They never imaged as well as the one shown in Photo 2, possibly due to the width of the enclosure or more likely due to component tolerances in the crossovers, which were identical despite differences in driver impedances.  The tweeter and mid-range units are identical and the woofers are of the same manufacturer's series, one being an 8" and the other a 10".  Other than the difference in the low pass filters for the woofers, the crossovers are also the same.  The woofers are Peerless 8" (220WR/8) or Peerless 10" (260WR/8), the mids are Focal 5K013L and the tweeters are Focal Tc90TDXT.  I still have the 12" Peerless units.

The units in Photos 1 & 2 were designed and built in 1993-94.

As a result, these incorrectly made cabinets sat around for about 20 years.  This changed after my checking the Thiele-Small parameters of several 10 inch woofers I had lying around and much to my surprise, I had two that would work well in 3 cubic feet.  They were the Bostwicks.

There are no photos of the finished system as it's not yet finished. I'm waiting for a friend to come by and get into that project which is to truncate the fronts of those cabinets.  It will start at the top at about 5" from each corner and 5" along each side from the front and taper down to about a foot from the bottom.  This will narrow the cabinet front from the with of the tweeter, mid-range and woofer.  The hope is that imaging will improve by reducing front panel reflections and raising the effects of diffraction higher into the audible band.

I could do it myself but since there's no big hurry, I decided to wait for his visit as he likes to get into that stuff.  Besides, I have no place to store the finished cabinets so they'll have to stay in the kitchen by the refrigerator and be covered with a mover's blanket.

A sample of one of the early crossover designs using the Dynaudio D-76 and a Zobel.  Since then, a computer program I wrote was used as it was realised that this was going to involve a lot of trial and error and repeated calculations that would get tiresome, not to mention the increased possibility of errors.

First of all, ignore the fella with the geetar.

The Fountek and the Vifa.

Originally, a DYNAUDIO D-76 was intended but it didn't have the low frequency output to 400hz.  Going lower would have been preferred but the Fountek couldn't produce enough output at 200hz to maintain the response I was after, flat within 1dB from 300hz to at least 15khz.  That little tweeter got me to 17khz.

Such a flat response from a speaker isn't the only thing to consider but it was done more to see if I could  do it.

The low frequency response below 300hz in the Fountek data sheet may have been a near field response.  Just about any speaker is certain to look a lot better measured at a distance of about 0.25" from the cone vs 1 meter distant.

The crossovers, three way, second order.  The notch filter can be seen on each side of the the light blue inductors, of which there are 4 in each crossover.  They are each 0.33mh wired in parallel to get the desired 0.0825mh.  On the left of those is the 22 ohms resistor, made from a 6.8 ohms and a 15 ohms in series.  Various 5% resistors had to be tried to get the 22 ohms, measured on a Fluke 8050A with a known 0.1% accuracy.  The 4uf capacitor on the right is a Dayton 1% and checked on an Agilent U-1241 with a 1.2% accuracy.  Actually, the capacitor was used to check the accuracy of the meter as it's the more accurate.

The resistance accuracy of the Fluke was checked with laser trimmed precision resistors of 0.01%.  Those things cost $$$.

My best inductance meter, the LMS has an inductance accuracy of 2%.  The beauty of the LMS is that capacitance and inductance can be measured at any frequency between 10hz and 100khz.

The attached wires were for test purposes.

With the notch filter, herein after called RLC-2, the response in the high end above 4000hz improved.  Further fooling around with the RLC-2 was abandoned due to the work involved.  It was hard enough getting the stop band where it was needed and the attenuation satisfactory, so well enough was left alone.

To get further attenuation in that band, it was decided to lower the low pass filter section of the mid-range from around 8000hz to 7000hz.  The peak in that band dropped a little so the filter section was set for 6000hz with a better result.  This was repeated until the low pass section was lowered to 3000hz.  Any further lowering showed this to be beyond the point of diminishing returns.  The actual low pass section is 3dB down at 2800hz. (the knee)  With a second order filter, this would be -15dB at 5600hz and -27dB at 11khz, right where it's needed.

