DIY cardioid monitors

It is fun for playing around - for results simulations without relevant datapoints are fruitless. Just get the tweeters and measure them in your baffle.

So are you talking about L-C-R circuits parallel to driver? For objectivist satisfaction and mostly out of laziness (it is significantly harder to dial in the filter without them) the impedance correction circuits appear to be a good idea.
For sound they give bad results - in my experience without exception.

Usually the tweeters have sensitivity to spare - then a parallel resistor directly on tweeter leads is magic - it lowers the effective Q.

 

BTW I'm still curious, even if the simulation looked worse and there's no way around the comb-filter.

The other option is way simpler
... just let the peaks be peaks and use a single series coil for the midrange as a crossover. No other parts on the midrange as I can't protect it from LF or the bass-mid crossover doesn't work. I still have a coil with a fitting value...

As for the tweeter XO parts - I think I have all that's needed from my simulation.

(The single parallel resistor didn't lower the Q enough, I had to actually notch out the peak at Fs to be less than linear.)
So yeah, I should really try to modify the tweeter somehow, heh.

I could of course just use a different tweeter, preferably one that's larger with a higher sensitivity (so the Q is lower) and lower Fs, but those are either silk or expensive as hell, usually both. And I want a ceramic tweeter, which basically limits my options to SB Acoustics or Accuton (which have too high a Fs anyway). And I don't want a ton of beaming past 10kHz, so a larger (say 34mm) tweeter would be dumb anyway...


Tbh, you're right. I should give it a shot. I only tried to see if it's feasible in theory, which seems to be the case.
Basically it's either:

• Widebander with two notch filters to make the treble linear

or

• Midrange with a series coil and a tweeter with a series cap and an LCR (or possibly mods and a single resistor) in parallel.

I suppose the result depends on parts quality and how bad the comb filter is vs the treble from the widebander. Dispersion even in the top octave shouldn't be too dissimilar based on the spec sheet. And I like the treble presentation, so I'm biased towards the 2 way approach vs 3 way. And of course it depends on how good the tweeter actually is. My experience with inexpensive tweeters is that they sound pretty terrible.


Anyway. Ordered two tweeters. Could be a nice XMAS project.
I can either sell them or use them in a different project anyway - they seem like solid tweeters for various purposes.

 

What is your reasoning to go with ceramic tweeter? If it is matching timber quality or ''material has a sound to its own and has to be matched'' then this is a fairy tale that does not die out because incompetent speaker builders with a loud voice can't do any better.
I used to also think that matching materials is important, until I tried all combinations. Mating to a soft midrange the best combination is still Be driver (even I am not thinking CVD-D for that ). Al sound so bad it has no place in hi-end. Ceramic is lifeless (lack in fluidity and microdetail) and dry (emphasizing macro detail). Some people dig it, I stay away.

IME hard membrane tweeters are not worth it unless you can swing to Be or CVD diamond. A well made silk dome beats the hell out of any Al, Mg and artificial sapphire for musical performance. Only reason to use cheap hard dome is for studio monitoring - you want to hear the faults - in the recording and are willing to take the crappy inherent qualities of the driver with it. This does not mean the equivalent soft dome would have less resolution or detail.

Agreed. The workaround in your case is that you can push the xo point rather high and get a small tweeter. The cheaper tweeters usually differ from their more expensive relatives in motor quality. Membrane is usually same. This means especially the low end suffers, very high end as well, as the inductance linearisation will be worse. In the middle it could sound quite good.

 

• Curiosity. I didn't really click with the Accuton ceramics that I heard, so I wonder how the SB compares. Plus I like the ceramic tweeter in the Reference 3K and I think it could mate well with the AL-sandwich widebander. And I absolutely despised the sound of their metal tweeters they had before. Though I suppose some crossover component changes also made it sound softer.
• What I heard from most soft domes always sounded a bit too "soft" for my liking, too. This is one of the advantages this speaker has over the large OBs I find nowadays. Sometimes the Voxativs are just too soft, maybe it's just from the lacking top octave energy.
• The Al version had really good measurements on hificompass, I find. Wide bandwidth. No comb-filtery dips in the off-axis response like we often see with soft domes. Distortion is low and has no weird peaks. Xmax isn't bad. Qms is high. It does look good on paper.

