FSAF (Fast subband adaptive filtering) measurement

[dcibel]

Whether I use the measurement Sweep, or FSAF, using a file-saved ESS stimulus, or a FSAF with ESS stimulus, this time saved at +20dB louder than the previous setting- I get the same frequency response:

View attachment 76468

I'm not sure what you've done here, but the frequency response is first indication that something is not right.

FWIW I exported two sine sweep measurement files from REW's generator, one at -12dBFS, the other at -3dBFS. Ran each through FSAF and got identical results. The Frequency response I displayed above, and now here's the noise floor:

 

[dcibel]

Looking at the noise floors shows something very different between the three:
View attachment 76467

Noise floor is shown in dBr, so it's just SNR being shown. Noise level is constant, but level is different.

If the boosted stimulus was actually LOUDER than the original, we'd expect more distortion. In this case, it is less TD+N.
View attachment 76466

My conclusion is that the level of the noise floor is affected the calculated TD+N. But all 3 recordings were done in the same session (after midnight

Look at the noise measurement above, you only have 40dB SNR on the red trace, it dominates the FSAF result.

@John Mulcahy would it make sense to include some level normalization for file playback in FSAF, to ensure files are played around the same RMS level?

Can I also ask why this preview shows a level just above 0 for a pink noise file with actual level set to -12dBFS (-3dBFS pk)? And why the preview shows any level >0?

 

[John Mulcahy]

would it make sense to include some level normalization for file playback in FSAF, to ensure files are played around the same RMS level?

REW will apply up to 17 dB of gain to produce file data with a nominal 0 dB rms level, after which the chosen replay gain is applied.

 

[John Mulcahy]

Whether I use the measurement Sweep, or FSAF, using a file-saved ESS stimulus, or a FSAF with ESS stimulus, this time saved at +20dB louder than the previous setting- I get the same frequency response:

If you want a different measurement level use the level control, that's what it's for.

Looking at the noise floors shows something very different between the three

With a dBr (or percent) axis setting the noise is being shown relative to the level of the linear part of the response. Use an absolute axis unit instead of a relative one.

 

[dcibel]

REW will apply up to 17 dB of gain to produce file data with a nominal 0 dB rms level, after which the chosen replay gain is applied.

Perfect, thank you. So as long as you don't do something silly @Tikkidy like use a file with level of -22 or -44dBFS, it will be normalized.

To avoid confusion, it would be great for the level shown on the little preview chart to be the same if the file is indeed normalized to 0dB. Below is the same signal from REW's generator, just at -12dBFS and -3dBFS levels.

 

[John Mulcahy]

Help me understand the amplitude levels in FSAF measurement and distortion tab.

The impulse response is a transfer function. Changing the measurement level doesn't change the transfer function until the system reaches either its saturation threshold at the high end or the noise floor at the low end. Within the linear range of the system measuring at different levels produces the same transfer function and so the same frequency response. Presenting the result in that way is quite counter-intuitive for people, however, so REW offsets the frequency response to account for the measurement level.

For sweep measurements the distortion graph shows how the transfer function's harmonic distortion relates to its linear response (labelled fundamental). That fundamental is the same as the frequency response shown on the SPL & Phase graph, perhaps barring cal file effects depending on the preference setting for whether distortion results include cal.

For FSAF measurements the situation is quite different. Any stimulus is allowed, and the stimulus may have more energy at some frequencies than others, important information to interpret the TD+N result, which is the difference between the system's total output and its linear output. The fundamental on the distortion graph shows the power spectrum of the linear part of the system's response to the stimulus. To maintain some commonality with the presentation of sweep results, that spectrum is shown at the SPL corresponding to the measurement level. For example, if white noise were the stimulus and the noise was playing at 80 dB SPL the power spectrum would sit at 80 dB SPL. That isn't the true spectrum of course, since 80 dB is the total SPL of the signal, the spectrum of that signal from a 64k FFT would sit at about 35 dB SPL. That would, again, be rather counter-intuitive, hence the chosen presentation.

 

[dcibel]

Thanks for the explanation, I will take some time to review and digest it.

 

[Tikkidy]

Part of my curiosity in FSAF was to see how a speaker might sound using the FSAF residual), as opposed to how it might appear appear (harmonics, IMD etc), and is there a relationship/correlation?

Essentially, FSAF would be a suitable method of hearing how it may sound (residual). However, it's far from straightforward.

If we are taking measurements of a driver with an exponential sine sweep/chirp, there are some special techniques that we can take to reduce/reject background noise. Like increase the length to 512K for indoors ~31.6cm, or 1024K for outdoors 4pi @1m... or 4 repetitions, which really helps with transient noises that can otherwise kill the measurement.

