Funny i just drawn a coaxial tweeter in my mid range 🙂 but since they are so small....impedance is a big issue 🙂 anyhow looks good !! looking forward to your findings!
I've done some more napkin math on the shading network.
My idea is to achieve the shading with 2 techiques.
The first is to run multiple coils and then terminatte them early. Since the force is the number of coils times the current, which will be constant then 4 coils equals 4x the force, 3 coils is 3x and so on.

The problem is that when I designed the above membrane, I was planning on using 6 mm wide coils, but have since switched to 3 mm wide coils. That limits the amount of coils I can realistically cut so I will most likely have to settle for 2 coils, which then can get me down to -3 dB. Since I need to get down to -12 dB I need another technique to achieve that.
And my idea for that is to, when there is a single long trace, to add parallel traces that terminate early like a parallel resistor. But unlike a traditional resistor in a crossover where the voltage is wasted as heat, I want to minimize the efficiency losses by using most of the parallel resistor trace to generate foce.
I can do this by first splitting the trace into regions with my desired X - dB step, then I split the trace into the green and the blue part.
The cut length can, at maximum, be as far the termination of the shading step before. Lets for now assume we set it at that maximum length minus a cm or so.
Then our next variable is what the width we should use for the green trace. And here comes the main part of my idea:
If we set the green trace to be minimal, then the resistance of the green trace will be high relative to the blue trace, thus the attenuation will be negligable.
If we instead set the green to a maximal width, then the resistance of the green trace will be low relative to the blue trace, thus the attenuation will be very large.
Thus, there exists a sweet spot of green trace width where I can get exactly -X dB attenuation in that given step.
And when I do some napkin math to calculate the actual trace widths then I start getting problems. Or at least I get problems when I try to have -1.5 dB steps. The problem I get is that for that to work, I need green trace widths of exactly 0.32 mm at a cut length of 850 mm, where even 0.01 mm wrong will mess up the resistance.
I do not think I can achieve such high precision with my cutter, so I think I will need to settle for larger steps like -3 dB, then the trace width doubles while the avaiable cut length also increases so in practice I can get away with ~ 1 mm green trace widths which is perfectly managable.
And it looks like -3 dB steps will still be plenty good enough based on this paper.
The following graph snippets are stolen from "APPENDIX 3. VARIATION OF CBT SHADING" of the paper linked above.
If we compare perfectly continous shading vs -3 dB steps it is still pretty close.
But we also need to keep in mind that my array will be truncated at -12 dB, which also messes up the response:
Keele himself, in the paper, based on the simulated responses, advocates for -3 dB steps & -12 dB truncation and as far as I can see for good reason, so that is probably what I will use.
My idea is to achieve the shading with 2 techiques.
The first is to run multiple coils and then terminatte them early. Since the force is the number of coils times the current, which will be constant then 4 coils equals 4x the force, 3 coils is 3x and so on.

The problem is that when I designed the above membrane, I was planning on using 6 mm wide coils, but have since switched to 3 mm wide coils. That limits the amount of coils I can realistically cut so I will most likely have to settle for 2 coils, which then can get me down to -3 dB. Since I need to get down to -12 dB I need another technique to achieve that.
And my idea for that is to, when there is a single long trace, to add parallel traces that terminate early like a parallel resistor. But unlike a traditional resistor in a crossover where the voltage is wasted as heat, I want to minimize the efficiency losses by using most of the parallel resistor trace to generate foce.
I can do this by first splitting the trace into regions with my desired X - dB step, then I split the trace into the green and the blue part.
The cut length can, at maximum, be as far the termination of the shading step before. Lets for now assume we set it at that maximum length minus a cm or so.
Then our next variable is what the width we should use for the green trace. And here comes the main part of my idea:
If we set the green trace to be minimal, then the resistance of the green trace will be high relative to the blue trace, thus the attenuation will be negligable.
If we instead set the green to a maximal width, then the resistance of the green trace will be low relative to the blue trace, thus the attenuation will be very large.