Also I'm not that well versed in tweeters

Honestly, that does sound like a pretty good combination to me. I think Be is different from other hard materials

Could be the sandwich diaphragm, but I find the small driver here to be an exception for sure. I'd take it over all the paper widebanders I've heard in this size range, heck I haven't heard one driver in this size range I prefer.

True, a smaller tweeter with a tiny faceplate would be ideal. SB has that tiny little 14mm tweeter with the small faceplate. Otherwise I know the Accutons have a small faceplate, but yeah...

The problem is the low-end response starts to matter so much with a 1st order XO that at least in the simulation the output at 1kHz matters a ton even for a 10kHz crossover. Although the crossover point (with all the peaks and dips of both drivers) isn't so clear cut, anyway. Point being that I need a driver with a low Fs and technically a low Q-factor. The shallower the knee, the better.
Also (and this is purely from an objective sense) - I wanted the bandwidth. Why bother with a tweeter if in the end we don't have any more extension than without it?

Maybe a ring radiator could be good. Not sure I liked what I heard from those, though. Foils/ribbons have to be large to go low enough. Fs of 500Hz would be ideal, doesn't need to be super efficient.
Vifa/Peerless XT25BG60-04? Though it's quite a bit more beamy and has a dip at 60 deg off-axis. But it's also cheap enough to give it a shot.


I've done a first order crossover in a coaxial and even that was super beamy. Frankly I don't think a truly successful 1st order MT XO can be done.

 

It came out blunt - I meant tweeters specifically. Aluminium (and most other hard stuff) for midrange and woofer is fine.

What compromises can you take.
For imaging precision and uniform soundstage, given equal care 2. order beats 1. order.
For plankton, involvedness and tonal density 1. order is superior.
It also heavily depends of drivers. Better drivers are easier to use with 1. order.

Not sure what kind of stuff are you sim-ing.
For a speaker with efficiency of 96 dB 1.order dome tweeter at 2 kHz will not work for sure.
For a 90 dB speaker you can easily find a tweeter that handles that.
Music content does not really have much energy above 1 kHz. Sim-ing with white (even pink) noise is off.

 

No, I got what you meant.

I mentioned the TB 4" since being a widebander it also serves as a tweeter and I did not like the other metal widebanders I've heard, though those really are just show impressions. There aren't many - I think I've heard the TB-W1337SDF, a Jordan and a metal Markaudio.
Either way the measurements aren't pretty for any of them. The same could be said for the 4" flat honeycomb TB I use, except that dispersion seems to be much wider with the flat diaphragm and the peaks can be taken care of.

That said it seems quite picky with regards to cables, even. Yesterday I compared the tinned copper in cotton to the TC in XLPO marine cable I used to use and it was not subtle.

Interesting, kind of reassuring to hear!
I'd expect the widebander to beat the two way in imaging precision/soundstage, to be honest. Especially this one as it's one of its strengths.
Depending on the XO parts I may just try both and compare against the widebander itself. I'm not expecting the three way to be superior to the two way for my preferences (where I don't care too much about a linear top octave measurement).

I'm not sure I understand what you mean by sim-ing with white or pink noise...

I didn't mean the power handling, which is fine. Power into the tweeter is negligible. I attached the curve I generally use for power dissipation.
I based the curve I use on an article from the german hifi selbstbau magazine:
https://www.hifi-selbstbau.de/index...yzer&catid=36:software--messtechnik&Itemid=66

It's about the summing. Currently it's a mess. It mostly averages out in the listening window response. But it's still not what I want to see. We'll see how it goes when I have the tweeter in a baffle. I'm already bracing for the worst.
Hey, at least Troels was surprised by the low distortion of the tweeters, I just found, heh.

 

I got the tweeters early (DHL was fast!), so I could run some experiments on the weekend already.



The build quality on those SB Acoustics drivers is merely okay. Also I'm not seeing anything extremely awesome in the distortion measurements. It's fine, but that's it. D2 seemed to hover around -55dB between 80 and 90dB and D3 at -70dB or so, but I got rather high D4, at around -60dB. I think I'll have to repeat the measurements since that's definitely not something I'd want to see. Sometimes a mic cable with too high an impedance can cause that.