With FSAF, with a music track of 15 seconds, it captures ALL the noise AND room reflections. Recently we had rain, thunder, frogs, traffic, children, all audible when listening to the residual with the gain turned up 36dB. But in your living room, listening to your music, this previously never distracted you.

This will make A to B comparisons between different people making measurements challenging.

The room gets in the way, yet again (in the measurement and the listening)

 

[dcibel]

Yes, measure in short distance and with a good mic, and in complete silence, it's not something easily achieved by everyone, including myself. Comparisons of your own speakers in your own home under similar conditions however can be useful and enlightening. I've learned a fair bit through the use of FSAF, however not the most enlightening revelations other than the reality of the sheer magnitude of modulated distortion products that bass distortion produces in a driver. It has also provided confirmation and validation of what I was already hearing. Last weekend I set up 3 rather different speakers to compare and ran them through FSAF to listen to the residual results, and that comparison can simply make the "nuances" between speakers seem a bit more obvious and apparent.

 

[Tikkidy]

Here's another driver, this time I'm going to let it run "full range" ie. without any filters.

The equivalent Sweep...

Any errors or (mis)representations is the author's own fault.
(small office sized room of 3.6x3.2mx2.4) Listen at your own risk!

 

[FSAF]

Very interesting! So much hype and so discouraging results:-) ... in a 2-way implementation. Could you re-run this test with HPF on 50/100/150/200/250 Hz - to see if it may behave any better in a 3-way configuration?

 

[dcibel]

One anomaly in the residual.

 

[dcibel]

Here's a comparison I made last weekend. Tale of 3 very different speakers compared under similar conditions. These are complete speakers with crossovers. In order to reduce room interaction, measurements were completed at 315mm. Unfortunately this makes for some very skewed frequency response from the big 3-ways, so comparison is primarily mid range to high frequencies, however the "3-way" benefit still comes through very clearly in these comparisons.

To avoid a very busy post with many images, I will simply attach a link to download the MDAT file and residual audio so you can view and playback as you wish.

All tests completed with mic located between midrange and tweeter, and level set for ~85dB/1m. Tests were completed with M-Noise and QDust track, see notes for specific filter configuration used. Background noise may include a refrigerator compressor from my kitchen or other random noises, sorry but you get what you get here.

Equipment used:
Motu M4
Line Audio Omni1
Boxem NC252MP amplifier (nCore OEM 250W/ch)

First speaker - "The Northern Pikes"
Woofer: Vifa M26WR09-08 woofer, with CSS APR12 passive radiator.
Midrange: Wavecor WF120BD03
Tweeter: Dayton AMT3-4 with my own personal modification to improve it's high order harmonic performance significantly.
Crossover: 500Hz and 2700Hz

Second Speaker - "Disco of the Floor"
Woofer: Scan-Speak Discovery 22W/8534 in a tapered transmission line enclosure.
Midrange: Eton 5-880-25Hex
Tweeter: Vifa XT25TG30 in a Visaton WG148 Waveguide
Crossover: 400Hz and 2kHz

Third Speaker - "Atari"
Woofer: Satori MW16P in a bass reflex enclosure.
Tweeter: Melodavid Be28 in a custom waveguide CNC milled into the baffle.
Crossover: 2kHz

Note: ESS via FSAF measurement for the Atari speaker did not have a BU12 high pass, while the other 3-way speakers did. My mistake, but the point of ESS measurement was to provide point of reference through the midrange anyway, and to confirm that FSAF provides a similar THD result to REW's sine sweep measurement given the same stimulus.

For comparing ESS vs ESS via FSAF, make sure "plot harmonics at harmonic frequency" is selected.

Download and compare here:

FSAF comparison Dec 8.zip

drive.google.com

Instructions: Copy residual files to the FSAF folder that you've configured in REW, then simply open the MDAT file and select "play residual" on the distortion tab.

Before you declare any of these results as "awful, horrible retched disasters", just know that my subjective opinion of all of these speakers is that they all are very enjoyable speakers. One may be more detailed and resolving than another, but none are "bad".

 

[Tikkidy]

One anomaly in the residual. View attachment 76630

Sure am glad this forum had editable features.

FSAF has forensic level of recording.

ESS vs FSAF + IBS,

Winner winner chicken dinner!

 

[FSAF]

@dcibel : the names of the files in the folder and the names of the files referenced in *.mdat do not agree. Could you specify explicitly which is which?

