Spinorama! (also known as CTA/CEA 2034 but that sounds dull, apparently)

[napilopez]

Sorry Guys, I haven't been keeping up! I don't have time to read through all of the above, can someone summarize any still unresolved issues? I'll try to clarify if I can. I also have a HATS file that does the Spin calculation, so I can look at that to get another reference.

@twelti Thank you so much for taking the time out to respond!

I'll summarize the two questions I have about the Early Reflections curve. As a recap for everyone, the ER curve is defined in CTA-2034A as an average of the following curves:

Floor Bounce: -20v, -30v, -40v
Ceiling Bounce: +40v, +50v, +60v
Front Wall: 0, ±10, ±20, ±30
Side Wall: ±40, ±50, ±60, ±70, ±80
Rear Wall: 180, ±90

So my questions are:

1) Does the rear wall portion only include the 3 curves listed above, or is it every curve from ±90 to 180 degrees (19 curves)?

The original 2002 Devantier paper specifically says the 'Rear' portion of the calculation includes 19 measurements. The CTA-2034A document is unclear as to whether to use 3 curves or 19. Toole's book implies 3, and @amirm confirmed above the Klippel software calculates it only using 3 curves as well.

2) Is the ER curve computed as a single average of all the above angles, or do you first average each 'section' of the curve and perform an average of all five results? The latter gives a bit heavier weighting to the vertical portions in the final curve.

The Devantier paper makes it clear the curve is an average of averages, but the CTA-2034A is more ambivalent. At least one popular speaker design tool (VituixCAD) calculates it as a single average of the 26 listed angles, which is also how others and I originally interpreted the standard.

Though there usually isn't that much of a difference regardless of how you calculate,or some speakers the ER and ERDI curves can change a fair bit depending on how you calculate, especially around crossover.

(Made a few edits for clarity)

 

[twelti]

Good question! Floyd says, "Rear wall bounces: average of 180 deg, +/- 90 horizontal". Looking at my HATS Spinorarma template, I see that all 19 rear horizontal curves are used. So, i think Floyd meant average of ALL measurements from 180 +/-90.

Again, looking at our template, I see that the different directions are averaged first (energy average) and THEN averaged together as you said. Perhaps the CTA standard needs to be rewritten a bit to clarify.

 

[napilopez]

Good question! Floyd says, "Rear wall bounces: average of 180 deg, +/- 90 horizontal". Looking at my HATS Spinorarma template, I see that all 19 rear horizontal curves are used. So, i think Floyd meant average of ALL measurements from 180 +/-90.

Again, looking at our template, I see that the different directions are averaged first (energy average) and THEN averaged together as you said. Perhaps the CTA standard needs to be rewritten a bit to clarify.

Thanks Todd! This is what I ultimately suspected was the intended process given the Devantier paper. This makes things a lot clearer.

That said, it seems Dr Toole might've had a bit of the same confusion I did, because on Page 133 of the third edition (I have the kindle version, not sure if the pages correspond correctly) he says:

"If the loudspeaker under test is a horizontally symmetrical forward-firing design the early-reflections calculation involves only 18 measurements to embrace the floor, ceiling, one side wall, rear and front wall reflection estimates. Asymmetrical designs require both side walls and both left and right front -wall data points, bringing the sum to 26 measurements. All but one of these (the 180 ° measurement) are in the front hemisphere. "

This is part of why I thought the rear portion just included three curves, as he says the ER curve is just 26 measurements total, which wouldn't work with 19 rear hemisphere measurements

Also, I'm not sure what the best way to address this is, but a small editing note that if the measurement did just include 3 rear curves, then then a horizontally symmetrical speaker forward-firing design would only require 17 curves, not 18.

@amirm You might want to inform Klippel; it obviously isn't a huge deal since your measurements have tracked Harman's pretty well, but they'd probably want to know for consistency's sake, and it probably matters more for speakers than have more going on in the rear hemisphere. Hopefully, the process can be made a bit more explicit in the CTA-2034B revision.

 

[Dave Zan]

...my HATS Spinorarma template, I see that all 19 rear horizontal curves are used....

That makes sense, the 3 measurements never looked correct to me, that was why I started this thread in the first place.
So thank you very much, nice to finally have this sorted out.

...Perhaps the CTA standard needs to be rewritten a bit to clarify.

There are a few other problems in the standard, we covered some of them in this thread, would it be helpful to send you a summary as a contribution to the next revision?

Best wishes
David

 

[amirm]

2) Is the ER curve computed as a single average of all the above angles, or do you first average each 'section' of the curve and perform an average of all five results? The latter gives a bit heavier weighting to the vertical portions in the final curve.