I was pleasantly surprised by the large copper cap in the magnet system. However the response of the unmodified tweeter in the baffle makes it almost impossible to build a first order XO. The Q is just too high. The first mod wasn't agressive enough, but as you can see it can be dealt with.


On-axis -10dB is around 60kHz which is okay. Could be better, could be worse.

I built a very rudimentary test crossover with a normal quality Mundorf air core coil taken from the PAP Trio 15 Voxativ and a vintage PIO cap. Part quality should be fine. Could be better, but it could be much worse.

FR looked fine at 1m at the right height, but that's about it. It also sort of averaged out at the listening position, but the response with a frequency dependent window (to capture the initial wavefront) at the listening position looked very concerning. Exactly what I had feared.

 

It was too bright for me, so I added a resistor. I used what I would call a decent wire-wound resistor. Maybe it's a little sharp sounding. Overall the thing that should suck the most plankton out of the XO parts I used now should be the series coil.
It would be a good coil for a commercial speaker, but it's not really up to DIY standards. Mundorf L140 coil. Its DCR lkely also limits the bass precision now.


(I like to use this Yello album for testing)

Average between the two way and the three way,
Multiple position average around the listening position:



I didn't average as many positions, so there's more of a room interference here, but I was surprised by how well my intuition and my ears worked.

Sidenote: For a while I've been meaning to change the values of the notch filters so there's more 8kHz and less top octave and the tweeter is even closer to what I had imagined than what it's like now. Very nice.

So why is it that I keep emphasizing parts quality?
Because frankly it sounds worse to me so far. I don't even think it's the tweeter itself. I think it strikes a nice balance, not sounding too hard or too soft. I like its tonality and impulse handling. I feel like it's a good tweeter, especially at its price.
I'm not sure if more transparent XO parts would help or if it'd take a better tweeter. Accuton D25 would surely sound better than this and likely also one of the many Be tweeters, but would it be good enough?
Maybe some better tweeter mods, maybe an actual baffle (lol). Maybe a better crossover. I don't know.

The good

• It does sound more linear in the treble, more extended too. So for simple music it works decently well. I even liked it for Jazz somewhat.

The bad

• The openness, the plankton, clarity, speed, imaging. It all takes a noticeable step back.
• Of course it's also less coherent and somehow there's less slam. But the integration between the two drivers already seems to be quite good.
• Vertical comb-filtering is there. I attached a pic of what it could look like at one position, but move up or down a bit and the notch changes in frequency and Q-factor.

The way I have it set up I can switch between the two configurations in seconds, so that's what I've been doing. However now I think I'll keep listening to them with the tweeter for a while and then switch back. It likely takes some time to get used to it.

The good thing is that I was able to come up with a workable crossover this quickly. I feared the worst and instead it went swimmingly.

 

Parts quality matters, but on the level you operate I doubt Mundorf L140 (I have these in my lab drawer) is the bottleneck. There are much better coils in existence but the difference would become significant with better drivers and better electronics. The wirewounds lack clarity and microdetail in highs - with a good tweeter you should be able to circumvent that. You are not using the parallel coil for the tweeter, right?
Look into the phases of your acuoustic output - I bet the fish is buried there. Do the fr measurement again but flip the phase on midrange driver. It would tell you some gross mistakes - if any. What is the xo point from bass to mid again?

 

Nope. Actually I didn't have the time to open the speaker and rip the crossover board out, so the coil is in series with both the mid and bass drivers. But that's fine, technically it's better that way for the bass-mid XO.
The tweeter is simply connected where the input of the coil is connected and then has its own little crossover.

XO from bass to mid is circa 180Hz, depending on the angle. Let's say 200Hz for simplicity.

Measuring the phase matching is not an easy task btw. Move the mic up or down three millimeters and already some of the treble will start to disappear. However I tried to measure it still, but instead of inverting the midrange I tried to look at the whole rest of the speaker as a unit and inverted the tweeter instead. Inverting the midrange will make it almost impossible to measure at 1m distance since it's supposed to sum properly at 2m distance. Even the baffle from the tweeter I introduced wasn't planned and might change things to the woofer to mid crossover.

