• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

How to make quasi-anechoic speaker measurements/spinoramas with REW and VituixCAD

OWC

Member
Joined
Apr 14, 2019
Messages
26
Likes
8
Audio Precision also had an excellent read on this matter.
Also REALLY check their references, some very interesting papers to read.
(just omit the plug of their own products)

In general I would advice for ARTA instead of REW.
There are some essentials, like a Burst Decay (= waterfall diagram but in periods) that can't be done with REW

The most important thing I would like to add, is that I still see a lot of people use a rectangular time window.
This type of window is completely useless for this kind of measurements.
One should use something like a Hann (25%) window or something.
With enough clearance, I am able to measure reliably from around 180-250Hz.
That is inside the living room.

In most cases this gives more than enough information to use the stitching method
 

Attachments

  • AudioPrecision-AppNote-LoudspeakerEA Measurements.pdf
    1.9 MB · Views: 67

fluid

Active Member
Joined
Apr 19, 2021
Messages
243
Likes
382
With enough clearance, I am able to measure reliably from around 180-250Hz.
That is inside the living room.

In most cases this gives more than enough information to use the stitching method
It might give enough data to stitch a nearfield response but the resolution gets progressively worse as frequency goes down which is included in the Audio Precision document (which is very good).

Even with a 10ms gate time narrow band resonances that are in the lower end of the frequency range will be smoothed out due to the resolution of the measurement.

This image from Floyd Toole's Sound Reproduction shows an anechoic vs 10ms gate. If the gate was 5ms which is typical for a domestic room then this effect would move up much closer to 1KHz.

Floyd Toole 18.9.png
 

OWC

Member
Joined
Apr 14, 2019
Messages
26
Likes
8
It might give enough data to stitch a nearfield response but the resolution gets progressively worse as frequency goes down which is included in the Audio Precision document (which is very good).

Even with a 10ms gate time narrow band resonances that are in the lower end of the frequency range will be smoothed out due to the resolution of the measurement.

This image from Floyd Toole's Sound Reproduction shows an anechoic vs 10ms gate. If the gate was 5ms which is typical for a domestic room then this effect would move up much closer to 1KHz.
Since we are doing some near-field measurements anyway, one could easily also have a look for any kind of resonances.
These can be either seen in the impedance curve, or in a so called Burst-Decay graph (waterfall shown in periods).
At these lower frequencies these will probably also show in the nearfield frequency response.

At these lower frequencies disturbances are basically always a result of some kind of internal resonance of the cabinet or something.
When designing or looking at at a loudspeaker, it's always wise to calculate (or get e good educational guess) were and what resonances to expect.

So in the end it's not an issue that some of these resonances/peaks/dips will be covered.
That is the whole reason to do a couple of different types of measurements.

Btw, I am familiar with that picture and example.
Keep in mind that even anechoic chambers aren't perfect.
So it could even be some kind of resonance created by something else.
I am absolutely not saying it's the case here, but a lot of people assume anechoic chambers are perfect.
They really aren't, if you measured in a couple, you often will see some issues.

In fact, these days I would even combine/stitch a gated measurement with an anechoic measurement.
 
Last edited:

fluid

Active Member
Joined
Apr 19, 2021
Messages
243
Likes
382
So in the end it's not an issue that some of these resonances/peaks/dips will be covered.
That is the whole reason to do a couple of different types of measurements.

Btw, I am familiar with that picture and example.
Keep in mind that even anechoic chambers aren't perfect.
There are many ways to get better insight into the data that is smoothed from the windowing like you mention. This thread is a guide on how to make quasi anechoic measurements and not all those reading will have the knowledge and experience to know how to perform those checks.
If it is an active speaker impedance is not easy to measure, if it is a commercial speaker it is not easy to know the internal cabinet structure.

So saying that smoothing the data is not an issue and doesn't matter needs more qualification. All measurement methods and locations have their own issues. What was missing from the screenshot was the accompanying text explaining how that resonance caused an audible issue that was missed during the design phase. They obviously did not perform a full suite of other measurements.

