Aspects of speaker design

[neildavis]

I tried a similarly ambitious software effort for loudspeaker design over 10 years ago, and I can attest that it is possible to make significant progress on your own, but that it is unlikely that anyone will join you in your efforts.

There are at least 6 stumbling blocks that will make it difficult to get others to join:

1) You must have an over-arching architecture in mind for the components. The component model that I used for PSD is shown below. That diagram defines how the components interact and the data that passes from one component to another. I had to refer to this diagram fairly often while building PSD and ASD.

2) You need to define a shared data model, and get every developer to use it and maintain it. The data model provides a common "language" for sharing parameters and defines the standard units. The data model I used for PSD is here: www.audiodevelopers.com/Software/Online_help/Shared/Object_model.htm

3) You will need to get some agreement on the implementation approach. For example, I started to play around with 3D audio modeling using Visual Studio and WPF, which has built-in 3D libraries. But then I started to get interested in the ray-tracing features of the new video cards, which would probably provide much needed acceleration of 3D audio rendering.. Obviously, there are other 3D libraries used in other disciplines, but getting people to agree on a common approach when there are so many new shiny technologies is a challenge.

4) Agreeing on a common HMI approach is always a challenge. The web-based HMI is good solution, but it's not how a lot people like to develop code.

5) It will really help to do some use-case analysis to get a clearer idea of how these tools would be used. Modeling is fine for simple, well defined problems, but for some complex designs experimental measurements might be the only accurate way to characterize loudspeaker behaviors. For example, the cabinet vibration problem has a very large number of variables, such as the type of wood, wall thickness, amount of bracing, amount of energy transfer through the cabinet feet, relative dimensions of the cabinet walls, number of corners, etc. Since it is impractical to design a model that accounts for all of these parameters, you would want to identify the "class" of problems for which the model would be useful. Those constraints would get identified in the use-case modeling.

6) A lot of people who might have joined you 10 or 15 years ago are now interested in other loudspeaker technologies that aren't on your list. For example, I think active speakers are the only way to go, and I am enamored with line array designs, especially ones with electronic curvature. Also, I am impressed by what people are doing with 3D printers--that's opening an exciting new set of possibilities for high SAF designs. Maybe 15 years ago I would be interested in modeling cabinet resonance modes, but I don't have any interest in that now. I like the idea of using more 3D analysis for audio design, but most of the topics on your list aren't ones that I still find interesting.

[LOUT]

We could use something as easy to use with real time graphical estimations like PSD-Lite, but allowing more complex filters. An added feature to graph excursion with the crossover would be slick. ( it is my belief, a great many designs over-tax the tweeter)

We could use a model for adding mass to a driver. What does it do to efficiency vs Vb, vs Fs and what does that mean to Vb? Graphical results of course.

I recently realized these couple of things at least seem to be in the free VituixCAD cabinet and XO modelling software.

A lot of the other stuff you mentioned would be pretty cool to have in free software with nice graphs to show what's going on.

[Chris Roemer]

tvr - while Edge doesn't do baffle edge treatments, it CAN do multiple drivers.

[tvrgeek]

It might have been 25 years ago, when we were still using slide rules and scientific calculators to design enclosures. Today it's just as irrelevant as those devices. For that matter it was when written.

I disagree. It helps to understand what the simulation is telling you, besides the explanation of how it works. There are details here not covered in a dumb sim.
Back in the slip-stick days, we never made mistakes on order of magnitude as we had to keep track! These days in calculators, it is not uncommon to be an order of magnitude off and not have the experience to recognize it.

Another aspect I was looking at is group delay. Never gave it much thought. I ran across this:
http://speakerdesignworks.com/group_delay.html

Lots of discussion, but the conclusion is quite enlightening.

Yes, I have a 12 inch Pickett, though I got though school with a 6 inch plastic wonder. Calculators were not permitted when I was a freshman as they were unfair. Only 3 or four in my class had one.

[andy19191]

I tried a similarly ambitious software effort for loudspeaker design over 10 years ago, and I can attest that it is possible to make significant progress on your own, but that it is unlikely that anyone will join you in your efforts.

Hmmm... The previous open source project appears to have confused things somewhat. The current software I am developing is part of my day job (i.e. how I earn a living) and is not part of my hobby interests like speaker design. This second project concerns writing words not code on aspects of speaker design that are currently relatively poorly understood by speaker DIY folk (but not engineers or scientists from the relevant fields). The software is to be used to provide detailed quantitative evidence around which to draw conclusions about how to go about designing DIY speakers. I am seeking to use a hobby interest to select more interesting worked examples to use in the development of the software.

The software is not a 0d/1d network solver of the kind in the previous project and most of the currently available speaker design tools. It is a general 3d solver for fluids and solids. It is a more appropriate approach for some problems like cabinet radiation and aspects of the other topics listed in the OP. 0d/1d network solvers are more appropriate for other tasks where 3d geometrical details have less influence.

