driver&bulletts, impedence and W

Well, what I mean is you need to build the physical box, and then take some acoustic measurements using your new soundcard and microphone (presumably a measurement microphone? The kind with a long narrow barrel on a wider body...) Or at least, make up a temporary baffle to mount your horns and bullets on, with approximately the same positioning as your final box. To find out what kind of passive crossover to use between the horn and bullets, you'll initially drive all the drivers from your LMS so that you can easily change filter types and frequencies. Obviously not everything the LMS does can be translated into a physical filter, so you can only find the best compromise.

By "2-channel measurement" I mean you use the method that involves sending the sound card's left channel output to your speakers and connecting the measurement mic to the left input, while looping the right channel output straight back into the right input, so that the software can show you the DIFFERENCE between the straight, flat signal and what your speakers reproduce. 

In 2-channel mode you can use pink noise for a signal, or some music - as long as the music contains a relatively balanced range of frequencies.

There is also the important step of telling the software exactly how far away the mic is from the speakers so that it can make deductions about the relative timing of the two signals. Phase is about time difference.

I recommend "ARTA" (www.artalabs.hr) as it is shareware, so you don't have to pay for it unless you need the functionality to save trace data; you can of course capture screen-shots for this purpose. Once you have it set up to work with your soundcard, spend some time playing with it, a microphone and a studio monitor or Hi-Fi speaker to get a feel for how it works. 

There are a lot of features and modes, but for your purposes you would mostly want to use "2FR" mode (two-channel frequency response). This gives you a graph with two traces, a green one for frequency response and a grey trace for phase angle. Don't forget to click "crosscorrelation/delay estimation" after taking your first measurement, so that it can apply an accurate correction for mic position to the phase estimation.

Then, to accurately phase-align drivers, you would first fix your microphone in place relative to the speaker, and ensure it does not get moved during any measurements. Now, apply your initial or first-guess crossover settings in the LMS. Probably don't bother with EQ or delay for now. You can't easily translate those into a physical crossover anyway.

Next, take a measurement with only the lowest-frequency driver running, and make the resulting graph as an "overlay" or "saved trace" (different software uses different names). 

Then, take a measurement of the next driver up on its own. You will be able to see the two frequency and phase curves on the screen at the same time. 

In the frequency region where you want the two drivers to meet, you want the phase curves to be as close to touching as possible. You can change the slope of the phase curve by changing the filter order (1st, 2nd etc) and frequency. You can move it left and right while keeping the slope, by adjusting the delay between the two drivers. Again though, you can't do that in the real world, except by physically moving the drivers. 

In terms of frequency response, you want both drivers measuring about 6dB down at the crossover point, to avoid any lumps in the total response. If they're both in-phase and -6dB, the output will sum to 0dB acoustically. This is likely to be one of the points you have to compromise on; it may not be possible to get good phase alignment with filter slopes and frequencies that put the output at exactly -6dB where you want it. You can compensate for this with EQ, later.

In this way you bring invisible acoustic information into the visual domain, where it is much easier to see the effects of your adjustments. You would repeat this process for each crossover point. You may need to adjust the driver you're currently measuring, or the one represented by the overlay trace, to get the desired result.

Finally, although it is tempting to tweak and tweak until the graph looks perfect - once you have them within 45 degrees, that is probably fine. If you can get them to perfectly line up using simple filters, so much the better, but if not, don't tear your hair out. You don't want to end up designing an impossibly complicated and expensive passive network.

Hope this gives you more understanding :)