Thus, there exists a sweet spot of green trace width where I can get exactly -X dB attenuation in that given step.
And when I do some napkin math to calculate the actual trace widths then I start getting problems. Or at least I get problems when I try to have -1.5 dB steps. The problem I get is that for that to work, I need green trace widths of exactly 0.32 mm at a cut length of 850 mm, where even 0.01 mm wrong will mess up the resistance.
I do not think I can achieve such high precision with my cutter, so I think I will need to settle for larger steps like -3 dB, then the trace width doubles while the avaiable cut length also increases so in practice I can get away with ~ 1 mm green trace widths which is perfectly managable.
And it looks like -3 dB steps will still be plenty good enough based on this paper.
The following graph snippets are stolen from "APPENDIX 3. VARIATION OF CBT SHADING" of the paper linked above.
If we compare perfectly continous shading vs -3 dB steps it is still pretty close.
But we also need to keep in mind that my array will be truncated at -12 dB, which also messes up the response:
Keele himself, in the paper, based on the simulated responses, advocates for -3 dB steps & -12 dB truncation and as far as I can see for good reason, so that is probably what I will use.
Last edited:
I did some tests of the segmented solid backing membranes from this post. I forgot to upload the results to the cloud, so images will have to wait until tomorrow.
But in short, it does not have the 300 hz distortion peak but the overall distortion is higher, so no free lunch. Overall, the normal gapped membranes is more smooth on the top end since there is now an extra 10 khz bump so no magic bullet.
But all of this fiddling with coaxial gave me a thought. My current driver uses 4 rows of holes and 5 rows of mangets, and will probably let me cross in the 350-400 hz region. In theory, if I half the radiating area, then for the same linear movement I should instead be able to cross in the 460-500 hz region.
500 hz would be low enough to be usable for my use case and thus the question would be if there would be a significant improvement in the 5-20k region with such a driver compared to my current wider midtweeter. Kind of makes me want to try to build a segment such that I can compare to my current one and find out 😆
But in short, it does not have the 300 hz distortion peak but the overall distortion is higher, so no free lunch. Overall, the normal gapped membranes is more smooth on the top end since there is now an extra 10 khz bump so no magic bullet.
But all of this fiddling with coaxial gave me a thought. My current driver uses 4 rows of holes and 5 rows of mangets, and will probably let me cross in the 350-400 hz region. In theory, if I half the radiating area, then for the same linear movement I should instead be able to cross in the 460-500 hz region.
500 hz would be low enough to be usable for my use case and thus the question would be if there would be a significant improvement in the 5-20k region with such a driver compared to my current wider midtweeter. Kind of makes me want to try to build a segment such that I can compare to my current one and find out 😆
Last edited:
Keele had a 8 segment variant shading by 3dB, and the max was 12 if i remember correct. i used the same on my shaded ribbon tweeters.
Remember that you need a good octave beyond the passband for a good behaving crossover region...
//
//
halving the surface area. at the same frequency is not the same as halving the xmax (theory it should.. but thats based on a piston , i mean halving the surface area might impact more then halving the xmax since a part of that foil cant hardly move, and that part becomes biger compared to the part that can). so in theory if you talk about air been displaced halving the width that would be instead of 350Hz capable to 440 give or take ? not sure if i do the math right.. i used 350*1.25. i calculated this many this but the other way around... down in Hz haha but it might be quite different since changing x max is different from surface area since the foil is not a piston. and this halve the frequency needs 4 times the displacement is based on a piston.
explaining something simple is aparantly not my thing. so is math (changing up a simple formula haha)
explaining something simple is aparantly not my thing. so is math (changing up a simple formula haha)
Remember that you need a good octave beyond the passband for a good behaving crossover region...
That would be one of the benefits of the current wider 350-400 hz crossover capable variant. Although the planar driver will have nice and uniform dispersion to allow good integration regardless of crossover frequency, the woofers will most definently not.
