#65: Basics of using FFT on an oscilloscope

This video briefly presents the basics of using a Fast Fourier Transform (FFT) function of a modern digital oscilloscope to observe the frequency or spectral content of a signal. It discusses how the setup of the scope (sample rate, record length, memory depth, etc.) influences the frequency range and resolution of the FFT result. It also shows how the "high-resolution" sampling mode of the scope can be used to increase the vertical resolution of the scope. The example used shows the result of measuring the audio frequency content of the noise created by two different electric drills.

[@jacobfaseler5311]

That explanation of a window function is golden. I came thinking “I know how to use the FFT - I’ve done it before, but this is good television”, but even in that regard I learned oodles about the parameters. I’d pay out of pocket for another degree if you were on teaching staff.

[@1959Berre]

You really explain things the best way. Anyone who takes the time to actually draw on paper with a pencil deserves a big thumbs up. Those 4 morons who did not like this video should be ashamed!

[@camiemengineer]

11 years on and you're still the best and electronic master. Just to let you know:- One of the finest videos you made was on the CMOS Phase Locked Loop, (CD 4046). That was truly a master class! Thank you Mr. Wolfe!

[@garywinders3707]

Thanks for a superb demonstration of FFT for audio frequencies. I've only recently upgraded from an old analog CRO to a DSO. I design, build & repair valve (tube) guitar amplifiers and was particularly interested in FFT for audio and your excellent, detailed demonstration has nailed it for me. Thanks again.

[@rioschad3284]

As a recent EE grad, I could only wish my instructors taught as well as you. Thanks for taking time to explain these fundamentals.

[@ykalidas]

Thank you for putting out this very helpful tutorial. The basic idea of how FFT would have been contemplated by early thinkers is clear now.

[@TonyG8WBI]

Well I never really understood what FFT was, so thank you very much for that. A picture paints a thousand words.

[@tonysfun]

Love your demos/videos very much! Whatever I didn't completely understood I now understand. Thank you again!

[@TheHarmonicOscillator]

Excellent video as usual! Thanks! I can see using FFT on a digital scope to sound proof a room or reduce sound as well as wear and tear in a machine.

[@thewii552]

Great job, and very interesting as always. It's always a nice day to see that you've posted a video, even if it doesn't even remotely apply to me.

[@krish2nasa]

Hey Alan, Excellent Explanation. Thank you very much for your time and efforts.

[@asifsurti9154]

Thanks a lot sir. Your videos are really easy to understand and helps me get the concept really fast. Really appreciate the practical nature of the demo. Thanks again!

[@sebrura]

Thank you! Just got a second hand R&S scope for the sole purpose of having frequency distribution from DC to 9 kHz, and couldn’t get the span narrower than 10 MHz. Though I had made a mistake. Seeing your vid made it all work out as expected. TY!

[@williamogilvie6909]

Your videos are all top quality. There are so many utubers who just do so much hand waving and yabba-dabba, on EE subjects. I produced an FFT programs a few years back. I took the code from Frank Press's book "Mathematical Algorithms" and got it working on a PC-104 486 board connected to a quarter res LCD. I used a DOS graphics library I had developed earlier. I really like the MDS4000 scopes. One company I worked for put one on my desk. I used it a lot and left a write-up on one of their products, complete with signal graphs. I wish I could afford one.

[@DrCassette]

This video was indeed very helpful. Thanks!

[@kkingkrylon]

I don't know how to thank you for all the good videos!! Thanks you so much!!!

[@kv3x]

Excellent introduction, thanks!

[@gregjohnson5194]

That was very helpful, thank you for you time on this topic.

[@mjlorton]

Thanks good man. Great tutorial as always.

[@AshokRajpilli]

Thanks I have struggled in this fft, but after seeing your video I understand how it works

[@mandomonica]

Thanks for posting. You have done a great job explaining things.

[@mikefinn2101]

Very Excellent explanation well done and well presented. Learned something new Great Job

[@Pwaak]

Thank you for your effort and sharing this interesting information!

[@bigcx21]

awesome video and explanation as always, thanks.

[@nnamrehck]

Excellent video!