To further flatten the response in the high end, the high pass section of the tweeter was raised from 5000hz to 6000hz.  This was repeated until reaching 8000hz.

While it may seem silly to have a mid-range stop at 2800hz and the tweeter come in at 8000hz, the result below (Photo 16) speaks for itself.

Curves 38 & 39 (light brown and black) are the "A" set of speakers and associated crossover measured at 1m and 2m, resp.

Curves 40 & 41 (dark brown and red) are the "B" set of speakers and associated crossover measured at 1m and 2m, resp.

The 6dB difference is due to the doubling of the distance.  Also, since the reflection distance approaches the on axis distance as the on axis distance is increased, the LMS loses its ability to obtain a good sample of the incident wave before the reflected wave is sensed.

It works like this, assuming the reader is interested.  At 2m, it takes a pressure wave 0.0058 seconds to traverse. With the mic 1.22m above the floor (midway between floor & ceiling), the nearest reflection is 3.15m. The reflected wave from the floor and ceiling will reach the mic in 0.0091 sec.  this is a time difference of 0.0033 sec.  The time of a 420hz signal is 0.0024 sec per cycle.  While there are two cycles at this frequency at 2m between the mic & speaker, the mic will be turned off by the system around 0.0033 sec or sooner.  This gives the mic at best, a sample of about 1.5 wavelengths at 420hz below which it becomes a little difficult to differentiate between the incident and reflected waves.  It will try many times, usually taking a minute or so before realizing it's futile and the system stops.  That erratic behaviour can be seen at the left ends of the response curves by the +/- 2dB variations.

Another source of such behaviour could be where the distance is measured; from the duct cap or flush with the annulus, a difference of 1" to 3", depending on the size of the loudspeaker (the depth of the cone).  Comparative measurements were done in the past but the difference wasn't found to be enough to completely account for the deviation. 

1", 2" and 3" differences amount to 0.00007, 0.00014 and 0.00021 sec. (0.07, 0.14, 0.21 mS)

This shows the phase relationship between the woofer, the nice waveform and the vent, the somewhat distorted one.  The phase shows about 95º  but one would expect 90º.  The error is due to a difference in the vertical position of the ch 2 X-axis.  This was noticed after the test was complete and things disconnected but it's close enough. This could have been corrected had it been noticed prior to disconnecting everything since this old scope has a baseline problem in ch 2; it moves up when changing voltage scale and repeats every third step.  I usually adjust it to the baseline but it seems it was overlooked this time.  This error can be seen by looking at the peaks of the somewhat straight edged waveform.  The positive peak is +8 sub-divisions and the negative peak is about -6.  By lowering the ch 2 vertical position so both peaks are at +7 and -7, the right slope of that waveform will move a little to the left, closer to the expected 90 degrees. The frequency used was 33hz, the tuned frequency, Fb of the cabinet..

This was done by setting the scope to display a full cycle between 8 divisions, using the variable time base.  Since one cycle is 360º, each of the 8 major divisions is 45º and each of the minor divisions is 9º.

The measurement is made between the points at which each waveform crosses the X axis.

The next two photos, 20a&b show the mics in position.

The cabinet under test.  The thin lines seen on them are the preliminary cut lines for the truncating.

The microphones are home made from a pair of mic elements from Radio Shack, a 10uf capacitor and a 9 volt battery.  They are more than adequate for making phase checks between two speakers using a scope.

The wood holders are 5/8ths inch dowels drilled to fit the mic element and jack.

The end result measured by LMS at 1w1m, gated.  The black curve and the green curve are the mid-range and tweeter only, respectively.  The thin red curve is the mid-range and tweeter simultaneously.  The thin blue curve is all three speakers, flat within 1.5dB from just a little below 350hz to about 17khz.

A response will be run outside this winter when the temperatures get into the 80's to find out how well the woofer performs.