2 way waveguide speaker build ABEC modelling

I have wanted to build a high efficiency speaker for a long time but something else has always seemed to jump the queue. camplo's 2 way thread and some of the ideas there inspired me to buy a pair of Faital HF146R drivers. I plan to use Faital 15PR400 woofers but I haven't bought these as yet. I found mabat's Ath waveguide tool at the same time and since then I have been trying various different waveguide designs and simulating them with ABEC. mabat's tool makes that very simple in an infinite baffle. To go further and turn that into a free space simulation in an enclosure is a bit more complicated.

I found a design which seems promising and have begun simulating this in an enclosure with various edge treatments to see the effect of those.

The waveguide is difficult to render and show the contour but the attached images should give some idea.

I'll post some more details and images of the simulations to show the sort of output you can get from ABEC. I enlisted the help of DonVK to help me get my head around how to drive it and it is now starting to make some sense.

Although at the moment this project is still quite virtual I do intend to build it in the real world. To that end I have bought a CNC kit which I am in the process of building up. Hopefully it will work well enough to allow me to mill the the majority of the cabinet by CNC.
 

Attachments

  • RectXRounded.png
    RectXRounded.png
    279.8 KB · Views: 1,956
  • RectXRoundedLTrans.png
    RectXRoundedLTrans.png
    184.4 KB · Views: 1,963
  • Basic Enclosure.png
    Basic Enclosure.png
    345.6 KB · Views: 1,953
There has been fun but frustration too ;) It has been a good learning experience.

This is the Horizontal Pressure response of the basic waveguide simulated in an infinite baffle.

attachment.php


This is a good way to compare waveguides to see which ones are worth pursuing further as the waveguide can be modeled with a fairly high degree of accuracy without the simulation taking forever. The number of elements in the model is an important parameter to determine how long the simulation will take. 2500 to 3000 elements usually takes my i7 laptop about 30 to 40 minutes to solve, 4000 somewhere around an hour, 5000 around one hour and 40 minutes.

The size of the elements is also important to make the model accurate and realistic to higher frequencies. The simulation above was done with elements between 5mm at the driver through to 10mm at the mouth. This seems to be a good compromise between time and accuracy.
 

Attachments

  • Infinite Baffle waveguide Horizontal.jpg
    Infinite Baffle waveguide Horizontal.jpg
    65.3 KB · Views: 4,332
A little off topic but looking at your waveguide renderings couldn't help but think of stamped copper.
I was trying out the available surfaces in Fusion and the red metal finish seemed to show the contour reasonably well and look quite nice. Reality will likely be more subtle :)

Oh Dear, I'm in the same case as the first sentence but wanted to do it with simple 3" cone driver... Compression driver seem more suited. Subscribed :) . active amps & DSP EQ ?

Yeah Faitals 15PR400 or 15FS520 :cool:
Cone drivers at the top end tend to limit the overall max spl and dynamic potential and horn loading doesn't change that. A compression driver usually produces a flat wavefront which matches well with an OS or OS derived waveguide. A conical horn is a better match for a driver that produces a more spherical wavefront. mabat's tool can be programmed to show this quite easily.

ABEC would be a good tool to see the effect of the waveguide on the cone driver but it would take a fair bit of effort to get the meshing of the various parts to line up properly. I'll show later why it's important to pay attention to the mesh positioning.

I like the look of the 15FH520 but it's not so easy to get hold of and quite a lot more expensive as I would have to get it from overseas. If TLHP had two in stock a few weeks ago I probably would have bought them. Who knows maybe they will be in stock before I need to buy the woofers.

I will start active with DSP as I am more familiar with that, I might well try a passive crossover if real life measurements make it seem reasonable.

It looks as though much effort went into creating this waveguide. Was all round polars for the square shape the goal, or was it a distributed diffraction oriented design or maybe something else?

There was a fair amount of effort that went into refining the waveguide to the point it is now. I was aiming for a smooth response on and off axis and to have the DI be fairly flat and smooth. The pattern is narrower in the vertical, more than the outline suggests.

