Understanding FFT in Audio Measurements

[amirm]

No idea how to compute the FFT gain.

It is pretty simple to approximate it if you don't include the window:

FFT Gain =10*LOG10N(N/2)

Where N = number of points. For 32768 points ("32K") it is 42 dB.

 

[amirm]

And also, does the FFT gain affect just the bottom edge of the noise floor, or the noise peaks?

Spikes only occupy one bit or so as a result, they are not subject to FFT gain and can be read as is. I do this all the time in my analysis of FFT.

 

[sabristol]

I also have a similar setup- an older QuantAsylum QA-400.
I usually run 32K points, and Hann windowing because.... actually I have no idea. Seems like a happy median setting. No idea how to compute the FFT gain. And also, does the FFT gain affect just the bottom edge of the noise floor, or the noise peaks? Maybe Matt from QuantAsylum can make some videos like this for his product? hint hint
I have to make an unsolicited and unaffiliated plug for him btw- it's a great product and tremendous value.

I change the FFT size depending on what I'm looking for. Usually start small as it gives a quicker response.

We could do the old one-two manoeuvre. I'll post a link to @amirm's video on QA's forum and you can reply asking for Matt to do some tutorials that follow on

 

[ted92]

Interesting video although a bit technical.
But we would not be here if we believed audio is not science.
Thanks Amir for helping us to better understand.

 

[Krobar]

Thanks for this video, was both interesting and informing for me.

 

[Mulder]

This video was a little bit harder to follow than the others. At least for me. For me this had probably been easier to read about. I understand that resolution makes different result, but then what is the proper resolution and what is used in the reviews?

 

[DualTriode]

@amirm,



Thanks for the peek behind the green curtain.

Yes please share move behind the scenes views.

I would like for you to shine more light on the idea of noise as it is measured in SNR, SINAD and THD+N. Do you like A-Weight filters?



For typical hard wired connections room noise and instrument fan noise are excluded from the measurements. For headphone measurements what do you do to minimize environmental noise like HVAC fan noise and noise from the cooling fan of your APx555.

I understand that closed back headphones isolate much of the environmental noise in the lab. I also know that folks like Head-Fi use a sound isolation chamber sitting on a 1000 pound plus vibration isolation base.

In my headphone tests with the 45CA I have found that moving the 45CA 20 feet away from the APx555 cooling fan with a couple of closed doors in between helps a lot.



Thanks DT

 

[sabristol]

I change the FFT size depending on what I'm looking for. Usually start small as it gives a quicker response.

We could do the old one-two manoeuvre. I'll post a link to @amirm's video on QA's forum and you can reply asking for Matt to do some tutorials that follow on

Posted on the QuantAsylum Forum.

 

[estuardo4]

Thank you @amirm . This is the first video of your channel that I watched from the beginning to the end, paused and rewatched. Most impressive.

I know you have written posts explaining FFT, but seeing in video is much better.

I don't like the debunking videos because you sound smirk at them (sorry for pointing out this). I don't like videos of smirk people laughing at my stupidity, so I guess many people won't be watching them or taking them seriously. I know it's not intentional, but I much rather prefer your teaching videos.

Thank you again for your useful video.

 

[Mnyb]

Nice extra video suggestions the algorithm vidoe kind of flew over my head a bit i view again . The 3 blue one brown video was easier to understand it brings the ball home to my previous line of work where i actually did FFT of the rpm of electrical drives (rotating systems) , the purpose there was to find mechanical issues like misaligned encoders and shafts or unbalanced rolls wheels whatnot or problems in gear boxes.

 

[NoSnakeOil2]

Thanks, Amir! Very clear explanation. Wish you had been my high school calculus teacher.)))) Hope you're feeling better.

 

[anchan]

I just experimented a bit with my setup, going form 2k points to 64k. I want to understand the math a bit later, but in super laymans terms, can I summarize what is happening as follows?

Increasing the number of points (bins) obviously improves the frequency resolution. So when the resolution increases, what once looked like a big hashy mess of noise, now looks more clarified into spikes. These peaks of the noise are still of the same height (dB), but the grass just thinned out is all. It doesn't fundamentally change anything except that the baseline of the noise floor is lowered. The hash just turned into more clarified frequency data.

 

[amirm]

You got it 100%.

 

[Chuck S]

More, please. Thank you.

 

[anchan]

You got it 100%.

Yay, thanks! Your video made me dive deeper and understand. So thank you for the video!

In a way one could say that with enough bandwidth, there is no more noise. Everything is understood

 

[SoundsGood2Me]

Thank You Amirm!

I've been reading John Atkinson's Stereophile tests for ~30 years now, and it, literally, took me 5 years to even begin to understand what he was talking about. Gradually, thru time, I could compare one graph from one amp, to the same graph from another, and began to understand his view.

NOW YOU TELL ME IT'S ALL ABOUT TESTING PARAMATERS?!?!!

Oy, my head hurts.

But I appreciate the truth, and your actual methodology helps me see what you have to do to get real numbers.

Does the whole Audio Community work off the same AP parameters / calibration?

Please KEEP IT UP!!!!

Thank You

 

[wpwoodjr]

@amirm you mentioned in the video that you could go between time and frequency domain and back again. I don't understand how this works; if the FFT is using 32k samples, then how do you stitch these back together into the time domain? Are consecutive input 32k samples converted into frequency domain and then consecutive 32k samples on output? Or do they overlap? Sorry if this is a dumb question.

 

[SIY]

Does the whole Audio Community work off the same AP parameters / calibration?

No, and the biggest problem is that they're often not disclosed.

Amir's policy (and mine as well) is that all parameters are disclosed and project files can be examined. In the couple cases that he's sent me his, our measurements under the same conditions track almost perfectly. I have full trust in his measurement quality and transparency.

 

[Ajax]

Hey Amir,

After many years purchasing HiFi equipment my belief, based on my own experiences and participating in audio forums, is that the biggest impairment to our hobby is bias.

It comes in many forms and causes otherwise reasonable people to behave irrationally. People who are now labelled as "subjective" are often chastised and ridiculed for their silliness, which is not helpful if our goal is to promote more transparent (read HiFi) equipment that is also good value.

I therefore feel that interviews with leading academics and ethical manufacturers that explain how psychoacoustic phenomenon works would be incredibly helpful. Armed with this information ASR members we could make greater inroads into converting the naysayers into a more rational approach and disperse some of the myths propagated by the not so ethical manufacturers.

If we want our hobby to attract the next generation(s) I believe it is time to stop ridiculing the subjectives and provide some well thought out reasoning as to why they "believe" they hear what they hear.

 

[milosz]

Maybe this has already been answered and I just didn't catch it - what are the differences in using a Hamming vs a Blackman vs Nutall window? What advantages would each convey is various situations? Seems this has something to do with not losing transient events in a longer FFT, but FA if I can figure out how this could be.

And- why do we call it FAST Fourier Transform? Is there also a use for a SLOW transform?