This is the section from the Audio Precision Document which covers it:

Limitations of Quasi-Anechoic Measurements
As mentioned above, quasi-anechoic techniques work by windowing out the portion of the impulse response containing room reflections. Sometimes, this windowing truncates the impulse response before it has decayed to a sufficiently negligible level. This truncation causes errors in the estimated frequency response, especially at low frequencies. For example, consider a situation in which the impulse response is windowed 5 ms after its peak value. This would be a typical value for a small (bookshelf-sized) speaker measured in a room with a 2.74 m (9 ft.) high ceiling. With a measurement distance of d = 1.0 m and the distance h = 1.36 m (4.45 ft) to the nearest reflecting surface, the first reflection would arrive about 5.5 ms after the direct sound. Hence the impulse response might be windowed to a length of 5.0 ms.

AP QA.png


Figure 8 illustrates the effect of windowing one loudspeaker’s measured impulse response at 5 ms after the peak. This data is from a measurement in a small acoustically lined hearing aid test box with an integrated full-range loudspeaker. Although this test setup would preclude an accurate measurement of this loudspeaker’s response at low frequencies, Figure 8 does show the dramatic impact of windowing the impulse response. The windowed response is radically different from the un-windowed response below about 300 Hz, and the smoothing effect of the window is apparent up to about 1 kHz. With a window width of T = 5 ms, the resolution in the frequency domain is about 1/T = 200 Hz. In the graph, the shaded region below 200 Hz is intended to indicate that the curve is unreliable in this region. But, note that the effect of windowing extends to a much higher frequency than 200 Hz.
 

OWC

Member
Joined
Apr 14, 2019
Messages
26
Likes
8
There are many ways to get better insight into the data that is smoothed from the windowing like you mention. This thread is a guide on how to make quasi anechoic measurements and not all those reading will have the knowledge and experience to know how to perform those checks.
If it is an active speaker impedance is not easy to measure, if it is a commercial speaker it is not easy to know the internal cabinet structure.

So saying that smoothing the data is not an issue and doesn't matter needs more qualification. All measurement methods and locations have their own issues. What was missing from the screenshot was the accompanying text explaining how that resonance caused an audible issue that was missed during the design phase. They obviously did not perform a full suite of other measurements.

This is the section from the Audio Precision Document which covers it:
Nearfield measurements can always be done and will reveal most of the issues.

When developing your own speakers, measuring impedance is just a must have, even for active systems, and an extremely easy circuit can be used for this.
You don't need to know exactly the internal structure, a discrepancy in the impedance will very often be result of some kind of resonance.
Once again, you can cross correlate that with the nearfield measurements.

Like you say as well, all measurements have their issues, even anechoic ones.
So once people start measuring things themselves, they must at least be aware of it.

Another good tip, is to simulate the speaker.
Or at least things like bafflestep etc (can be done with Edge or VituixCad these days).
Get a sense what to expect, so you also know what to look for.
They can always ask for help, advice or assistance.

But just measuring things without really knowing what you're doing not only results in disappointment, but also will result very often in wrong conclusions.
One only has to know a handful of things, what to look for and where they come from.
Most important is to be very consistent and systematic.
Take 5 minutes more, instead of doing things "quick and dirty"

Another good tip is to simulate the speaker.
Things like bafflestep etc. (Edge or VituixCad these days)
So you get a bit of an idea what to look for and what to expect.
These programs won't be 100% accurate, but will give decent ballpark figures.
 
Last edited:

fluid

Active Member
Joined
Apr 19, 2021
Messages
243
Likes
382
Most important is to be very consistent and systematic.
Take 5 minutes more, instead of doing things "quick and dirty"
Indeed, for myself I want more resolution with as long a gate as I can manage. I use all the techniques you mention along with BEM simulation.
 
OP
N

napilopez

Major Contributor
Forum Donor
Joined
Oct 17, 2018
Messages
1,923
Likes
7,290
Location
NYC
Audio Precision also had an excellent read on this matter.
Also REALLY check their references, some very interesting papers to read.
(just omit the plug of their own products)

In general I would advice for ARTA instead of REW.
There are some essentials, like a Burst Decay (= waterfall diagram but in periods) that can't be done with REW

The most important thing I would like to add, is that I still see a lot of people use a rectangular time window.
This type of window is completely useless for this kind of measurements.
One should use something like a Hann (25%) window or something.
With enough clearance, I am able to measure reliably from around 180-250Hz.
That is inside the living room.

