I understand Angelo's method cos I came up with it independently circa 1995. Angelo has papers explaining it from several viewpoints. In the presence of noise (ALL environments have noise), his method achieves a given accuracy in the theoretically shortest possible time.
But I still don't see how Mike's method does this and especially how it does THD.
Very well-written and illustrative documentation about FSAF is available in 4 parts here (as PDFs along with other files) if anyone wants to understand more about it: https://www.mathworks.com/matlabcentral/fileexchange/83363-fast-subband-adaptive-filtering-fsaf
(200+ pages to read)
It is unfortunate that Mike chose to leave this forum..
(200+ pages to read)
It is unfortunate that Mike chose to leave this forum..
Member
Joined 2003
FSAF is great stuff, it's been a long time since we've seen any innovation in measurement and analysis techniques for distortion evaluation. Special thanks for Mike Tsiroulnikov and @JohnPM for bringing this to the public domain for all to use. I am still working on comprehension of the results, I think it will take some time to allow for comparison of many speakers under similar conditions to fully understand it. However, without any comprehension of the charts and graphs, it is amazing to be able to test with real world dynamic signal such as music tracks, and separate stimulus and residual distortion + noise into a separate file that can be listened to. This functionality alone allows people to better correlate the charts and graphs to real world audible effects. Side effects for REW is that it now allows for impulse measurement with pink noise, so you don't have to listen to that annoying sine sweep for an impulse measurement anymore.
Where are the 200+ pages? Is it in the *.m files? Do we need MATLAB to read them? Where are the PDFs?
So under which circumstance would we see a difference between the good ol' sweep and a misc stimuli... after all, music contains a lot more jolts and transients....
If they always match as well as we have seen so far, whats the point? 😎
//
For example, to prove to some people that music is a stationary signal. That is, all laws applicable to linear circuits apply to music.
Well, the most interesting thing for me is that now you can find out the frequency response of the studio equipment on which the music was made.
Last edited:
Thanks tktran303. But that's just measuring response. You can do that with ANY wideband signal ... even music as mark100 said in #9. I could do that party trick even in da 90s so I'm not surprised that Live Sound people now do this.
But measuring THD is different. The signal needs to have well defined properties otherwise you can't distinguish between THD and a slightly different impulse response IR. Loadsa B&K papers dating back to the 70s on the subject. That's why I'm wading through mikets42's stuff, chock full of buzzwords & obfuscating TLAs, to see if his claims hold water.
I've loadsa caveats but has anyone done THD with Angelo's method and compared them with THD using FSAF with the same length of signal? What level do you assume you are operating the speaker? Actually you can get around that by using one of Angelo's sweeps in FSAF. But let's try also using FSAF+pink noise.
Tranh, you going to tell us which is the Red curve? My guess (and its only a guess as I dunno da parameters of da white noise signal) is it's the white noise measurement using FSAF
Last edited:
The strict answer is that Angelo's method (and I think FSAF also) is for weakly non-linear systems. ie we assume THD is 'small'.
If THD gets too big, eg when the signal clips, or there is too much noise, this assumption breaks down and we get big differences. Angelo's method is better than TDS and the 'noise' type systems like MLS (which can't measure THD at all and breaks down completely)
That was the point:
REW’a FSAF implementation can measure frequency response using a snippet of noise (white, pink or brown) or music, just as accurately as the traditional sine sweep, as demonstrated in the initial post.
What I will demonstrate is that FSAF will also allow you to hear the distortion residual.
Soon you be able to test AND listen to the residual Total Distortion + Noise of all manner of things:
From virtual audio cables to audio interfaces to various (eg. tube) amps, drivers from subwoofers to woofers to full rangers to tweeters, effects on cabinet bracing/damping, cone resonances, using your own favourite music clips.
Just measure and publish the results!
Caveat: the dynamic range of your measuring setup depends on the noise floor of your room and distortion of microphone.
The idea is not only can we measure distortion and visualise them on a graph. But we can also hear them
“…we subtract the result of filtration from the mic. The result is called "residual". Ideally, it should be just microphone noise... but it is not. It also contains distortions. When we listen to the residual alone, we can better understand what was bothering us.”
