[bjs]

I tried out WinISD today. Modelled a sealed CSS SDX15 and compared it against Ilkka's actual test results of the same.

The results were poor. For example, WinISD modelled a maximum spl at 20Hz as 107dB while Ilkka's tests only produced 99dB. The frequency response curves were quite different from each other too. This was a simple sealed box, vented box would be worse especially with large signal behaviour.

So the question is do you believe your modelling? People don't seem to check their "as built" results. :gulp:

 

[Jailer]

My MJ-18M build came in real close to what it modeled in WinISD. I'm real happy with the results I ended up with.

 

[ec114]

Ilkka's numbers are 2m groundplane, or ideally, 1m freespace. If WinISD is like UniBox (not sure if it is), it's equivalent to 1m groundplane. So you'd need to subtract 6dB before comparing to Ilkka, and then you'd be close...

 

[ironglen]

Don't know about real output vs modeled, but playing test tones verify my sub came between 14 and 15 hz tuning when I modeled and built for 14.5 hz...

 

[bjs]

Thanks ec114, I hadn't read the manual and just assumed WinISD was 1m freespace...according to the help files it is indeed half space.

As for the rest, I was thinking less about validating the tuning frequency and more about the actual output. I didn't find any threads that took the time to actually do that (although I probably missed some).

 

[Moonfly]

When AE finally pull their fingers out and deliver my drivers, and then I finish my sub, I'll be sure to test the end product and post back here in the future for reference :T

 

[Blaser]

Ilkka's measurments are 2 meters GP while winISD indicates 1 m. Doubling the distance is indeed 6 db less, which looks consistent...

 

[Kevin Haskins]

T/S models are small signal level models. Within that context, they work 100% nut-on.

If you try to use them to predict something that they were never intended for..... then you shouldn't be surprised when they are not accurate.


Kevin Haskins
Exodus Audio

 

[Yad]

Yes. Sure )))

 

[Blaser]

T/S parameters enable to have at least an idea about a driver behavior, don't they?

 

[Ricci]

T/S models are small signal level models. Within that context, they work 100% nut-on.

If you try to use them to predict something that they were never intended for..... then you shouldn't be surprised when they are not accurate.

See above for the answer.

 

[Blaser]

Ricci,

Is this addressed to me?

 

[DrWho]

T/S models are small signal level models. Within that context, they work 100% nut-on.

If you try to use them to predict something that they were never intended for..... then you shouldn't be surprised when they are not accurate.

Couldn't agree more.

Btw, WinISD can be made to predict non-linear behaviors and all sorts of other things as well, but you gotta know what you're doing and how to interpret the results. It's a very manual process.

If my memory isn't failing me, I believe you can account for your 2dB discrepancy with power compression by modeling a 200C temp rise for the SDX. WinISD models the effects of temp rise a bit weird, such that you need to manually add the compression (knowing that no frequency should be louder than the no temp rise model).

 

[Ricci]

Ricci,

Is this addressed to me?

No. I was addressing the main title question of this thread. Kevin's post was spot on.

If my memory isn't failing me, I believe you can account for your 2dB discrepancy with power compression by modeling a 200C temp rise for the SDX. WinISD models the effects of temp rise a bit weird, such that you need to manually add the compression (knowing that no frequency should be louder than the no temp rise model).

200 celsius :yikes:? I don't think the vc and glues will survive that will they? I normally don't go over about 190deg F to represent a heavy load on the driver.

 

[Kevin Haskins]

I didn't mean to be mouthy. The model just doesn't have enough data to predict what happens at high signal levels. LSPCad has some high-level signal tools but they are a crude estimation of what happens at those levels. That is 'ok', it gives you some idea of what is going on but you have to know that it is an estimation and it is better yet to know why the model is doing what it does.

To get an accurate high-level model, you would have to have to have many more data points. Voice coil temp is difficult to predict just based upon power because it depends on the type of signal along with the box design. You would need a way to model it with both steady state sine wave, and maybe a toggle button that could show it with music/movie content (1/8th power as an estimate). You would need more data about the coil+former, and how the system dissipates heat. I can easily smoke a driver in a resonant enclosure if I run a sine wave near the tuning frequency. I can run the same system all day long with the driver moving so that would have to be accounted for in the model.