FYI this is what Klippel system does:

 

[Dave Zan]

Now that we have the rear hemisphere calculations corrected there are a few other aspects of CTA 2034 and the Klippel NFS that I would like to discuss, if anyone is interested.

My own speakers are intended to be placed in the room corners, directly in contact with both walls.
This was motivated by Roy Allison's research, I didn't have the benefit of Dr Toole's book because it was not yet published when I started.
I use horn loaded mid/hi and a 15" direct radiator woofer.
So the woofer is constrained to a 90 horizontal arc, could be viewed as partially horn loaded by the walls.
How will CTA 2034 and the Klippel NFS handle this?
I appreciate that the Klippel scanner can measure the free field behaviour of my speaker but this probably doesn't accurately reflect its performance in a room.

David

 

[amirm]

It could be measured by building a 90 degree cage around it on one side to simulate the two walls.

 

[Dave Zan]

It could be measured...to simulate the two walls.

Does Klippel discuss this issue in their NFS documentation?
Is the documentation, user manual or whatever, for the NFS, available on-line?
If not then would Klippel be OK for it to be posted?
I think it could only be in their interest to create discussion of their device but I appreciate it's their intellectual property.

If the documentation is not available then can you tell me what is the maximum radius you can scan?
As radius increases there may be issues, with S/Noise ratio and so on.
Is it possible to limit the scan to just within a quadrant, or in other ways?

Best wishes
David

 

[amirm]

Does Klippel discuss this issue in their NFS documentation?

No, that was me thinking out loud. They do have something similar for measuring drivers by mounting them in a board and how edge diffraction doesn't impact measurements.

 

[amirm]

If the documentation is not available then can you tell me what is the maximum radius you can scan?
As radius increases there may be issues, with S/Noise ratio and so on.
Is it possible to limit the scan to just within a quadrant, or in other ways?

There is such a configuration but it seems to change the density of points, rather as documented, the scanning surface. I will need more time to play with it and/or ask them about it.

 

[napilopez]

Just thought I'd show an example of how different a speaker might look when calculating the Early Reflections curve using a simple average vs when using an average of averages, as discussed previously.

From my D&D 8C writeup:

White is doing it as I was doing it before, and how VituixCAD currently calculates the early reflections curve - just an average of the 26 angles listed here. Blue is doing it as clarified by Todd Welti above, taking the separate ceiling, floor, front, side, and rear averages first and then averaging those five results together.

There's a clear difference in this case; it goes from a "nearly perfect" ERDI curve to a "very good" one thanks to the added impact of the vertical reflections.

 

[Dave Zan]

...an example...
There's a clear difference in this case...

This worries me a little because my initial impression is that simple Ave, makes more sense than the Ave. of Ave. - which would put me on the opposition side of the debate once more!

I had the confidence to query the rear reflection calculation because I had a clear idea of the physics and acoustics and it didn't make sense.
It was plausible that all of the CTA standard, Floyd Toole's presentations and Klippel's software were incorrect because a simple, copied typo provided a common explanation.
Whereas in this case I have a less definite idea and there is a definitive statement by the researcher involved.
So I hesitate to dispute the subject but your initial version of the calculation is the way I would have done it too.
Of course it's important to have consistency so it's nice to have the current standard clarified but it bothers me.
Do you, or anyone else, have any comments on this to help me think it over?

Best wishes
David

 

[NTK]

The average of averages to me is to apply different weightings to each group of measurements. There may be a logical reason for it, but I haven't seen an explanation.

 

[Dave Zan]

....reason for it, but I haven't seen an explanation.

Exactly, nice to know it's not just me.
Nor can I see any obvious reason so an explanation really would help.

Best wishes
David

 

[napilopez]

This worries me a little because my initial impression is that simple Ave, makes more sense than the Ave. of Ave. - which would put me on the opposition side of the debate once more!

I had the confidence to query the rear reflection calculation because I had a clear idea of the physics and acoustics and it didn't make sense.
It was plausible that all of the CTA standard, Floyd Toole's presentations and Klippel's software were incorrect because a simple, copied typo provided a common explanation.
Whereas in this case I have a less definite idea and there is a definitive statement by the researcher involved.
So I hesitate to dispute the subject but your initial version of the calculation is the way I would have done it too.
Of course it's important to have consistency so it's nice to have the current standard clarified but it bothers me.
Do you, or anyone else, have any comments on this to help me think it over?

Best wishes
David

Thank you for having the initiative to question that assumption in the first place! One of the wonderful things about this forum: you pointed out an inconsistency and we were able to get some clarification through some investigation.

