REW (V5.40 beta) by @JohnPM can make loudspeaker measurements using the Fast Subband Adaptive Filtering process, developed by Michael Tsiroulnikov @mikets42
This FSAF implementation can use noise (white, pink, brown) or a segment of a user-supplied file e.g. MUSIC (!), to measure the frequency response, as well as total distortion (harmonic + intermodulation) + noise.
Reference:
https://www.roomeqwizard.com/betahelp/help_en-GB/html/fsafmeasurement.html
Shall we test this out? Here is a picture of the setup:
Here is the amplitude response of this driver in box (ie. without filters or crossover)
Frequency responses were measured by
A- logarithmic Sweep (aka chirp (devised ~2000 - link: AES)
see attached .ZIP file for "1 M Measure sweep 0Hz to 22050Hz"
B - white noise via FSAF, devised 2020- link: MATLAB Central File Exchange
see attached .ZIP file- " White noise -12dBFS"
C - "Put Your Hands" music clip via FSAF. This music clip is courtesy of Dr Stéphane Pigeon, a signal processing engineer. You can listen to the clip here - scroll down to near the bottom of the page to section and click on Play/Pause button:
EDIT 19/10/2024
Distortion residuals added
TBC - the graphs, and sounds of distortion from this speaker, and other speakers...
This FSAF implementation can use noise (white, pink, brown) or a segment of a user-supplied file e.g. MUSIC (!), to measure the frequency response, as well as total distortion (harmonic + intermodulation) + noise.
Reference:
https://www.roomeqwizard.com/betahelp/help_en-GB/html/fsafmeasurement.html
Shall we test this out? Here is a picture of the setup:
Here is the amplitude response of this driver in box (ie. without filters or crossover)
Frequency responses were measured by
A- logarithmic Sweep (aka chirp (devised ~2000 - link: AES)
see attached .ZIP file for "1 M Measure sweep 0Hz to 22050Hz"
B - white noise via FSAF, devised 2020- link: MATLAB Central File Exchange
see attached .ZIP file- " White noise -12dBFS"
C - "Put Your Hands" music clip via FSAF. This music clip is courtesy of Dr Stéphane Pigeon, a signal processing engineer. You can listen to the clip here - scroll down to near the bottom of the page to section and click on Play/Pause button:
EDIT 19/10/2024
Distortion residuals added
- Virtual audio cable (digital internal loopback)
- Focusrite Scarlett 2i2(3rd gen) (line out to line in)
- *Scarlett 2i2(3rd gen) line out -> Hypex UcD400MP -> Scarlett 2i2(3rd gen) line in
*baseline
TBC - the graphs, and sounds of distortion from this speaker, and other speakers...
Last edited:
I for sure find it fascinating to use the intended signal (music) for such measurements. A colleague once made an analogy that the purity of the sine wave used (in such measurements) is like the focus on a microscope. Now we can do it using a signal more appropriate to the goal of the thing being measured, versus the measurement.
Cant wait to try it on a guitar, where I drive a signal from a Dayton DAEX25FHE exciter into the bridge. Apparently now I can use "guitar sound" as the driving signal, instead of pink or a sine sweep. Now if I could only play well, I imagine testing would be less bothersome for those nearby!
Cant wait to try it on a guitar, where I drive a signal from a Dayton DAEX25FHE exciter into the bridge. Apparently now I can use "guitar sound" as the driving signal, instead of pink or a sine sweep. Now if I could only play well, I imagine testing would be less bothersome for those nearby!
Smoke on the water spinoramas?🙂
If music can be the source the measurement procedure could likely just tap into what ever is playing and just pop out the response and no one ever noticed a measurement happened? Add in automatic EQ and walk around the room with mic in the hand would be fun experiment 🙂
If music can be the source the measurement procedure could likely just tap into what ever is playing and just pop out the response and no one ever noticed a measurement happened? Add in automatic EQ and walk around the room with mic in the hand would be fun experiment 🙂
Live sound routinely measures transfer functions with music in real time, and doesn't even require a file for comparison.
(Using dual-channel, reference vs measurement.)