You would have to have data on the Kms curve of a driver, you would have to have more Le data, and you would need more information about the shape/limits of the BL curve. None of those things are represented in the specs used today. We just crudely estimate them with Xmax.

So... the current models work for low signal levels, with pretty much 100% certainty. Just don't expect them to accurately portray what goes on at the limits of a driver, because they simply don't have enough data to give you an accurate answer.

Kevin Haskins
Exodus Audio

 

[Blaser]

Thank you RICCI.

I agree in a way. Nevertheless T/S parameters still give quite accurate information to the extent of their accuracy to begin with. Ilkka has measured a lot of drivers that behaved not so far from the predicted curves using WinISD. So accurate T/S parameters from manufacturers will not only give an idea but will go quite a lot further in terms of subwoofer behavior at higher signals ... as far as we're talking "instantaneous" though. And my point is : is it worth modeling a sub using T/S parameters? It is definitely YES.

 

[DrWho]

200 celsius :yikes:? I don't think the vc and glues will survive that will they? I normally don't go over about 190deg F to represent a heavy load on the driver.

http://www.klippel.de/pubs/Klippel%20papers/Nonlinear_Modeling_of_Heat_Transfer_03.pdf

They're showing in excess of 100K temp rise with 15W of real power...Celsius is just K + 273, so 100C temp rise is easily attainable. Speakers are highly reactive so the power factor isn't going to be too great. I don't know what 15W comes out to in terms of amplifier power...maybe about 50W or so?

If you want to get into serious large signal modelling, then Klippel is the way you wanna go. Here's a bunch of other very interesting articles on large signal behavior:
http://www.klippel.de/pubs/papers.asp

 

[Kevin Haskins]

The Klippel data is useful for designing transducers. It is less useful to an end-user. But... as an example of what we are looking at when we do design one.

You want parameters that are stable with power.

Obviously, if you see large changes in Qes, Qts it indicates that the other T/S parameters will not be the same with output. With most transducers, these graphs look much more parabolic with stroke, indicating that the parameters are shifting. The parameters will shift differently, with every transducer so you would need to know how much each parameter changes with output in order to accurately model it.

The coil temps inside the coil reach crazy high numbers. They do so in a transient nature, where the temp inside the coil skyrockets and then quickly dissipates the heat into the atmosphere.

In terms of power into the coil. Most subs are 1-2% efficient. That means if you have 100W into the device, at least 98W is dissipated as heat. Most of that thermal loss to the outside atmosphere occurs via convective heat loss, through air moving over the coil & former. It is different for every transducer but at the very least, 80% and probably closer to 90% of the heat in the system is lost to the outside atmosphere through the air passing over the coil during movement of the former/coil. That is why the coil temp is significantly different based upon the signal, 20Hz vs 80Hz has a significantly different impact on the movement of the coil, type of test signal (continuous sine wave vs. burst).

The bottom line is that the output at full power depends upon how you measure it (what type of signal) along with the physical parameters of the device. So that would have to be a part of any model that accurately described high-signal level output.

Kevin Haskins
Exodus Audio

 

[DrWho]

The Klippel data is useful for designing transducers. It is less useful to an end-user. But... as an example of what we are looking at when we do design one.

I might argue that the large signal behavior should affect the cabinet design choices - even for DIY when you have no control over the driver design (except for knowing which drivers to purchase).

In terms of power into the coil. Most subs are 1-2% efficient. That means if you have 100W into the device, at least 98W is dissipated as heat.

I agree with the 1% efficiency number, but I don't buy that 98W is being dissipated inside the motor. , it's not trivial to build a heatsink that can dissipate a full 98W, especially one without fans.

I've been meaning to measure the power factor of some of the speakers I've got laying around, but just haven't had the time. I couldn't see it being much higher than 10% (which would mean only 9.8W are being dissipated inside the speaker).

Regardless, even 9.8W is gonna create a load of heat.

 

[bjs]

...the current models work for low signal levels, with pretty much 100% certainty. Just don't expect them to accurately portray what goes on at the limits of a driver, because they simply don't have enough data to give you an accurate answer.

I see Unibox includes more accurate modelling to include voice coil Eddy currents. Does that amount to much with modern subwoofers? Perhaps it helps explain the peak in the real world SDX15 frequency plots versus the WinISD modelled results (for small signal).