BTW, res-e9fe4ecc-3d42-43bb-a6c3-166fc8ce9e2a.wav is pretty bad.

 

[dcibel]

They are the files REW creates on measurement. Open the MDAT, select the measurement, go to measurement info and look down at the bottom where it says "UUID", that will point you to the corresponding file. You will find the file you referenced in the Qdust track for the Atari speaker, with 20Hz HP, or effectively full range for that one. File "res-f19f3977-3b1b-460e-bcea-213233af821e" will be the same speaker and audio with 200Hz BU12 HP.

Or follow instructions to play back directly via REW

 

[FSAF]

Yes, you are correct - I used an old beta. After I downloaded the latest, the problem disappeared.

 

[Tikkidy]

my quick and dirty summary (learnings so far)

FSAF for measuring speakers

-is highly dependent on room/environment- can vary between recording sessions (more than DAC/ADC/microphone)
-The two 3-way speakers are superior to one 2-way speaker
-M-noise is just punishing for 2-way speaker
- I can see the difference in harmonics- but I can't hear the difference, as least not through casual listening to the FSAF residual.


Anything to add? @dcibel

Not sure how we use this information to design "better" speakers...

 

[dcibel]

My commentary is simply that if you are looking for some validation of an opinion that 3-ways are simply "better", this is it.

FSAF has highlighted some of the obvious listening notes I already had on the speakers, but I will provide some subjective impressions to go along with the measurement results.

Atari: Despite the bass presentation highlited by FSAF, I describe this speaker as "fantastic", with a clarity that I simply don't get from any other speaker I've built. No doubt a result of the high end low distortion drivers used and with a great implementation. This speaker paired with a subwoofer could be near perfect. You will notice especially in the 200Hz High passed result, the clarity through 200-1kHz range actually surpasses the other speakers here, that definitely comes through audibly in the midrange presentation.

Disco on the floor: These measurements may not highlight it well, but the bass performance of this speaker is excellent, I prefer it over the 10" Vifa M26 by a lot. The Eton midrange has a simple motor and is the highest 3rd order harmonics through the midrange. I don't find this speaker to be off-putting at all to listen to however, as there is consistency in the distortion throughout the frequency spectrum. It's swings in distortion that appear to be more problematic in listening, where some frequencies will just stand out an create harshness or sibilance, etc. FSAF result of this speaker shows what the added 3rd order distortion does to the presentation, perhaps a bit more "nasal" or honky sound is present that shouldn't be there.

Northern Pikes: Great speaker all around, this speaker was a real challenge to get the crossover balance right with the woofer right at floor level. A close second to the Atari subjectively, however it's problems are more in linear performance than non-linear, with the floor level woofer and large AMT rolls off the top end a little early so it's just missing the lively treble that ther Atari speaker provides.

 

[Tikkidy]

I am not looking for validation, but I am making observations based on listening to the our (limited) FSAF residual data.

These findings may or may not be consistent with other designer's opinions. More than one designer has stated that "a real woofer playing is better than a 2 way + subwoofer". Does our limited alpha/beta testing of FSAF support this, or not? I have several of "yet another boring 6.5" 2-way speaker" that I have enjoyed...

Another designer has said (s) he couldn't hear any differences with cabinets with curved edges and straight edges, which I thought was surprising. Yet, other designers have said that they prefer NOT to use waveguides. It's not clear whether FSAF can make the case to support or refute this. It seems to me. that recordings of direct sound (and reverb), and not able to exactly measure how the sound is projected into 3D space.

It's interesting that you describe the Atari speakers as "fantastic", which I can only assume that you are most impressed with this speaker. I would like to think that all the efforts of selecting "low distortion drivers and with great implementation" (of non-resonant cabinets, careful crossover design etc) has something (a lot) to do with this. Otherwise we could just cobble together a 3-way with curbside scrap speakers and off-the-shelf textbook crossover and call it a day.

The subjective comments from loudspeaker designers are what I'm listening to these days, people like yourself, Kimmo, Dennis etc. Rather than the reviewers who are listening to various speakers in their rooms. My personal bias is that I believe that the room imparts a fingerprint on the sound. Again this is only low level anecdotal evidence from a person (me) who has moved 7 times in 10 years... and one of the reasons I prefer not to give commentary on how something I design sounds (it's conflated with my (current) room)

The wonderful mysteries continue...

PS. Gorgeous cabinets, btw! If nothing else, building cabinets is one of my first loves...

 

[dcibel]

I am not looking for validation, but I am making observations based on listening to the our (limited) FSAF residual data.

