There is a tiny dusting of ReMount sticking to the actual membrane left over from the cutting mat. So technically, I don't need to fold over the top and bottom edges.
And 160 might work, this is probably easier if your oven has a hot air fan to help circulation but you can slightly open / close the oven door to regulate the temperature. It is what I did to regulate it to 150 +- 1 degree instead of 150 +- 10 degrees.
The shaded membrane is starting to take shape:
This is a much shorter membrane than the real one, 1/6 of the total lenght but otherwise the proportion should be correct. It is shorter to make it easier to view because otherwise it is too long to fit in my monitor 😆
The red numbers are, in theory, the amount of current that, in theory, should flow in each section.
The idea behind them is to split the trace into a parallel resistor to attenuate the signal. But unlike a parallel resistor, since the parallel trace has current flowing in the same direction as the rest of the coil, it will still generate useful force. Because of that, the overall membrane will be shaded, and loose max power output, but the per watt efficiency losses should be very small.
What the code generating the above membrane is trying to do is to model the resistance of the traces incrementally. So in this example, my model estimates that when measuring form red to red, then the resistance of the blue trace should be exactly the same as the resistance of the green trace.
So right now the target resistance ration is 0.5, 0.5 but I can set other ratios if I want, it is all parametric.
In practice, what it does is that I run multiple calculations. The first calculation is where the initial widths are the same, such that the trace width of the blue and the initial green part is the same. Then I look at the the comparative resistances and look at the error. If the green resistance is too big, then I halve ratio for the next iteration and try with 0.25, 0.75 widths. This will also be wrong, but for each step I will get a little bit closer.
This method of course already has a name, it is binary search where fore each iteration, the delta error relative to our target is halved. I run 20 iterations which gets very close to the ideal width.
But this is all very nice in theory, I still do not know how well my theroretical model actually matches with reality. So before I start making the real membranes I need to cut some smaller tester pieces and then progressively cut the membrane into parts and measure the resistance of each part with my 1 mOhm multimeter. If all my calculations are correct (which they probably aren't) then the theoretical resistance will match the actual resistances, but if not then I can use it to find where the errors are and fix them and then try again.
This is a much shorter membrane than the real one, 1/6 of the total lenght but otherwise the proportion should be correct. It is shorter to make it easier to view because otherwise it is too long to fit in my monitor 😆
The red numbers are, in theory, the amount of current that, in theory, should flow in each section.
The idea behind them is to split the trace into a parallel resistor to attenuate the signal. But unlike a parallel resistor, since the parallel trace has current flowing in the same direction as the rest of the coil, it will still generate useful force. Because of that, the overall membrane will be shaded, and loose max power output, but the per watt efficiency losses should be very small.
What the code generating the above membrane is trying to do is to model the resistance of the traces incrementally. So in this example, my model estimates that when measuring form red to red, then the resistance of the blue trace should be exactly the same as the resistance of the green trace.
So right now the target resistance ration is 0.5, 0.5 but I can set other ratios if I want, it is all parametric.
In practice, what it does is that I run multiple calculations. The first calculation is where the initial widths are the same, such that the trace width of the blue and the initial green part is the same. Then I look at the the comparative resistances and look at the error. If the green resistance is too big, then I halve ratio for the next iteration and try with 0.25, 0.75 widths. This will also be wrong, but for each step I will get a little bit closer.
This method of course already has a name, it is binary search where fore each iteration, the delta error relative to our target is halved. I run 20 iterations which gets very close to the ideal width.
But this is all very nice in theory, I still do not know how well my theroretical model actually matches with reality. So before I start making the real membranes I need to cut some smaller tester pieces and then progressively cut the membrane into parts and measure the resistance of each part with my 1 mOhm multimeter. If all my calculations are correct (which they probably aren't) then the theoretical resistance will match the actual resistances, but if not then I can use it to find where the errors are and fix them and then try again.
Last edited:
I cut a test membrane, fetched my 1 mOhm precision ohmmeter and then started measuring and progressively cutting away the membrane to see if the expected resistances match my theory.