[@CorbbinGoldsmith]

That makes so much sense!!! And, THANK YOU SO MUCH!!! You've just been put on my "best friends forever" list! Lemme go try this! thanks again! I have Native Instruments Reaktor to check this on, so it's pretty much like programming, but with modules and in real time.

[@marcdraco2189]

Now that is how to teach! I wish this guy had been one of my tutors.

[@hyedenny]

Very nicely presented!! Thank you!

[@muppetpaster]

Great idea!!!! Let's take your scope to a hardwarestore to examine the "noise"of your soon to be new drill

[@sdmoonlitsea12]

Thanks for this great tutorial on fft. I am taking a receivers class this semester and I need to know how all this works.

[@HiFiWright]

Awesome video. Video instructors like yourself are a godsend & especially to people like me who are trying to modify and build amplifiers with no electrical engineering background or former electronics education other than what we have been able weave through on the internet. Thank you very much! Still trying to decide whether to purchase a used but working/calibrated analog scope (I lost a bid the other day on a GW 20mhz semi-modern analog scope w/digital buttons for $42 lightly used *sigh*😣) or one of the modern cheap digital units that does a simple 100khz or similar simply for the math and fft functions like in the video. Im torn lol i like the simplicity and real time accuracy of the analogs ive seen...but i also really want to be able to have fft capability as well as the simplified time saving math functionality. Oh, plus at this very minute my budget is limited to maybe slightly over $100 (the biggest handicap of the matter 😅) to spend on such - so used is thr only real option 😬

[@JAKOB1977]

That was solid and comprehendible.

[@MikeOBrien0]

Love your vids! Thx!

[@richpaul8132]

Very Good! Even an entry-level hobbyist like myself understood that entire video! I wish I knew why I understood it now, but when I read about the FFT function, I thought to myself "...I must be dumber than a box of hammers because I have no idea what I just read..." Thanks for the video and the reaffirmation in my own mind that I may not be as stupid as my dad (rest his soul), said I was, lol. Rich

[@BM-jy6cb]

I can't be the only one who winced at the thought of a drill so close to that scope!

[@funkyironman69]

Great video, thanks.

[@yizhang2646]

Very nice explanation

[@nelsongracia]

Hi , I would like to know if it is possible to measure in a specifit frecuency the exact value of the ohms in a crossover 2nd order ? thank You

[@CorbbinGoldsmith]

Oh! I looked at that video a while back, now it makes more sense. Thank you so much. I was sweeping the frequency of a bandpass filter instead of a the frequency of a sinewave. *facepalm*

[@leonverycool]

thanks you that's a good video you made

[@HansVanIngelgom]

I still vividly remember how I struggled to understand all that. I hope I have given you some clues that point you in the right direction.

[@jadonbull1601]

Hi Alan, Thanks for the video, I am using a similar oscilloscope to the one in the video (Tektronix DPO400 series) and trying to measure values between 30Hz-150kHz using the FFT mode. When comparing the results of the measurements at two test frequencies (120Hz and 125Hz), I noticed significant difference in FFT max value vs. the Vrms value of the time domain signal at the same input value.. At 120Hz, I'd see 6.35Vrms for both FFT and time domain Vrms. At 125Hz, I'd see 5.39Vrms for FFT and 6.35Vrms for time domain. I'm assuming this has to do with what you were describing when talking about windowing and the signal beginning and ending at the same value. How would I go about solving this issue in this case? Thanks, Jadon

[@jagardina]

Thanks, this was a very good refresher. Very clear and concise. Of course you would use a better microphone with a known frequency response if you really wanted to measure accurately.

[@dbeierl]

Nyquist limit applies to every frequency in the signal you're examining, not just the ones you're interested in. Those high-frequency components you don't care about will still give you aliases. In this case though I imagine the speaker is rolling off the higher frequencies pretty well.

[@Boretoto]

Excellent, does is need a 1k Ohm resistance on the cables between the sound card and the computer or oscilloscope with FFT function? Thank s

[@nathanhaltman7235]

It is worth to note the dynamic range of the measurement. Pending on scope settings, the spurious dynamic range is only about 40 to 50 dB. Most oscilloscopes are limited by a 8 bit analog-to-digital converter. However, the scope may show a noise floor well under 50 dB with erroneous harmonics.

[@DavePKW]

Good stuff. Thanks.

[@pollada1]

Thank you sir!

[@CorbbinGoldsmith]

Cool, I'll check them out as well!

[@californiakayaker]

I'm also thinking of incorporating a audio sweep generator into along with fft from my scope. What sort of things would you see ? What would you need to consider ? What if you were to use a noise generator ? N6GRG

[@AboubakrA]

Sorry for popping up randomly in the comments of your old videos, but I recently discovered your channel and I wish I had discovered it earlier, it's awesome, thank you. One question on the video: can you actually determine the rpm of the drill, maybe form the lowest frequency peak ?

[@saurabh8998]

First of all let me take a pleasure to appreciate this video. I am neither designer or musician, I am engineering student struggling with my Digital signal processing(DSP) course. Everything seems perfect in theory class but I am still doubtful as to how I am gonna use this in my project that aims at building voice controlled security system, recording particular voice in say more noisy region. (thinking of this for my final year project). I need some help in visualizing all these DSP tools...

[@GOWTHAMKRISHGowthamShanmugaraj]

can u make a video on how a to d and d to a "audio" converters work???

[@rajeevmech9]

Very good video

[@CorbbinGoldsmith]

If you can, could you explain how FFT works? I've been trying to wrap my head around the windowing methods. like, how it figures out each of the frequencies. If you could, that'd be great! Thanks again!

[@KX36]

I can't find much by googling about the window factors involved in frequency resolution mentioned in this video (freq res = w.f. / t_duration), only about window correction factors to correct amplitude or power of peaks for different window functions. e.g. for Hamming 1.85 for amplitude and 1.59 for power. Are they the same thing?

[@w2aew]

Here are a few blog posts that discuss the function and purpose of the window functions: www.tek.com/blog/spectrum-trace-processing-rsa www.tek.com/blog/window-functions-spectrum-analyzers

[@AnandMMisra]

can you please recommend a book on exploiting an oscilloscope for its functions including that of a FFT

[@janwatters635]

thanks very helpful

[@montegary19]

hey, guess its kindda ages after you have posted this video... but a really good job giving the bacics of FFT..i really wud love to find some more stuff on FFT can you refer me to some of that? Thanks

[@nopenadanowaynohow]

Thank You !!

[@CorbbinGoldsmith]

Thank you so much!!! :D

[@David_94]

Thanks for the video Alan, I have a question, how do I measure the dynamic range of a signal with the FFT function in the scope? How do I know where is my noice floor and how to improve it?

[@w2aew]

The noise floor is affected by two things - sample rate and record length. A higher sample rate results in an FFT that covers a wider frequency range - thus the scope's noise floor is spread out over a wider range, which make it lower at any given point in the FFT result. A longer record length improves the frequency resolution, essentially a narrower resolution BW, thus the power in each of the resulting trace points (FFT bins) is lower.

[@mlynch001]

I have 3 scopes with this feature and now I know how to use them.

[@HansVanIngelgom]

Concerning the negative value: the phase could be 180 degrees off. In fact to correctly do an FFT you not only need to multiply it with a sine, but have to do the same again with a cosine (something I had omitted because explaining an TTF in 500 chars is a bit challenging :)). These two values then make x and y values of a vector where the the angle is the phase, and the magnitude is the signal strength of that frequency.

[@ornithopterindia]

👍Thank you sir.

[@Arijit_VU3ICT]

Thank you sir..

[@harrisaastamoinen]

You are also getting the magnetic field inductively, mixing into the signal. The effect of that may (or may not) be quite significant.

[@quantomic1106]

Hey Alan! This might be a dumb question; but can you do a bode plot using FFT? If so, can you show us how. Thanks!

[@w2aew]

You can, but only if you apply a noise source to the device under test, or apply a sweep. If you apply a sweep, you could just show the device response vs. the sweep time, which will basically be the frequency response.

[@quantomic1106]

***** Great! Thanks! Maybe you should do a small video about it. I think it will make a good topic. Thanks again and please keep us educated : )

[@jaguar65100]

Nice explanation of using FFT on a scope. Just a couple of questions, Is the window effectively a built in apodization function? I always looked at the FFT as capturing the frequency content of a time domain signal at a specific instant in time. I am curious when we look at the audio spectrum as typically 20 Hz - 20 kHz, if we take a sample longer than 50 microseconds don't we begin losing resolution of the signal at high frequency since the signal is possibly changing this quickly? What I guess I am getting at is that if we sample at more than double the Nyquist frequency for the frequency band in question, and we capture 400 ms of samples the FFT could change dramatically over that time with a complex audio (music) signal, so is the oscilloscope just averaging all of these samples? Thanks for your instructive demo!