[tvrgeek]

Andy,
I see one of your issues you are highlighting. That is so many try to jump in to this hobby without putting any effort to learn the very basics. Many seem to think they can build a megabuck system for pennies by getting others to do all their work and design that violates the laws of physics. " Design my cabinet, design my crossover for me, help making a 4 inch sub in a 1/2 l cabinet be a sub for my HT"

I do not expect the majority of beginners to understand the s-plane, but should understand what "impedance" means. Not every beginner would understand why system impedance matters to amplifier stability and distortion. One should at an absolute minimum understand what the TS parameters mean. So, for all the beginners out there, PLEASE start by picking up one of the basic books on speaker design. PE sells one. Learn the syntax. Learn the most basic fundamentals. Your first project will go a lot smoother and likely much better results.

Now Andy, again I go back to suggesting you pick one topic from your list.

Reading the details of the software you want to write. There are some very advanced fluid dynamics simulators out there. Many based on MATLAB. I am sure that behavior of a port can and has already carefully modeled. I am vaguely familiar with them from understanding intake and exhaust systems as they apply to my old British sports cars, starting with the original Smith and Morrison work to modern very complex fluid analysis. There are also many examples, again using MATLAB for panel resonance. The problems with these are requirements to understand the materials and physics way above the hobbyist, and by the many so-so commercial designs, above most professionals too. The hobbyist is not going to know the density or modulus of any given piece of MDF and moving to laminates, really really complex. A lot of work was done trying to understand violins. (I have seen some simulating vibrating sand on a panel.

I suspect a simulator that could graphically and numerically predict "chuff" based on the port shape might be slick. Draw a profile, graph the predicted spectrum across velocity. Not just flair, but shape. Round slot, corner etc. Numerically, what does a 1/2 inch radius do for a corner port, vs a 3 or 4 inch deep flair? Does the flair profile make a significant difference as it does with horn radiators? We know from manifold design, the most efficient entry and exit is a 7 degree taper with a 270 degree radius, though a 1/2 inch radius to a flat plane is about 95% as effective.

[neildavis]

It wasn't clear whether you were looking for help on writing software or had other intentions. I wanted to give you some feedback on my experiences in attempting a group project, but I wasn't sure whether that experience was relevant or not. Sounds like it isn't relevant, but I meant it to be constructive rather than disparaging, so I it posted anyway.

A while back Jeff Candy wrote an article on solving the 3D Helmholtz equation for calculating baffle diffraction. It's the "right" way to solve the problem, but most of us can only get our heads around "brute-force" ray-tracing or other approximations to this 3D problem. I'd love to see more applications of CFD tools to loudspeaker designs, but this stuff is not for the mathematically challenged and requires significant computing resources (Jeff''s calculation required about 2 seconds per frequency on a modern workstation).

I had written down some topics I was interested in exploring with 3D modeling a while back, but I lost some files in my last computer upgrade. I'll explore my backups and see if I can find those topics and the associated reading material that I saved . I had a couple of interesting articles on room reverberations and 3D stereo imaging...

[tvrgeek]

The following is an advertisement, but does touch on some of the topics mentioned herein.

http://usenclosure.com/Our%20New%20T...abilities.html

Also, search: " manifold flow simulations "

[tvrgeek]

Let's here it for modern computers. My first Spice simulation ran on a 286, DOS and had 2 op amps. Took over two days. A full analysis on my current desktop takes a few seconds.

Would the 3D Helmholtz model be able to resolve edge diffraction accounting for radius and profile? We assume a round profile as we can buy router bits of that shape, but is it optimal, or at least is is "close enough for government work" ? I also found audible differences when I radiused the BACK edges of my cabinets. Surprised the heck out of me. I did it originally for visual symmetry.

[neildavis]

Would the 3D Helmholtz model be able to resolve edge diffraction accounting for radius and profile? .

The answer is "yes", but if you read Jeff's article you would see that his analysis is for a simple six-sided box, and you can see how he modeled those surfaces mathematically. It is possible to add rounded corners to the model, but I wouldn't be able to set up the necessary equations since I don't even understand the equations for the flat surfaces. The calculation would probably take a lot longer since you would have to use piecewise integrations and the equations for the rounded sections would be more complex.

Other popular baffle diffraction tools such as Edge, Jeff Bagby's Baffle Diffraction Simulator and PSD-Lite all use simple ray-tracing algorithms. PSD-Lite actually accounts for baffle edge rounding, using the method described by Paul Verdone. This feature is discussed in the Help file for the Diffraction Model in the PSD-Lite Baffle Design tool. An updated link to Paul's BDS manual in which he describes the algorithm is here. I didn't implement the chamfer shape in PSD-Lite, but it would be easy to add that variant. Paul's assumptions about the weighting values for points along the roundover make sense, but I don't know whether those weights were validated experimentally..

BTW, the biggest difference between the Edge program and the Diffraction modeler used in PSD-Lite is the ability to change the summation point. In PSD-Lite it is at a fixed angle, although you can change the distance ("z-axis"). And the reason you can't change the angle of the summation point in PSD-Lite is that I refused to add this feature without using 3D visualization tools. What I wanted was a "3D walkthrough" capability to visualize the intensity of the audio for all frequencies, and to be able to see how the stereo image might change with varying frequency content. That's why I started looking at WPF, and I even started playing with Unity to see how hard it would be to implement that 3D walkthrough with a gaming engine.