Anyway, here's the measurement. I tried to measure at 1m distance in my room.


We would really need more low end bandwidth from the tweeter or a lower Q rolloff. We still see less output between 500Hz and 2.3kHz with the inverted tweeter. There's also a spot between 4 and 5kHz. This should be around where they cross over. Otherwise however the matching is quite a bit better than what I had anticipated.

The tweeter mods were mainly back chamber damping. Stock there's no damping material inside, so I tried a couple different damping configurations. The one I have now sounded the best, but doesn't limit the low end extension as much as the one above. There is definitely a greater sense of opennness now. I experienced a similar thing with the midrange, where a more closed enclosure just didn't sound as open, but in that case we also change the polar pattern at the low end.

As far as parts quality goes, there's really no better quality part than not having any parts is kind of my philiosophy. But I'm probably too much of a widebander nut anyway. So for now I just blame the lack of plankton on the fact that there used to be no parts on the widebander at all and now suddenly there's a coil and a tweeter with a cap and a resistor!
And the tweeter might simply not be as resolving as the widebander itself.
And the vertical off-axis lobing which is quite severe. Wavelenghts are 5-7cm in the range of interest. Center to center spacing right now is 14cm. By removing the tweeter faceplate and cutting through some of the midrange faceplate we could reduce it to 10cm, maybe 9cm. Even that is too much, but we can't really do any better than that. The dZ also becomes an issue and we might get quite a bit of diffraction from the midrange basket.

The initial part of the step response doesn't look any better than it did without the tweeter - I think from the 24kHz resonance. Maybe a soft dome (ring radiator?) would've indeed been the better choice in general. The excess phase also isn't as perfect anymore. I now get -5° of phase shift at 20kHz. But that's really more an observation than anything else.

 

Your first clue that there is work to be done in XO.

You could do that. Surely lets you go off easy. There certainly is not such argument that adding (parasitic) components would increase resolution.

My money is still on filters needing more work. The worst situation is when two drivers are fighting each other. The output could be linear on axis and tonality passable - but the resolution is worse than either driver by it's own. 1. or 2. -th order filters are very shallow, they muck up a large area in the xo region when wrong. You have nailed the XO when you hear similar level of effortlessness and resolution as the individual drivers playing.

 

Don't get me wrong, I'm not expecting the 2nd revision of this crossover to be magical. But currently I don't know how to improve it:

1. The XO point is where I want it to be (5kHz).
2. The listening axis FR is where I want it to be and the phase response is perfect.
3. Most importantly: the averaged FR at the listening position is spot on to where I want it.

And it even took as few parts as possible. So for now I'm not sure what to change. Except the filter order, but your comment above makes me feel I'd prefer first order and using RePhase I could hear the excess phase introduced by 2nd order.

I don't think the comb filtering can be improved from a first order crossover. Which is why I mainly built a first order out of curiosity as to how it sounds, not as the correct solution. There are certainly worse offenders for tonality changing sitting vs standing. Even Marten speakers, for example. But it's logical: The Accutons have less bandwidth and their center to center spacing is larger. And I'm generally a fan of the 1st order Marten speakers.
[The Mingus Septet might be similar in terms of sitting vs standing due to its smaller midrange with more HF extension.]

If time permits I will build a 2nd order crossover. I have a feeling I'd prefer it. 1st order between M and T is nuts IMO.

BTW it was impossible to gauge the level of resolution of the tweeter playing on its own. It sounded neither overly smooth, nor did it sound harsh. It didn't sound super fast. But that's pretty much all I could tell.


For what it's worth my dad loves it, is already thinking about building a standmount (to replace his Canton Reference 3K?), etc. He sort of immediately fell in love with the sound . He's used to 3 ways, I'm used to widebanders. Maybe I just need to get used to it.

Spoiler: Tangent

It's not like me and my dad generally disagree, either. We got to listen to the flagship Canton Reference GS speakers this weekend driven by some local boutique tube amps in a decent sounding room and with all the time in the world to play music of our own chosing. We got to chat with one of the speaker designers, the amp builder, etc.