These findings may or may not be consistent with other designer's opinions. More than one designer has stated that "a real woofer playing is better than a 2 way + subwoofer". Does our limited alpha/beta testing of FSAF support this, or not? I have several of "yet another boring 6.5" 2-way speaker" that I have enjoy...

I won't speak for "other designers", but my though process for a 3-way has been to keep the midrange operating above the baffle step. Specific frequency will vary depending on the cabinet, but usually around 400Hz for a tower speaker. This reduces the power and excursion requirement significantly, which a small 12cm driver like WF120 doesn't have a lot of, so make the woofer do as much of the heavy lifting as possible.

You won't get a 12cm driver to play at any high SPL comfortably to cross to a subwoofer at 100Hz range, but a 18cm driver is a lot more capable. Even so, FSAF will show there is still a benefit to use a crossover at 300-500Hz over 90-120Hz.

Another designer has said (s) he couldn't hear any differences with cabinets with curved edged and straight edges, which I thought was surprising. Yet, other designers have said that they prefer NOT to use waveguides. It's not clear whether FSAF can make the case to support or deny this. It seems to me recordings of direct sound (and reverb), and not able to exactly measure how the sound is projected into 3D space.

As always, take opinions of DIY amateurs with a grain of salt. Problems you are describing are primarily linear distortions, may not be easily identified via FSAF. "Reverb" requires longer delays and decay time than what occurs with edge diffraction.

I do have another 3-way tower that is without waveguide and with sharp corners. Perhaps over Xmas break I'll lug it upstairs and measure it to be compared with the others, but don't expect much to be learned simply because of the lack of waveguide, it's different in many ways so hard to make any conclusions based on any differences observed.

It's interesting that you describe the Atari speakers as "fantastic", which I can only assume that you are most impressed with this speaker. I would like to think that all the efforts of selecting "low distortion drivers and with great implementation" (of non-resonant cabinets, careful crossover design etc) has something (a lot) to do with this. Otherwise we could just cobble together a 3-way with curbside scrap speakers and off-the-shelf textbook crossover and call it a day.

There's more to excellent sound than raw distortion performance. There's a fine balancing act of many aspects to achieve excellent power & DI in addition to low distortion. One could argue that power response is of greater importance than distortion, perhaps one reason why I enjoy the Disco/Eton/Vifa combo so much despite its "okay" distortion results.

The subjective comments from loudspeaker designers are what I'm listening to these days, people like yourself, Kimmo, Dennis etc. Rather than the reviewers who are listening to various speakers in their rooms. My personal bias is that I believe that the room imparts a fingerprint on the sound. Again this is only low level anecdotal evidence from a person (me) who has moved 7 times in 10 years... and one of the reasons I prefer not to give commentary on how something I design sounds (it's conflated with my (current) room)

The wonderful mysteries continue...

PS. Gorgeous cabinets, btw! If nothing else, building cabinets is one of my first loves...

I am lucky to have a pretty good room to listen in, where I can get a good 5m between myself and the speakers, carpeted and lots of plants. The carpet won't last much longer though, it will be wood eventually and I'll need some thick rug and tapestries to keep the room reverb to a minimum.

...and thanks for the kind comments on my cabinets. I'm an amateur woodworker at best, I swear.

 

[Tikkidy]

Measuring bass is challenging, particularly with another radiator eg. port or passive radiator in close proximity with a midwoofer.
Numerous methods, including near field, ground plane, or anechoic chamber all have their own challenges.

Furthermore, I can hear birds singing, frogs croaking, or crickets chirping due to the forensic level recording ability of FSAF.
These transient noises can also contaminate ESS. but they can very effectively discarded with averaging multiple shorter sweeps.
eg. 256K samples x 4 repetitions vs 1M samples.

Are there any special techniques we need to take into practice when measuring complete transducers or speakers with FSAF?

 

[dcibel]

Move to a colder climate - much less birds, frogs and crickets in the dead of winter

 

[Tikkidy]

I was going to take some in-the-middle-of-night trax for Mike. But then...

 

[Tikkidy]

@dcibel

Moving somewhere that’s less prone to hurricanes, war planes doing military drills and hotted up cars from C20 roaring down the street is in the 5 year plan.

In the meantime, how can we use FSAF to see if the residual is not coming from the (rear mounted) passive radiator? The box? The room?

If the answer is to build a better enclosure Eg. sealed box, another project…



The ESS has some advantages in that multiple sweeps can reject plosives
I’m afraid FSAF will be misused and abused

Any experience with windsocks?