You obviously haven't read about the sweep sine method developed by Angelo Farina in the early 2000 and which is used in REW 🙂. The Sweep sine technique used by REW allows to separate the linear and nonlinear part of impulse response and eliminate the effect of loudspeaker nonlinearity. What you see in the IR window in REW is the linear part of IR and from a mathematical point of view it is completely reliable.
Was that really "developed" by him?
It's kinda obvious way to be perfectly honest.
Swept since methods have been used for many many decades, knowing how to pull a IR from there as well and how to get distortion from there too?
It's just that all of that was a bit much for most computers in the 80s and early 90s, but the idea that it could be done was well known.
At least that's what I always learned from my (digital) data acquisition books.
On the my knowledge, Angelo Farina first suggested to use the nonlinear convolution technique to separate linear and nonlinear part of DUT, for this purpose he used exponentialy sweeped sine (as far as I remember). The key point in his proposal was to use the nonlinear convolution.
Interesting, but even without convolution techniques, one can dissect the distortion from it?
I mean, don't get me wrong doing it his way is a lot more elegant obviously! 🙂
I've been under the impression that Farina realized if using a log sine sweep, mathematical arrivals before the impulse peak provide the levels of harmonic distortion.
Maybe that was harder to glean, or mathematically compute from using linear sweeps, which I think were more prevalent......dunno..
Or maybe I'm plain barking up a tree....
What comes out of your speakers and goes into the microphone is totally irrelevant.
Fact is that we can make an IR from it, a step response, or do other stuff.
It's all exactly the same thing, just a different representation.
A MLS signal, pink noise or even a pulse could also be used instead of a since wave.
Problem is that the SNR from those signals is total garbage.
And a step response is not usable for speakers.
So again, I am totally missing what you're trying to say here?
I was talking about system errors, meaning how a system behaves.
What has that to do with data presentation errors?
I also have to get back to his findings to know all the ins- and outs.
I don't see why a linear sweep would be problematic, since convolution only cares about the signals coming in?
A log sweep only goes much faster obviously and as far as I know they have been doing log sweeps also for many decades?
Cross-correlation and other techniques can also be used for example.
Last edited:
I have to admit I never pay much attention to driver details like this...I always try get drivers with the greatest BL2/Re i can and leave it at that , regarding BL.
Is the efficiency gain in the knee area simply due to the motor being unable to, and therefore unasked to, completely control the mass?
Iow, underdamped by low BL?
Btw. Diyaudio member mikets42 developed a very interesting and promising measurement technique based on some kind of adaptive filtering that allows filter out nonlinear distortions from a real musical signal. I did not go deep into the technique, those interested can read the thread
https://www.diyaudio.com/community/...ns-measurement-in-matlab.388032/#post-7066888
https://www.diyaudio.com/community/...ns-measurement-in-matlab.388032/#post-7066888
To be perfectly honest, I actually don't know.
Mostly because I don't care to much about the why in this case, haha 😀
Fact is just that BL shifts that line up, very similar to like an amplifier and its gain/sensitivity vs low end frequency response.
So I guess if we take the equivalent circuit again, we can find our answer in there, since the BL is nothing more than a ratio.
I don't know how that will end up in math, but I bet you will find your answer there.
But again, for me this is something like I know it's there, but the question why, isn't really that important for any practical sense.
If that makes any sense? 🙂
For subwoofers, I don't really look at parameters that much anymore.
Just needs to have plenty of cone excursion, low Fs, power and a Qt that is low so we don't need an enormous cabinet.
Fact is that you have to just simulate things and within 30 seconds you know how it will perform in a specific cabinet with a specific target curve and SPL etc.
That's easy, if something isn't audible, then it is "good enough".
He was the first, although similar techniques had been used in underwater sonar, but he was the first to suggest their use in room acoustics.
🤷♂️ Apparently not, because you devoted an entire paper on it, and it still seems to be open for debate?
If we can't quantify our constraints, it's also not possible to determine if a system is linear enough or not.
All we have are some vague (subjective) ideas.
So unfortunately also not easy....