EEVblog #845 - Oscilloscope FFT Comparison

[EEVblog]

Dave compares the FFT modes on 7 different oscilloscopes:
Rohde & Schwarz HMO1202 Series
Tektronix MDO3000
Keysight 3000X Touch
Lecroy WaveJet 354 Touch
GW Instek GDS-1104B
Rigol DS1054Z
Rigol DS2000
Which is the best?
Which one sucks the most?

[chris_leyson]

Dave, my 30+ year old TEK 7L12 will do better than any of those scopes, 70dB to 80dB dynamic range and 1.8GHz, albiet very very very slowly. Try an Analog Discovery or Red Pitaya, 14-bit A-D, I don't why manufacturers bother putting FFT software in scopes with 8-bit or 10-bit digitizers.

OK, to be fair FFTs are a damn sight faster than a swept 7L12 and you get much better frequency resolution, but I'm thinking 16-bit A-D plumbed into 1st IF (105 MHz) or 2nd IF (10MHz). 

[wraper]

When Dave said that FFT sucks in the cheap scopes, for some reason didn't mention that instek's 1000B series are even a bit cheaper than Rigol 1000Z (for 50 MHz model if comparing with 1054Z). Also from actual FFT performance it's certainly looks to work better than R&S and Agilent (except fiddly UI) and is second after Tek MDO.

[chris_leyson]

Forgot to mention that FFTs can't handle transients very well, and aliasing, that's always fun if you get your sampling rate wrong, but it can be a useful feature sometimes. They're good if you can fill your 50 to 60dB dynamic range with signal, fair enough. I've got the extra math module plugged into the back of my HP54610B so it will do integration, differentiation and FFTs, only used the FFT function to try and figure out where some noise was coming from, a few frequency domain clues are always useful.

If you could convolve or correlate two channels that would be a useful function, I don't know if this feature is available on Keysight or Tek scopes.

[karoru]

With any algorithm it's "garbage in, garbage out" principle. FFT from 8-bit ADC in the scope will be bad, no matter how clever you try to analyse the data (unless they invent some oracle chip that will extract the additional information using heaven microwave link). We can compare mainly how software guys managed to put algorithms into software so it at least don't make you cry when after turning the knob it takes 10 sec to return the scope to responsive state.

Otherwise you can just dump the data from your scope and put it on your random Matlab/Octave/Scilab/whatever and do the data processing by yourself.

[Froese]

Sorry if this is a dumb question - I'm not a DSP-guy.

Why do these scopes need this huge number of FFT points?   Basically, if you have 1024 horizontal pixels on the screen, shouldn't a 1024-point FFT (+/- some slack) be good enough?  Couldn't they mix the selected window down to  DC (like a direct conversion receiver), filter, maybe resample  and then a 1024-FFT over DC to span-width?  Software mixing/filtering/resampling/1k-FFT should be much cheaper than a direct 128k-point FFT, shouldn't it?

[EEVblog]

When Dave said that FFT sucks in the cheap scopes, for some reason didn't mention that instek's 1000B series are even a bit cheaper than Rigol 1000Z (for 50 MHz model if comparing with 1054Z). Also from actual FFT performance it's certainly looks to work better than R&S and Agilent (except fiddly UI) and is second after Tek MDO.

Yes I completely forgot the price of the GW-Instek, and stand corrected.

[Muxr]

I remember having to use the FFT on my Rigol DS2k and how frustrated I was. It really is as useless as your video showed.

[EEVblog]

I remember having to use the FFT on my Rigol DS2k and how frustrated I was. It really is as useless as your video showed.

You can see a signal is there, but that's about it.

[wraper]

I remember having to use the FFT on my Rigol DS2k and how frustrated I was. It really is as useless as your video showed.

I once switched on FFT on my hacked DS2072, thought wtf is this and did not even bother to try using it again. IIRC even on my old Instek GDS-1152A it worked better. Donated to my friend about a year ago, very similar with Rigol DS1000E series but IMO better quality. ADCs were of the same model but originally higher speed grade 100 MHz vs 40 MHz in Rigol. They were overclocked too but only by 25% not 150% as in Rigol, and there were four of them at 100 -> 125 MHz vs five of 40 -> 100 MHz in Rigol.
EDIT: and BTW in the 1152A CPU/FPGA/ADCs were on the daughter board too (GDS-1000B teardown) . So it might be common how instek design the things.

[Muxr]

I remember having to use the FFT on my Rigol DS2k and how frustrated I was. It really is as useless as your video showed.

