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

[fluid]

The room certainly dominates the in room low frequency response but as a method for evaluating one speaker against another or checking that a design performs as intended these techniques still have value.

napilopez I am wondering if you have tried the ARTA method, as in the application note it is claimed that their simplified baffle step equalizer is more accurate.

"Note: ARTA and STEPS uses the previous expression for the estimation of the diffraction for spherical or rectangular baffled boxes. Some CAD and simulation programs are using a high-frequency geometrical model for the estimation of box diffraction at low frequencies. Such models can give larger errors on low frequencies than the simple model that is presented here"

https://www.artalabs.hr/AppNotes/AN4-FreeField-Rev03eng.pdf

You have a lot of data to compare against NFS and anechoic graphs so I wonder what the difference is.

 

[napilopez]

The room certainly dominates the in room low frequency response but as a method for evaluating one speaker against another or checking that a design performs as intended these techniques still have value.

napilopez I am wondering if you have tried the ARTA method, as in the application note it is claimed that their simplified baffle step equalizer is more accurate.

"Note: ARTA and STEPS uses the previous expression for the estimation of the diffraction for spherical or rectangular baffled boxes. Some CAD and simulation programs are using a high-frequency geometrical model for the estimation of box diffraction at low frequencies. Such models can give larger errors on low frequencies than the simple model that is presented here"

https://www.artalabs.hr/AppNotes/AN4-FreeField-Rev03eng.pdf

You have a lot of data to compare against NFS and anechoic graphs so I wonder what the difference is.

Interesting -- I have not used ARTA, as I started out with freeware and there isn't really in arta that I need that I can't do with REW and VCAD.

 

[fluid]

ARTA is free to use in a demo mode where the graphs can be viewed and screenshots taken but to compare the results saving the results as impulse or as text would make sense. I have a full version of it and could process the files if it was something you wanted to try. You might like ARTA more if you built their automated turntable as they have a function with it to make taking polar measurements automated.

 

[Shives]

So I just stumbled upon this post. To the OP, thanks! This looks like hours upon hours of good work, that now I can make sometime to go through. So, thanks.

 

[changer]

Hello napilopez,
I have another question. In VituixCAD under the options panel for DI directivity index you include vertical, is that right? I have some issues with a well known DIY-speaker topology, the so called Econowave, a JBL 4425/4349 remake with focus on constant directivity. After measurements I was astonished to see that although the concept holds its promise on horizontal axis, DI seems to be thouroughly messed up. Seemingly vecause of the choosen waveguide, which cannot keep the pattern on the vertical axis downwards. Some data is here. Now I wonder if almost any speaker measured on this forum has a smoother DI than the speaker which was praised as constant DI (above 500 Hz), or if vertical DI is somewhat excluded. Can you tell me if DI and so on in the measurements commonly feature vertical? Often, because vertical is so much chopped up, I have the feeling they must be weighted differently with very smooth DI graphs.

 

[napilopez]

Hello napilopez,
I have another question. In VituixCAD under the options panel for DI directivity index you include vertical, is that right? I have some issues with a well known DIY-speaker topology, the so called Econowave, a JBL 4425/4349 remake with focus on constant directivity. After measurements I was astonished to see that although the concept holds its promise on horizontal axis, DI seems to be thouroughly messed up. Seemingly vecause of the choosen waveguide, which cannot keep the pattern on the vertical axis downwards. Some data is here. Now I wonder if almost any speaker measured on this forum has a smoother DI than the speaker which was praised as constant DI (above 500 Hz), or if vertical DI is somewhat excluded. Can you tell me if DI and so on in the measurements commonly feature vertical? Often, because vertical is so much chopped up, I have the feeling they must be weighted differently with very smooth DI graphs.

Yes, as far as I know, vertical DI is always as part of the Sound Power DI calculation unless otherwise stated. The weighting is always the same for both Horizontal and Vertical as far as I know as well. Certainly, all of the spinoramas posted on this forum do it this way.

The Early Reflections DI, which I am more partial to, does weigh horizontal and vertical data differently, giving particular emphasis to a handful of vertical angles.