I think it's clear that we should be calculating it as an average of averages. It makes more sense to me logically, it's based on the original devantier paper, and the spinorama comes from harman; as twelti clarified above, his HATS files takes an average of averages

As for how many rear curves to use, there's some respite in that if using the average of averages method, none of the speakers I've measured have shown a significant difference using just 3 or 19 measurements. I assume this would be a bigger factor in speakers with substantial rearward radiation, and of course my measurements are only useful down to 200Hz, but either way it's a small enough portion that I think it doesn't have too big of an effect. It's really the weighting of the vertical measurements that has the biggest effect.

The average of averages to me is to apply different weightings to each group of measurements. There may be a logical reason for it, but I haven't seen an explanation.

Yes indeed, that appears to be the goal; the vertical portion has a minimal impact otherwise. The logic comes from the devantier paper methinks but the paper itself doesn't go deep into the specifics of the weightings too much. It does note that while the choice of vertical angles was clear, the division of the horizontal ones were "after some experimentation and debate."

That said, regardless of the specifics of the horizontal angle weightings, the average of averages makes sense. Early reflections are basically the direct and single-bounce sounds arriving at your ear soon after the direct sound. That's roughly 7 soudns in a typical home (direct, 4 walls, floor, and ceiling). From the study:

So you have the direct sound portion (which also includes a bounce from the wall behind you), the sidewall portion, the rear wall portion, and then the floor and ceiling. I can't speak to the lumping of different angles, but it does seem you should try to lump them some of them together in some way.


But the most important thing is ensuring consistency among different sources. After all, that's why it's supposed to be a 'standard' Considering a dip/bump around the crossover is one of the things we look at the most in the ER and ERDI curves, it's important to get that right.

 

[napilopez]

By the way, I pinged VituixCAD's creator over on DIY audio about the proper method for calculating the ERDI yesterday, and he promptly issued an update, so the latest version of vituixcad now calculates the curves as clarified by Todd Welti above. I now get the same exact result now on both the ER and ERDI (offset to show similarities):

Which I'm glad about because it saves me a ton of time!

Tagging @BYRTT , since I know you often use VituixCAD =]

 

[napilopez]

@amirm Per the discussion in this thread and how you said the Klippel calculated the ER curves, I decided to check for myself and use the klippels own data to calculate the ER curves manually using the process described by Todd Welti above. I had noticed that the klippel-generated ER and ERDI curves looked usually a fair bit smoother than the Harman ones around the crossover, and this shows why.

Here is the M16 ER and ERDI curves as generated by Klippel (white) vs the process used by HATS(blue) mentioned by Todd Welti above:

Not too much of a difference, but you can see the notch/bumps in the blue curves are worse and match Harman's a little more closely than the klippels.

Now on the Revel F35:

A more prominent crossover issue, and once again, closer to the HATS result:

The bump on the ERDI curves are now nearly identical.

While probably not something worth going back to fix, it's something to perhaps keep in mind with measurements going forward, and that I hope klippel is informed of. Once again, it could be the difference between a good and great DI curve, especially on a speaker with particularly egregious vertical problems. And it would be good for the CTA-2034A standard's process to actually, you know, be standardized .

As an interesting contrast, while almost every speaker I've tested gets worse with the HATS process, the Harbeth actually seems to get better. It has less of a tilt in the ER and smoother directivity. The vertical and horizontal dips actually do a better job of balancing out with this interpretation.

 

[amirm]

Here is the M16 ER and ERDI curves as generated by Klippel (white) vs the process used by HATS(blue) mentioned by Todd Welti above:

I have given them this feedback already. But I have not heard back as to whether they agree and will make the correction. Thanks for doing those graphs. I will provide it to them. In general, Klippel has been hugely responsive to issues I bring to them.

 

[spacevector]

I have given them this feedback already. But I have not heard back as to whether they agree and will make the correction. Thanks for doing those graphs. I will provide it to them. In general, Klippel has been hugely responsive to issues I bring to them.

They oughta hook you up with ALL the software modules in exchange of your QC feedback to them and the shout-outs in every speaker review.

 

[napilopez]

I have given them this feedback already. But I have not heard back as to whether they agree and will make the correction. Thanks for doing those graphs. I will provide it to them. In general, Klippel has been hugely responsive to issues I bring to them.

Awesome, thanks! Hope the graphs help. It seems this all stems from the vague language in CTA-2034A.

All I had to do to generate the 'correct' ER curve was do a simple rms average of the existing front, ceiling, floor, side, and rear curves that klippel already calculates in your files.