Diffraction

Status
Not open for further replies.
I know of 4 widely available diffraction simulators: Paul V's BDS Excel program, Jeff Bagby's Baffle Diffraction and Boundary Simulator, The Edge from Tolvan Data, and my program. All of them are full 3D simulations, where the phase and amplitudes of the rays from the source and reflections are summed at a point in 3D space. My program does over 8 million ray summations for a typical 8" driver--it is a lot of processing. 3 of the programs in this list use some sort of approximation to calculate the effect of roundovers on the baffle edges, and 3 of these programs can accurately predict the effect of the backwave interacting with the sources on the front for a dipole configuration. So your account of the limitations of these models isn't accurate.
I think you missed my point about not being true 3D simulations - I'm not talking about summing the response at a 3D point in space, obviously that's mandatory to get any kind of meaningful result at a non-infinite listening distance.

I'm talking about a failure to model the cabinet and more importantly the driver as 3D objects. For example The Edge by Tolvan is the one I usually use, (although I've also used Paul V's spreadsheet somewhat) and it's not even possible to enter the depth of the box. It's just a baffle that is either open baffle or not.

More importantly the 3 dimensional shape of a driver is not modelled in any of the ones I have seen - the driver is represented by only a single parameter - diameter, or in the case of a rectangle width and height.

This is simply not enough information to model the directivity of a driver and how its radiation will impinge upon a baffle at different frequencies. Treating a cone driver as a flat disc made of a large number of point sources and tracing rays from each point to each point around the edge of the baffle is not accurate at high frequencies no matter how many million rays you trace.

Even if you were to provide a cone shaped geometrical model for the driver that allowed the user to enter the depth of the cone and diameter of the base of the cone so that the conical shape could be approximated, this model could still only be accurate in the piston range of the driver.

Even with a full 3D geometric model of the driver the results above cone breakup would be inaccurate because there is no way for the app to know what the breakup behaviour of the driver is in this region, without providing it a full set of Klippel data.

For example, as the effective radiating diameter of the cone reduces with frequency, the outer part of the cone starts to horn load the inner active part of the cone. None of this is taken into account. Another example would be diffraction from the surround of the driver at high frequencies. A driver with a reverse roll surround will have a different baffle diffraction effect than one with a forward roll - also not taken into account.

Yet another example is a wave-guide loaded ribbon tweeter, how would you model that in your program exactly ? Unless you accurately model the functioning and internal geometry of the wave-guide itself (possible, since programs like horn-resp do it, but far from trivial) the results will be erroneous.

I'm not trying to be critical of any of these programs, simply drawing attention to the fact that they are useful but very simplified models of reality, with plenty of fudge factors thrown in to work well with "typical" circumstances, and while they can do a reasonable job of modelling the box edge diffraction, modelling real drivers as a source is very challenging, and above the piston range of the driver, impossible with the amount of data the user can manually enter.

If you use these models with small cone drivers and dome tweeters, you'll get fairly accurate results. If you use it with more difficult to model drivers like wave-guide loaded tweeters and large full range drivers the results will be dramatically in error.

There are a number of ways in which the models can be improved, but Jeff Bagby did a lot of comparisons of his model with actual measurements and found that his model is already fairly accurate. One possible improvement is factoring in the effect of re-radiation from the edges. That is, the reflected waves can be viewed as sources which have their own diffraction signatures. None of these models currently address this, and it there is some published work that suggests this effect might be significant. Another improvement is factoring in the 1/N^2 losses due to different lengths of the rays. I added this factor to my model, and it results in a somewhat less than 6db overall baffle step and a somewhat less pronounced peak at the step frequency. I believe that this model is more accurate, particularly for predicting the response at close distances (summation point less than 4 feet, for example).
Hmm, the fact that 1/N^2 losses were not taken into account surprises me, and certainly casts some doubt on the results...I'm also concerned that when you added this in you get less than 6dB - with an infinite baffle it should be exactly 6dB, not nearly 6dB. What result do you get if you enter a very large baffle ?
 
Last edited:
The Operations Manual has extensive discussion of the theory and even includes the Excel equations for most calculations: http://www.pvconsultants.com/audio/diffraction/downloads/opmanrel6.07.exe Read that and you will have a better understanding of the underlying theory.

Sounds like there is some very useful stuff in there.

