely does not exist, but you'll rather need to make individual compromises between different criteria, for different use cases. What exacly are you after? What exactly do you want to see or measure, and at which accuracy, resolution, etc.? Without detailed specification of the objectives it is hard to optimize. If your aim is limited to "just some spectrum display of my signal", then you already did hit this aim.

[peter-h]

Let me put it differently: how come, on the Tek scope, you just press Math then FFT and you see a spectrum which looks about right and which is a reasonable starting point.

And they don't need the timebase shrunk so it is displaying hundreds of cycles of the waveform, since nobody uses a scope like that (it is completely useless).

So there must be some simple way to set up the defaults. For example, earlier in this thread you can see that if the timebase is set to any less than 10ms/div, the FFT doesn't work, but at/above 10ms/div it does work, and if you go much above 10ms/div the result doesn't change all that much.

[Momchilo]

Maybe the FFT tutorials from Lecroy will help you.

https://teledynelecroy.com/doc/tutorial-fft-analysis
https://teledynelecroy.com/doc/setting-up-an-fft

I didn't use the FFT function on modern Tek scopes. So I can't tell what they do.
But here is a basic FFT video for a Tek scope made by w2aew:

You have to set it up properly too like your Lecroy scope.

You need some understanding of what you expect in the spectrum analysis. It's similar for a spectrum analyzer. The device doesn't know what you want to see. So you have to set up the parameters of the device like span and RBW.

[gf]

Let me put it differently: how come, on the Tek scope, you just press Math then FFT and you see a spectrum which looks about right and which is a reasonable starting point.

Chcked the manual of the TDS2004. Seems that it has a (fixed) record size of only 2500 samples. This implies that the sampling rate must start dropping when the timebase becomes slower than 250ns/div, otherwise the buffer would overrun.

On the LeCroy you obviously used a record size of 10M. That is 4000 times as much. At 2GSa/s, this means you can store 5ms of data in the buffer w/o overrun. In other words, you can slow down the timebase until 500µs/div, and the sampling rate can still stay at the maximum of 2GSa/s w/o overflowing the buffer. Only if you slow down the timebase even more, the sampling rate must drop in order to avoid overflow.

Consequence: If you want to deliberately drop the sampling rate, then you need to select a pretty slow timebase.

Did you set the record size to 10M, or is this the default?

[Someone]

it's not because I wanted people to tell me to read the manual, which I have done

The manual answers your exact question, directly, in the section for FFT analysis, a section that is extremely short and to the point.

Can any of these do an FFT on just one cycle of the waveform?

You have not read the manual if you are asking this.

Let me put it differently: how come, on the Tek scope, you just press Math then FFT and you see a spectrum which looks about right and which is a reasonable starting point.

Show some examples of that behavior, on other scopes if you only capture several cycles of waveform you get the same low resolution FFT.

And they don't need the timebase shrunk so it is displaying hundreds of cycles of the waveform, since nobody uses a scope like that (it is completely useless).

You say useless, others say important. Different scopes have different ways of working, and you need to read the manual.

So there must be some simple way to set up the defaults. For example, earlier in this thread you can see that if the timebase is set to any less than 10ms/div, the FFT doesn't work, but at/above 10ms/div it does work, and if you go much above 10ms/div the result doesn't change all that much.

The FFT is still working, its just not showing you what you want to see (and you still have not been able to explain exactly what that is you are wanting). There are no "defaults" that readjust other settings of the scope, the FFT function is taking what you have captured, and displaying what it can. You are in control of what the scope is doing.

garbage in garbage out

[Someone]

FFT also uses windowing

Can/may/might/could/possibly use windowing depending on the application. Windowing is not inherent to FFT.

Well this may be a matter of the point of view²⁾. If we consider that FT and DTFT are actually defined for an infinte time line, then....

Bold assumption, when many would take an FFT as explicitly periodic.

[Mechatrommer]

And they don't need the timebase shrunk so it is displaying hundreds of cycles of the waveform, since nobody uses a scope like that (it is completely useless).

You say useless, others say important. Different scopes have different ways of working, and you need to read the manual.

well just to be fair to OP... i dont own WaveSurfer 3000 series, so i can only do screenshots... the 1st screenshot is interesting.. in zoom tab we can manually enter vertical and horizontal setting. if we established the FFT parm/variables i explained earlier, we'll have a clear mind what setting need should be.. for example if our FS is 1GHz, we can set horizontal center to 500MHz @ 100MHz/div, vertical setting to suitable signal power level, generally maybe 0dBm center @ 20dBm/div will be a good start to see all levels.. then you should have see complete view on screen of FFT from 0-1GHz.. we can then zoom in to freq of interest further by reducing horizontal setting, but we need to keep our FFT formulation in head, yes DSO is not perfect SA, so we need a clear and sharp head setting FFT parm in DSO. about why some DSO can show the correct view at first click well i guess some cars are aimed for automatic gearshift where users dont have to think and click anything, but unfortunately some manufacturers still like manual transmission setting where you need to be like a F1 driver, before electronic age... ymmv.. (manual link: http://cdn.teledynelecroy.com/files/manuals/wavesurfer-3000-om.pdf)

[gf]

Bold assumption, when many would take an FFT as explicitly periodic.

I possibly didn't express myself well. I consider DFT itself periodic, too, of course. And yes, windowing is not inherent to FFT algorithm itself, but rather a pre-processing step. But windowing is an inherent part of a (discrete) Short Time fourier Transform (STFT), where a continuous stream of samples is split into junks by a sliding window (that's the "windowing" part of the STFT) and then DFT is applied to each chunk. And that's basically the analogy I'm thinking of: After sampling the input signal with the ADC we get a continuous stream of samples. The scope captures just a finite time window of this sample stream, and DFT is applied to the samples in this window. So what we get is more or less equivalent to a shapshot of a STFT, for one particular window on the time axis. [ What get is of course not the Discrete Time Fourier Transform (DTFT) of the whole signal, since this would require an infinte number of samples. For signal components which are known to be periodic, the DFT of the samples in the window may be a good enough approximation for their DTFT, though, depending on the chosen window size and window function. ]

[voltsandjolts]

If you haven't explicity applied any windowing function then you are are using the default rectangular window of all 1's

[gf]

Yes of course, that's what I alredy mentioned earlier. And even the rectangular window corresponds to a particular frequency reponse of the spectrum analysis filter, when we consider the windowed DFT from a filter bank point of view. In fact the corresponding bandpass (which gets centered at and applied to each frequency bin) has high side lobes and much spectral leakage.

[peter-h]

I have read the manual for the FFT and I can't say I would be any wiser if it wasn't for this thread.

The key is to have lots of cycles on the screen; 100+. After that, everything is secondary.

Now I am trying to get something else to work: ARINC429. Nothing in the manual. I will start a new thread.

[2N3055]

I have read the manual for the FFT and I can't say I would be any wiser if it wasn't for this thread.

The key is to have lots of cycles on the screen; 100+. After that, everything is secondary.

Now I am trying to get something else to work: ARINC429. Nothing in the manual. I will start a new thread.

I recommend to anybody learning FFT on a scope these two short writeups.
They are simple but explain basic relations between sample size, sample rate, sampled data time period, and basics of why windowing is needed and fact that different windows give different results. They are great for building basic intuition of how to use FFT on a scope.

[peter-h]

Thank you 2N3055. That 1st one is a good summary of what one has to do.