I thought my recent speaker project might be of some interest to the members of this forum. This is largely copy-pasted from my post at diyaudio.com.
The waveguide profile is oblate spheroidal with an angle- and curvature-matched Euler spiral (a.k.a. clothoid) termination starting at 40% of the total depth. Throat angle is 10°, which matches the compression driver I'm using (Celestion CDX1-1747). Here's the completed speaker:
The cabinet is 73cm x 41cm x 30cm (approx. 29in x 16in x 12in). Obviously inspired by Earl Geddes' designs.
It took me a while to figure out how to make an accurate waveguide of this size since I don't have a big lathe or CNC router. I have a 3d printer, but I wasn't too keen on the idea of printing the whole waveguide due to the time required and the fact that the finished product would be a bit flimsy unless a lot of material was used. The basic process I arrived at is gypsum cement (basically plaster of paris, but much harder and stronger) over an MDF substrate coated with a suitable bonding agent. After some experimentation I found that type II water resistant yellow (carpenter's) PVA glue works remarkably well as the bonding agent (more on this later).
I shaped the gypsum cement using 3d-printed running molds that pivot on a 1/2in stainless steel rod held in place by this jig (which is also 3d-printed):
Here are the running molds:
Two of the running molds have a wide angled edge which helps to spread the plaster, kind of like a plasterer's trowel:
I applied the plaster in two coats, similar to two coat veneer plaster. The basecoat is about 3mm thick, cut 50/50 by weight with fine silica sand and the finish coat is 1mm neat. Here's the bare plaster waveguide. You can see the stainless rod in the middle:
This particular one was actually a failed attempt as the plaster didn't bond properly to the substrate. Due to the fact that gypsum plaster expands slightly as it sets, you need a bonding agent that provides some bond strength even when damp to prevent the plaster layer from lifting off the substrate near the waveguide mouth. For the attempt above, I sealed the MDF with a waterproof primer and then coated that with Elmer's Glue-All (a white PVA glue without water resistance). The problem is that the Glue-all turns back into a PVA emulsion on contact with the wet plaster and thus has no bond strength until dry. What I found works well is two coats type II water resistant PVA (I used Gorilla wood glue (NOT the polyurethane stuff), but Titebond II or any other similar product should work just as well), sanded lightly between coats, and one coat of Elmer's Glue-All. The Glue-All is probably not required, but I figured it would improve the bond once dry. Type II PVA adhesives get a bit soft when exposed to water, but don't completely re-emulsify. This property appears to be what makes it work so well for this application.
The gypsum cement I used is a G-P product called Densite (K-5, I think), which you can buy in small quantities at dickblick.com. USG Hydrocal White should also work well. Do not use USG Ultracal 30 or any other gypsum cement with portland cement (check the SDS). Based on my experience with Ultracal 30, it will shrink slightly over time and crack.
So how does the final product measure? Very well, in fact. The data seems to agree pretty well with the ABEC3 simulations I did. The measurements were done inside with a short window, so resolution is limited. Data below 500Hz or so are invalid due to the time window. Also, there's a slight problem with the vertical measurements, which caused the small peak off-axis at around 650Hz. I'm planning to do new, better measurements before too long, but I haven't gotten around to it yet. Needless to say, having a Klippel NFS would be nice . Anyway, here's the data I have:
I generated these graphs using a python script I wrote. The crossover is LR6 acoustic at 1050Hz, implemented with my DSP software.
The waveguide profile is oblate spheroidal with an angle- and curvature-matched Euler spiral (a.k.a. clothoid) termination starting at 40% of the total depth. Throat angle is 10°, which matches the compression driver I'm using (Celestion CDX1-1747). Here's the completed speaker:
The cabinet is 73cm x 41cm x 30cm (approx. 29in x 16in x 12in). Obviously inspired by Earl Geddes' designs.
It took me a while to figure out how to make an accurate waveguide of this size since I don't have a big lathe or CNC router. I have a 3d printer, but I wasn't too keen on the idea of printing the whole waveguide due to the time required and the fact that the finished product would be a bit flimsy unless a lot of material was used. The basic process I arrived at is gypsum cement (basically plaster of paris, but much harder and stronger) over an MDF substrate coated with a suitable bonding agent. After some experimentation I found that type II water resistant yellow (carpenter's) PVA glue works remarkably well as the bonding agent (more on this later).
I shaped the gypsum cement using 3d-printed running molds that pivot on a 1/2in stainless steel rod held in place by this jig (which is also 3d-printed):
Here are the running molds:
Two of the running molds have a wide angled edge which helps to spread the plaster, kind of like a plasterer's trowel:
I applied the plaster in two coats, similar to two coat veneer plaster. The basecoat is about 3mm thick, cut 50/50 by weight with fine silica sand and the finish coat is 1mm neat. Here's the bare plaster waveguide. You can see the stainless rod in the middle:
This particular one was actually a failed attempt as the plaster didn't bond properly to the substrate. Due to the fact that gypsum plaster expands slightly as it sets, you need a bonding agent that provides some bond strength even when damp to prevent the plaster layer from lifting off the substrate near the waveguide mouth. For the attempt above, I sealed the MDF with a waterproof primer and then coated that with Elmer's Glue-All (a white PVA glue without water resistance). The problem is that the Glue-all turns back into a PVA emulsion on contact with the wet plaster and thus has no bond strength until dry. What I found works well is two coats type II water resistant PVA (I used Gorilla wood glue (NOT the polyurethane stuff), but Titebond II or any other similar product should work just as well), sanded lightly between coats, and one coat of Elmer's Glue-All. The Glue-All is probably not required, but I figured it would improve the bond once dry. Type II PVA adhesives get a bit soft when exposed to water, but don't completely re-emulsify. This property appears to be what makes it work so well for this application.
The gypsum cement I used is a G-P product called Densite (K-5, I think), which you can buy in small quantities at dickblick.com. USG Hydrocal White should also work well. Do not use USG Ultracal 30 or any other gypsum cement with portland cement (check the SDS). Based on my experience with Ultracal 30, it will shrink slightly over time and crack.
So how does the final product measure? Very well, in fact. The data seems to agree pretty well with the ABEC3 simulations I did. The measurements were done inside with a short window, so resolution is limited. Data below 500Hz or so are invalid due to the time window. Also, there's a slight problem with the vertical measurements, which caused the small peak off-axis at around 650Hz. I'm planning to do new, better measurements before too long, but I haven't gotten around to it yet. Needless to say, having a Klippel NFS would be nice . Anyway, here's the data I have:
I generated these graphs using a python script I wrote. The crossover is LR6 acoustic at 1050Hz, implemented with my DSP software.
Last edited: