Hardware?

[jstma]

I like what you're putting together in this repository.

I'm considering designing an open hardware design similar to eDRUMin. It would be a combination of:

1. signal conditioning for the piezo inputs
2. a socket for some cpu som (like esp32)

The idea would be piezos in, usb midi out. Maybe even ble midi out.

A reasonably simple main board that could be made at OSHPark and then hand soldered. Designed in KiCAD.

Before I start on a design, I was wondering if you've got plans for hardware.

[corrados]

Sounds very interesting. Yes, I was thinking of creating a PCB and I already installed KiCad but I have no experience with that tool and I have so many things to improve in software that I do not see my own PCB being created sometime soon. What trigger software did you have in mind when you started thinking about your project?

[gustavsl]

I can also contribute designing a PCB in KiCad. I think it's crucial to decide which module will go in the hardware (esp32 or teensy?).

[corrados]

Great :-). Yes, before designing a PCB some decisions have to be made and some test have to be done. Right now I am concentrating on the Teensy 4.0.

[corrados]

As you might have already noticed, I have done some schematics created in KiCad: https://github.com/corrados/edrumulus/tree/main/hardware/prototype2

Are you experienced with KiCad? I have never worked with that tool and I am just trying out things here. Could you please take a look at my design and review it?

[corrados]

Regarding ESP32 vs Teensy: I think it makes sense to provide schematics and maybe even a PCB layout for both options.

[jstma]

Sorry what do you mean by trigger software? The hardware would produce midi out over usb and maybe ble.

I think the processor/adc situation is critical to figure out first, obviously, since that will impact the piezo conditioning circuitry necessary.
I personally want at least 16 piezo inputs and that's assuming I can make positional sensing work on my L80 cymbals. The eDRUMin supports 20 piezo inputs and 2 opto/fsr inputs. So I would aim to do the same. And I hate to keep bringing up eDRUMin, but:

eDRUMin uses a digitally controlled analog gain system, allowing it to support
an extremely wide range of signals while maintaining optimal sampling resolution
throughout its entire range.

Sampling with 1024 levels of precision at 10K samples per second, eDRUMin
provides natural, smooth, and nuanced triggering even when applying
velocity curves.

So he's using 10-bit adc at 10ksps with digitally controlled analog gain. I'm not sure that's required since 12-bit adc will give 12dB more range. But that comes down to the hardware since some adc's have a pga in front.

And, of course, latency. Needs to be low.

The other thing I would like to see is a wifi interface for configuring the modules (with client side rendering). Then it will work anywhere. No software specific to the hardware will need to be installed anywhere. That's the dream anyhow.

I have doubts that the ESP32 is capable of processing all of that data and providing a web interface. But that depends on how positional sensing is implemented. When I played with something similar I was using cross correlation which was heavy and it didn't work all that well.

So the first thing I'm trying to decide on is dedicated muxed adcs or a microcontroller (like the STM32) that has 20+ adc available.

And if you're not familiar with eDRUMin:
https://www.youtube.com/watch?v=v9_w32TWKZ0

It's well thought out hardware and software. If it was only open source ;p

[jstma]

For hihat control we need to handle two cases in hardware:

• 2 terminal device: potentiometer, fsr, or switch
• 3 terminal device: optical, hall effect

For 2 terminal devices, they form the other half of a resistor divider.
For 3 terminal devices, we send them power and they return an active signal.

I had intended to keep the hihat on a separate TRS jack. This way that adc input could still be used for single zone piezo or both 2 and 3 terminal hihat devices. I haven't given it a lot of thought yet, but I was hoping that by using a mono or stereo cable we could get power in the correct place when the mono cable shorts the ring and tip together. Now that I think of it, that would be a bad idea for any kind of switch input.

With a usb design 5V is always available and could be used to power the device. Or if we get fancy we can provide for either 3.3V or 5V power to the jack. If the hihat device can output more than 3.3V an external resistor divider could be used. Or if we get fancy we can use a pot (analog or digital) on the input to do the same job.

More thought will be required. Simpler is better, but I want to support hall effect and optical.

[jstma]

I roughly drew out all the configurations I could think of:

I came to the conclusion that a pot, digitial or otherwise, on the device should handle all of the situations. Won't work for piezos though. But that's why the hihat input is on a separate jack.

The question is how other manufacturers have wired their hihat devices. Do they always use tip and shield? Would it be safe to put V+ on ring?