The goal is that, at every junction where two traces are in parallel, that the resistance of each side should be the same. Hence half the current should flow to each trace. I can adjust the ratio with a parameter in my python code but for now it is 50% for each side.
First I measured the resistance with both traces in parallel, then I cut into the membrane and measured the inner and outer trace individually.
And here is the measurements inserted into a diagram of the membrane with error percentages. And it looks to be pretty spot on so I am very pleased!
It makes sense that the error is larger at the teeny tiny resistances, but at such a high attenuation it isn't as important and it is still pretty close. And the errors seem pretty random, not biased towards the inner or outer so I can probably leave it as it is. All in all this is definently usable! But with these errors in mind I think it is a good pragmatic choice to stay at -3 dB steps instead of going down to -1.5 dB steps.
I will probably redo this with a full length membrane and see if the errors are still low enough or if they need adjustments. My guess is that the errors will be lower than this small scale experiment but I still want to know 🙂.
So, shading network is pretty much done for now. Next is to finalize the attachment on the bottom where I will glue the membrane to a PCB and solder the terminals. Then it is time to start scaling up to full size!
The goal is that, at every junction where two traces are in parallel, that the resistance of each side should be the same. Hence half the current should flow to each trace. I can adjust the ratio with a parameter in my python code but for now it is 50% for each side.
First I measured the resistance with both traces in parallel, then I cut into the membrane and measured the inner and outer trace individually.
And here is the measurements inserted into a diagram of the membrane with error percentages. And it looks to be pretty spot on so I am very pleased!
It makes sense that the error is larger at the teeny tiny resistances, but at such a high attenuation it isn't as important and it is still pretty close. And the errors seem pretty random, not biased towards the inner or outer so I can probably leave it as it is. All in all this is definently usable! But with these errors in mind I think it is a good pragmatic choice to stay at -3 dB steps instead of going down to -1.5 dB steps.
I will probably redo this with a full length membrane and see if the errors are still low enough or if they need adjustments. My guess is that the errors will be lower than this small scale experiment but I still want to know 🙂.
So, shading network is pretty much done for now. Next is to finalize the attachment on the bottom where I will glue the membrane to a PCB and solder the terminals. Then it is time to start scaling up to full size!
Last edited:
I think I am pretty much done with the membrane, at least for now.
I plan to cut it in 2 layers, which is neccesary with the combination of coaxially driven and shaded, they overlap if done on a single side.
First is the top layer
Next is the bottom layer
Here is both at the same time
And the traces on the left that stick out of the rest I plan to cut and fold back onto itself. This the rear will fold to the front, and the front will fold to the rear:
After folding, and with a PCB it will look something like this
The membrane will be glued to the top of the PCB, but folder such the red pads will wrap the fold and connect to the red pads on the PCB.
Then the speaker wires will connect to the rear of the PCB. Thus there is much less risk of future ripped traces since I have a conventional PCB as a middle layer in between. And since it is a normal copper PCB it is much easier to solder.
I haven't yet decided if I should solder wires to the back of the PCB or add a terminal. I am leaning towards wires since it will be inside the speaker anyway.
Next up is to finalize the PCB so I can order some. And then finalize the 3d printed parts that for example the PCB will be attached to. And also to finalize the other 3d printed parts of the speaker. And then it is time for full length testing!
I plan to cut it in 2 layers, which is neccesary with the combination of coaxially driven and shaded, they overlap if done on a single side.
First is the top layer
Next is the bottom layer
Here is both at the same time
And the traces on the left that stick out of the rest I plan to cut and fold back onto itself. This the rear will fold to the front, and the front will fold to the rear:
After folding, and with a PCB it will look something like this
The membrane will be glued to the top of the PCB, but folder such the red pads will wrap the fold and connect to the red pads on the PCB.
Then the speaker wires will connect to the rear of the PCB. Thus there is much less risk of future ripped traces since I have a conventional PCB as a middle layer in between. And since it is a normal copper PCB it is much easier to solder.
I haven't yet decided if I should solder wires to the back of the PCB or add a terminal. I am leaning towards wires since it will be inside the speaker anyway.
Next up is to finalize the PCB so I can order some. And then finalize the 3d printed parts that for example the PCB will be attached to. And also to finalize the other 3d printed parts of the speaker. And then it is time for full length testing!