In most cases this gives more than enough information to use the stitching method

Arta is a significantly less user friendly software imo, not the least because it costs money. It's hard enough to get people to buy a measurement mic, let alone have them pay for more software =]

As for burst decay, it just so happens this was added to REW today =] See version 5.20 RC15.
I should also note REW recently added a very nice feature for spinoramas: a timed multi-measurement mode. So you can measure, turn your turntable, and it'll automatically measure again. This'll make my life a lot easier. That's another reason I like REW: the update pace is relentless.

Also not sure what the issue with the rectangular window -- going for a 25% hann just yields less accurate results in my experience with no benefit other than making the data look prettier, which I don't consider a benefit. It correlates less with the anechoic data and I've matched quite a few speakers to anechoic results quite closely by now. But I know little about signal processing, so I'm happy to hear an explanation.

Btw, another simple, kind of brute force method I use to investigate resonances in the lower frequencies is to take a whole bunch of ungated measurements at different close-ish mic and speaker positions. If a bump is consistent throughout most of the measurements, it points to an audible resonance. It's a 'fuzzy' method but it's a quick and dirty way to check bigger resonances quickly. If designing a speaker, one would of course want to use further methods to investigate. I'm curious to try the burst decay in REW now.
 
Last edited:

dc655321

Major Contributor
Joined
Mar 4, 2018
Messages
1,003
Likes
1,219
Also not sure what the issue with the rectangular window -- going for a 25% hann just yields less accurate results in my experience with no benefit other than making the data look prettier, which I don't consider a benefit. It correlates less with the anechoic data and I've matched quite a few speakers to anechoic results quite closely by now. But I know little about signal processing, so I'm happy to hear an explanation.

Isn't the default gating window in REW a Tukey window (as in Tukey of FFT fame)? Those have a parameter to vary shape from rectangular at one extreme to Hann (raised cosine) at the other. The usual tradeoffs apply: amplitude vs. frequency resolution, leakage.
 

OWC

Member
Joined
Apr 14, 2019
Messages
26
Likes
8
Basically any "weighted" window is better than rectangular.

Rectangular windows are only useful for pure sine waves, or when there is a very big pulse with no or very small reflections/disturbances afterwards.
It might sound a little silly, but just use google and there is absolutely millions of articles out there to explain all of this.
And I don't say that just to give a lame and easy answer, but there are really so many articles explaining all of this.

ARTA is free as well, data can't be saved in the free version, but can still be exported and used in programs like VituixCAD.
I don't understand the less user friendly argument, in fact I find REW pretty user unfriendly and a total mess in itself.
(with that nasty ugly always smooth graph thing on, that Microsoft Excel started to introduce after Excel "97, talking about making data look prettier)

I don't take the whole making data look pretty argument really serious.
Than you clearly missing the point of the stitching method. (and makes me doubt if you ever worked with it in general, no offense)
And for the record, it is not about "less accurate results", it's about results that represent enough data to say something useful about what is being measured. In certain areas of physics measurements are WAAAAY less accurate, yet still consistent and very accurate conclusions can be made.

Going down to the last difference in mili decibel and micro frequency is just really silly.
 
OP
N

napilopez

Major Contributor
Forum Donor
Joined
Oct 17, 2018
Messages
1,923
Likes
7,290
Location
NYC
Basically any "weighted" window is better than rectangular.

Rectangular windows are only useful for pure sine waves, or when there is a very big pulse with no or very small reflections/disturbances afterwards.
It might sound a little silly, but just use google and there is absolutely millions of articles out there to explain all of this.
And I don't say that just to give a lame and easy answer, but there are really so many articles explaining all of this.

ARTA is free as well, data can't be saved in the free version, but can still be exported and used in programs like VituixCAD.
I don't understand the less user friendly argument, in fact I find REW pretty user unfriendly and a total mess in itself.
(with that nasty ugly always smooth graph thing on, that Microsoft Excel started to introduce after Excel "97, talking about making data look prettier)

I don't take the whole making data look pretty argument really serious.
Than you clearly missing the point of the stitching method. (and makes me doubt if you ever worked with it in general, no offense)
And for the record, it is not about "less accurate results", it's about results that represent enough data to say something useful about what is being measured. In certain areas of physics measurements are WAAAAY less accurate, yet still consistent and very accurate conclusions can be made.