We definitely need to prevent ring from shorting to shield with a mono cable. Not sure how just yet. We could just make it switchable in software (on/off) and if a mono cable is shorted the usb host port should shut things down. However, if powered by some other device we don't have that protection.

A simple solution might be a PTC thermistor (resettable fuse), but they may be hard to source.
Could put a skinny trace on the board to act as a fuse which is free. One time use though.
Could also use a small 20 ohm current limiting resistor (5V 0.250A), but it would need to be sized for 1.25W. That's a big resistor.

Actually ptc fuses are listed on JLC. Not as basic parts, but there's a large selection of 0805 with reasonable current limits. 150mA should be fine to protect things, but the voltage to the jack still needs to be switched on/off. This is easy to do with a p-channel mosfet with 3.3V to the jack. But with 5V to the jack we'll need an additional n-channel mosfet to turn it off with 3.3V logic. I dislike how many extra parts this adds. It's 2 different parts only used once.

3.3V should work with most optocouplers. Hall effect sensors need 5V probably.

Have I overlooked anything?

[corrados]

@ignotus666 Thank you for sharing the pictures. You drum set looks very professional. Good job. After looking at your pictures, started a new thread where people can share their motivation for building edrums: #4 (comment). If you like, you could share yours there.

After years of endless tinkering the kit has remained stable for the last 5 years or so and plays quite well. I use it regularly on Jamulus ;-).

Since my Edrumulus prototype is now getting better and better (of course, there are still a lot of things to improve...), I am also using Jamulus with that prototype from time to time. It is a good way of testing the performance of Edrumulus when playing together with other musicians. E.g., I found out about a dynamic issue just by playing with others. If I played alone, it was not so obvious that there is an issue. So, Jamulus is very useful ;-).

[SpintroniK]

I roughly drew out all the configurations I could think of:

I came to the conclusion that a pot, digitial or otherwise, on the device should handle all of the situations. Won't work for piezos though. But that's why the hihat input is on a separate jack.

The question is how other manufacturers have wired their hihat devices. Do they always use tip and shield? Would it be safe to put V+ on ring?

I checked on my TD-4, TD-6, and TD-17, they only use tip and sleeve. I took apart an old FD-8 a while ago. I just checked the PCB, the jack socket actually has a ring connector, but it's not connected. So it would seem that's OK for the Roland e-drums.

We definitely need to prevent ring from shorting to shield with a mono cable. Not sure how just yet. We could just make it switchable in software (on/off) and if a mono cable is shorted the usb host port should shut things down. However, if powered by some other device we don't have that protection.

A simple solution might be a PTC thermistor (resettable fuse), but they may be hard to source. Could put a skinny trace on the board to act as a fuse which is free. One time use though. Could also use a small 20 ohm current limiting resistor (5V 0.250A), but it would need to be sized for 1.25W. That's a big resistor.

Actually ptc fuses are listed on JLC. Not as basic parts, but there's a large selection of 0805 with reasonable current limits. 150mA should be fine to protect things, but the voltage to the jack still needs to be switched on/off. This is easy to do with a p-channel mosfet with 3.3V to the jack. But with 5V to the jack we'll need an additional n-channel mosfet to turn it off with 3.3V logic. I dislike how many extra parts this adds. It's 2 different parts only used once.

3.3V should work with most optocouplers. Hall effect sensors need 5V probably.

Have I overlooked anything?

If you're going to make it a product, I think it's best to have a good level of protection.
I mean, it'll sure add some diodes to every input, but that's better than having "fragile" device.
I had a look inside the TD-6 module yesterday, there are inductors and diodes everywhere, everything is very well protected.

[jstma]

The question is how other manufacturers have wired their hihat devices. Do they always use tip and shield? Would it be safe to put V+ on ring?

I checked on my TD-4, TD-6, and TD-17, they only use tip and sleeve. I took apart an old FD-8 a while ago. I just checked the PCB, the jack socket actually has a ring connector, but it's not connected. So it would seem that's OK for the Roland e-drums.

That's very helpful, thanks!

If you're going to make it a product, I think it's best to have a good level of protection. I mean, it'll sure add some diodes to every input, but that's better than having "fragile" device. I had a look inside the TD-6 module yesterday, there are inductors and diodes everywhere, everything is very well protected.

Well not exactly a product, not on the same level as anything that's sold. It would have none of those markings: CSA/UL, FCC/CE, etc.

However, I will do my best to make sure that it's not a fire hazard and that it's not fragile.

