I've been listening to the speakers with the test crossovers quite some time now, and they are really quite good. That's why I wasn't in a real hurry to finish these speakers. Which is a good thing, since that gave me the time to learn how to measure speakers myself and also acquire the right equipment and software to do so. Now that I have a nice speaker measurement jig and measurement software (Sample Champion) I went ahead and measured one speaker.

Here are the individual on axis responses:

The scale is a little different than on the previous page, but the resemblance is quite obvious. The two main differences are the Y-scale (I need to find out how to get absolute SPL on the Y-axis) and the phase. The measurements taken at BD-Design were not compensated for the distance between speaker and microphone, these are.

Same goes for the woofer: The resemblance is clearly visible. The small "jump" in phase around 600Hz is due to the merging of near and farfield measurements.

I stuck to my original choice for a crossover point at 2500Hz, only this time, I would make sure that it actually stayed at 2500Hz. When you inspect the summed frequency response on the previous page closely, you can see that the XO point has shifted up to around 4KHz (!!). I also wanted less components in the signal path for the woofer. For two reasons: Soundquality and cost. High quality inductors are pretty expensive and if they're avoidable you can gain sound quality and save cost.
In the individual responses, there are several things that I noticed. Two peaks in the response of the woofer. The peak at around 600 Hz is due to a standing wave in the enclosure. It seems that the wool I put in the box to damp this standing wave, isn't totally up to it's task. The woofer also has a peak at around 4000Hz. This is the cone breakup, which was to be expected. The tweeter has a large dip around 10KHz. This really surprised me, because it shouldn't have this. But it's real. The dip is visible in the BD-design measurements (taken on the other speaker) and in my own. I also have some trouble explaining this dip, because the wavelength of this frequency is really small. The only reason I can think of for this dip is the gap in which the ribbon is mounted. But if this is so, this dip should also be visible on measurements taken by others.

I chose a third order electrical, 4th order acoustical filter. The third order electrical is to protect the tweeter from low frequency signals, because it is quite sensitive to this. I Left the dip alone, since there isn't much you can do about it.

For the woofer, I designed a first order electrical-4th order acoustical crossover. It has two notchfilters, to remove the two peaks mentioned above. I'm quite pleased with this, because this allows me to purchase a very high quality inductor to put in series with the woofer, because I only need one :)

The resulting frequency response is quite good, if it weren't for the irregularities in the 10KHz and up region. Once the filter is finished and installed, I will do some further testing. I hope I will find out what's causing the tweeter to behave like this.

I tried this filter out, using the aural emulator built into LspCad (I borrowed a friend's pc, with this version of LspCad). It sounds pretty good (better than my current testcrossover), so I decided to go ahead and purchase the parts.
The aural emulator, built into LspCad, also gave me an opportunity the listen to the speakers with a Linkwitz transform. Allthough the diagram, displaying both the target and achieved transfer functions showed a big discrepancy between these two bass improved markedly. A small boost turned out to be the best choice. Because of the way the speakers are positioned and my listening position, I get a lot off bass for free. Only a small amount of boost is sufficient to give them enough extension for my taste. After the passive filtering is done, I will finish the linkwitz transform filter (I already have all parts soldered onto the boards, but it needs a nice housing)

The parts arrived in just a few days, yippie. I used mkp capacitors throughout the filter, because they're relatively affordable. I thought about using a higher grade capacitor to put in series with the tweeter, but that would almost triple the cost for the total filter. (a 10µF mkp costs €4,70, a mkp-plus, costs €40.) I chose a tritec inductor to put in series with the woofer, because of their low resistance value and I used a foil inductor to use parallel to the tweeter.

First, I had to see if everything would fit onto the board (which had to be small enough to fit through the woofer mounting hole). Luckily, it did. I shifted everything around a little and this is the layout that seemed to work best. Because there are four inductors in each filter, it isn't possible to mount them in such a way that they won't influence each other. So, I placed the two inductors that are positioned the same as far away from eachother as possible.

I marked the position of each part on the board. After that, I drilled holes for the tie-wraps. I placed every component on the board, using hot glue and a tie-wrap. Everything sits in place very firmly.

There's just enough room to place the filter, but that's all that's needed. I glued the board in place using a n. Mconstruction glue that (according to the label) is solvent-free and resistant to vibrations (how usefull!)

After all was done, I took some measurements to check how it worked out and how the simulations compared to the real thing.
The graph below is a comparison of the simulated FR and the measured FR on axis. I think the matching is good enough.

The graph below is a comparison of the FR with and without grille. The notch at 9Khz deepens a bit and the small hump at 20Khz is lowered a bit. Not really big issues. The protection of the grille is necessary, and luckily it doesn't harm the sound that much.

The graph below is the horizontal off axis frequency response.

and last but not least: a waterfall plot, form 500Hz and upwards:

I also measured intermodulation distortion at four points: 900Hz and 1Khz to measure the IMD of the woofer, this was a decent 0,4%. Also 2250Hz and 2500Hz, to measure IMD in the crossover region, this was at a less decent level of 2%. The !khz and 10Khz measurment was to measure IMD between woofer and tweeter and was 0,3%. The last measurement at 10Khz and 11Khz was to measure the IMD of the tweeter and was about 0,3%.

Well, how do they sound? Pretty darn good, actually. At first, I was little nervous if LspCad had simulated the level of the tweeters properly. But after a few minutes listening, I decided that the tweeters had the proper level. This filter sounds better that my testfilter, and the difference isn't subtle, but very audible. Also, the highs seem to have improved quite a bit. I'm hearing a lot more detail now, which scared me a little at first: I thought something was broken, but it appeared to be on the cd. All in all, I'm very proud of my first completely finished speakers and I'm convinced that more will follow!