I was reading some of those earlier posts, quite interesting.
The offset along the axis of the waveguide for an angled throat is in fact small as is the radius difference, but these should still be corrected. People were quite correct in pointing out that holding these small differences in wood and fiberglass is quite difficult.
In fact this became a real problem in production because we could measure these variances in the throat matching in the acoustic response. I have since changed the process so that the last step is to cast the throat geometry into the device with epoxy. This yields a dimensionally correct shape that lines up perfectly every time. I can't imagine how you would ever hold sufficient accuracy in wood. Fiberglass can't.
Once I saw the trick to the math for the throat it was quite easy. Sometimes you just don't see the easy road right off, so you take the one that works and forget about looking for anything simpler.
The offset along the axis of the waveguide for an angled throat is in fact small as is the radius difference, but these should still be corrected. People were quite correct in pointing out that holding these small differences in wood and fiberglass is quite difficult.
In fact this became a real problem in production because we could measure these variances in the throat matching in the acoustic response. I have since changed the process so that the last step is to cast the throat geometry into the device with epoxy. This yields a dimensionally correct shape that lines up perfectly every time. I can't imagine how you would ever hold sufficient accuracy in wood. Fiberglass can't.
Once I saw the trick to the math for the throat it was quite easy. Sometimes you just don't see the easy road right off, so you take the one that works and forget about looking for anything simpler.
JoshK said:
Many of the CAD packages can do this. I use Smartsketch from intergraph and it can do scale drawings 1:1. But the printers vary. My laser is pretty bad, but I have an inkjet thats pretty accurate. You have to check the printed sheet for accuracy, you just can't assume it is because the computer said 1:1.
My molds were done with CNC and there were no drawings only codes to the CNC cutter head, so this was not an issue.
Anyone aware of a modeling package that will do FEM simulation of acoustic wave propagation? I've read references to such simulations in some papers but no mention of specific publicly available or commercial software.
KSTR said:
Too much like my day job.
KSTR said:
Good catch.
KSTR said:
Reassuring.
I don't know any CAD programs, at least I've never used any before. That is what I was hoping others could help with.
If there is interest, for EAGLE (Schematic/PCB-Layout CAD software) I could write a C-quicky which would take the input from a coordinate text file (cut/&pasted from XL) and generate the script commands for a printable template. This could be used with the built-in CAM-processor for outputs into various "devices" (printers/plotters, TIFF-file), but even the direct printout from the program gives quite accurate results IME (depending on the printer and/or windoze's settings).JoshK said:
- Klaus
If you want to visualize the wavefront, CARA might be an interesting program to view the wavefront propagation.
CARA is not industrial grade acoustic modeling, you have to pay serious cash for those, but it is under $100 and you can master in about an hour.
You could probably do a 2-D approximation. I don't know if you import a curve or you will just have to draw many,many tiny segments.
I have only played with it briefly, so I do not know how accurate it is.
CARA is not industrial grade acoustic modeling, you have to pay serious cash for those, but it is under $100 and you can master in about an hour.
You could probably do a 2-D approximation. I don't know if you import a curve or you will just have to draw many,many tiny segments.
I have only played with it briefly, so I do not know how accurate it is.
I have two purposes. One is for simulation of speakers so I can test various speaker geometries to house a plasma driver; I'm not interested into the coarser room acoustics stuff.
I also need something accurate enough to simulate an HRTF from input of laser scanned heads, so that I can then use the HRTF in DSP. There are papers that have used this method to find HRTFs because actual measurement is very complex and expensive, requiring an anechoic chamber, many speakers, and a long measurement time. A laser scanner can be had for a couple of thousand, and it takes minutes to get a decent scan. However, I didn't find software available for download.
As a last resort, I can write my own and it would probably be faster (I'm very good at optimization), but that is quite time consuming and I'd rather pay up to a couple of hundred.
I also need something accurate enough to simulate an HRTF from input of laser scanned heads, so that I can then use the HRTF in DSP. There are papers that have used this method to find HRTFs because actual measurement is very complex and expensive, requiring an anechoic chamber, many speakers, and a long measurement time. A laser scanner can be had for a couple of thousand, and it takes minutes to get a decent scan. However, I didn't find software available for download.