I once switched on FFT on my hacked DS2072, thought wtf is this and did not even bother to try using it again. IIRC even on my old Instek GDS-1152A it worked better. Donated to my friend about a year ago, very similar with Rigol DS1000E series but IMO better quality. ADCs were of the same model but originally higher speed grade 100 MHz vs 40 MHz in Rigol. They were overclocked too but only by 25% not 150% as in Rigol, and there were four of them at 100 -> 125 MHz vs five of 40 -> 100 MHz in Rigol.
EDIT: and BTW in the 1152A CPU/FPGA/ADCs were on the daughter board too (GDS-1000B teardown) . So it might be common how instek design the things.

Yeah DS2072A is what I got (it's my main scope until I take delivery of my new to me RTM1054 in a few days). I am not getting rid of DS2072A. It really isn't a bad scope for time domain stuff it has served me well, but that FFT is just a major letdown.

edit: All these Chinese scopes seem to have major deficiencies, some are better than others in certain areas, but overall, and that could be just based on popularity, Rigol seems to do a ok enough job on most of it. I do think they should address the FFT though, because the audience they cater to, most likely doesn't own a dedicated Spectrum Analyser. Spectrum Analysers are generally pretty expensive, for this demographic, and having a usable FFT function should probably play a major role in deciding which first scope to get.

[orion242]

14:43....its like "dick n' balls"

LMAO!!

Its that off the cuff commentary that makes your videos great.  Keep up the good work!

[wraper]

14:43....its like "dick n' balls"

LMAO!!

Its that off the cuff commentary that makes your videos great.  Keep up the good work!

Dave just confirmed after a moment my first thought what it looked like, when this thing appeared on the screen (I think that I even said it loudly on my language). Then I started to laugh loudly. It really looks like this.

[nixfu]

Most of us hobby players are often interested in using FFT becuase we cant afford a proper SA for the bench.  And sometimes FFT is at least a chance to get some sort of data in the frequency domain that is better than nothing.

If there was just a little more competiton in the SA market and more decent options than the only one now (Rigol 815), then maybe more benches could have a real SA sitting on them and FFT on scopes would hardly be worth using.

Someday someone(Owan, Siglent, Hameg, GW)  surely will come out with something to compete in price and functiom against the Rigol 815 eventually.

[Someone]

Sorry if this is a dumb question - I'm not a DSP-guy.

Why do they these scopes need this huge number of FFT points?   Basically, if you have 1024 horizontal pixels on the screen, shouldn't a 1024-point FFT (+/- some slack) be good enough?  Couldn't they mix the selected window down to  DC (like a direct conversion receiver), filter, maybe resample  and then a 1024-FFT over DC to span-width?  Software mixing/filtering/resampling/1k-FFT should be much cheaper than a direct 128k-point FFT, shouldn't it?

One power of the FFT with a large number of points is when you zoom on its horizontal scale (frequency) and can see that your original signal contains not one but two or more closely located frequencies, similar to how you might have a memory depth of 1M but the screen shows the overview in 1k points until you zoom in. The other reason to use lots of points is that the noise in the final result is reduced which is why you can see 80-100dB of signal to noise ratio and turn on dB scaling for the vertical axis even though in time domain the 1000px linear in the vertical axis was enough to see the limits of the ADC.

[Muxr]

Most of us hobby players are often interested in using FFT becuase we cant afford a proper SA for the bench.  And sometimes FFT is at least a chance to get some sort of data in the frequency domain that is better than nothing.

If there was just a little more competiton in the SA market and more decent options than the only one now (Rigol 815), then maybe more benches could have a real SA sitting on them and FFT on scopes would hardly be worth using.

Someday someone(Owan, Siglent, Hameg, GW)  surely will come out with something to compete in price and functiom against the Rigol 815 eventually.

Of all the options for an SA I am leaning towards the Signal Hound the most. It has higher bandwidth, and it covers audio frequencies. It's USB based but from what I understand the software is quite usable.

[scopeman]

I know that this is about low cost scopes but below is a link to a good introduction to FFT on LeCroy scopes.

Deep memory on a software based FFT is a must. The Chinese scopes are no match for the FFT and math functions on even the older 9300 series LeCroy scopes with deep memory. LeCroy had deep memory scopes with FFT probably 10 years before anyone else. Even before HP got into the business with their scopes that were lucky to do 6 bits and did not have FFT, and before Tek had anything that would do FFT.

With the advancement in processing speed they are more responsive than ever.