[@w2aew]

The window function is a selection available as part of the FFT math function. The FFT cant measure frequency content at a specific instant, for the same reason that you can't tell how fast something is moving from just a snapshot. There has to be an observation time to see the variations over time. Longer acquisitions give you better frequency resolution in the FFT. the sample rate determines the frequency range given by the FFT. In order to see 20kHz in an FFT, the sample rate has to be greater that 40kHz. If the signal is changing during the acquisition used for the FFT, then the FFT will show basically all of the frequency components that appear in the acquisition. The amplitude of each frequency component will be proportional to how long that component lasted with respect to the acquisition length used for the FFT.

[@arlindoqueiroz2723]

jaguart65100

[@dierenfilmpjes5136]

Fantastic

[@alexmessina7652]

HI there, this was very interesting for a novice. I use a scope in automotive applications but I'm not expert. I am however trying to work out if I can use the FFT function to detect a momentary frequency drop out in a constant frequency signal. My expectation is that I would see a spike at the low end of the spectrum? Appreciate your thoughts. Alex

[@w2aew]

Probably not because good resolution requires long record lengths which would mask any short dropouts.

[@alexmessina7652]

w2aew damn! Thanks for that ... didn’t realise the long record time was required

[@Popart-xh2fd]

Nice video. In Acquisition Mode 9:19 what is the difference between Hi Res and Average?

[@w2aew]

Average mode will average the waveform samples on successive acquisitions - so it requires several triggered acquisitions in order to work. Hi-Res mode is used when the scope's sample rate is faster than what is needed to produce the waveform record requested - it averages a group of successive sample points into the waveform points - thus it works even on a single acquisition. These modes are more fully explained in this video: kzbin.info/www/bejne/e2Gcep2Kr9ifd5I

[@xiaosun9111]

we bought a oros nvgate for vibration and noise measurement, in the manual i saw selections such as rms, max, or min, are these the time domain displayed signal? how can i get these from the original value? and the frequency domain has inst. spectrum and avg. spectrum, are these the same?

[@w2aew]

I am not familiar with the Oros Nvgate product, so I can't comment for sure.

[@yaidontknowwhattoput]

I cant seem to unify the FFT from the scope with one that I am doing in python. It is off. Like really off. Used hanning, and converted to dBV. In what should I have confidence?

[@w2aew]

It's difficult to say. I don't know your scope, or how you implemented your FFT. I would tend to believe the scope since it was designed by a team of engineers. If you want them to match, then you'll have to match the FFT length, including any zero padding, as well as the window applied.

[@yaidontknowwhattoput]

that's what I'm thinking. I'm running noise through the scope and exporting voltage and fft. I have tried to mimic the specs of the scope as best I could. I'm sure most manufacturers use similar algorithms to do the fft. Sample rate, aquire time, reference voltage, windowing are all the same. If the fft is run on the previous window data, I expect to get little/no correlation (meaning cross correlation) between the two. you seem to know your stuff and I wouldn't ask if I hadn't tried all the tools in my arsenal. thanks for the quick reply.

[@w2aew]

If the scope is truly doing an FFT, then you'll have to determine what the scope does in order to make the record length equal to a power of 2 in length. Is it truncating the waveform, or is it zero padding? Is it applying the window to FFT vector prior to zero padding or after zero padding? Lots of little subtle details that can affect the results.

[@jj-js5sx]

Thanks for your great video. Question: Why is it so difficult to find a "swept-tuned" frequency analyzer for audio? My scope does FFT, but I would like to use swept-tuned. It seems they all want to start at 9KHz and go up, not down. (I'm guessing cost is a factor, but why?)

[@w2aew]

Not enough of a market demand. Can easily be done with s/w and a soundcard, in realtime.

