Please note: this is not a construction article. It describes the process of building a large plywood horn, but the information is only provided for your interest and inspiration, and as an example on how to build horns.
The background for building this horn is actually quite complicated. Back in 2003-2004 I came in contact with Thomas Dunker, who was one of the proponents of horns (and tube amps) in the "horn and triode revival" of the 90s. One of our main topics of discussion was how to build a good midrange horn, covering the major part of the human voice fundamental range. One horn that did this, was the Western Electric 15A, covering 80Hz-5kHz (approximately) with a single large horn and a small compression driver. How was this possible? Common horn wisdom says that a horn will cover approximately three octaves (or one decade).
So we set out to gather as much information on horns and horn theory that we could, especially from the time when horns were full range devices, for use with mechanical phonographs or theatre speakers.
The project evolved and expanded to become a far larger undertaking than either of us had imagined. Hundreds of articles, books, theses and patents piled up. Software was written, measuring equipment built. And as a more or less direct result of our investigations and research, I decided to start on a MSc in Acoustics, and now I work on my PhD.
Back to the midrange horn project. First of all we needed to select a suitable driver. A pair of JBL 2470s for each horn (using a Y-split) was considered, but later rejected. We ended up with a modified version of Altec 288B. This driver is quite close to the field coil 287, which was designed for a cross-over frequency of 300Hz. 288B is specified for 500-16000Hz at 40W input. It has an underhung voicecoil, and somewhat larger clearing to the phase plug than 288C and later versions. Which means less HF-response, but better LF performance. As I have mentioned before, 288C is a better choice than 288B in terms of HF response.
The 288B also has a larger diaphragm than WE 555W, which was used down to 60Hz, without crossover. It should be suitable for a lower midrange horn at home listening levels, provided of course, that the horn has good resistive loading of the driver in the working range.
Since proper acoustic loading of the driver, especially near cutoff, was important, this was taken into account in the design. One measure of the horn's loading properties is the power factor, or the cosine of the angle between the resistance and the reactance of the throat impedance. The power radiated by the horn is proportional to the power factor, so the power factor should be as close to one as possible in the working range. Near the cutoff frequency, all horns have a large mass reactance. But the shape of the resistance and reactance curves can be manipulated by changing the way the horn flares at the throat end. The profile will be similar to a hypex horn, which flares slower at the throat end.
The principal parameters for the horn were:
And since the internal flare of the compression driver is very much part of the horn, the design has to include that portion too. The 288B internal flare has a cutoff of about 208Hz, which is too high for a horn that should start out with a low flare. So we rebuilt the drivers with a new internal throat section of constant diameter. This means that the horn, as it is, will not work with any stock driver on the market!
The profiles of the horn are shown below. The first part is Hyperbolic-Exponential with T=0, to match the driver plane wave throat well. The next parts are exponential. Total horn length is 95.5cm, and the mouth is 65 by 65cm.
Below are simulated throat impedance and power factor for the horn, mounted in free field.
The power factor is remarkably flat from 400Hz and up. Of course, this is only a simulation. Throat impedance measurements have not been done on this horn yet.
When the basic outline of the horn was decided, and the interface dimensions set, design of the specific parts began.
The middle section was designed first. This was expected to be a constant factor in the experimentation to follow. The profile was corrected for the curving of the wave fronts, the wave front areas were calculated by numerical integration. The horizontal profile was taken to be conical, since this would simplify the calculation of wave front areas, and also make the actual wave fronts behave in a simpler way. The main part of the expansion was to be in the horizontal plane for the same reason: to keep the wave front as little curved as possible.
The throat section will do the transforming of the cross section from round to rectangular. Based on an old Bell Labs patent (which was the basis for the WE 31A horn), this transition is made by going through an eye-shape, as shown below.
The mouth section is slightly exponential in the horizontal plane, and purely exponential in the vertical plane. Due to the strong curving of the wavefronts, it was not attempted to correct the shape here; it proved too difficult and time consuming at the time, and the project had already taken too long. We decided on a pure exponential horn, with a square mouth of 65 by 65cm.
Construction began with the middle segment. Since the walls of a rectangular horn do not meet each other at a constant angle, we had to cut that angle correctly. For the middle segment this is not too complicated, since two of the side walls are conical. Here is how we did it:
1) Make a jig with a slope equal to the angle between the horn axis and the wall.
2) With the side wall on the jig, cut the profile on a band saw.
3) Then mount the side walls on a jig, and start fitting the top and bottom walls.
4) When the first panels are in place, laminate several layers of plywood on both sides. We used 4mm 3-ply wood for the curved sides, and 8mm 5-ply for the straight sides. Build up to 16mm thickness. Use stiffeners inside the horn to prevent sagging of the walls due to the tension in the plywood.
The middle segment is then fitted with flanges that fit the throat and mouth segments. These flanges are made from 16mm plywood covered by 2mm aluminium at both sides to get a smooth surface. Pegs are used to align the segments.
The throat segment is quite complicated. It consists of a welded steel shell with flanges that fit the 288B at one end, and the middle segment at the other. Inside this steel shell, the transition from round to rectangular cross section is cast in tin. Here is how it was done:
1) The shape of the transition is calculated at every 5mm. The shape is exported to AutoCAD and printed in 1:1 on paper. The paper is glued to "architect cardboard", 5mm thick foam-filled cardboard, and cut out. A hole is drilled at the center of each piece. The pieces are then stacked on top of each other on a rod.
2) The shape is smoothed, filled with putty etc, to make it nice and smooth.
3) A paraffin positive mold is made from this cardboard plug.
4) A plug of RTV silicon is cast in the paraffin mold.
5) This plug is then centered in the steel shell, and hot liquid tin is poured around it.
6) When the tin has cooled, the plug is removed, and we have the transition.
7) The extra tin is sawn off, and the inner surface primed with a thick filler/primer (we used Capalac Unigrund).
Next up is the mouth segment, which is curved in both directions. This is much more complicated than the middle segment, and requires a good jig. Below is a picture of Thomas Dunker and me holding the middle segment and throat segment, with the mouth segment jig on the ground.
The jig is required to hold the force of several layers of plywood pressed towards it. We started with the least curved sides. A massive amount of bracing and clamps is required, see below. Also note that we had already made the mouth frame, to hold everything in place. We laminated 4 layers of 3-ply wood. After this was done, a similar procedure as with the middle segment was used to cut the wall corners so that the two last walls would fit properly.
To prevent the walls from sticking to the frame or jig, we used some thin plastic foil between the parts.
When the side walls were laminated and cut, the top and bottom walls were laminated in a similar way. As can be seen from the photo, this was a quite complicated procedure. Threaded rods, wing nuts and plywood braces, in addition to clamps were used. We had to use all we could gather of clamps, big and small. Note the plastic foil to prevent the parts from being glued to the jig or braces.
The finished mouth segment.
The next step is to bolt the segments together and make a smooth transition. Then the horn is painted with a primer/filler, and finally two layers of strong paint. For the final mountings, a thin foam gasket is used between each segment. 12 6mm bolts are used to connect the throat and middle sections, and 14 6mm bolts are used for the middle/mouth segment connection. Pegs are used to align the parts.
The modified 288B driver ready (with gasket) for mounting on the horn.
The horn was measured outdoors, on a jig made from an office chair for rotation.
Below is the frequency response for various driver configurations. The blue and red lines are for one set of 288B, with and without rear cover respectively. The black curve is for another pair, I think with rear cover. It is evident that there is a difference between the drivers. Some drivers have a silver shorting ring, evident in the better HF response in the black curve.
You may wonder why the LF response is falling off so early, even if the throat impedance is high in the region, and that I have shown here that the 288B with plane wave exit has a flat frequency response down to 100Hz. Well, it is because of the directivity of the horn.
The near field measurements also show this, here the response is flatter.
Distortion at 100dB SPL at 2m distance. The typical rising 2nd harmonic distortion is visible. Higher order distortion is comfortably low.
It was of course very exciting to hook up these horns and play music on them for the first time. And we were not disappointed. They sounded very nice. It turned out that having the conical expansion in the horizontal plane was the best, and this gave a very wide sweet spot. The horn fills the room better than the AH425s, and having a large part of the voice fundamental range in one horn certainly has its advantage.
The project started somewhere in 2006, and finished in 2010, a month before I moved to Trondheim and had to pack them down. They are perhaps the part of my old system that I miss the most.
Thomas has finished his pair, though, and they are playing beautifully. Below is a picture from his setup. The low frequency system is JBL 4560 with JBL 2220A woofer, and the HF system is University 4408 horns with P.Audio WN-D34 modified for 0.5" exit. Crossover frequencies are around 300 and 3000Hz, tri-amplified with active crossover (Behringer DCX2496).
Update, Fall 2013: The horns are up again in my system.
©2006-2012 Bjørn Kolbrek and Thomas Dunker