It was awesome. Except it didn't sound that resolving. It seems that for these larger companies every cent counts. MKTs, electrolytics, cheap MOX resistors, cheap coils arranged for maximum crosstalk, plastic driver baskets, etc. And of course it's not just Canton. I saw the same thing in Burmester speakers costing many times what the Canton does.

Given that the crossover parts quality is similar between the new series and the Reference K series I think the culprit (other than that the system matching wasn't too great) is that these new dark grey drivers are too damped. Supposedly the resonance doesn't peak above baseline.

 

Just spent way too much time measuring the vertical off axis response ±45 degrees. Only within ±5° it stays smooth, past that we get into comb-territory:

(sub 500Hz is really only there to fill the graph. They're gated and theres no value to that data)

Not too far off from my simulation using tweeter FRs from the internet, LOL:
I think center to center spacing ended up being a little less than what I had anticipated, so it maybe looks a bit better.


For reference: If we sit at 90cm ear height at 3m distance and then decide to stand up we are now at +15 degrees and will be missing quite a chunk of the treble.
Those blue holes are also super nasty. They're very deep holes in the FR and if we could hear what it sounds like at that angle (which we can't in a room) it would sound absolutely terrible.

FWIW averaging through all the angles nets me the attached result. Matches the in-room RTA quite well.

 

I'm sort of spamming this thread now, but I want to take you guys along the ride (hopefully not just Priidik lol), documenting my thoughts during the process - also for future reference.

Quick apples to apples comparison between the 2 way and the 3 way:

(This time a more sensible 10Hz-40kHz range. I used the larger microphone for the 2 way XO11 measurement, so getting any sort of accuracy past 40kHz is a bit of a pipe dream, despite of course compensating for the HF rolloff up to 100kHz.)

This is more of a sanity check than anything else. The full 180° measurement would be more telling, but it's likely a couple months before I can find the time to do so. That could take multiple days, depending on how good the initial measurement turns out to be, the weather, etc.
In the end it's the sound that counts anyway. I just want to make sure both the listening axis and in-room response is the same when comparing the two configurations and at least that seems to be the case already.

Two thoughts:

• I will probably try a smaller resistor value for the 3 way to raise the range from 2kHz on somwhat.
  However I'm not sure it will be better since the in-room looked remarkably close already.
• As mentioned the 8kHz notch on the XO11 was a bit off, which is why I had planned XO12 with more 8kHz and less of a bump in the top octave.

The tweeter appears to be quite beamy past 17kHz. That peak at 24kHz also didn't look as bad in any of the graphs online. On axis there's a dip here. Not good. (These kind of sharp dips normally sound awful)
The tweeter also doesn't have a ton of extension when taking the off-axis response into account. It's only from 50kHz on that the tweeter has more output than the widebander!

I consider these subjectivist studio monitors, which is why I put so much emphasis on the FR, directivity, etc already from the beginning on.

 

I made a little tweeter housing to better gauge what the tweeter can do:

(In this image trialling a baffle without a step for lower diffraction - I sit lower than tweeter height, but a small gap is better)

The center to center spacing is now a bit larger, which isn't a plus. But I don't notice any difference in lobing or imaging, really. You don't notice the lobing much, anyway.
I also went through a couple different cap and resistor values, but I ended up going to back to the original "XO2" version. Seems to work out well. It does sound better now. I also added an extra ground wire to the magnet's top plate and I have no idea why it works, but it always does. Sounds more precise, both in terms of imaging and in its transient handling.

Aside from the frequency of the dome resonance, which is a bit low at only 24kHz the tweeter seems to be very good objectively and in its construction. It even seems to have a Ti voice coil former, as far as I can tell.
Dispersion is very even, no weird dips. However most importantly distortion is very low and it's especially nice to see that D3 seems to be much lower than its Al counterpart. Some ceramic drivers still have high D3. This one is basically on the level of the very best tweeters with D3 hovering around -75dB at 90dBSPL past 2kHz.
The low 3rd order distortion matches my impression of a rather smooth sound for a hard dome. Generally I find drivers with low D3 to also have good resolution and generally they will also have good power handling. It does have good power handling and it does outresolve the tweeters I've tried in DIY projects or the LS50 Meta for example.