I have some experience on measuring low resistances. Down to u Ohm. Not easy!! But this is how it is done:
Connect a current source @ 1.0000 Amps, Start with the Current output set to 0A, Measure uV with your probes with zero current. Null the uV voltmeter when the probes is attached (This is the important part, since you create a battery with the different plating of the probes and the measured object, there is also a thermo electric effect). Start the current, and read the uV voltage directly before current has heated up the resistor.
Connect a current source @ 1.0000 Amps, Start with the Current output set to 0A, Measure uV with your probes with zero current. Null the uV voltmeter when the probes is attached (This is the important part, since you create a battery with the different plating of the probes and the measured object, there is also a thermo electric effect). Start the current, and read the uV voltage directly before current has heated up the resistor.
Cool! Although in my case the 4-wire ohmmeter I have seems to be precise enough. It is rated at 1 mOhm precision and based on how consistent enough it measures I believe that.
On the topic on my project, right now it is the calm before the storm (full scale testing)...
I have a 700 mm wide x 35 m roll of 23 um mylar on the way, and I am also waiting for the PCBs. All packages will arrive by May 9:th so by then I plan to start cutting full length membranes.
In preparation I am fixing some mechanical oversights in my current 3d printed parts.
One of the issues I have fixed is that the lengthwise sections can shift ~ 0.5 mm to the side. Since I want the membrane to be exactly in the middle, I wanted to fix this:
My fix was to add some indexing knobs:
Works great so far.
Also while adding the knobs I finalized the top and bottom parts, they will be 25 mm high on the top and bottom and the PCB will be 12 mm wide and 70 mm long.
The next problem I have fixed is that in my final speaker, the front and rear steel plates will overlap in a staggered way. Thus when all parts are screwed together, it will be very strong but when only the rear is assembled, it is pretty weak and is held together by threaded inserts in 3d printed plastic parts.
To solve that I have made an arc I can place the bottom plates on with the screws facing upwards such that I can first screw all the bottom plates to each other, then mount the membrane to the edges and then progressively add the front parts one by one.
It is not a beauty, but it will do the job just fine.
And while at it, I wanted to validate my corrugated vs uncorrugated lengths such that the final membrane length is correct.
So to do that I took one of my baked membranes and rolled it flat, only to realize that it instantly re-corrugated itself. The baking truly works!
Corrugated, the length was 27 cm. Stretched flat and rolled with a roller, the length was 30.5 cm. And then afterwards, it re-corrugated itself to a length of 28.5 cm!
So yeah, baking the mylar is amazing at making it hold corrugations!
In fact, the corrugations hold so well that when I previously used the EPDM rubber suspension, which previously crushed the corrugations flat under the rubber, the baked 23 um mylar was strong enough to resist this crushing force and kept the corrugations. It also reverted to the same shape after removing the EPDM rubber where the kapton + alu membranes had permamently crushed edges.
I will still use the soft foam though, since it measures better, but it is cool that the baked mylar is so elastic while also holding the corrugations so well!
So to do that I took one of my baked membranes and rolled it flat, only to realize that it instantly re-corrugated itself. The baking truly works!
Corrugated, the length was 27 cm. Stretched flat and rolled with a roller, the length was 30.5 cm. And then afterwards, it re-corrugated itself to a length of 28.5 cm!
So yeah, baking the mylar is amazing at making it hold corrugations!
In fact, the corrugations hold so well that when I previously used the EPDM rubber suspension, which previously crushed the corrugations flat under the rubber, the baked 23 um mylar was strong enough to resist this crushing force and kept the corrugations. It also reverted to the same shape after removing the EPDM rubber where the kapton + alu membranes had permamently crushed edges.
I will still use the soft foam though, since it measures better, but it is cool that the baked mylar is so elastic while also holding the corrugations so well!
I have both had some progress and some setbacks:
A success i that I made a more simple reference signal generator. Previously I had a separate amp with a speaker driver placed on the floor which made the setup more complicated. Now I have built a mini spaker from headphone drivers which on the bottom has a 3/8 threaded insert + I have ordered a clamp mount such that I can mount it below the measurement microphone.