It is perhaps the term 'constant directivity' that leads to some misunderstanding as it is a little misleading. As far as I know, speakers that are constant directivity almost always refer to the horizontal directivity. Some manufacturers may try to optimize for vertical directivity as well, but it usually is 'sacrificed' by nature of being a vertically aligned speaker. I've not seen a constant directivity speaker with perfect verticals.

I'd not worry too much about having a perfect DI though. In the example you shared, it may not look pretty but it's unlikely to be that much of a problem. As we know the issue primarily comes from the vertical data, it means it's unlikely to have much of an effect on soundstage. This means that the jagged DI will most likely be heard as a slight midrange recession that keeps the speaker from being tonality perfect, but it certainly won't be that major of an issue.

This is why I think it's important to separate horizontal and vertical data and I've begun to include 'Horizontal ERDI' in my graphs.

 

[napilopez]

It seems even some of the Harman engineers have acknowledged this flaw, as I've noticed recent spins from JBL include the Horizontal DI (not Horizontal ERDI, but still welcome).

For example, JBL 4349:

JBL L82:

At some point I hope either Horizontal ERDI or Horizontal DI becomes standardized for the spinorama, as it adds a lot of useful context for 'at a glance' evaluation. I know the standard is currently being reworked for CTA-2034B, but I'm not sure if any of the included curves are being changed.

 

[Kadent]

Very informative write up! Will definately get around to testing once I clean the garage out. Question: How do you get the overlay in REW to say Bass, mids, upper mids, ect? I've been trying to find the option in the settings.

 

[napilopez]

Very informative write up! Will definately get around to testing once I clean the garage out. Question: How do you get the overlay in REW to say Bass, mids, upper mids, ect? I've been trying to find the option in the settings.

In REW's 'All SPL' window, tap on 'Controls', and select 'show frequency bands'

 

[Tom Danley]

Hi
Just as with computer modeling of loudspeakers etc, the actual value of a program is it's ability to predict what you then build and go ahead and measure and part of that prediction is based on valid data as inputs. Having done essentially all the crossovers active and passive at the website under my name, I can't over stress the importance of testing what was built based on predictions.
Often enough at least with more complicated speakers it turns out there are little fudge factors and unaccounted for things you need to add to make what is predicted close to what you measure.

I use ARTA often but I would offer that in a normal living room, if you put the speaker at half the room height and more or less centered in the room and the mic say 2 feet away (too close unless it's a tiny speaker) , you can see that best case, the floor and ceiling reflections arrive behind the direct sound at the microphone but that short gap is your only "reflection free zone" as Heyser called it.

So if you gate up to the first reflection, one could work out a mid and hifh crossover in room but usually around 200-300 Hz and below the data isn't usable. That's where the great anechoic outdoors on a quiet day can be handy when you really have to know what the loudspeaker alone does (because that's the only part you can change with a crossover)
Noise effects (external and internal) measurements, to the degree there is variation (especially in phase response at low frequencies) measurement to measurement, the process benefits from the noise reduction trick used in astrophotography, stacking or averaging.

In the case of ARTA, one can do that automatically by averaging say 16 measurements, I think REW also allows that too) which in theory reduces the noise component as much as -12dB.

With impulse based systems you can usually see where the effective measurement noise floor is at the low frequency end of the response.
Where the speaker is known to roll off at say -24dB per octave, somewhere often around -20 to -30dB, the curve flattens out and often has a deep notch a bit below that.
All of that "extended bass" is normally measurement noise, the speaker still rolls off at that -24..
Sometimes, increasing the mic gain helps this issue.
Anyway, I love seeing people making speakers!

Hey if the measurements taken that way can be converted to the CLF format, the viewer is pretty cool and fairly widely used in commercial sound.
http://www.clfgroup.org/viewer.htm

Tom Danley
https://www.danleysoundlabs.com/

 

[napilopez]

Hi
Just as with computer modeling of loudspeakers etc, the actual value of a program is it's ability to predict what you then build and go ahead and measure and part of that prediction is based on valid data as inputs. Having done essentially all the crossovers active and passive at the website under my name, I can't over stress the importance of testing what was built based on predictions.
Often enough at least with more complicated speakers it turns out there are little fudge factors and unaccounted for things you need to add to make what is predicted close to what you measure.