Going down to the last difference in mili decibel and micro frequency is just really silly.

Fair enough on the window thing. I will do further research. In any case, as noted by @dc655321, the default behavior in REW is a Tukey 25% window anyway so it shouldn't be an issue unless you actively change it =].

We're also kind of splitting hairs. In most cases the difference between a tukey 25 window and a 0.01 window are not that big, if you are measuring in a quiet environment and gating it prior to the reflection. Most of where they do differ is going to be below the splice/blend frequency anyway.

For some comparisons so people know how small the differences are that we're talking about, here's the Q Acoustics 3020i, a speaker that Amir has measured on the NFS, with a 6ms window. In Red is Amir's measurement, just for an anechoic reference.

Full Hann window:
1630776592942.png

Tukey 0.25 (which is flat then the last 25% hann):
1630776626666.png

Tukey 0.01 (close to rectangular):
1630776716425.png


But oh no! God forbid you use the last measurement, which as you said, is 'completely useless!' Not at all representative of the speaker in question, no sir!

Joking aside, I do recognize that the rectangular-ish window can introduce more potential for mistakes to creep up at the bottom of the measurements It usually doesn't as long as you're reasonable about keeping your environment quiet and free of stray reflections, but I can change the images in my OP to show the 25% window instead to stay on the safe side. By the way, you can also turn off the graph aliasing in REW to look less smooth. It's in the 'view' menu of REW settings. You can also turn off thick traces and you'll get something closer to ARTA.

Anyway, it's fine if you prefer ARTA, I just not sure why the need to harp on REW. I don't consider having to export the data and being unable to import data in the free version user friendly, especially as we often import and compare measurements from others on this forum. I'm quite confident if we asked people to choose which of the two apps is more user friendly REW would win out. Not the least because many forum members are already familiar with the app and use it. But it's good people be aware of the option if they want something different.

Unless I am misunderstanding you, it's strange to accuse me of not having "ever worked with" the stitching method, . I am a little offended! :) I have shared more stitched measurements of commercially available speakers on this forum -- or any other forum, that I'm aware -- than anyone. Especially measurements that have been compared to anechoic results. I don't expect everyone that's ever been on this forum to know that, of course, I'm just pointing out that it's not as if I'm coming from a place with no experience.

I apologize for saying "making data look prettier," that wasn't really what I was getting at. More specifically it is that the tradeoff in resolution by using a smoothed window means that lower midrange anomalies become obscured. Yes, these can be investigated with other methods, but I like having them visible in the actual spinoramas too. I will usually try a few different windows to see if anything looks weird about the sharper ones. But if it does, it usually just indicates a problem with a stray reflection in my measurement setup.
 
Last edited:

3ll3d00d

Member
Joined
Aug 31, 2019
Messages
42
Likes
55
or when there is a very big pulse with no or very small reflections/disturbances afterwards.
this is a description of a (clean) quasi anechoic measurement

certainly worth understanding how different windows affect the resulting response and hence when as you might choose to use one window or another though
 

OWC

Member
Joined
Apr 14, 2019
Messages
26
Likes
8
this is a description of a (clean) quasi anechoic measurement

certainly worth understanding how different windows affect the resulting response and hence when as you might choose to use one window or another though
It is typically used in acoustics, say when an hammer hits a surface very hard.
The initial peak is so much louder than the rest that any stuff that shows up later isn't really relevant anymore.
Basically down in the background noise, so to speak.

As for which window.
ARTA only uses Hann, but these days one could pick whatever suits in VituixCAD.
So it is actually very easy to import any impulse responses into VituixCAD.

I don't think it's splitting hairs. @napilopez
That's also not the point I am getting frustrated with, I am getting frustrated when people get pedantic about things that don't matter.
But missing the bigger picture of it all.

A good scientist is smart with the way he's measuring stuff.
One doesn't need a 100k Klippel system to get accurate, consistent and repeatable data.
Those systems are nice to have, no discussion about it, but clearly aimed for companies who have the funds for it to back-up the time an engineer has to spend on measurements.
Back in the day when I was doing my study of applied physics, we had the crappiest gear one could think of.
Yet we were able to measure things extremely precise, consistent and accurate.
It didn't look pretty, but it got the job done.