But to make a design robust means adding cost. A good example is ESD (static discharge). Most electronic components are usually rated to 4kV. That's not enough for a robust product, but it might be okay for this. Look at how many people are handling esp32 or teensy boards and I haven't heard of rampant ESD damage. I will probably add the necessary parts to the schematic, but may not populate them.

Inductors are probably there for EMC reasons (either radiated or conducted emissions). Without taking the finished device to a lab for testing ($$$$), those extra components only drive up the cost. So I'm choosing to omit them until I know there is a problem. Ignorance is bliss, right? ;p

[jstma]

The choice of enclosure is a tough one due to the size of these jacks. But at the same time the number of jacks is a problem too. And supporting 16 channels means we can't use fewer jacks. I don't really want to compromise on the number of inputs.

I've picked out a single enclosure that I think can be used for both 16 and 32 jack designs. The extrusion is quite common, but the length isn't so it may one day not be available. It should be possible to get them cut to length though, at least from that company. I'll look into that. The end plates won't be machined and I'm sure that could be done for extra cost. Something else I'll enquire about.

Pcb pricing is driven by size. 100x100 or less seems to have a "special" price. The size 141x91 is what's needed for the enclosure. Minimum size for this design could be as small as 115mm so the board will have to be 26mm wider.

Min Qty is 5. Prices in USD. hasl is leaded.

141x91 pricing:

layers    finish    price
 2         enig      $27.5
 2         hasl      $9.2
 4         enig      $57.3
 4         hasl      $41.1

100x100 pricing:

layers    finish    price
 2         enig      $19
 2         hasl      $2
 4         enig      $23.40
 4         hasl      $8.00

shipping for 5:
$2.99 (10-18 business days)
$11.80 for fedex intl (4-6 business days)

enig might not be important to have as long as hasl is done with smobc. I need to confirm that because JLCPCB doesn't say. If it's possible to stick to 2 layers and hasl, then $9.2 vs $2 isn't a huge deal to go 4 layer.

I've got one enclosure in my cart (CAD$23 including shipping) and a bunch of jacks. But before I order anything I thought I'd ask if perhaps a different configuration would work.

Specifically 3.5mm jacks. They're significantly smaller. It would require special cables, or 3.5mm to 3.5mm cables with an adapter, but they would pack in a lot tighter and the pcb could be less than 100x100 so special pricing. Dumb idea? Be honest ;)

Alternately, could use a roland style db25 connector and then source a roland harness. Need to buy cables anyway so this might not be a terrible idea and each harness has 17 zones it seems.

I think 1/4" jacks is the best way to go, but it does add expense to the project. Using the cheapest shipping options:

The cheapest jacks: $15
The cheapsest pcb: $17
The enclosure: $23

subtotal so far is at CAD$55 or USD$43. To that we still need to add:

endplates with holes in them
an esp32 or teensy 4/4.1
some headers
some passive components (diodes, resistors)

And that's basically a kit that you solder together yourself. But it's pushing USD$100 already with a teensy.

Since the stuff costs what it costs, it seems fair to me, but I don't know about how other purchasers might feel.

I have no idea what MegaDrums cost. The eDRUMin is USD$329 including shipping for basically 22 adc inputs.

Would anyone want to build their own (besides me)?

At the same time to expand to 32 inputs would be really cheap. Would just need to buy the jacks and an esp32/teensy since there are 5 pcbs and two of these boards should fit in that single enclosure.

[corrados]

FYI: I just fixed the incorrect schematic on the Edrumulus main page: https://github.com/corrados/edrumulus/blob/main/algorithm/images/edrumulus_testing.png. Now the hi-hat controller input uses 10 kOhm resisters (in the previous version of the schematic there was 10 Ohms which was incorrect).

[jstma]

I found a really nice part: MCP4351. It's a digital potentiometer with some nice features and it's pretty cheap. It's got a wide tolerance of 20% and I'm not sure how that will play out. A 100K pot could be 80K to 120K. One useful feature is that it can disconnect all 3 terminals. Using that feature I think I can combine a hihat and piezo input with some caveats. Here's a schematic to talk about. I'm using regular potentiometer symbols for illustration because it's more compact.

The top half shows the circuit. The RV reference designators represent the digital pots. I've connected 3.3V via p channel fet and a polyfuse to the ring terminal of the jack. This way a gpio can be used to turn power on/off for hihat electronics.