One requirement for accuracy is that the music be full-range, and stimulate the entire spectrum. A well known 'trick' is to mix inaudible pink with the music.
For FSAF, I imagine the need for full-range with whatever music file is used, holds for it too.
It's really cool how FSAF can also measure distortion.
You mean like this ?
Unfortunately not. My post grad studies have taken me far from away from maths- I would be what you would call an EE drop-out (left after yr2). Just talking about Matlab gives me goosebumps.
I’m just a man with a mic.
All I can offer is recordings of distortion of various drivers/speakers. I shall measure and publish graphs and audio files to listen and compare.
Perhaps I’ll revisit that SB12PFC-4 that @mikets42 tested in (standard) voltage drive mode…
If anyone has his email, can you tell him Thanh from Australia wants to know if he can come back and play? As @kgrlee says, our single brain cell does need time to adjust to what he’s writing/saying….
Last edited:
sorry Jan, I’d need to use Regularized Least Squares. And one needs at least a PhD to play with Volterra series.
Someone in BC needs to send some chocolates and/or flowers and ask Michael to come back to play… of course I would say that because I am of the arty/farty persuasion.
A man of numbers might respond better to a petition.
@DcibeL ?
Someone in BC needs to send some chocolates and/or flowers and ask Michael to come back to play… of course I would say that because I am of the arty/farty persuasion.
A man of numbers might respond better to a petition.
@DcibeL ?
There are 3 curves in 3 colours in the attached graph. Two are very similar but the Red curve is different below 200Hz. Which of A, B or C is the Red curve?
What were the parameters for Angelo's Sweep? How long? You can't do a 0Hz to 22050Hz Farina sweep. Has to start at non-zero.
I'm obviously missing something. Do the comments refer to another page? https://www.roomeqwizard.com/betahelp/help_en-GB/html/fsafmeasurement.html doesn't seem to be it. ??!!
Last edited:
@WhiteDragon Sorry, it's a trick question.
The (Farina) log sweep is just the usual way one would measure the frequency response and harmonic distortion of a driver (or complete speaker)
This is what ARTA, Klippel, have been doing.
The FSAF measurement process does the same- it can measure the frequency response, and distortion of a driver, or a complete speaker.
However, in addition, you can HEAR the residual (difference between input and output)
Perhaps you need to do it yourself. Or I need to create a YouTube video to explain...
@kgrlee
If you are not familiar with the REW software package, that page might throw you in the deep end... The frequency response was measured with a 1M sample sweep from 0 to 22,050Hz, with a sample rate of 44,100Hz. The audio interface is the Focusrite Scarlett 2i2(3rd gen), amplifier Hypex UcD400MP with drive level 2.83V, and the B&K 4191+2669 microphone @ 31.6cm.
The (Farina) log sweep is just the usual way one would measure the frequency response and harmonic distortion of a driver (or complete speaker)
This is what ARTA, Klippel, have been doing.
The FSAF measurement process does the same- it can measure the frequency response, and distortion of a driver, or a complete speaker.
However, in addition, you can HEAR the residual (difference between input and output)
Perhaps you need to do it yourself. Or I need to create a YouTube video to explain...
@kgrlee
If you are not familiar with the REW software package, that page might throw you in the deep end... The frequency response was measured with a 1M sample sweep from 0 to 22,050Hz, with a sample rate of 44,100Hz. The audio interface is the Focusrite Scarlett 2i2(3rd gen), amplifier Hypex UcD400MP with drive level 2.83V, and the B&K 4191+2669 microphone @ 31.6cm.
I'm a dinosaur whose DSP stuff is sorta 1980s & 90s. Anyone got some good public domain links to Subband Filtering?
The best I can grok so far is that it is a way of doing long FIRs without HUGE latency by breaking the spectrum into subbands. But how do you do the Filter Banks? FIR, IIR, black magic?? How do you recombine?
The best I can grok so far is that it is a way of doing long FIRs without HUGE latency by breaking the spectrum into subbands. But how do you do the Filter Banks? FIR, IIR, black magic?? How do you recombine?