Here's a measurement without a low pass filter using the harmonic frequency as reference:


However that doesn't mean that the distortion of the whole speaker will be as low in the upper midrange. But since the widebander seems to have very low D3 already, it's not that big of a deal.

With all the tweaks in place the general characteristics between the 2 way and 3 way don't change much. I still feel there's more liveliness, microdynamics and more focused imaging without the tweeter.
On the other hand sibilance is slightly more natural with the tweeter in place and it no longer sounds like ice picks in the top octave when sitting on axis. Somehow there seem to be less treble behind the speaker, I feel.

I would say the widebander is more resolving than the tweeter. So in general Priidik was probably right.

I quickly made a comparison table including the modded HD600 and HD800 for reference, just for fun. Attached.

EDIT: Attached some more measurements.

 

Naw man, keep the updates coming! Love to see the process and a lot I can learn about the speaker design process.

 

Have you tried lower? The tweeter can easily handle 1.5 kHz in that small speaker. Dispersion would be more uniform.
Only downside is below 2 kHz distortion starts to rise. In a bigger speaker the argument could be made that macro dynamics will suffer, not the case here. Your woofer is by far the limit here.
At 5 kHz the acoustic matching from driver to driver is madness. (expect ´weird imaging and phase issues)
Not impossible, but it is hard work and some luck involved.
First issue is many of the imperfections around driver baffles (edges, grooves, membrane groove, even screws) start to mess with the wave propagation.
Another thing, in your speaker spacing from tweeter to mid is way to long for that xo point. You are in half wavelength zone here!
I say no more than 2.5 kHz gauging from the picture.

 

Bit of a lengthy response, sorry. But to recap, there were basically three main goals for this speaker:

• good step response aka minimum phase behaviour
• cardioid behaviour from the upper bass on
• full-range sound

So anything other than a true acoustical first order crossover is out of the question for this speaker due to the first (and most important) goal/requirement.
For a different speaker project - maybe. But then I would chose different drivers, enclosure dimensions, etc. Probably a standmount, too.

This is not to say that I'll never try a second order crossover in this configuration. It should be easy enough to set up. And I'm curious myself. But it's not how it'll stay.

Spoiler: lengthy response

So... the goal was to trial a 1st order acoustical crossover from M to T, which I thought could not really be done without sounding off and it turns out that it does sound normal. So in a sense I think I achieved what I had set out to do.

Maybe 2.5kHz could work from a 2nd order acoustical crossover. But I'm not interested in that as I know how such designs sound. It's not what this speaker is supposed to be - my own variant of something like the Dutch&Dutch 8C or the Kii Three, with the same "perfect" step response, cardioid response and 20Hz bass extension, but fully passive.
(In this case the excess phase will be lower since it's done without DSP. It's not just an approximation.)
In a way I wanted a fully passive variant of this new breed of "objectivist" speakers.

The woofer is rated to 120W, the widebander to 30W. In this design we're exceeding the widebander power limit first. However I'm not too worried about that since I've pumped 75VA (as measured) into the whole speaker already without damage and while still sounding surprisingly good. I'm more worried about hitting Xmech from the widebander, but I've not managed to do so, either.
Exceeding Xmax is a different story, however the distortion will be so far down due to the acoustical (not electrical) crossover that I think it hardly matters.
For all intents and purposes I'd say these speakers go loud and low enough. Something like 20Hz and 105dB in-room, depending on the room.

Or if you're talking strictly macrodynamics subjectively... Well, it's probably still the small widebander that's the limit here, but considering how inefficient it is, it's incredible. Other small speakers I've heard don't compare.
It'd likely still benefit from a better woofer, though.

Yeah, the phase matching is madness with a crossover this high and spacing this large. And of course with a first order we're not just talking about 5kHz, but rather two octaves below and above. From ca 1kHz all the way to 20kHz we're in the crossover region, where each driver has an effect on the other. It's not possible to put two drivers of this size close enough, even if the diaphragms were literally touching at the edges.
I measured the proper offset at the listening position and while the response was always reasonably smooth, the best response was at 3mm offset, not 2 or 4mm!