And it seems to work great!
Now the actual progress on the planar driver, which is not perfect...
I have had huge problems connecting the steel plates together, because naively thought that it would be good enough to mount it into the 3d printed plastic spacers on the edge. That is nowhere near strong enough, and I have problems that they shift and don't align properly.
The current workaround is that I enlarged the edge holes to threaded M4 which let me use the connecting pieces to hold it together. But this is not a perfect solution since they are too small to curve easily so the holes don't alignt perfectly so I had to enlargen them somewhat which makes the alignment problem worse. I think I can solve it, but the solution is finicky and not as simple as I would prefer.
I have realized the following problems with the current driver:
Because of all those problems, and since I am a bit of a perfectionist and not in a hurry, I am considering scrapping the current iteration and making a new one with all those problems fixed.
And while I am at it, since coaxial driving worked so well, it would be fun to investigate it further. Say I test a variant where I have the current setup with 4 rows, the outer rows are lowpassed at 7 khz and the inner run full range, but I add 2 extra rows on each side so I get 8 rows in total. It would have a lot more low end extension, would probably let me cross in the 200-300 hz range. If I get down to 200 hz then that would be low enough that I could skip the midwoofers on the speaker and just have subs + this driver.
A success i that I made a more simple reference signal generator. Previously I had a separate amp with a speaker driver placed on the floor which made the setup more complicated. Now I have built a mini spaker from headphone drivers which on the bottom has a 3/8 threaded insert + I have ordered a clamp mount such that I can mount it below the measurement microphone.
And it seems to work great!
Now the actual progress on the planar driver, which is not perfect...
I have had huge problems connecting the steel plates together, because naively thought that it would be good enough to mount it into the 3d printed plastic spacers on the edge. That is nowhere near strong enough, and I have problems that they shift and don't align properly.
The current workaround is that I enlarged the edge holes to threaded M4 which let me use the connecting pieces to hold it together. But this is not a perfect solution since they are too small to curve easily so the holes don't alignt perfectly so I had to enlargen them somewhat which makes the alignment problem worse. I think I can solve it, but the solution is finicky and not as simple as I would prefer.
I have realized the following problems with the current driver:
- The rear plates have an alignment problem, as I described.
There is also an alignment problem to the edges where I mount the membrane, they also shift a bit.
The perfect solution would be to have a single steel plate for the rear. And then extend it slightly such that the edge pieces where I clamp the top and bottom membrane pieces all connect to a single plate. Then there would be no alignment problems at all. It would be a lot more annoying to glue the magnets though 😆 - I forgot to sandpaper the neo magnets before gluing so the bond is not as strong as I would like. I have had to re glue a few magnets here and there if I knock them too hard or especially when I accidentally let 2 pieces smash together without a spacer then lots of magnets got loose when I pulled them apart.
- The driver curve is not perfect because each 1/6 segment is curved but the supports that hold them together is not.
- I should have added indexing dowels to the steel plates + the 3d printed spacers. Right now the screws hold them in place but it would be better if the plastic was shaped in such a way that I don't have to rely on the screws for sideways indexing. Although if I continue with the current design, I do have more screw holes than I need for strength so I could repurpose some to use as dowel pins.
- I found an error in one of my mathematical functions that calculates, given a circle radius + a distance along the edge of the circle, how many degrees is that?
I had functions both to get L, given R, and R given L.
If I have the angle, it is easy, I just calculate 2 * radius * math.pi * (degrees / 360). But the problem was with the reverse function, for that I used math.degrees(math.acos(1 - ((distance * distance) / (2 * radius * radius)))) and I am not certain if my math is wrong or if the precision of the library is not good enough but it works reasonably good at small lengths. Then the error is less than 0.1% but at larger distances, like along the whole planar radius then the error gets as large as 5%!
Here is an image of a model with the previously incorrectly calculated degrees. Notice that the yellow supports are not at equal lengths, the top spacing is too large.
Here is with the function corrected:
Because of all those problems, and since I am a bit of a perfectionist and not in a hurry, I am considering scrapping the current iteration and making a new one with all those problems fixed.