Measurement conditions are similar to what you would do for far-field measurements, but to minimise biases
- measure speaker driver at distance >2* cone diameter, or 2x baffle width for complete speaker.
It is not necessary to measure at 1m
- speaker should be well away from first reflections ie. usually close to midpoint of room between floor or ceiling
- When using readily available electret condenser microphones, many electret condenser microphones tend to exhibit their own distortion when observing around ~95dB. If you your microphone is at 31.6cm, this is equivalent to 85dB/1m.
- as quiet an environment as practically possible
Answer key
Red: FSAF white noise
Blue: sine sweep
Black: FSAF music
😊
Are these the 4 zip files in #1?
Can you tell us what the "puthands" files are supposed to be?
Member
Joined 2003
PutHands is the song...
Virtual Audio cable is software loopback
Other is loopback of Scarlett 2i2 interface
Other is loopback of Scarlett 2i2 interface and UcD400MP.
For the most part, the only difference between the 3 audio files that I see is the level of the noise floor and some very top end rolloff.
I am curious of the use of white noise however, perhaps it would make sense for an amplifier test, but audio spectrum more closely resembles pink noise, so a white noise test is sending a lot of excess energy to the upper frequencies. White noise would follow the power density of a linear sine sweep, while pink noise would follow an exponential sine sweep, however they are being displayed as equal here. There is some magic in FSAF, it seems to care not about the power density of the frequency spectrum of the measurement, and somehow determines correct frequency response in relation to pink noise / ESS spectrum regardless.
Virtual Audio cable is software loopback
Other is loopback of Scarlett 2i2 interface
Other is loopback of Scarlett 2i2 interface and UcD400MP.
For the most part, the only difference between the 3 audio files that I see is the level of the noise floor and some very top end rolloff.
I am curious of the use of white noise however, perhaps it would make sense for an amplifier test, but audio spectrum more closely resembles pink noise, so a white noise test is sending a lot of excess energy to the upper frequencies. White noise would follow the power density of a linear sine sweep, while pink noise would follow an exponential sine sweep, however they are being displayed as equal here. There is some magic in FSAF, it seems to care not about the power density of the frequency spectrum of the measurement, and somehow determines correct frequency response in relation to pink noise / ESS spectrum regardless.
Last edited:
If so, this I believe it could actually progress the industry and the SOTA.
Ground braking as I see it.
//
Last edited:
Interesting stuff.
I've been looking for a way to test what happens to speakers as you turn the levels up and up and up. Sine sweeps subject HF drivers to sustained high power, which isn't present in music, so I've burnt a few tweeters (don't worry, nothing expensive or interesting) when testing like that.
Using real-world test signals will be much better, since HF stuff tends to be more transient.
Meyer Sound has devised their own test signal for this stuff: M-Noise. It's similar to pink noise, but has a crest factor that more closely matches music.
I'll probably use that.
Can't wait to get some testing done with this.
Chris
PS - Being able to listen to the excess distortion and other rubbish is, I believe, SOTA. Well done to those involved in the development.
I've been looking for a way to test what happens to speakers as you turn the levels up and up and up. Sine sweeps subject HF drivers to sustained high power, which isn't present in music, so I've burnt a few tweeters (don't worry, nothing expensive or interesting) when testing like that.
Using real-world test signals will be much better, since HF stuff tends to be more transient.
Meyer Sound has devised their own test signal for this stuff: M-Noise. It's similar to pink noise, but has a crest factor that more closely matches music.
I'll probably use that.
Can't wait to get some testing done with this.
Chris
PS - Being able to listen to the excess distortion and other rubbish is, I believe, SOTA. Well done to those involved in the development.
Absolutely fantastic. I think this is the biggest invention I have experienced in the field.
The noise build up in Scarlett (subtle) but you start to hear the music traces in the Hypex one... 36 dB amplified but still...
//
I noticed that the last listed implementation mentioned in the first post is grayed out, or is it only my browser?
If not, any reason why?
If not, any reason why?