I use ARTA often but I would offer that in a normal living room, if you put the speaker at half the room height and more or less centered in the room and the mic say 2 feet away (too close unless it's a tiny speaker) , you can see that best case, the floor and ceiling reflections arrive behind the direct sound at the microphone but that short gap is your only "reflection free zone" as Heyser called it.

So if you gate up to the first reflection, one could work out a mid and hifh crossover in room but usually around 200-300 Hz and below the data isn't usable. That's where the great anechoic outdoors on a quiet day can be handy when you really have to know what the loudspeaker alone does (because that's the only part you can change with a crossover)
Noise effects (external and internal) measurements, to the degree there is variation (especially in phase response at low frequencies) measurement to measurement, the process benefits from the noise reduction trick used in astrophotography, stacking or averaging.

In the case of ARTA, one can do that automatically by averaging say 16 measurements, I think REW also allows that too) which in theory reduces the noise component as much as -12dB.

With impulse based systems you can usually see where the effective measurement noise floor is at the low frequency end of the response.
Where the speaker is known to roll off at say -24dB per octave, somewhere often around -20 to -30dB, the curve flattens out and often has a deep notch a bit below that.
All of that "extended bass" is normally measurement noise, the speaker still rolls off at that -24..
Sometimes, increasing the mic gain helps this issue.
Anyway, I love seeing people making speakers!

Hey if the measurements taken that way can be converted to the CLF format, the viewer is pretty cool and fairly widely used in commercial sound.
http://www.clfgroup.org/viewer.htm

Tom Danley
https://www.danleysoundlabs.com/

Thanks chiming in! I defnitely agree with everything you wrote. I just wanted to note that this guide was written primarily for evaluating speakers, as I have no experience building my own Although I hope to soon change that!

As for auto-averaging measurements, REW can do this, but unfortunately, the technique does not work for the USB mics (like the MiniDSP Umik-1 and dayton umm-6) most people are using. That said, you can simply use a longer sweep for similar effect. I tend to use a longer sweep for the listening window measurements and then shorter sweeps for the off-axis ones in the interest of time and not annoying my neighbors too much =]

 

[NTK]

Thanks chiming in! I defnitely agree with everything you wrote. I just wanted to note that this guide was written primarily for evaluating speakers, as I have no experience building my own Although I hope to soon change that!

As for auto-averaging measurements, REW can do this, but unfortunately, the technique does not work for the USB mics (like the MiniDSP Umik-1 and dayton umm-6) most people are using. That said, you can simply use a longer sweep for similar effect. I tend to use a longer sweep for the listening window measurements and then shorter sweeps for the off-axis ones in the interest of time and not annoying my neighbors too much =]

When you are measuring using the Farina sweep (logarithmic sine chirp), the preferred method to improve signal-to-noise ratio is to use longer sweeps. Professor Farina described it in his paper.
http://pcfarina.eng.unipr.it/Public/Papers/226-AES122.pdf

 

[kimmosto]

Could you modify this thread so that Bagby's Excel sheets are replaced with VituixCAD Diffraction and Merger? Excel sheets don't support off-axis measurement sets, directivity simulation and directivity transfer from FF to NF=LF so they are obsolete, slow, difficult and should not be recommended and mentioned anymore. Thanks.

 

[MrPeabody]

Thanks chiming in! I defnitely agree with everything you wrote. I just wanted to note that this guide was written primarily for evaluating speakers, as I have no experience building my own Although I hope to soon change that!

As for auto-averaging measurements, REW can do this, but unfortunately, the technique does not work for the USB mics (like the MiniDSP Umik-1 and dayton umm-6) most people are using. That said, you can simply use a longer sweep for similar effect. I tend to use a longer sweep for the listening window measurements and then shorter sweeps for the off-axis ones in the interest of time and not annoying my neighbors too much =]

For someone who has never built a loudspeaker you certainly do have a very intimate understanding of how to measure them!