This previous experiment, which used GRS PT2522 tweeters worked pretty well but failed because the woofers performed great up to 800 hz, but the tweeters needed to be crossed at 1400 hz, so no overlapping region.
If I use the same setup with my diy planar driver, then the wider driver would allow pretty much perfect integration with 1 oct of margin which nice overlap, the half width driver would need to compromise on the overlap.
Best sound seem to materialise when nothing is stressed and work completely effortless - to dimension driver to reach just to the point where it starts to lose the "grip" will not satisfy that... Sd is king.... 😉
//
//
halving the surface area. at the same frequency is not the same as halving the xmax (theory it should.. but thats based on a piston , i mean halving the surface area might impact more then halving the xmax since a part of that foil cant hardly move, and that part becomes biger compared to the part that can). so in theory if you talk about air been displaced halving the width that would be instead of 350Hz capable to 440 give or take ? not sure if i do the math right.. i used 350*1.25. i calculated this many this but the other way around... down in Hz haha but it might be quite different since changing x max is different from surface area since the foil is not a piston. and this halve the frequency needs 4 times the displacement is based on a piston.
explaining something simple is aparantly not my thing. so is math (changing up a simple formula haha)
Yup, I converted Linkwitz old excel spreadsheet into a google spreadsheet and use it as basis for my xmax calculations, here is a snippet of it:
the 4x planar at 350 hz is equal to the 2x planar at 440 hz.
EDIT: In case anyone wants to copy the spreadsheet for your own use, you should be able to view and copy it through this link.
Last edited:
Best sund seem to materialise when nothing is stressed and work completely effortless - to dimension driver to reach just to the point where it starts to lose the "grip" will not satisfy that... Sd is king.... 😉
Yup, there is a beauty in not having to push the limits to make it all integrate well. That is the main benefit that would point me towards continuing with 4 rows.
That is the most beautiful thing with my full range SB65 based CBT, there is no separate midrange or tweeter so no messy integration. The only limit is on the low end but it can still be crossed at 140-170 hz which is good enough. It is not a perfect tweeter by a long shot, hence my current experiments, but good enough and nothing on it is pushed close to the limit.
Last edited:
but does this take into account its not a piston ? for rough ideas i usually use a online calculator and just apply it to what i already have. but with allot of margin 🙂 since my planar is not a piston. so not exactly same rulls apply , but it does if i make them longer , it wont if i make them wider or less wide https://www.baudline.com/erik/bass/xmaxer.htmlYup, I converted Linkwitz old excel spreadsheet into a google spreadsheet and use it as basis for my xmax calculations, here is a snippet of it:
View attachment 1419560
the 4x planar at 350 hz is equal to the 2x planar at 440 hz.
EDIT: In case anyone wants to copy the spreadsheet for your own use, you should be able to view and copy it through this link.
its for woofers in a closed box but that does not matter the rules apply the same. if you compare it with something you already build as long as you go longer or less long. wider and less wide might be different because the foil wont move as much on the sides. what it would do on a piston
but does this take into account its not a piston ?
No, that spreadsheet only calculates volume displacement required for a given SPL into account. Close enough to be useful as long as one does not look too close but rather use it as a guideline for further experimentation rather than an absolute truth.
i agree. i use the calculator that is for pistons. to compare it to a panel i already have. so lets say i made a panel and i wanted to go half that frequency its a a ok tool. since we compare more apples to apples . then again i take some margin especially when panels are less wide it does not work out always 🙂No, that spreadsheet only calculates volume displacement required for a given SPL into account. Close enough to be useful as long as one does not look too close but rather use it as a guideline for further experimentation rather than an absolute truth.
I now have the measurements from the weekend. First EPDM-coax with fabric and yarn in the 2 outer holes:
And then foam suspension coax without any fabric or yarn:
All in all it looks pretty good, but to get more precise high frequency measurements I should probably setup an acoustic timing reference and shorten the window length.