And while I am at it, since coaxial driving worked so well, it would be fun to investigate it further. Say I test a variant where I have the current setup with 4 rows, the outer rows are lowpassed at 7 khz and the inner run full range, but I add 2 extra rows on each side so I get 8 rows in total. It would have a lot more low end extension, would probably let me cross in the 200-300 hz range. If I get down to 200 hz then that would be low enough that I could skip the midwoofers on the speaker and just have subs + this driver.
Hi OllBoll ... Might I suggest that you present this:
to either ChatGPT or Grok 3. In my experience these AIs are able to find solutions to many challenges - not least when used in their "Deep Research" mode (chatgpt).
Cheers, Jesper
to either ChatGPT or Grok 3. In my experience these AIs are able to find solutions to many challenges - not least when used in their "Deep Research" mode (chatgpt).
Cheers, Jesper
Yup, it will be 🙂
I did some more calculations and I think 9 rows of magnets are overkill. Even with 9, the expected volume displacement is still not enough to cross at 200 hz so I still need the woofers. And in that case i might as well try half way and go with 7 rows of magnets i.e. 6 rows of traces.
With 6 rows of coils I could also get away with wider suspension, from 6 mm wide now to 10 mm which would let me use use the dust sealing strips without cutting them, which would be a lot easier to manifacture.
Also, with 6 rows I could add coaxial drive of the membrane without having to run alu traces both on the front and the rear, so I could cut the whole membrane in one go which simplifies it by a lot when going big.
And I think that with 6 rows, then crossing at 300 hz should not be a problem. And such a low crossover like 300 hz would let me place the woofers behind the planar in a force cancelling setup where with a 450-500 hz crossover, I have to place them at each side of the planar in a MTM setup to get the best performance.
I am in the process of buying a huge stockpile of 12x5x3 magnets from my previous supplier so I have enough magnets to do some 6 magnet width experiments. After I have ordered the magnets I might order some more steel sheets.
And at the same time my 3d printer broke
. But I have a started a support ticket such that I can ressurect it.When printing I heard a noise, and then the diagonal axis is now messed up and semi randomly shifts during the print. It is a Core XY printer and the problem is just with one diagonal axis so my guess is that one or more of of the following parts is broken and need to be replaced:
- The belt controlling the axis
- The motor for the axis
- the controller for the motor
Some progress on new models. The significant changes are:
While I do want to try 6 trace rows, 4 might still be better so as my models are parametric I will also have some tester plates for 4 rows made. But I already know 4 rows work great, so in this case I would mostly test the mechanical fit, if I want to change something before scaling upp to full size again.
The main question of 4 vs 6 trace rows is if I can add more width without compromising on the top end response by driving the membrane coaxially.
And an update on my 3d printer, I believe I have found the problem. I think the problem is that the electric motor handling the problematic axis is broken. When the printer is turned off, it should present a very low resistance and allow me to move the printer head around but the for the problematic axis it takes a lot of force to move it around. And when moving it around it acts as a generator and makes a small LED flash bright. So definently not expected behaviour 🙂.
If I am right it is not an expensive part to replace, it costs just € 30 and is stocked within the EU so should ship fast enough. They also have an excellent guide of how to replace it which is nice. But I am still waiting for an answer on my support ticket so I will wait until that before ordering spare parts.
- The bottom steel plate is now in one piece, and covers the whole arc of the planar, including the edges where I clamp the membrane. Thus eliminating side alignment problems. Being in one piece will also make it very easy to get a smooth curve across the whole planar. The segmented variant was slightly more flat between the segments because the connecting steel plates are too small to fit in my roller.
- Cut semicircles on the edge of the steel & corresponding filled area on the yellow 3d printed plastic parts. The goal is to have them act as dowel pins and index such that the pieces align exactly as they should.
- Swapped some even holes in the 3d printed parts for extra dowel pins. This might be overkill though and if I keep this depends on how cleanly it prints and if it is really neccesary.
- I am also considering fastening the yellow 3d printed edge parts to the steel with some superglue. Just enough at the dowel pins to keep it in place and make it easier to asseble. My previous solution had threaded inserts but they were prone to getting loose and would then rattle, glue is easier and more reliable. And since it is plastic to steel, if I want to rip it off I can still do that.
- This variant has 7 magnet rows and 6 trace rows. The foam suspension gap is also 10 mm wide which should allow me to use the 9 mm wide dust sealing strip /orange in the model) without cutting it.
While I do want to try 6 trace rows, 4 might still be better so as my models are parametric I will also have some tester plates for 4 rows made. But I already know 4 rows work great, so in this case I would mostly test the mechanical fit, if I want to change something before scaling upp to full size again.
The main question of 4 vs 6 trace rows is if I can add more width without compromising on the top end response by driving the membrane coaxially.
And an update on my 3d printer, I believe I have found the problem. I think the problem is that the electric motor handling the problematic axis is broken. When the printer is turned off, it should present a very low resistance and allow me to move the printer head around but the for the problematic axis it takes a lot of force to move it around. And when moving it around it acts as a generator and makes a small LED flash bright. So definently not expected behaviour 🙂.
If I am right it is not an expensive part to replace, it costs just € 30 and is stocked within the EU so should ship fast enough. They also have an excellent guide of how to replace it which is nice. But I am still waiting for an answer on my support ticket so I will wait until that before ordering spare parts.
Last edited:
I just ordered laser cut steel parts to try out the latest models, both in 4 and 6 trace widths so I can see how well it works.
And since the 3d printer spare parts were so cheap I decided to stock up on a motor in case I am right in addition to some more common spare parts that get worn. Even if I am wrong and the motor is fine, it doesn't hurt to have a stockpile of parts because in the long run stuff will break, and when that happens I now just might have a replacement part close at hand.
And I got another idea. Since the the non sandpapered magnets are pretty easy to knock off, I might as well plunder some of my previous pieces for magnets and reuse them for the new test drivers. Then I can save my stockpile of new and clean magnets for when I build the full scale driver.
And since the 3d printer spare parts were so cheap I decided to stock up on a motor in case I am right in addition to some more common spare parts that get worn. Even if I am wrong and the motor is fine, it doesn't hurt to have a stockpile of parts because in the long run stuff will break, and when that happens I now just might have a replacement part close at hand.
And I got another idea. Since the the non sandpapered magnets are pretty easy to knock off, I might as well plunder some of my previous pieces for magnets and reuse them for the new test drivers. Then I can save my stockpile of new and clean magnets for when I build the full scale driver.
Some questions is not that polite. In social forums you never ask about weight or age. In hobby forums.... time and money is... sensitive.
In some cases like this one.... the goal of the project is set quite high. Or lets say "cost is not an issue" kind of. How good can it possible be? Where is the limits on how good this principle can perform? I have not seen anything like it... this is a combination of several technologies that is taken to it´s extremes.
Now I also want to know how much this will cost ..!! 😎
In some cases like this one.... the goal of the project is set quite high. Or lets say "cost is not an issue" kind of. How good can it possible be? Where is the limits on how good this principle can perform? I have not seen anything like it... this is a combination of several technologies that is taken to it´s extremes.
Now I also want to know how much this will cost ..!! 😎
Currently probably somewhere around €2500 until now for the DIY planar so far.
Itemized it is roughly:
- €1000 for laser cut steel
- €500 for the magnets
- €300 for many different thicknesses of aluminum foil
- €200 for mylar of different thicknesses
- €300 for 2x rolls of kapton. Kapton is expensive! 💸
- €200 other things like spray adhesive, suspension foam, PETG filament and so on
The goal is absolutely to make the best possible. The reason i started was because I was disillusioned with the performance on the available planars I can buy. I looked at a line of GRS Neo8 clones + a line of GRS Neo3 clones. They would cost roughly €3300 just in drivers and even then I tried it in small scale and it wasn't even good. In fact it measures way worse than my full range CBT. I can affoard that price, but I I am going to spend that amount of money it better be great. And since I cant buy it I guess I have to make it myself 😆
But a big contributor is how approachable my 3d printer has made this type of manifacturing in combination with ordering laser cut steel and wood parts. I can build stuff pretty easily and reliably that I could only dream of 10 years ago.