 

[napilopez]

Could you modify this thread so that Bagby's Excel sheets are replaced with VituixCAD Diffraction and Merger? Excel sheets don't support off-axis measurement sets, directivity simulation and directivity transfer from FF to NF=LF so they are obsolete, slow, difficult and should not be recommended and mentioned anymore. Thanks.

So this guide was written with the assumption that most people using it aren't going to be creating full vertical and horizontal measurements. I assumed most people would use it for an on-axis measurement and perhaps a few off-axis measurements rather than creating full spinoramas, which is why it's divided into parts. Using VCAD for a single on-axis measurements seemed a little overkill

Don't take this the wrong way, as VituixCAD is definitely the more powerful tool and I appreciate the incredible amount of work you've put into it, but I find the Bagby sheet quicker to use, while providing practically identical results for these purposes. For example, here's the on-axis and power response of a recent speaker using both the Bagby sheet and VituixCAD

I'm not sure if you read through the guide, but I am using the VituixCAD merger tool for those who choose to perform the full spinorama with off-axis simulation. As far as I can tell, doesn't the merger tool do the same thing as the directivity option in the VCAD Baffle tool?

There are definitely times the VituixCAD is more useful though, so I will add a guide for how to use it as well.

For someone who has never built a loudspeaker you certainly do have a very intimate understanding of how to measure them!

That's what happens when you spend a year researching and then next two measuring a few dozen of them

 

[kimmosto]

So this guide was written with the assumption that most people using it aren't going to be creating full vertical and horizontal measurements. I assumed most people would use it for an on-axis measurement and perhaps a few off-axis measurements...

I have some difficulties to understand this kind of assumptions on any forum, but especially on ASR which nowadays lives with spinorama, and DIY forums where off-axis data is modern, safe, cheap and fast way to control design before it's built. Why should any instruction contain limited and/or slow tools which "force" people to study something else and different when number of angles increases from 1 to 10 (hor 0-90) or 19 (hor 0-180) up to full dual plane 360 deg. My negative conclusion is that some authors (moderators on diyaudio) think that the others are as frozen in the past as they are or share their preferences biased by limited experience or perspective so it's best to recommend the same old tools and methods other forums such as techtalk @ parts-express have listed more than a decade ago. Just like beginners would deserve the simplest but worst.
So the same instructions will work for both single axis and full dual plane.

Merger noticed, but screenshots of Bagby's sheets stressed and stole my attention. Note that Merger has two modes:
1) Merging of near field LF responses multiplied by baffle effect response (simulated with Diffraction) to multiple far field responses (HF). In this mode near field uses directivity of far field HF responses at LF. This works quite well when directivity acts smoothly at LF i.e. can be captured with time window. This works also with few off-axis responses helping to avoid step at transition frequency with data - not with blending alone. Doesn't have to be half space or full space angle coverage to get advantage.
2) Merging of multiple LF responses to multiple HF responses matched with angle coding in filenames. This is designed for off-axis LF responses with simulated directivity or measured with different method such as ground plane. Directivity of conventional boxes can be simulated with Diffraction.
Both modes especially with more than single response should be faster and easier by far than Bagby's sheets.

I have also made kinda promise on HTguide to improve if some feature in VituixCAD is not good enough or worse than somewhere else. No suggestion received so VCAD might already be as good as the others. It's not all inclusive so some other tools are needed especially for simulation of other than direct radiators and external shapes.

 

[Tom Danley]

Thanks chiming in! I defnitely agree with everything you wrote. I just wanted to note that this guide was written primarily for evaluating speakers, as I have no experience building my own Although I hope to soon change that!

As for auto-averaging measurements, REW can do this, but unfortunately, the technique does not work for the USB mics (like the MiniDSP Umik-1 and dayton umm-6) most people are using. That said, you can simply use a longer sweep for similar effect. I tend to use a longer sweep for the listening window measurements and then shorter sweeps for the off-axis ones in the interest of time and not annoying my neighbors too much =]

"Although I hope to soon change that!"