I assume the .exe part is because it is a self-extracting archive? Would someone be so kind as to extract it and provide me with a not-self extracting archive (the exe does not play nice on my Macs)

dave
 
I thought .exe meant executable file. Which is funny 'cause it's not the case on a mac. 🙂

It does. A manual is unlikely to be a standalone program, but often (usually by those that forget that Windows is not the only operating system) a zip will be made into a self extracting archive so that the windows user does not have to go looking for their unzipper. (ie the same functionality Mac users get by just double clicking the zip file)

dave
 
I so wish that Siegfrieds site would load but it never has and I am now on my third computer and fourth service provider.
That blue progress bar gets 1/4 of the way, then stops forever.
Really strange 😕

Never had any problems here through Sky Broadband, and I checked it on my phone (Three) and that worked too, who are you using, and which browser ? Perhaps a browser problem or an over-enthusiastic internet security / firewall program on your PC ?

Siegfrieds website is very plain HTML and images, so shouldn't present a problem to any browser...and its a great resource, so it would be a shame if you can't access it.
 
Last edited:
More importantly the 3 dimensional shape of a driver is not modelled in any of the ones I have seen - the driver is represented by only a single parameter - diameter, or in the case of a rectangle width and height.?

It would be very easy to model the additional path length to the center of the cone. However, the simpler approximation of a radiating disk pretty much eliminates the high frequency ripples from the diffraction response. Adding in the additional path length from the cone would probably result in even greater randomization of the phase, so the summation would probably show even less ripple. However, when you click on the button for point source vs disk source in my program, you can see that the disk source has ripple components that are very small--well less than .5db. So the additional accuracy of a 3D driver model is probably not worth implementing.

A much greater issue than what you bring up is the effect of diffraction from adjacent drivers, and the amount of absorbtion as the sound wave from one source passes by an adjacent driver that is not a perfectly reflective surface. Also, there is air friction along the baffle that can should be accounted for in a "perfect" model. But at some point you need to decide when the model is "good enough" for modeling the design.

Yet another example is a wave-guide loaded ribbon tweeter, how would you model that in your program exactly ? Unless you accurately model the functioning and internal geometry of the wave-guide itself (possible, since programs like horn-resp do it, but far from trivial) the results will be erroneous.

Sure, the geometries here are quite different from the simple driver-on-a-baffle topology. The ray-tracing methodology can still be used, but the algorithm to sum the various paths will be different.

Hmm, the fact that 1/N^2 losses were not taken into account surprises me, and certainly casts some doubt on the results...I'm also concerned that when you added this in you get less than 6dB - with an infinite baffle it should be exactly 6dB, not nearly 6dB. What result do you get if you enter a very large baffle ?

I think you've got it backwards. With a large baffle you will get less than 6db. The larger baffle will result in a lower baffle step frequency and a longer path from the edge of the baffle than from the center of the driver. Since the path length from the edge of a large baffle is longer, there will be greater attenuation from the reflection due to 1/N^2 loss, hence less overall contribution.

Anyway, the impact of using the 1/n^2 losses is typically less than 1db, so I can understand why this effect wasn't considered in the other models. There are some measurement conditions where the effect is greater (nearfield), so it is nice to have in the model.
 
Really strange 😕

Never had any problems here through Sky Broadband, and I checked it on my phone (Three) and that worked too, who are you using, and which browser ? Perhaps a browser problem or an over-enthusiastic internet security / firewall program on your PC ?

Siegfrieds website is very plain HTML and images, so shouldn't present a problem to any browser...and its a great resource, so it would be a shame if you can't access it.

It's Virgin and I use a Mac with Safari (first time I tried with IE on MacOS 8) and no firewalls or other internet security.
I've tried for years by now but it never once loaded, it is the ONLY site I have ever had problems with.
 
I recently thought of an intuitive model for thinking about diffraction that helped me. Strictly speaking it's not quite accurate, but I think it shows how baffle step and edge diffraction are variations of the same thing.

The key idea is that wavefronts have a minimum size based on their wavelength.

Imagine a point source. A wavefront is going to have a minimum size around this point, which we might imagine as a sort of sphere of compression or rarifaction. Now imagine the point size on a baffle. At high frequencies, the minimum sphere around the source is quite small, so a pattern of clean hemispherical wavefronts form and propagate away from the source. Now imagine lengthening the wavelength until it's larger than the baffle. The minimum size sphere around the source is now larger than the baffle. If we pictured this minimum sphere as a ballon taped to the baffle, it's easy to imagine seeing it peeking around the edges of the baffle/box if you looked at it from behind. That the key to the diffraction: the space behind the baffle can "see" some portion of the sphere.

Likewise, thanks to Huygens' principle, we can imagine any wavefront in space as a bunch of point sources scattered over the wavefront's surface. If we imagine the point sources of a high frequency wavefront at the edge of a baffle, again, we can imagine these minimum size spheres peaking around the edge.