I would like to say that I was able to make Ath do exactly what I wanted but that is not the case. The effect of the supershape and the corresponding parameters is not that intuitive and sometimes it's just a happy coincidence. This was close enough to what I was aiming for to move on to the next stage and see what happened.

This is the predicted spinorama style chart from Vituix CAD. The data used for this is from the normalized polars which were normalized to 20 degrees as this is intended to be used in a geddes style setup with the speakers toed in. The listening window (light green trace) is pretty good and the reference axis is dead flat from the normalizing. The DI is pretty flat up to 4k and rises smoothly from there. Predicted in room response is orange hiding behind the early reflections curve in light blue.

attachment.php
 

Attachments

  • Vituix Spinorama RXRounded.jpg
    Vituix Spinorama RXRounded.jpg
    111.3 KB · Views: 2,520
Last edited:
diyAudio Moderator
Joined 2008
Paid Member
The effect of the supershape and the corresponding parameters is not that intuitive
So I see this was a shape in the Ath library, one that reminds me of the JBL M2. IIRC Earl once suggested that the goal with this would have been to create diffraction in consideration of the precedence effect with a pseudo random spread in time and space.
predicted spinorama style chart from Vituix CAD
I could see the benefit of running diagonal polars, and I could also see the spinorama data here as not being on point. What you may be trying to show isn't easy or straight forward as I've watched others attempt to do the same.
 
This was my first try at a freestanding simulation, it was a total disaster. The model contained 6500 elements due to only being able to use half symmetry and took over 12 hours to solve only to give me complete garbage.

The waveguide did not meet the baffle exactly and the vertices of the waveguide edge, baffle and interface (boundary between the interior and exterior sub domains) were not all aligned.

It doesn't look terrible in the screenshots but a tiny gap or misalignment causes real problems solving for the boundary elements. The results were so bad I didn't save the VACS graph.


860262d1594625634-acoustic-horn-design-easy-ath4-freespace-jpg




860263d1594625634-acoustic-horn-design-easy-ath4-freespaceguide-jpg
 
So I see this was a shape in the Ath library, one that reminds me of the JBL M2. IIRC Earl once suggested that the goal with this would have been to create diffraction in consideration of the precedence effect with a pseudo random spread in time and space.


I can see why you might think that but no this was not a shape that was included in the Ath demo files. There is some similarity sure but there are some key differences.

This is the JBL Style file included with Ath 4.5. It has "knuckles" that protrude out. I haven't run the simulation for this demo properly to show what it looks like but the same sort of thing in previous versions of the tool was a little ragged. Where and how the bumps stick out make quite a difference,


attachment.php



My waveguide doesn't do that, it's hard to see exactly what is going on without being able to revolve the 3D model. There will be diffraction but it simulates as well controlled to improve the outcome.

I could see the benefit of running diagonal polars, and I could also see the spinorama data here as not being on point. What you may be trying to show isn't easy or straight forward as I've watched others attempt to do the same.

I have ran diagonal polars and they are very similar to the horizontal and vertical, there is nothing horrible hiding there. I am much further along than I have shown so far but I am trying to post it in some semblance of a chronological order. So when I get to that you can tell me what you see.

In what way is the spinorama data not on point? What do you think I am trying to show and why is it not easy or straightforward?
 

Attachments

  • JBL Style Mesh.jpg
    JBL Style Mesh.jpg
    144.7 KB · Views: 2,114
I went back to 1/4 symmetry to speed things up while trying to work out the kinks in the simulation. Don helped me to understand that the waveguide mesh needed to be used to cut the baffle in the cad model to make sure that there were no gaps and that the mesh and baffle were conformal.

I exported the mesh from ABEC as an STL, imported this to Fusion, converted the mesh to a BRep surface and used that to cut the baffle. This made the corners line up to the mesh points and made the baffle fit tight to the waveguide mesh.


attachment.php



I was still trying to save some elements by making the baffle mesh less dense.