As a conclusion I want to remake the measurements and compare the following setups:
Then based on the winning membrane setup, play around with the alu weights and see if a lighter membrane performs better. Lets play with the idea of with a coaxial membrane, using thinner aluminum foil for the inner two coils which will be driven without a lowpass. Then the top end should be bumped somewhat increasing the off axis response. And as long as both coils are run in series then the difference in resistance should not be a problem, no level matching needed.
And then foam suspension coax without any fabric or yarn:
All in all it looks pretty good, but to get more precise high frequency measurements I should probably setup an acoustic timing reference and shorten the window length.
As a conclusion I want to remake the measurements and compare the following setups:
- Yarn in the outer holes vs coaxially driving (with a 8 khz 1st order lowpass) or both?
- Gapped structural fill of non driven aluminum only around the traces or in larger segments like this?
- Final showdown of EPDM rubber vs the white square foam.
Then based on the winning membrane setup, play around with the alu weights and see if a lighter membrane performs better. Lets play with the idea of with a coaxial membrane, using thinner aluminum foil for the inner two coils which will be driven without a lowpass. Then the top end should be bumped somewhat increasing the off axis response. And as long as both coils are run in series then the difference in resistance should not be a problem, no level matching needed.
I found measurements of a kapton + full solid sheet of 14 um aluminum backing + 50 vs 30 um aluminum coils on the other side of the kapton:
They were measured on different days so don't stare too close, but the 9 khz dip is much more severe on the 50 um version, and as expected the 30 um variant has more top end efficiency.
So running the inner coil in a coaxial setup on say 30 um and the outer on 50 would definently be interesting.
They were measured on different days so don't stare too close, but the 9 khz dip is much more severe on the 50 um version, and as expected the 30 um variant has more top end efficiency.
So running the inner coil in a coaxial setup on say 30 um and the outer on 50 would definently be interesting.
Are you using any gating ?
if you want anything above 10 khz, lighter is the way to go. i think you overdamped it pretty hard. tailored towards low end. while the thing gets larger thenj it is right now.. so it would be tailored towards low end anyways. if you make it 2 times or 3 times as long (not sure how much longer they will be) you will lowere distortion in the low end by a huge amount. and even lose more on the top, of witch you dont have that much right now
if you want anything above 10 khz, lighter is the way to go. i think you overdamped it pretty hard. tailored towards low end. while the thing gets larger thenj it is right now.. so it would be tailored towards low end anyways. if you make it 2 times or 3 times as long (not sure how much longer they will be) you will lowere distortion in the low end by a huge amount. and even lose more on the top, of witch you dont have that much right now
Are you using any gating ?
if you want anything above 10 khz, lighter is the way to go. i think you overdamped it pretty hard. tailored towards low end. while the thing gets larger thenj it is right now.. so it would be tailored towards low end anyways. if you make it 2 times or 3 times as long (not sure how much longer they will be) you will lowere distortion in the low end by a huge amount. and even lose more on the top, of witch you dont have that much right now
+- 10 ms. But the measurements are done indoors where, when turned around the 90 degree measurements are closer to a wall that can reflect. Ideally I'd measure outside, then the measurements are much cleaner and I can resolve the low level 90 degree levels. It is a bit too cold out now though but I'll have to see if adding an acoustical reference that is independent of the rotating driver & shorter windows can clean things up.
Also worth noting that those measurements are EQed with a - 12 dB high shelf @ 0.5-1 khz, so it is not quite as grim as it looks. The sharp downturn above 10 khz is something I hope to at least mitigate with lighter membranes, we will have to wait and see if they work as well in a coaxial setup as they do when the whole membrane is the same. Since with a too light membrane then efficiency suffers, but if I can have more low end efficiency by having thicker alu on the outer coils & lighter on the inner then the end efficiency should be somewhere in the middle, while at the same time helping to improve the top end since the weight difference will act as a lowpass filter. Or at least it will, in theory, do that 🙂
Last edited:
yeah thats what i was thinking, if they get bigger it will look allot better in the lows. witch already looks pretty nice must say !Also worth noting that those measurements are EQed with a - 12 dB high shelf @ 0.5-1 khz, so it is not quite as grim as it looks.