I hope you do also, you have gone farther than many do before making sawdust.
If there is anything i might be able to help with.. ask
Best Regards
Tom

 

[Tom Danley]

When you are measuring using the Farina sweep (logarithmic sine chirp), the preferred method to improve signal-to-noise ratio is to use longer sweeps. Professor Farina described it in his paper.
http://pcfarina.eng.unipr.it/Public/Papers/226-AES122.pdf

View attachment 130715

Hi
I have used this approach as well but when S/N is a problem at low frequencies, TDS in practice actually has the greatest noise immunity.
TDS is something like a how an AM radio rejects adjacent signals with it's intermediate frequency conversion. Conceptually it's more like the AM transmitter frequency sweeps and tied to it is the receiving frequency on the receiving end.
For example, noise immunity is so much better that one can use a vacuum cleaner nearby or take TDS measurements while playing music. This does not free one from consideration of the "reflection free zone"
Like the Beta tape vs VHS tape issue, although one was superior, the other was widely licensed

 

[napilopez]

I have some difficulties to understand this kind of assumptions on any forum, but especially on ASR which nowadays lives with spinorama, and DIY forums where off-axis data is modern, safe, cheap and fast way to control design before it's built. Why should any instruction contain limited and/or slow tools which "force" people to study something else and different when number of angles increases from 1 to 10 (hor 0-90) or 19 (hor 0-180) up to full dual plane 360 deg. My negative conclusion is that some authors (moderators on diyaudio) think that the others are as frozen in the past as they are or share their preferences biased by limited experience or perspective so it's best to recommend the same old tools and methods other forums such as techtalk @ parts-express have listed more than a decade ago. Just like beginners would deserve the simplest but worst.
So the same instructions will work for both single axis and full dual plane.

Merger noticed, but screenshots of Bagby's sheets stressed and stole my attention. Note that Merger has two modes:
1) Merging of near field LF responses multiplied by baffle effect response (simulated with Diffraction) to multiple far field responses (HF). In this mode near field uses directivity of far field HF responses at LF. This works quite well when directivity acts smoothly at LF i.e. can be captured with time window. This works also with few off-axis responses helping to avoid step at transition frequency with data - not with blending alone. Doesn't have to be half space or full space angle coverage to get advantage.
2) Merging of multiple LF responses to multiple HF responses matched with angle coding in filenames. This is designed for off-axis LF responses with simulated directivity or measured with different method such as ground plane. Directivity of conventional boxes can be simulated with Diffraction.
Both modes especially with more than single response should be faster and easier by far than Bagby's sheets.

I have also made kinda promise on HTguide to improve if some feature in VituixCAD is not good enough or worse than somewhere else. No suggestion received so VCAD might already be as good as the others. It's not all inclusive so some other tools are needed especially for simulation of other than direct radiators and external shapes.

I appreciate the feedback and you are right that it might be a bit shortsighted to assume people will use the guide a certain way. I suppose my thought was that it took me a long time to gather all this information as someone with no 'hard' science background and I wanted to make it as accessible as possible. And indeed part of the reason I wrote this guide in the first place was because the existing ones tended to be too limited or used old expensive software that isn't accessible to many enthusiasts.

As I intend to explore finally making some speakers of my own over the next few months, I'll be familiarizing myself more with VCADs tools and using it more like how it was intended to be used I will mess around more with VCAD's merger and diffraction tools in the coming weeks.

 

[dominikz]

I appreciate the feedback and you are right that it might be a bit shortsighted to assume people will use the guide a certain way. I suppose my thought was that it took me a long time to gather all this information as someone with no 'hard' science background and I wanted to make it as accessible as possible. And indeed part of the reason I wrote this guide in the first place was because the existing ones tended to be too limited or used old expensive software that isn't accessible to many enthusiasts.

As I intend to explore finally making some speakers of my own over the next few months, I'll be familiarizing myself more with VCADs tools and using it more like how it was intended to be used I will mess around more with VCAD's merger tool in the coming weeks.

I for one applaud and thank you for putting together this guide. A few months ago I first dipped my toes in quasi-anechoice loudspeaker measurements and I have to say I struggled for a while before setting up the process, understanding it and finally getting reasonable results. Since I at least have (some) audio and EE background I can imagine many who don't struggle even more and your guide surely helps them immensely. Maybe it even motivates those who otherwise wouldn't to try their hand at measuring.
Having this kind of guide back when I started would have made my life much simpler for sure!