My approach is to adjust [phase] and [time alignment] by FIR filter with one time.
It seems I misunderstood your design approach. You are designing 8 FIR filters for the drivers. Each filter to correct time delay between drivers and phase rotation within the driver? My understanding of this approach is limited as I have never used it. It is a very popular approach. I would have tried it, but just don't own the needed equipment. Please take the following thoughts accordingly.
I now see the 2x4 can accommodate a 6144 tap size per output channel. This is very limiting particularly in regard to dealing with delays. Considering that you have set delays in the 2x4 to removed most of the timing offset by aligning the IR peaks. That would greatly reduce the tap size wasted on timing delay. You did indicate though that the midrange driver was not time aligned properly. That would suggest that some additional 2x4 delay adjustment on that driver would resolve that problem. The closer the delay timing is to ideal using the 2x4 delay feature, the smaller FIR filter needs to be.
With only 6144 taps the low frequency phase rotation cannot be fully removed so we need to expect that the phase rotation can only be corrected to be flat for the higher frequencies and not for the bass range. That would explain some of the low frequency phase rotation present in your measurement.
So my suggestion would be that due to very limited tap size, readjust the delay of at least the midrange horn and recreate that FIR filter. That may improve the overall phase rotation greatly. We should still expect some of that low frequency phase rotation to exist due to the limited tap size.
More experienced users of this method and this equipment may have a better input for you.
[Minimum phase] curve is created based upon SPL property, you said a year ago as far as I remember??
Yes, the minimum phase is calculated on the SPL measurement without regard to the measured phase. It reflects the causal phase variation that correlates with the SPL response. The difference between the actual phase response and the calculated minimum phase response is the excess phase response. See Wiki or other more technical references as my understanding doesn't extend very deeply.
What is the difference between [Phase] and [Minimum phase] on the [SPL/Phase screen]. The [Phase] with t=0 at IR peak situation and [Minimum phase] the same? Those look to me different.
They are different; different by the impact of the excess phase in the system. In your measurements you have successfully eliminated most all the excess phase in the higher frequencies. There remains only a minuscule amount there.
Is it the minor 70° spread in the response at 20k that is concerning you? That is practically nothing in group delay due the small wavelength. It is no practical concern. I suspect it may be due to the measurement extending to 22kHz where the SPL is falling off rapidly. I am unsure as to the exact cause, but one source could be that the inductance of the TW voice coil is adding excess phase at those frequencies. It could also be related to the impact of other aspects of the driver design or even the horn design. Drivers are basically minimum phase devices, but not entirely. They are likely to show excess phase particularly as they are pushed out of their bandpass range.
I designed my FIR filter to correct the actual phase response without regard for the minimum phase. It just makes sense to me that we want the sound to arrive at LP in its proper timing regardless if the SPL falls off due to the house curve or system bandpass limits. This is different from the generally offered advice to target minimum phase. I have not seen others suggesting this though. I just have not yet understood why minimum phase is preferable from a theoretical perspective. In practice, either approach should be perfectly fine. We should all agree that it is indistinguishable at the high frequencies. There may be an arguable difference in the bass range where some report hearing differences.