Likewise, we can imagine a horn, where the minimum size sphere may fit within the horn. It's clear why we gain efficiency: the pressure that normally would be part of the sphere outside the horn is reflected back into the area of the sphere that is within the horn. So as the sphere grows in relative size to match the horn's dimensions, the fraction of the sphere reflected onto itself shrinks, hence the horn loses it's efficiency gain.

In any case, this isn't strictly in line with the math as I understand it, but I find the mental image intuitive.
 
After a wee bit more thinking I get it now.

The mistake I made in the visualization was jumping from long wave-lengths straight to short ones. It became clear once I visualized a smooth increase in wave-length.

Still would like to see Siegfrieds site though, I feel like I'm losing out.
 
It's Virgin and I use a Mac with Safari (first time I tried with IE on MacOS 8) and no firewalls or other internet security.
I've tried for years by now but it never once loaded, it is the ONLY site I have ever had problems with.
I'm using Firefox on a Mac. I tried Safari and that worked fine too. Still puzzled why you're having problems. Perhaps try installing Firefox or Chrome to see if there is something wrong with your Safari installation ? Or emailing virgin technical support to see if they have inadvertently blacklisted his web-hosting provider ? Can't think of anything else..
 
It would be very easy to model the additional path length to the center of the cone. However, the simpler approximation of a radiating disk pretty much eliminates the high frequency ripples from the diffraction response. Adding in the additional path length from the cone would probably result in even greater randomization of the phase, so the summation would probably show even less ripple. However, when you click on the button for point source vs disk source in my program, you can see that the disk source has ripple components that are very small--well less than .5db. So the additional accuracy of a 3D driver model is probably not worth implementing.
It's more than just additional path length to the centre of the cone due to the cone having depth, the shape of the cone fundamentally alters the radiation pattern of the driver relative to a flat disc regardless of the presence or absence of any baffle, finite or infinite, particularly in relation to radiation along the plane of the baffle.

If the geometry and directivity of the driver isn't modelled properly (and practically speaking the directivity can't be modelled above the piston range of the driver without taking measurements, in which case why not just measure the diffraction directly) then the baffle contribution can't be modelled correctly either.

Is it worth creating a detailed 3D model ? Probably not. It would be an extreme amount of extra complexity for a modest improvement in accuracy. As I said, typical baffle step simulators are useful for what they are - an estimation that gives a good idea of where peaks and dips are likely to occur, but an estimation that must be followed up with measurements, especially in the case of more exotic drivers, where the simplified model really isn't that accurate.

Fortuitously, baffle diffraction simulators almost always make things look worse than the really are, and when they are in error the measured response is usually better than predicted, not worse. 🙂 Understanding the limitations of simulations is key to making good use of them and not ascribing more accuracy to them than what they really have.


A much greater issue than what you bring up is the effect of diffraction from adjacent drivers, and the amount of absorbtion as the sound wave from one source passes by an adjacent driver that is not a perfectly reflective surface.
Yes that's an issue too, something that could only be solved by a geometric model, particularly the diffraction from the edge of an adjacent driver where the cone starts to form a hollow.
Sure, the geometries here are quite different from the simple driver-on-a-baffle topology. The ray-tracing methodology can still be used, but the algorithm to sum the various paths will be different.
But do any of the current diffraction simulators model this type of driver properly ? Not that I'm aware.
I think you've got it backwards. With a large baffle you will get less than 6db. The larger baffle will result in a lower baffle step frequency and a longer path from the edge of the baffle than from the center of the driver. Since the path length from the edge of a large baffle is longer, there will be greater attenuation from the reflection due to 1/N^2 loss, hence less overall contribution.
No, I think it's you that has it backwards 😉 The difference between full space and half space with an infinite baffle is exactly twice the sound pressure level, eg 6.02dB. Period. If your model doesn't converge on precisely twice the SPL as baffle dimensions approach infinity there is a flaw in your model or assumptions. That's why I expressed concern about the accuracy of the model. Large baffles don't give "less" than 6dB total shift.
 
If small radius round overs can do the job, were the large "wings" certain Infinitys utilized back in the 80s overkill to help increase sales to picky audiophiles or just marketing-based i.e. they looked special so stood out in a crowded demo room?

My understanding is the radius size is linked to frequency.

So, a smaller radius will only function at higher frequency.

What I have read and heard is a 2" to 3" radius is meaningful as far as diffraction reduction. A 3/4" radius is much less so.
 
what about absorption?

Here is a brief article about using absorption to reduce or eliminate the effects of diffraction, as well as a data sheet on a sheet foam product. The article states that even horns have diffraction problems at their edges and benefit from acoustical absorption material.
 

Attachments

Status
Not open for further replies.