As a last resort, I can write my own and it would probably be faster (I'm very good at optimization), but that is quite time consuming and I'd rather pay up to a couple of hundred.
abzug said:
I did my PhD in FEM in acoustics. Here are my experiences.
FEM tend to be extremely limited in frequency. One needs a lot of elements per wavelength to get good results. Hence the size of the problem grows very very fast with frequency. As I started to use FEM for real problems I found that it was far too limited in frequency to be useful. Thats why I searched for analytic solutions to the waveguide problem. Its no coincidence that FEM has never really been useful for the waveguide situation, while with an analytic solution I was able to break through a lot of barriers.
While FEM has gotten better in the last few years and other techniques like BEM can work well, they all suffer from the "geometry leads to solution" problem. You input geometry and get a solution. FEM can never tell you what geometry is correct only what a certain geometry will do.
After study it and mastering it, I have never actually found it to lead to any breakthoughs. I don't use it all anymore.
i.e., it's just a matter of speed and memory requirements.
However, with current and ever improving hardware this is a limited problem. Rather than using FEM/BEM, for example, a simple FDM simulation is a problem trivially parallelizable and easy to implement on a GPU.
Aren't analytic solutions only possible for simpler geometries?
Well, in the case of finding HRTF by simulation, that is exactly what I need. I'm not trying to optimize the input geometry, which is the shape of my head...now that would be something 😀
Coupled with a nonlinear or stochastic optimizer, such as conjugate gradient, on an intelligently reduced space of parameters, simulation might not take too many runs to get convergence (and in many cases an good enough rather than a perfect answer might be sufficient).
Josh - nice work.
I'm getting ready to start building a few of these WGs too but I've been able to cheat a little - I've got a full blown CAD/CAM system. 🙂
I wrote the basic OS formula into a post processor meant for generating NC GCode and had it output XYZ coordinates to a comma delimited text file that could be read back into the system as a series of points on the curve. I set it up so I could vary the step size for as fine a resolution as needed, and then just ran a spline thru the points after reading them back in. After that, it was pretty simple to create a line at the appropriate angle tangent to the curve, add the appropriate mouth radius, and then simply shift the whole thing to where the angled section intersected the throat diameter at 1 inch or whatever. Not very intellectual, but it works great.
I took a close look at the area of the throat and I agree with Earl - if you're gonna build these out of wood, good luck holding the tolerances down there. A mold of aluminum or steel is a better way to go.
At 6 degrees (12 degrees included) that's a pretty small change of < .003 per side and not much more than a mm in length. I'm not gonna argue about it's importance, but that's gonn be a bear to do with a manual wood lathe. 🙂
I'm getting ready to start building a few of these WGs too but I've been able to cheat a little - I've got a full blown CAD/CAM system. 🙂
An externally hosted image should be here but it was not working when we last tested it.
I wrote the basic OS formula into a post processor meant for generating NC GCode and had it output XYZ coordinates to a comma delimited text file that could be read back into the system as a series of points on the curve. I set it up so I could vary the step size for as fine a resolution as needed, and then just ran a spline thru the points after reading them back in. After that, it was pretty simple to create a line at the appropriate angle tangent to the curve, add the appropriate mouth radius, and then simply shift the whole thing to where the angled section intersected the throat diameter at 1 inch or whatever. Not very intellectual, but it works great.
An externally hosted image should be here but it was not working when we last tested it.
I took a close look at the area of the throat and I agree with Earl - if you're gonna build these out of wood, good luck holding the tolerances down there. A mold of aluminum or steel is a better way to go.
An externally hosted image should be here but it was not working when we last tested it.
At 6 degrees (12 degrees included) that's a pretty small change of < .003 per side and not much more than a mm in length. I'm not gonna argue about it's importance, but that's gonn be a bear to do with a manual wood lathe. 🙂
abzug said:
To get HRTFs to 10 kHz. for a laser scaned head would be, I suspect, not doable with numerical techniques.