This is the simulated output, not bad till 5K but not right either.



attachment.php
 

Attachments

  • Freespace sym Horizontal.jpg
    Freespace sym Horizontal.jpg
    59.2 KB · Views: 1,414
  • Freespace sym Mesh.jpg
    Freespace sym Mesh.jpg
    192.7 KB · Views: 2,046
Don recreated the CAD files and meshes and ran a simulation showing much improvement. He used a flat interface instead of the one created from Ath and made it so that every waveguide and baffle vertex were aligned with each other. The vertical edge is rounded so the vertical looks smoother than the horizontal.


attachment.php



This is the result from that simulation, much better. I am using the Horizontal curves as a comparison because they are the most representative and show the differences between the different tries. I have all of the different graphs and outputs but it becomes tiresome to keep snipping them and attaching when the differences are all relative.


attachment.php
 

Attachments

  • Donv3.jpg
    Donv3.jpg
    56.2 KB · Views: 1,390
  • Donv3mesh.jpg
    Donv3mesh.jpg
    199.9 KB · Views: 1,500
I see there's a lot of struggle you are going through with these models... It will be possible to create an enclosure like that for a free-standing situation with the Ath tool automatically. I've just been quite busy last weeks. With Gmsh library it is easy to make the mesh element sizes vary so it can be denser at the front and get progressively less dense on the back sides (which is what I'd consider optimal). I only can't promise any deadline.
 
Last edited:
... He used a flat interface instead of the one created from Ath and made it so that every waveguide and baffle vertex were aligned with each other.
I've read this many times (about the flat interface) but it still doesn't make any sense to me. The whole purpose of the extruded interface is to avoid too sharp angles between boundaries at the mouth egde, i.e. too narrow spaces - this is even recommended by the ABEC author and I can see the point in that. Otherwise the shape of the interface should be just irrelevant.

Your second result is more stable at higher frequencies because the mesh elements around the mouth are smaller. And of course it will be better because the egde is rounded. Good job!
 
Last edited:
I wanted to try and get a similar result using my own CAD process. I spent a lot of time getting Rhino to Duplicate the mesh edge as a polyline where there was a point at every vertex. I was able to export that as a DXF and import it into Fusion to cut my baffle. Fusion still joins all the straight lines into one so it was no different to using the STL to BRep.

I created my own flat interface and meshed the interface and baffle only at 10mm as the edge of the waveguide used that resolution. This caused all the vertices where the waveguide baffle and interface joined to be very well aligned.

This is what the mesh looked like


attachment.php



This is the mesh before ABEC adds the symmetry


attachment.php



This shows the alignment between the three surfaces


attachment.php
 

Attachments

  • Abec Base Model.jpg
    Abec Base Model.jpg
    79.5 KB · Views: 1,505
  • Abec Model Symmetry.jpg
    Abec Model Symmetry.jpg
    164.2 KB · Views: 1,495
  • Alignment.jpg
    Alignment.jpg
    63.4 KB · Views: 1,457
Yeah, the nice thing about using the Gmsh library (when used alone) is that it aligns all edges automatically - you just don't have to care about that. It "simply" creates all the vertices from defined geometry and mesh densities (at each control point) and joints them into one cohesive structure.
 
Last edited:
Here are the simulated results for the baffle with a 36mm roundover on the horizontal as seen before. Left to right : Horizontal polar 20 deg normalized, Horizontal Curves, Vertical, Diagonal and curves exported from the normalized polar.

What is nice is that the numerical instabilities that showed up in some of the other graphs are virtually gone and my conclusion is that it had a lot to do with the alignment of the vertices and having the important objects meshed at the same resolution to enable that to happen.
 

Attachments

  • Round Baffle H Polar Curves 20 Deg Norm.jpg
    Round Baffle H Polar Curves 20 Deg Norm.jpg
    97 KB · Views: 209
  • Round Baffle Diagonal.jpg
    Round Baffle Diagonal.jpg
    56 KB · Views: 149
  • Round Baffle Vertical.jpg
    Round Baffle Vertical.jpg
    65.7 KB · Views: 164
  • Round Baffle Horizontal.jpg
    Round Baffle Horizontal.jpg
    65.2 KB · Views: 219
  • Round Baffle H Polar 20 Deg Norm.jpg
    Round Baffle H Polar 20 Deg Norm.jpg
    34.9 KB · Views: 211
Thanks for your input mabat, without your tool I would not be undertaking this at all. I would have just used the XT1464 Horns I have sitting in the shed. I do have hope that all this effort will give me a better result.

