Boxsim Simulation Speaker Design Example.
This was initially just intended as an example of a Boxsim simulation, but I did actually build one version and include a few details at the end of the page.
I have been learning how to use Visaton's free Boxsim program, which I am told gives reasonably accurate simulation results including baffle effects. I already have a pair of Visaton FRS8 full-range speakers which could be used with the Visaton W170S 17cm bass unit. I have the 4 ohm version of the FRS8, but that may turn out to be good because of the low sensitivity of the 8 ohm type compared to the W170S. The baffle-edge diffraction however reduces the bass driver sensitivity. The Boxsim response apparently includes an assumed 6dB low frequency edge diffraction effect, and with the speaker fairly close to a wall and floor there will be something like a 3dB bass boost. An enclosure size of 25 litres works well enough, but reducing to 20 litres makes little difference, about 1dB less at very low frequencies. The FRS8 needs its own small enclosure, I used a volume of 1 litre for the Boxsim simulations, and for both speakers I left the absorbent stuffing density as 'loosely filled' specified as 6.7g/litre.
The front panel size chosen is 20cm wide by 40cm high. This is not ideal, there is a widely recommended 'golden ratio' for box dimensions to spread out the internal resonance frequencies, but with absorbent material filling the entire volume that is not so essential. In addition the FRS8 has its own enclosed space near the top of the enclosure, and this reduces the vertical height of the bass driver enclosure, so it will not be exactly a 2:1 ratio of width to height. By making the FRS8 enclosure non-rectangular that will further reduce resonance effects for both drivers. The depth of the enclosure can be chosen to give the required volume, something around 25cm may be far enough away from the 20cm width.
The speakers are positioned on the centre-line of the front panel, with the centre of the W170S 15cm above the lower edge and the centre of the FRS8 29cm above the lower edge.
The benefits of active crossovers are well known, so a passive crossover would only be justified if it was simple, cheap, and had a similarly flat frequency response, and the next example just about meets these conditions. A suitable 1mH inductor made by Monacor is available from CPC with a ferrite core and 0R3 resistance. CPC also have two air-cored 1mH inductors made by Visaton, the one with resistance 0R4 can be used with no change to the filter, but is far more expensive.
The crossover frequency is higher than originally planned, one consequence being an increase in maximum SPL from 150Hz to 250Hz because the FRS8 is no longer the limiting factor. The vertical off-axis response however has a dip around 2kHz because of the higher crossover. Placement close to a wall can compensate to some extent for the reduced output below 200Hz.
This looks like an asymmetric filter, the extra capacitor in series with the FRS8 gives a second-order high-pass, and we would need an inductor in series with the W170S to give a second-order low-pass. The bass speaker actually has a voice coil inductance of 1.2mH, effectively in series, but in any case an asymmetric filter is not necessarily a bad thing, it can provide some compensation for a lack of time-alignment of the drivers, and in practice it appears to work well. I would have preferred a parallel crossover rather than a series crossover, but failed to achieve comparable flat response with any variation I tried with a single inductor, it needed two inductors. I have included the simplest second-order passive version I found at the end of this page.
The frequency response with just rectangular baffle edges revealed a problem around 3kHz, with a significant dip. The FRS8 has no similar dip in this area according to my previous experience with this driver, and some experimentation with the baffle edge chamfer options confirmed a baffle effect. The addition of a 3cm chamfer at both sides and 6cm at the top eliminates the dip, and is included for the above plot. Searching the various support documents reveals that a 'chamfer' is just a 45 degree cut off of the corner, but it is generally agreed that a rounded corner will work equally well or better.
I originally intended to use a 400Hz crossover, but then with a simple 2nd order crossover filter the maximumm sound level dips at 150Hz because of the FRS8 reaching its limit. Curiously the Visaton website has a design called the 'Stella Light' which uses the same drive units with a 200Hz first-order crossover. I checked that crossover and again found seriously reduced maximum SPL at 150Hz, so not a good choice I think. The peak at 10kHz may not be as bad as the plot on the FRS8 page would suggest, I used this driver in a previous design and although not flat above 10kHz there was nothing I would regard as needing compensation, mostly dips rather than peaks.
The response shown next is achieved with 2nd order active filters. The high-pass is 500Hz, Q=0.5 and the low-pass is 540Hz and Q=0.5. There is also a phase reversal for the FRS8, otherwise there is a dip at the crossover frequency. For this simulation the FRS8 has a series 50uH inductor with a parallel 3R9 to flatten the high frequency response, but of course the same effect can more easily be included in the active filter circuit, or left out altogether, the effect being only around 1dB. The frequencies specified in Boxsim for the filters are not always the -3dB frequencies, for Q=0.5 the gain is -6dB at the specified frequency.
With equal frequencies for high and low-pass and Q=0.5 this would be the 2nd-order Linkwitz-Riley response, and that would give a flat response if we could use time-aligned drivers with flat frequency responses, but for this design the frequencies need to be offset a little to give a reasonably flat response. The reduction in bass output below 200Hz can to some extent be compensated by positioning close to a wall.
The inversion needed for one of the drivers can be achieved by using a non-inverting Sallen-Key filter for the high-pass and a multiple feedback inverting circuit for the low-pass as shown next. The alternative of Sallen-Key low-pass and MFB high-pass is not such a good idea, with similar component values the noise will be higher. Another alternative is to use Sallen-Key for both filters and just reverse the connection to one of the drivers.
I did actually build something. Here is the best I could do with a parallel second-order passive crossover followed by the resulting frequency response. The inductors can be the two 1mH Visaton air-cored types, with 0R4 and 1R2 series resistances. I used three 10uF film capacitors in parallel for the 30uF and a bipolar 82uF electrolytic. The 0R73 resistor is three 2R2 in parallel. (I actually wound my own 1mH 0.4R inductor, but used too many turns and measured 1.18mH, and had to increase the 15R to 30R to compensate.)
I found the simulated result was almost identical with a simpler construction without the rather extreme chamfers originally used, by the simple method of moving the FRS8 off-centre. I include a few photos here but not as a recommended speaker project, though I think it does sound quite good. I just wanted something to use as a guitar practice speaker so I only made one. Maybe the offset driver is less ideal for stereo use, even with a mirror image pair, but anyway I am entirely happy with the result.
Note the FRS8 has its own 1 litre enclosure, and the cutout is as large diameter as possible for that enclosure size, with the drive unit mounted on a thin piece of hardboard glued onto the front panel. This is based on previous experience with the FRS8 which suggested it interacts badly with a thick panel, even when chamfered, and needs as much space as possible around it inside the enclosure. The enclosure spaces for both drivers were lightly filled with long-haired sheeps wool. The high-pass part of the crossover can be seen in a photo, the low-pass part is attached to the base of the enclosure to keep them well apart to reduce interaction. The wood panels are 15mm thick, but with a double layer on the back panel and eight 8mm holes spaced out on the back to give a 'distributed port' effect. A cross-brace was added between the side panels for extra strength.
Update: I noticed that a few of my other attempts had greater level variations at and above 10kHz, and accidentally found out why. Going to 'File'-'Project Properties' and clicking 'Ok' changed the frequency response, although I left all the settings unchanged. I found that the 'Number of frequencies' is being changed by this action, the number is shown as 126, but to get back to my original frequency response I had to reduce this to around 90. Somehow this setting is getting changed, without the figure in the settings changing. This looks like some sort of bug in the simulator, and it can make the high frequency response look better than it should.