Some progress and some stumbles:
Using an acoustical reference worked great! In the long run I might do a more fancy setup and add a small speaker under the mic mounted on the stand but for now this was good enough.
I made 2 identical copies of a new membrane, 14 um on the non driven aluminum fill, 30 um driven aluminum for the inner coil and 50 um aluminum for the outer coil. The idea was to explore having reduced weight in the middle thus enhancing the coaixal response without having to adjust the signal.
And it kinda worked! The first measurement is a normal gapped membrane from this previous post.
New multi weight membrane
It is less smooth, when it drops off it is sharper but there is no denying that the drop off is higher. 9 khz on the old conventional membrane and 12 khz on this new membrane.
This can be extended even further with a 8 khz first order lowpass on the outer coil. Not too shabby and interestingly enough it makes the transition more smooth too.
And the reason I made 2 membranes is because I also wanted to test with EPDM rubber suspension, because while the soft white foam is nice and all it is very annoying to attach to the steel. I tried my scissor cutting jigs but even with them, I have to reject half of my foam strips and it is so slow. Because of those problems, I want to continue investigating EPDM and try to tailor make membranes to work around the problems.
Something probably went wrong when mounting or this type of membrane is just not suitable for the EPDM suspension, here is the polar plot for the new membrane with EPDM:
My preliminary findings suggest that the EPDM works best when the membrane is pretty stiff, so it works great with a solid 14 um backing. But now since I want to try coaxial driving then one solution that seems to work well is to use segmented backing like in this post.
So the next thing to try is to make a segmented backing membrane like in that post but to vary the coil thickness, and maybe also the backing thickness. Say I use 16-20 um alu backing under the outer coils, 12 um alu on the inner, and 25 or 30 um alu on the inner coil and 50 um alu on the outer coil. That might be the project for next week.
Using an acoustical reference worked great! In the long run I might do a more fancy setup and add a small speaker under the mic mounted on the stand but for now this was good enough.
I made 2 identical copies of a new membrane, 14 um on the non driven aluminum fill, 30 um driven aluminum for the inner coil and 50 um aluminum for the outer coil. The idea was to explore having reduced weight in the middle thus enhancing the coaixal response without having to adjust the signal.
And it kinda worked! The first measurement is a normal gapped membrane from this previous post.
New multi weight membrane
It is less smooth, when it drops off it is sharper but there is no denying that the drop off is higher. 9 khz on the old conventional membrane and 12 khz on this new membrane.
This can be extended even further with a 8 khz first order lowpass on the outer coil. Not too shabby and interestingly enough it makes the transition more smooth too.
And the reason I made 2 membranes is because I also wanted to test with EPDM rubber suspension, because while the soft white foam is nice and all it is very annoying to attach to the steel. I tried my scissor cutting jigs but even with them, I have to reject half of my foam strips and it is so slow. Because of those problems, I want to continue investigating EPDM and try to tailor make membranes to work around the problems.
Something probably went wrong when mounting or this type of membrane is just not suitable for the EPDM suspension, here is the polar plot for the new membrane with EPDM:
My preliminary findings suggest that the EPDM works best when the membrane is pretty stiff, so it works great with a solid 14 um backing. But now since I want to try coaxial driving then one solution that seems to work well is to use segmented backing like in this post.
So the next thing to try is to make a segmented backing membrane like in that post but to vary the coil thickness, and maybe also the backing thickness. Say I use 16-20 um alu backing under the outer coils, 12 um alu on the inner, and 25 or 30 um alu on the inner coil and 50 um alu on the outer coil. That might be the project for next week.
- Home
- Loudspeakers
- Planars & Exotics
- DIY midtweeter planar, physically curved and shaded to be used in a dipole CBT