I see there's a lot of struggle you are going through with these models... It will be possible to create an enclosure like that for a free-standing situation with the Ath tool automatically. I've just been quite busy last weeks. With Gmsh library it is easy to make the mesh element sizes vary so it can be denser at the front and get progressively less dense on the back sides (which is what I'd consider optimal). I only can't promise any deadline.
I would say the struggle was to understand how to go about it, some of that was down to not being a CAD expert and having to try and figure it out as you go. Once that part is over then it is much less effort to do it all over again. That was my motivation for posting my journey from start to finish in the hopes that it might help someone else who is trying to do the same thing.

I look forward to the day that Ath can do all of this without my intervention. As that day has not yet come and me not liking to wait for a smarter person to do it all for me I got stuck in ;)

I've read this many times (about the flat interface) but it still doesn't make any sense to me. The whole purpose of the extruded interface is to avoid too sharp angles between boundaries at the mouth egde, i.e. too narrow spaces - this is even recommended by the ABEC author and I can see the point in that. Otherwise the shape of the interface should be just irrelevant.

Your second result is more stable at higher frequencies because the mesh elements around the mouth are smaller. And of course it will be better because the egde is rounded. Good job!
I can offer no good explanation to say why the flat interface worked better with a finite baffle, all I can say is that I tried it with the interface produced by Ath and a flat interface at 10mm exactly and the latter gave the output seen above with very little numerical issues.

Don used a flat interface that had higher density in some places and that did not work as well in this particular case as the one that was at 10mm.

I am now more convinced that the spikes seen in some simulations that we discussed in your thread are somehow mesh alignment/density related and not non uniqueness issues.

Yeah, the nice thing about using the Gmsh library (when used alone) is that it aligns all edges automatically - you just don't have to care about that. It "simply" creates all the vertices from defined geometry and mesh densities (at each control point) and joints them into one cohesive structure.
Very true, I can't reconcile that against what I just said above though as some gmsh library produced files have suffered from spikes.
 
This post should just about catch up to where I am now. I have compared a flat baffle to one with a 36mm roundover on the horizontal edges. I won't rehash all the curve graphs but you can see the diffraction off the sharp baffle edge as a less smooth response overall. It is not horrendous as the waveguide has a very gentle transition to the baffle itself but it is there.

The best way to show it is with the observation fields. I have put them all in one image side by side. Left side is the flat baffle right side is the rounded edge. The image is pretty big to keep decent resolution so I have not put it in line. The pattern can clearly be seen to be smoother overall with less raggedness to it in the rounded edge version. It is visible from about 100Hz upwards but more pronounced at high frequencies.

It seems that the back corners might also benefit from being rounded although the difference is likely small in reality. I might draw up some more observation fields to show the effect of that more clearly.

I do have a mitre rounded baffle and fully rounded baffle and carcass created in CAD but not put into ABEC projects yet. Time will tell if I think it is worth the effort to see what they show.

The big image doesn't seem to have saved very well so I added some full size comparisons.
 

Attachments

  • H Obs Compare.jpg
    H Obs Compare.jpg
    145.8 KB · Views: 216
  • H Obs 7180.jpg
    H Obs 7180.jpg
    53.7 KB · Views: 150
  • H Obs 3944.jpg
    H Obs 3944.jpg
    49.9 KB · Views: 150
  • H Obs 1989.jpg
    H Obs 1989.jpg
    48.4 KB · Views: 155
  • H Field 10112.jpg
    H Field 10112.jpg
    70.3 KB · Views: 151
  • H Field 7180.jpg
    H Field 7180.jpg
    68.1 KB · Views: 155
  • H Field 4296.jpg
    H Field 4296.jpg
    64.1 KB · Views: 183
  • H Field 1989.jpg
    H Field 1989.jpg
    61.3 KB · Views: 192
  • H Obs 10112.jpg
    H Obs 10112.jpg
    55.7 KB · Views: 170
Last edited: