The transition into the region where the tweeters become more directional is pretty abrupt. But I have not seen any other dipolar speaker with regular dome tweeters where the behavior is smoother. Experiments have shown that 0.75" domes will not help since the transition depends on the blooming. Perceptionally, however, the highs are well integrated into the mid range.
The wider than desired dispersion in the mid range is also a function of the baffle size and I still plan to improve it (these are the first measurements I could take for this speaker since a comparably long time). So far, up to 1KHz the response is fine and also looks pretty much sorted beyond 7.5KHz.
One thing should be said about the diagrams, though. They suggest that under angle the midrange is louder than the rest. Perceptionally, however, the behavior is as follows when you play pink noise while slowly turning the speaker you sit in front of: It is not that the midrange becomes somehow louder under angle or something is peaking out. It is only that there is increasingly less bass as it should be. And it takes quite some angle until the highs will fade as well. I performed this experiment indoors with the speakers in listening position but one sat on the turntable. Note that this experiment turns sound that is normally indirect into direct sound.
So the relative imbalance is there and audible with music, although it requires increased SPL (stronger reflections) and strongly depends on the program material but for the ears and the brain the effect seems less severe than you would guess from the pictures.
This may be a case were experiments are useful or even required to understand how visual data translates into audible significance and this point has to be kept in mind when looking at diagrams.
For now I conclude from the experiment: As long as the radiation under angle does not at all or not significantly (>1, 2 or 3dB ?) exceed the on axis SPL you have achieved something. In this case the 30° response exceeds the 0° sound pressure by 1.7dB @ 1700Hz (see cursor position in the FR diagram).
Sure, a perfect figure eight would be nicer but a speaker with such a behavior has to be invented first.
Constant directivity with dipoles is anyway history as soon as D/l exceeds 0.25 (actually it is even lower but 0.25 would already be a tremendous practical achievement). To put it into perspective: 0.5 is where the first peak is, which is practically a good aim if you consider other things like number of x-over points, vertical polar response etc. Priorities have to be set.
I also assume that the blooming has an effect on the ASW (apparent source width). But this will require a comparison with another dipolar speaker that is more perfect on paper. In absolute terms there is no impression that anything is "displayed" too big, overly wide or distorted in any way.
Possible ways forward for the polar response:
Reducing the baffle size for the mid woofer is an option. The 18WU has enough reserves. For the tweeters this measure is not convincing. You can gain improved dipole behavior with nude and faceplate-less drivers but at the same time it becomes increasingly difficult up to impossible to maintain a proper transfer function in conjunction with the cross over because you cannot easily equalize the dipole cancellation electronically. Also, the tweeter should kick in at 2KHz the absolute latest because otherwise the response at the back of the mid woofer has already rolled too far off. So “unbaffled” tweeters seem to be a dead end. Here is some more background information.
Wave guides for this speaker and these drivers would probably have to be too big in size to be effective if you want to keep the back-fire tweeter.
Planar tweeters could be an alternative. But the real good ones are too expensive (at this stage anyway). A Neo 3, which would not even line up quality wise with the other drivers, cannot be the only answer to the issue.
As one of the last resorts, room treatment should not be forgotten although treating a given room to a precise transfer function is virtually impossible.
That leaves only one practical and economical way to explore for now if different drivers are to be used: Small (< 3”) full range drivers as tweeters that can withstand some moderate equalization.
Additional comments about the diagrams:
Personally, I prefer the FR response plots because I can draw the most information from them. The sono plots seem to be kind of modern right now but provide a good overview. So I create and publish them. The directivity patterns are my second favorite but they are easily visually overloaded.
Also, there is one thing about them that I cannot explain: As stated above, the sound pressure @ 30° and 1.7KHz exceeds the on axis SPL by 1.7dB. That is the worst data point. Now, how come that the normalized and equally smoothed (but gated) directivity pattern looks as if the excess is closer to 3dB than to 1.5dB @ 1.6KHz and 30°. So much about visual data...