But an analytical solution to say a sphere or a spheriod would be. And as you said yourself "and in many cases good enough rather than perfect answer might be sufficient".
When I was doing HRTF's some ten years ago or more I used a spherical solution and got very good results.
So the question is always "Obtain an approximate answer to the exact geometry, or an exact solution to an approximate geometry?" There is never an absolute answer here.
At any rate, this thread is about waveguides not HRTFs and for waveguides, analytic wins hands down over numerical solutions.
AJ said:
I would be interested in having you do some prototyping work for me if you can contact me at egeddes@gedlee.com
14 kHz with hybrid approach: http://www.aes.org/e-lib/browse.cfm?elib=14005gedlee said:
Experimenting with impulse responses recorded from various subjects I downloaded from the HRTF database, in a convolution plugin for my music player, gave different audible results for me when trying data from different subjects, showing that small differences in fact create a significant effect. Indeed, even the individual variations of the shape of the pinnae need to be captured for a complete effect, especially to be able to distinguish sound source locations spaced in the vertical direction. So I don't see how a spherical solution can be considered sufficient.
The approximation quality achievable in the former case is a function of ever improving hardware; in the latter case, it's not, unless someone develops an amazing mathematician AI.
OK. I'm sorry for getting the thread too much sidetracked.
I was under the impression that CARA, being used for room design, gives you the wave path vectors. Does it really give you wave front shape if you gave it a wave guide type of structure?mbutzkies said:
Member
Joined 2003
Josh,
I apparently don't understand how to use your spreadsheet. For example, if the throat radius is changed to .5 or .7, nothing happens in the chart. Same for changing the throat exit angle...nothing seems to happen in the chart, even for very high exit angles. Theta etc seems fine...and I really like the mouth roundover function 😎 .
In John K's spreadsheet, the same changes produce what I would expect to see...throat radius changes, high throat exit angles approach a conical flare, etc.
What might I be doing wrong?
Paul
I apparently don't understand how to use your spreadsheet. For example, if the throat radius is changed to .5 or .7, nothing happens in the chart. Same for changing the throat exit angle...nothing seems to happen in the chart, even for very high exit angles. Theta etc seems fine...and I really like the mouth roundover function 😎 .
In John K's spreadsheet, the same changes produce what I would expect to see...throat radius changes, high throat exit angles approach a conical flare, etc.
What might I be doing wrong?
Paul
abzug said:
The technique that you describe has not been proven to be accurate. Just because someone can do the calculations does not mean that they are accurate.
And I was pointing out that you yourself were being and still are being hypocritical - saying at one point that high accuracy was required and at another that it may not be. I have done a lot of HRTF work and I know the issues. For extremly accurate subjective effects you need extremely accurate HRTFs. Spherical models work well for lower frequencies but not for very detailed subjective perceptions.
Nothing is going to beat using measured HRTFs for the individual in getting highly accurate subjective impressions. But from a practical standpoint this is unrealistic for anything other than academic interest. One needs to ***** the goals and the available solutions.
As the attached paper says, the problem of HRTFs at high frequencies is huge. No one ever discussed the errors in processing such extremely large arrays of numbers.
Using "multipole fiting of the HRTF's" would be a form of using the best of both worlds - the numerical stbility of the analytical solution with the arbitrary geometry capability of the numerical one. It makes perfect sense to me as a way to go, but I have not seen any Proof that it works well in real situations.
I have a patent on multipole fitting of loudspeakers from measured data. It works well in many cases, but there are cases where the matrices are near singular and troubles can occur.
soongsc said:
Thats why I don't paint the throat!!
And from experince, waveguide as shown won't work all that well. The reason, its coverage is too wide. The OS waveguides loose their control ability when the coverage angle gets large. Thats why I developed the Bi-Spheriodal waveguide in order to get wider coverage with low HOMs.
There is more to making things work right than just drawing lines on paper. You have to understand how they work and where the tradeoffs go against you. Wide directivity is deffinately NOT the OS's forte. Fortunately I always try to get as narrow a coverage angle as feasible which plays right into the OS strengths.