This is why my first response to a tweeter was that it would objectively be worse due to the comb filtering. But I realized that it doesn't sound as bad as I imagined. In the end it's a matter of preference. I still prefer the widebander with the two high Q notch filters. But without the notch filters it might be a different story.

In the end it doesn't really matter. The widebander has smooth treble as is and doesn't need help from a tweeter for my preferences. Sure, good tweeters - of which there aren't many - will have better treble. But at what cost?
In a way these are a standmount version of my larger speakers, I find. It sounds odd, but that's how I feel.

You do have a very good point with the baffle imperfections. I'm sure it could be improved. I think I need a flat baffle due to how the front and back wave meet. I don't think I see anything in the CSDs at the listening angle, however the on-axis response looks a little worse and we do see some diffraction effects.

I attached two representative CSDs. Obviously at a different angle they will be different.
Despite the 5kHz crossover, the response has a dip at 1.5kHz caused by the tweeter LF rolloff.

 

I have Satori ring dome in one of my speakers mated to Satori 6'' mid - a 1. order filter.
XO point is at 1.2 kHz (acoustic) ! This is 3-way with hefty 12'' woofer and 90+ dB / 2.83 V sensitivity.
The woofer spills it's guts way before the tweeter even breaks sweat.
I measured close to 300 VA continuous going into the speaker playing some Pig&Dan and Clark (with the used particular track the limit of woofer mech excursion) - for a few minutes and I was afraid the voicecoil would melt. All drivers survived.
*powered by a EV amp that does 1.2 kW per channel.

There is an article somewhere - studies from era of our grandparents - the music energy density was found highest in 200..600 Hz range, and dropping 10dB/dec after that (same slope as pink noise ET). This was the time of jazz, blues and orchestra. Nowadays I wager the thing is tilted more to the bass range.
We have the outlier of pop music with nasty female autotune, but are you really willing to listen to that loud?

The point is, when I simulated power usage by that curve of tweeter in the scenario I got about 5 W rms when woofer does 250 W.
At one point I even attempted to measure the power tweeter uses with music content, but did not get consistent results - only in Slayer you can trust. Can't recall what the numbers were but I do remember not believing the measurement - as it was so low. High frequencies in music are rather intermittent.
Mid is getting about 75 W. And the mid is crossed at 270 Hz. The mid's voiceoil will burn way before the tweeter gets even warmed up.
Tweeters die with crappy electronics emitting HF oscillations. That is a real thing.

 

Like you I'm not worried about the tweeter hitting its power limit. Looking at my power dissipation graphs I had to go looking for the tweeter, it was that far down compared to M and T! Buried somewhere along with its resistors (which makes sense). I'd hit Xmax from the tweeter(!) before ever burning out its voice coil.

That sounds absolutely badass and I'd like to hear that at some point.

Anyway, what I meant is that it doesn't work from a 1st crossover. Sure, the network may be 1st order, but the overall filter order won't be 1st order. Here's an example:

The black and green line are what happens with this cap and coil crossover from M to T at between 1 and 1.5kHz. Now if I invert either M or T the thin orange and green lines emerge.
An almost perfect 2nd order XO. Same as what Troels once found. It's still a lot better than 4th order where we don't flip the polarities to me, however.
Accordingly the phase response will now be flipped 180° for high frequencies. I simply plotted the excess phase for what happens if we don't flip the polarities, hence the straight line.

So yes, 2nd order is possible around 1kHz. 1st order is possible from 4kHz onwards.

I could get an even deeper notch to make it a real 2nd order by crossing even lower. But the directivity index already has the familiar "tree shape". At least that's how we call it in Germany (Tannenbaum). There is a major widening around 2kHz as the tweeter takes over from the midrange. It won't be nearly as bad when the midrange isn't cardioid and with a cardioid midrange I would have to mate it to a rather large waveguide. Or instead just build a traditional 3-way like you have done, like I mentioned above.

I think I have similar XO parts as needed, so I might just give it a shot. But it's 2nd order effectively and I know the flipped polarities are audible to me.

Sometimes I am
Well, not at 100dB or so. More like 85dB. Taylor Swift isn't really any worse than the Talking Heads in that regard.