Redesigning JmLabs Electra905 crossover

[pjpoes]

Hi guys, I have been working on this for, well on and off, years now, trying to find a good solution. I bought a pair of electra 905's from the Focal distributor around 7-8 years ago, maybe longer. I got a good deal, but at the same time, I spent so much, I can't bring myself to move on to a new equal level DIY project. I've built smaller projects using focal drivers, some seas projects, etc. I like certain attributes of the Focal's, certain others not so much. Anyway, I decided to finally test my crossover skills, or lack of, on the factory crossovers. I asked the distributor for the factory crossover schematic, which he sent. I measured the drivers in the factory enclosure without a crossover, one at a time. Unfortunately, when I pair up the midbass drivers, I get a really strange measurement, and I'm not really sure how to properly do this. I talked with someone else who mentioned that I could get a reasonable approximation by just telling the crossover software I have two of these midbass drivers, but only measuring the one. My concern is that, well they aren't the same, one is on top, the other on bottom, so the distance between the two is over a foot, and it seems like this isn't the best method. None the less, I do think I am getting a reasonable approximation, so I put the factory crossover into my software, and came up with a graph, using my actual measured responses. I then tried to come up with my own crossover, adding some baffle step compensation, a better contour, and even a notch filter. Before building this, I wanted some opinions on what I have. First I will post my response charts, maybe someone can tell me if, from those, something looks funny. Then when I get a schematic for the crossover together, I can post that.

[pjpoes]

here is my modified crossover's response. Keep in mind that, though the response doesn't look super smooth, it is an actual measurement in a room on the floor, not an anechoic chamber or outside (Too cold right now). However I compared my measurements of the 5W 4252 to both the factory specs, traced with SPL tracer, and to another persons measured response, and it matched both very closely. The biggest difference is that I have a strange dip in the response around 1800hz that I think may be an artifact of the measurement system. There is a slight one in other measurements of this driver, but mine is larger. Also, at first I thought these responses looked terrible, but then after looking closer, I realized it actually is within say +/- 3 decibles. So again, first, does my response look better than the factory one, or just to my eyes? The factory one looks really bad, and to my eyes, it almost seems like I must be doing something wrong for it to be that bad. It's a straighforward 3rd order with no zobels, no baffle step compensation, no correction cuircits, not even a proper l-pad, just a series resistor. Not only that, but with my measurements, adding no more components, I felt I could get a better response by just changing some values. Could this actually be a problem with how I have the crossover software orienting the two woofers?

[pjpoes]

hmm I may have partly answered my own question, ok so here is another graph with the drivers offset using numbers suggested in the manual for their project. I will have to measure mine to get an exact set, but I think this is, again, a reasonable approximation. I see the factory response improves considerably.

[pjpoes]

Ok here is the factory schematic.

[pjpoes]

well I was really hoping someone would have some thoughts on how to improve the crossover for this speaker, and what I need to be mindful in designing an MTM around these drivers. There is a large expense in this, but the TLR's that Zalytron has, or even the TD120's are so much better sounding than the TD90's, that I've been very tempted to just design around those instead.

Anyway, let me know what it is I could supply that would help someone of you aid me in the designing of a better crossover for these speakers. I could stick with the basic setup they chose and just add a zobel network and maybe some baffle step compensation, alter the values from there to flatten the response back out. Does that seem like a reasonable approach?