[@jj-js5sx]

@@w2aew Thanks for the quick reply. Interesting that higher frequency analyzers don't also use s/w and fft vs hardware swept-tuner. However, your answer still makes sense- there's a preference in the market and soundcards and pc's make it simple for s/w to manage the audio range analysis. Your videos are always superb. I've watched many many and learned from all.

[@w2aew]

@@jj-js5sx Actually, many new analyzers DO employ digital IF processing, realtime FFTs, etc.

[@fredriko.zachrisson9711]

I would like to see how you find the spatial resonant frequency of a inductor using the fft function of the oscilloscope.

[@w2aew]

I'm not familiar with the term "spatial" resonant frequency. Are you referring to the self-resonant frequency?

[@fredriko.zachrisson9711]

w2aew i actually mean spatial resonant frequency. There is a guy called Dr Ronald Stiffler that has been developing technology called SEC Exciter. Are you firmilar with it? Like the Slayer Exciter, but the original tech if you will. Finding the self resonant frequency with the FFT function would be very interesting too ofcourse. :)

[@AnandMMisra]

Can an oscilloscope with FFT do a spectrum analysis for power frequency voltage and currents? I am typically interested in analyzing motor currents(50Hz) to look at components of 2x, 3x, 4x and other higher-order harmonics.

[@w2aew]

Yes, but you will need to adjust the scope horizontal scale to give you a few hundred milliseconds of capture (at least) in order to get sufficient frequency resolution in the FFT result.

[@AnandMMisra]

thanks for your reply. I had forgotten that I had asked you this question and I was struggling with the poor frequency resolution of the FFT waveform. on increasing the record length to 250ms, the resolution improved, as you have rightly pointed out in your reply. thanks very much.

[@steelinhank]

Allen, is it possible to use the FFT to measure phase noise in 1 Hz bandwidths of a oscillator or transmitter?

[@w2aew]

In general, no, for a few reasons. In order to get a 1Hz resolution BW, you would need more than 1 second of time capture. If you're looking at an RF signal, then the sampling rate has to be much greater than the RF frequency. This translates to the need for a lot of memory (high sample rate capture for a long time), which most scopes don't have. Second, most scopes are 8-bit samplers, which won't have the dynamic range to see very low phase noise. Third, most scopes will have their own internal phase noise which will be greater than the device you're testing. There are exceptions (expensive exceptions) to this, but most affordable scopes can't meet that requirement.

[@ruhnet]

Never thought about the "audio quality" of my drills before. 🤯🤣

[@vishnudas9476]

Hi allen. Why do you need a spectrum analyzer-If we can get the spectral components through FFT. How is spectrum analyzer better than FFT ?

[@w2aew]

There are many ways that a spectrum analyzer is better than an FFT: - A spectrum analyzer will almost always have lower distortion, lower spurs, and better SFDR. - A spectrum analyzer will have much better selections for RBW, detector types, trace types, etc. - A scope FFT is fully dependent upon the acquisition settings on the scope (input BW, sample rate, acquisition depth), all of which affect the type of results you get. - Most scope FFT have limits on the FFT vector length which places a lot of restrictions on resolution. I could go on, but you get the idea.

[@vishnudas9476]

Oh ok.So the scope spectrum is not as accurate.Its only as good as how well you are able to represent the actual signal. Thank you :)

[@BruceNitroxpro]

It's almost as though you knew what you were talking about! LOL Good show!

[@w2aew]

...almost! ;-)

[@BruceNitroxpro]

@@w2aew , So cool. So calm and reassuring compared with the mixed messages from the EEBlog... LOL

[@yoramstein]

Thumbs up.Thanks

[@qwaqwa1960]

I've written a couple of introductory FFT blogs at Scope Junction...may be of use.

[@mamaurax25]

Thanks man. If it was not you. I would no longer high marks on this

[@Dr_Xyzt]

Does the FFT speed up if you go to increments of say, 10 hz?

[@w2aew]

FFT speed is primarily driven by the size of the vector or array of data being processed - i.e. waveform record length. This is determined by the sample rate and time duration. The waveform sample rate will determine the FFT frequency range, and the waveform time duration will determine the FFT frequency resolution (bin size). Slower sample rate and shorter time duration leads to faster FFT processing.