[dlr]

The software is not a 0d/1d network solver of the kind in the previous project and most of the currently available speaker design tools. It is a general 3d solver for fluids and solids. It is a more appropriate approach for some problems like cabinet radiation and aspects of the other topics listed in the OP. 0d/1d network solvers are more appropriate for other tasks where 3d geometrical details have less influence.

I would be extremely interested in seeing visual representation of wave radiation across the baffle surface and its change at baffle edges having to do with diffraction. Whenever I experiment with diffraction control, I try to make a mental visualization of this. I have a lot of questions related to that. Diffraction due to other things such as a nearby driver would be equally interesting, but I suspect that it would be far more difficult due to the rather random shapes in various drivers.
dlr

[dlr]

BTW, the biggest difference between the Edge program and the Diffraction modeler used in PSD-Lite is the ability to change the summation point. In PSD-Lite it is at a fixed angle, although you can change the distance ("z-axis"). And the reason you can't change the angle of the summation point in PSD-Lite is that I refused to add this feature without using 3D visualization tools. What I wanted was a "3D walkthrough" capability to visualize the intensity of the audio for all frequencies, and to be able to see how the stereo image might change with varying frequency content. That's why I started looking at WPF, and I even started playing with Unity to see how hard it would be to implement that 3D walkthrough with a gaming engine.

I may be at a point soon to take up a diffraction app myself. We've discussed this before. I have a complete 3D section in WinPCD that I can use as the start point for front hemisphere calculations. I've got my own ideas on UI for it, so I may contact you later on this if you have some additional info. I don't have Jeff's diffraction algorithm. Of course the most difficult part for me will be the 3D graphical UI for drawing the baffle surface and translating that into useful code for calculations. I have not written any of that before and will be starting from scratch. I don't know if there are any good open source resources for that, either.

dlr

[neildavis]

I'd be happy to send you the code I have, but it is VB.net and much of it is using the drawing library to create the baffle and box shapes. I don't know how easily that would translate to C#. I translated some of the utilities I have for the DSP version of the code to C# and I've thought about redoing some other pieces of VB.net code, but this module would be difficult....

My diffraction modeler was a completely new program. I wasted an enormous amount of time trying to convert the spreadsheet equations for WBCD to VB.net, and even put the original cell numbers in comments so someone else could trace it. But because of that faithfulness to the original spreadsheet logic, the code is ugly and difficult to maintain. When it came time for implementing the diffraction model, I looked at Jeff;s BDS and said "no way". My diffraction modeler does essentially the same thing as Jeff's BDS, but the code is much easier to follow and it is well documented, The number-crunching is a little confusing because I split the processing into 4 separate threads to take advantage of multi-core CPU's, but once you see how that is done the logic is very easy to follow.

Unfortunately, the bulk of the diffraction modeler is user interface code: drawing shapes from vertices, processing mouse moves and clicks and checking bounds. It's probably over 90% user interface and less than 10% number crunching. The code seems to work fairly well for building a lot of interesting shapes, but I kept thinking that all of the drawing should be done by a 3D drawing tool and that I needed to figure out a way to import data from a tool like Sketchup or use a third-party 3D drawing library. Given that I still think an external 3D drawing tool is a better way to go, I have no intention of updating this diffraction modeler code.

Stupidly enough, the biggest problem I had with that code was figuring out how to calculate the vertices of a the inner walls from the vertices defined by the outer walls of the box, where the corners can be at any angle. I needed that capability to calculate the box volume from any wall thickness and to calculate the effect of the roundovers.
:

[andy19191]

Would the 3D Helmholtz model be able to resolve edge diffraction accounting for radius and profile?

The Equivalent Source Method described in the paper is fiddly and not widely used for engineering simulations. The dominant engineering tool for something like a speaker baffle is the Boundary Element Method. There are several good commercial examples and a few basic open source examples. The open source one I used briefly for cabinet radiation was acousto. It works but is a bit basic and cumbersome. Here is an example of it being used for a speaker baffle but the original seems to have been moved or taken down. With a typical computer you are likely to struggle to accurately resolve upto 20 kHz (I would have doubts about the level of accuracy at 10 kHz in the simulation shown given the low order element being used) but a few kHz should be no problem.

[andy19191]

I would be extremely interested in seeing visual representation of wave radiation across the baffle surface and its change at baffle edges having to do with diffraction. Whenever I experiment with diffraction control, I try to make a mental visualization of this. I have a lot of questions related to that. Diffraction due to other things such as a nearby driver would be equally interesting, but I suspect that it would be far more difficult due to the rather random shapes in various drivers.

If you use a general method like BEM then it isn't really a question of difficulty but more a question of adequate resolution. Visualisation tends to be handled by tools like paraview which has some animation support. It handled animating mode shapes for the cabinet in a kludgy manner (the implementation was not done by the paraview team but a third party). It might do what you want. I ended up using a webgl library and a bit of javascript but wasn't happy with it as a long term solution. Things are evolving fairly rapidly and hopefully the next version of gltf or something similar will do what is needed in a standard way.