The whole magic in the stitching method, is that de lower part with bafflestep is VERY predictable.
Only when this goes into diffraction problems, simulations go very wrong.
Mostly because that part is close to non-linear acoustics (well, kind of , but let's not get to silly about details).
So basically one only needs decent data up to 2-3 times the lowest window frequency (time), and you're fine.
The lower part can be done and corrected with simulations.

The biggest reason why "weighted" time-windows are useful, is in less optimal rooms.
It just gives you that extra couple of tens-hundred Hz that can be useful to fit the data well enough.
The reason why I said what I said, is because I don't follow at all the "frequency resolution" argument at all.
For the reasons I already described, and to summarize than again; there are a lot of methods to find any disturbances around that frequency range.


In the end, maybe people think I sound grumpy.
That is not at all my intention with all of this.
(second, it is just the way I sometimes type and come across, my apologies, it's a non-stop work in progress.....)
But what I see a lot is basically see to many parrot stories about things that can't be done.
My message is just to be smart with the tools you have, and you can get VERY far.
Maybe not to 0.5dB accuracy, but who needs that anyway?

I have created very good sounding speakers with a ton of reflections in the measurements.
Just because at that time, or on that specific location there simply wasn't anything better.
As long as you know what to look for, they don't matter.
Nothing a normal human being could pick up anyway.

Not for creating speakers but also not for comparing speakers.

Some say that "measuring is knowing" but that only counts for when you know (have a idea, or feeling) what you're measuring.
Otherwise it is still meaningless ans useless.
biggest trap for young ,as well as old, players.

Above all, I am mostly hoping that people get engaged, get involved, fact check certain things, look into more articles etc.
I am the first one to change his mind if somebody can find a better method or explanation etc.
Mostly, try things out themselves !!!!
 
Last edited:
OP
N

napilopez

Major Contributor
Forum Donor
Joined
Oct 17, 2018
Messages
1,923
Likes
7,290
Location
NYC
It is typically used in acoustics, say when an hammer hits a surface very hard.
The initial peak is so much louder than the rest that any stuff that shows up later isn't really relevant anymore.
Basically down in the background noise, so to speak.

As for which window.
ARTA only uses Hann, but these days one could pick whatever suits in VituixCAD.
So it is actually very easy to import any impulse responses into VituixCAD.

I don't think it's splitting hairs. @napilopez
That's also not the point I am getting frustrated with, I am getting frustrated when people get pedantic about things that don't matter.
But missing the bigger picture of it all.

A good scientist is smart with the way he's measuring stuff.
One doesn't need a 100k Klippel system to get accurate, consistent and repeatable data.
Those systems are nice to have, no discussion about it, but clearly aimed for companies who have the funds for it to back-up the time an engineer has to spend on measurements.
Back in the day when I was doing my study of applied physics, we had the crappiest gear one could think of.
Yet we were able to measure things extremely precise, consistent and accurate.
It didn't look pretty, but it got the job done.

The whole magic in the stitching method, is that de lower part with bafflestep is VERY predictable.
Only when this goes into diffraction problems, simulations go very wrong.
Mostly because that part is close to non-linear acoustics (well, kind of , but let's not get to silly about details).
So basically one only needs decent data up to 2-3 times the lowest window frequency (time), and you're fine.
The lower part can be done and corrected with simulations.

The biggest reason why "weighted" time-windows are useful, is in less optimal rooms.
It just gives you that extra couple of tens-hundred Hz that can be useful to fit the data well enough.
The reason why I said what I said, is because I don't follow at all the "frequency resolution" argument at all.
For the reasons I already described, and to summarize than again; there are a lot of methods to find any disturbances around that frequency range.


In the end, maybe people think I sound grumpy.
That is not at all my intention with all of this.
(second, it is just the way I sometimes type and come across, my apologies, it's a non-stop work in progress.....)
But what I see a lot is basically see to many parrot stories about things that can't be done.
My message is just to be smart with the tools you have, and you can get VERY far.
Maybe not to 0.5dB accuracy, but who needs that anyway?

I have created very good sounding speakers with a ton of reflections in the measurements.
Just because at that time, or on that specific location there simply wasn't anything better.
As long as you know what to look for, they don't matter.
Nothing a normal human being could pick up anyway.