[@Dr_Xyzt]

I see. Does a, 93ms sample length sampled at about 100khz run fairly fast on a machine like this? I'd like to get a Tektronix 1052B and an inverter so I can check pulse width modulation on cars, and then audio equipment for frequency response/distortion, so I'll need a reasonable FFT. You know the oscilloscope app you can get for your phone? I need that kind of speed, but better resolution. For microcontrollers, a slow FFT is fine.

[@w2aew]

93ms sample length sampled at 100kHz results in about a 9300 samples. FFT lengths have to be a power-of-2 long, so this would result in a 16k sample FFT - certainly not the fastest...

[@Dr_Xyzt]

I see what you mean. Search in bands. I can live with that. 0.1ms sample at 100Khz.

[@kevinvanlandingham8938]

Would you be able to upload your Notes to this video? Thank you so much for your Videos!

[@w2aew]

Unfortunately this video is 8 years old and my paper copy of the notes are long lost/gone. However, the notes from another video on FFTs might help you out: www.qsl.net/w/w2aew//youtube/FFTonTDS2000.pdf

[@user-su5sq5ib3i]

Can the fft function be used like a spectrum analyzer? For RF?

[@w2aew]

To some extent, yes. The scope bandwidth has to exceed the RF frequency you want to measure. The main problem is that it may be difficult or impossible to get nice low RBW around a specific RF frequency due to sample rate and memory limits.

[@yoramstein]

Thanks

[@HansVanIngelgom]

PS: That's one reason why complex numbers are often used in DSP: one complex number can hold both the phase and the amplitude. You have to envision a complex sine wave as a spiral going along the time axis. Another advantage is that you can define both positive and negative frequencies (going clockwise or counterclockwise). E.g. doing AM demodulation of a signal using complex numbers is trivial: you just take the amplitude of each complex sample and you're done! Try that using just real numbers.

[@msk19991]

Thank you thank you & thank you ^_^

[@streamcnoremorse3323]

Great video but I can not see difference in spectrum of drills.

[@abdulrahimevli260]

Can we see original Signal power on FFT in the oscilloscope.i see DB but i must see dbm

[@w2aew]

The scope FFT doesn't show dBm because it has no knowledge of the termination impedance, which determines the signal power.

[@ReginaldPierreEE]

I used to do this in the lab

[@divyanshmittal772]

Hi, I have designed my own FFT on Verilog HDL and I want to verify that, how I can do that?

[@w2aew]

Feed it with sampled sinewaves at different frequencies and verify that the result appears in the appropriate FFT bins.

[@divyanshmittal772]

@@w2aew Thankyou for your reply, Can you tell me about more resource, because I an designing 1536pt fft on Hardware and I want to gain as much as theoretical and a practical knowledge as I can

[@HansVanIngelgom]

The windowing is not essential to the FFT, it is the method for 'cutting' a piece of signal out of a continuous stream. See it like a fade in/fade out. One important property that enables Fourier analysis is that if you take two sine waves, multiply them, and then integrate the resulting signal, you get 0 if the two frequencies are different. This is easy to check e.g. with a spreadsheat (make two columns with sine wave values, multiply the columns and sum the results).

[@byronwatkins2565]

A speaker works as a microphone but not "just as well." The inertia and spacial extent of the cone attenuates high frequencies and the coil couples directly to the motors' changing magnetic fields.

[@gearstil]

Why is the frequency span half the sample rate of the scope at that particular measurement?

[@w2aew]

A signal must be sampled at least twice per cycle in order to properly determine its frequency. This is called the Nyquist Criteria. In order to have at least two samples per cycle, the sample rate has to be greater than twice the signal frequency. If you don't meet this criteria, the signal will be "aliased". A visual example of this is the old western movies on TV where the wagon wheels appear to be rotating backwards. Since signal frequencies above 1/2 the sample rate won't be displayed properly, they are not displayed at all.

[@gearstil]

@@w2aew thanks for the quick and detailed answer!

[@SurvivalSquirrel]

Where does the scale and reference come from on the FFT?

[@w2aew]

Frequency scale from an fft is from DC to half the sample rate. Vertical reference level comes from the volts/Div setting on the channel.

[@CorbbinGoldsmith]

That makes a lot more sense now! Thanks! I'll try this with real numbers! just kidding.

[@CorbbinGoldsmith]

I was talking about qwaqwa1960. But okay. kool. thanks!