So the problem is that the waveguide above will give a very reasonably flat wavefront. For the device to work properly however, the wavefront needs to have some curvature, but still less than what it would be with only the outer walls of the waveguide present, and not the inner plug.
On Speakerplans i got the suggestion to make the inner cone/plug elliptical instead of circular.
So the design was redrawn and the equation set in Matlab rechecked for constant cross area expansion along the length of the device.
Matlab:
NX:
working it out further
Checking again to see if NX is correct with the equations.
I do have some doubts about the plane i use to cut the waveguide with. Right now i simply used the minor axis angle of 21.28 degrees.
One can imagine however that once the eccentricity of the ellips becomes large and thus the minor axis angle much smaller than the major axis angle, that the pathlength differences become large as well.
I'm not sure yet what the pathlength difference will be with this implementation of the ellips if you consider the position on the slot as function of height. I cannot wrap my head around what shape it will be, if it will be an arc or something completely different.
Also cannot really think of a formula that will help me determine it. I will let the simulations provide me with the answers to see if this idea really works.
So them some time past and i ran some comsol simulations for various values of the eccentricity of the ellipse.
I tried various values of the eccentricity E between 0.4 and 0.8. Any lower values will make it almost circular and any higher and i might as well leave out the central plug.
Below a table with the values of the eccentricity E and the major and minor axis angles.
E ThetaMajor ThetaMinor
0.4 33 30.8
0.5 33 29.4
0.6 33 27.5
0.7 33 24.9
0.8 33 21.3
So i ran a lot of simulations and these are my main conclusions:
At E is about 0.4 and 0.5 i could not really see what the effect of the elliptical shape is. That might have to do with the fact that the difference in angle and therefor path length between the major and minor axis is actually still very small.
At E is 0.8 the curvature of the wavefront is too high for it to still work with 3 drivers side by side.
somewhere between 0.6 and 0.7 seems to be the optimal wavefront curvature
At higher frequencies, the wavefront curvature is the best defined. The device actually works better at these higher frequencies than at low frequencies.
At lower frequencies i can clearly see standing wave patterns in the radial direction at constant height. That means there is a running wave and thus a radial velocity component always present in the air. This leads to damping and dissipation as well as probably screwing up the impulse response at lower frequencies.
To avoid this, i might put some ribs parallel to the major and minor axis. These could then also serve as a means of suspending the inner plug to the outer wall. I will need to investigate this later.
All waveguide designs i have simulated so fart however, have shown these pressure cells that i believe are standing wave patterns. This has got to be improved upon.
Anyways; less talk more images.
The images below are all for E = 0.7 at different frequencies.
5kHz
7kHz
10kHz
and a picture of the pressure cells/standing wave patterns at 4kHz. Note the outer sphere is totally green i.e. SPL is around 0 Pascal. This is because you are looking at the outer layer, which is a perfectly matched layer meaning it will absorb nearly all the impeding sound pressure such that there will be no reflections and we are really simulating free space but in an enclosed geometry. This is also why i provide the slice views instead of these surface plots.
I also tried to change the mouth width from 25mm to 35mm but at 10kHz i can already see the results become unacceptable. 25mm looks good and it still gives me the desired area expansion factor (about 5 degrees when you translate cross sectional area of the waveguide to a conical section)
So whats next?
Well, i need a faster computer for starters. The results i get at 10kHz are aceptable, but at 11kHz i can clearly see the mesh appearing and the results becoming very unphysical. To improve this, i need to shrink the mesh size, but my computer only has 8GB of ram memory and i need more to prevent comsol from giving a memory overload when its decomposing matrices.