So here is how you can do it (among many other possible scenarios) :
- For each driver (with several measurement per driver, as discussed above), use minium-phase EQ to get the amplitude reasonably flat within the pass band (the more you can trust your measurement(s), the more precise you can go, hence the reasonably)
- Use the "compensate" mode in the minimum-phase filters tab to flatten the natural high-pass and low-pass of your driver by trial and error (you need a measurement with a low noise floor, as it will quickly realize when playing with that feature...).
- at that point you should have a linear amplitude and phase (in the pass band and around, depending on your noise floor). If you don't then adjust your "compensate" settings, and also play with the "time offset" option in the measurement tab. You should not have to use phase EQ.
- Do not operate your driver with this kind of correction of course: this is only a temporary state!
- Apply the desired linear-phase high-pass and low-pass filters, and make sure you do not exceed the capabilities of the driver (excursion down low, breakups up high, directivity, etc.).
- Check the correction curve with the measurement bypassed to make sure it does not get too high in amplitude (for example if the target high-pass filter is much lower or with a shallower slope than the natural one...).
- For good measure, use the main volume attenuator in the "general" tab and make sure your correction does not exceed 0dB (amplitude offsets will have to be dealt with at some other place, for example in the crossover engine or in the amplifier...).
- Always use complementary slopes for your crossed-over drivers (ie LR of identical slopes on both sides, "reject high" on both sides, "reject low" on both sides, etc.). Try to avoid brickwall filters as these will add additional constraints for complementarity (same number of taps, etc.). If you need steep slopes you will be better off with high order LR "shapes".
When generating the impulse, if you do not have constraints on the number of taps (which should be the case if you are using jriver on a descent computer) you should use the "middle" centering option, and a large number of taps (64k should be more than enough for any realistic situation). With that many taps you can use a gentle windowing algorithm such as Hann, Blackman or Nuttall, without loosing much precision or steepness. You can also handle the delays inside rephase, directly specified as distances, eg "middle+3cm" to compensate for your driver's geometrical offsets (you can check that afterward with the "inverse polarity" method, seeking for the deepest null at the crossover point)
Once each driver is EQed and filtered that way you can add them together in your convolution engine.
I think Jriver will require a different set of impulses for each sampling frequency it might have to handle...