Not for creating speakers but also not for comparing speakers.

Some say that "measuring is knowing" but that only counts for when you know (have a idea, or feeling) what you're measuring.
Otherwise it is still meaningless ans useless.
biggest trap for young ,as well as old, players.

Above all, I am mostly hoping that people get engaged, get involved, fact check certain things, look into more articles etc.
I am the first one to change his mind if somebody can find a better method or explanation etc.
Mostly, try things out themselves !!!!

I do agree with you completely with everything you said above =] I harped on the fact that you said rectangular windows were useless, but at the end of the day I don't really disagree with you on anything! The reason I wrote this guide was that there wasn't a thorough yet (hopefully) easy to follow tutorial with commonly used tools, but other than getting started with the bagby paper, toole's book and a few other resources, almost everything I know has been through trial and error.

So I definitely encourage anyone following this guide to explore different ways of doing things and to build on what's been written here. As noted in the second paragraph of my OP, I don't have any technical expertise or engineering background, I was just lucky to have access to a lot of speakers to experiment with!
 

OWC

Member
Joined
Apr 14, 2019
Messages
26
Likes
8
It's not useless, but with other types of windows you can just squeeze out a bit more.
Think about 100Hz or so, just that extra to determine the bafflestep just a little better.

Or in some cases you will always have a bit of interference residue.
Is it the end of the world? no

But why wouldn't use the tools if they are there?


As we are on the subject anyway.
It really REALLY beats me why companies with an anechoic chamber don't use these methods as well.
Just as an addition to the measurements.

Some of the datasheets of drivers (and speakers) still look very crappy.
Even from some well reputable brands.
Really easy to get these just as nice as it should be.

In some cases it is so bad that things like surround resonances (the famous dip) can't be seen.
In theory one should be able to do some auto- or cross correlation techniques I think?
 

witwald

Active Member
Joined
Jul 23, 2019
Messages
197
Likes
183
The gist of it is to make a sine sweep in REW and truncate the impulse response such that REW only includes the data from right before the first major reflection 'hits' the microphone (it's much easier to do than it sounds!).
@napilopez If your are performing a sine sweep, then isn't the measured response essentially a sine sweep, not an impulse response as mentioned?
 

morpheusX

Member
Joined
Nov 7, 2019
Messages
59
Likes
47
Location
Portugal
When a speaker has each component in its own box, is it possible to measure each box on its own (only the box measured will be placed in the stand), or do we still need to have the full speaker (all boxes) in place.

I'm asking as i'm building a set of speakers which are composed by 3 boxes, one for each component, and each box weights between 10 and 20kg, and as such, if it is possible to measure each box by its own would help immensely (moreover for vertical measurements).

Please note that this question is within the scope of performing a set of measures for an active crossover design!
 

fluid

Active Member
Joined
Apr 19, 2021
Messages
243
Likes
382
The baffle formed by the three boxes combined won't be the same as with them separated so the results will not be completely accurate if measured separately. Measuring the top two together if possible would most likely give enough accuracy with the wavelengths of the lowest box being less affected.
 

gags11

Senior Member
Forum Donor
Joined
Feb 8, 2019
Messages
302
Likes
430
My God…. Thank you @napilopez !!!!

I can’t believe I’m just finding this thread. All this time using REW has been a self taught trial and error learning.

I got my DIY speaker ready and was wondering how to test. This would be perfect. I have KEF R11 to test as a control. But first, need time to read, digest and test.

…again, thank you Napi and to all the asr gurus and contributors!
 

morpheusX

Member
Joined
Nov 7, 2019
Messages
59
Likes
47
Location
Portugal
I'm scratching my head on how to be able to test my speakers in a secure way.
As mentioned, each speaker is composed by three boxes, with the full speaker being 98cm (H) x 40cm (W) x 35cm (D), and weighting around 30kg without drivers, and around 50kg with drivers:

cNXrgVr.png


I have a pair of Atacama 24 speaker stands, they are very robust, and i can fill them with sand, but they are too small for the full speaker:
Hd5AdTJ.png




I'm even considering building a custom stand that would be able to hold this speaker securely, but it will delay this project even more (it has started 3 years ago ...)

Any ideas?
 
Top Bottom