Scopes that do the FFT in hardware (dedicated hardware memory with dedicated FFT processing in hardware as opposed to a mixed approach) have the advantage of speed (Rohde & Schwarz does this on their high line scopes) but limitations in effective resolution bandwidth due to the fixed number of bins.

The ability to adjust the depth of the sample memory to optimize the required frequency resolution and or the processing speed is quite handy. Once you get your hands on a deep memory scope with FFT and see the advantages of variable depth sample memory you will never want to go back to wimpy or fixed memory scopes again.

My old 9354AL can vary the sample memory from 500 samples to 2M samples and I can do an FFT to zoom memory and then do additional math on the FFT like FFT average. Math on Math on Math is one of the great features of these scopes and they were built in the 1990's. The follow on scope generations are built on the same proven architecture with faster processing, more acquisition memory and even more math power.

These scopes trace their lineage back to Walter LeCroys' involvement with measurements in high energy physics and that is why they were way ahead of the curve in all matters of mathematical processing, including FFT.


See:

http://teledynelecroy.com/doc/setting-up-an-fft

It is still true that there are many occasions where you do need a good RF spectrum analyzer but for a lot of uses FFT on a good scope is still quite handy.

Sam
W3OHM

[daqq]

Thanks for the comparison Dave!

It would be a great idea for a Fundamentals Friday to show how Fourier (and Fast Fourier) transformation actually works - the mathy bits.

[Lukas]

Yeah, I know, everyone hates USB oscilloscopes, but leveraging a modern PC's processing power for FFTs could really improve things. My 2015 ultrabook i7 can easily do a 10M-point FFT (using libfftw) every second. No need for FPGA or ASIC wizardry. Compare this to the MDO3000 where a 1M point FFT took like 10 seconds.

Take a closer at how various oscilloscopes respond to turning the knobs while they're number crunching: Some scopes like the instek one don't give priority to the user and just keep going on crunching, making the oscilloscope seem unresponsive. OTOH the Keysight one suspends the acquisition in the instant the user turns the knob and waits some time after the user has done the last interaction. I'm wondering why they need to do this, since they claim to have hardware FFT. So changing the FFT parameters should just be poking some registers...

[EEVblog]

Yeah, I know, everyone hates USB oscilloscopes, but leveraging a modern PC's processing power for FFTs could really improve things.

This has always been a potential advantage of USB scopes. Particularly when they quite often have >8bit converters.

[EPTech]

Oops,

I just bought a Rigol DS4204. I did not buy it specifically for it's FFT features but I hope they are better than what I saw on Dave's review.

Anyway, it will arrive next week and I will put it through the test and get back to you. I only have an analogue function generator running max 2 Mhz.
What is the best benchmark using that? Maybe a square wave to get some harmonics?

[Mosaic]

Too bad about the Rigol....I have one, but also a DSA 815 TG...so i use that....I also have an 1102D for Q&D fft at low freqs.

I did some quick bandwidth testing on the DS2072A up'd to DS2302a...seems to be around 310Mhz , 3dB down....using a Tek SG504 leveled sinewave gen. via 50ohm RG58 cable. I need to up that cable to an RG400 or LMR400 for  more reliability. Getting a couple megaphase orange 4Ghz rated cables soon, so I can use those.
I can do a proper test soon as I got an HP8753D 3Ghz VNA with full cal kit. I can tap off a minicircuits coupler and sweep the Rigol  for leveled sinewave and compare it to the VNA for  a precision check.

Let me know if that data useful to anyone.

[Karel]

Dave, on the DS1054z, did you set the "FFT mode" to "Trace" or "Memory"?

In "Trace mode" it's only 1200 pts. In "Memory mode" it's 16384 pts.

From the video it's not clear to me which mode you selected.
Could it be that the FFT mode selected was "Trace"?

[Stupid Beard]

Yeah, I know, everyone hates USB oscilloscopes, but leveraging a modern PC's processing power for FFTs could really improve things. My 2015 ultrabook i7 can easily do a 10M-point FFT (using libfftw) every second. No need for FPGA or ASIC wizardry. Compare this to the MDO3000 where a 1M point FFT took like 10 seconds.

You don't need a USB scope to do that. You can just get the waveform data out of any modern scope and do it. Someone has already done that for the rigol scopes; I think http://rheslip.blogspot.ca/2015/09/software-spectrum-analyzer-for-rigol.html is the right URL. I've never used it so I don't know how useful it actually is.

[TiN]

14:43....its like "dick n' balls"
LMAO!!

 

TDS7000 (on example of mine CSA7404) is similar for FFT like MDO, superslow, but pretty accurate when it gets there.