The circuit connected to the tip terminal of the jack is what I'd like comments on.

The bottom half of the picture shows two cases for which the digital pots can be configured. The first is for a piezo input. The wiper of RV1 is set to 50% so that there is a voltage divider formed at the wiper. That will bias the piezo. RV2 can be adjusted from 100K down to it's minimum of 180R. This will increase the load on the piezo and lower the voltage. However, the minimum impedance will be limited to the parallel combination of the biasing resistors plus the minimum of RV2. That works out to 25K or so and that may not be enough to reduce a "hot" piezo.

The second case shows the configuration for a hihat input. That terminal disconnect feature comes in handy here because RV1 goes from resistor divider to pull up resistor. RV2 now goes to minimum impedance of 180R. Also, if we don't want any pull up (say for hall effect input), then we can disconnect the wiper from RV1.

This circuit would only be used for jacks 8 and 9. Jacks 1 through 7 would use fixed resistor dividers of 22K instead of RV1. RV2 would still be used to control the load on the piezo.

Have I overlooked anything?

This technique is only going to work if increasing the piezo load reliably reduces the peak voltage enough to accommodate hot piezos. And I have no idea to test this. @ignotus666 are you able to measure the peak voltage from your hot piezos? Are they normal 27mm piezos or the bigger 35mm? And how could I duplicate your setup?

[jstma]

Let me know when you've had a chance to try it.

This is great stuff! Thank you so much :-)

You're welcome ;)

I just recorded test data of my PD80R pad. Here is the commit: c9e7406

Looks like you have some clipping on the first part of the waveform on the louder hits.

The piezo is wired into A10, but because there's no bias the snare is alwasy ghosting and when I tap the piezo it turns into a roll, plus rimshot, plus bass drum.

You have to have the bias, otherwise you will get exactly these incorrect triggering you observed. If you just want to test the first pad (snare), then simply set this parameter to 1:

edrumulus/edrumulus.ino

Line 38 in c9e7406

const int number_pads = 8;

Then it does not matter if all the other Teensy inputs are not connected.

Good to know.

I would like to leave the L80 cymbals the way they are and do positional sensing. I don't expect your algorithm to cope at all ;)

Yes, I know what a L80 cymbal is. I am also not expecting my positional sensing algorithms to work. But it is worth a try :-). How do you intend to support positional sensing? Do you want to attach multiple piezos on that cymbal?

I'm looking into other ways of doing positional sensing. Particularly looking for how the waveform propagates (remember drums anywhere?). Based on the waveforms I've seen, the discontinuities in the waveform tell a story. I just need to figure out how to read it. I think all the information will be available in the first couple of zero crossings. Also important is that there is a bit of delay between when the hit occurs and when the cymbal resonances catch up at least with the piezo placed at the bell.

Anyway, when I get anywhere with it I'll start an appropriate discussion ;)

eDRUMin has figured out how to use positional sensing on plastic/well damped cymbals so that is worth exploring. There's a feature called "bell sense" which could mean less inputs. It would also be good to look at edge/bow/bell using a single appropriately placed piezo. But I haven't looked at cymbal stuff at all yet.

[corrados]

What I have on my list (and that list is very long ;-) ) is to add another piezo on the side of the mesh head to my PD-120 pad. That will be interesting to see what can be gained from adding another sensor.

[corrados]

Looks like you have some clipping on the first part of the waveform on the louder hits.

Exactly. I already had the impression that I am clipping the signal with that pad. Having a clipped signal will definitely destroy the positional sensing. So, my hardware design is not appropriate for this pad. But if I lower the value of the 22 kOms resistors, I fear that I cannot detect the very low level hits anymore...

[jstma]

Looks like you have some clipping on the first part of the waveform on the louder hits.

Exactly. I already had the impression that I am clipping the signal with that pad. Having a clipped signal will definitely destroy the positional sensing. So, my hardware design is not appropriate for this pad. But if I lower the value of the 22 kOms resistors, I fear that I cannot detect the very low level hits anymore...

The 22k shouldn't affect much. It's that 100k that's dominating the impedance.

If you need more dynamic range then a 16-bit adc is what's needed. At the same time eDRUMin appears to work very well with 10-bit adc (at least there are no major complaints). Now we know that the input front end for eDRUMin has some compression. Maybe that's a solution...

[corrados]

I think it makes sense to discuss specific topics in separate forum threads. I have created one for the anti-aliasing filter: #6 and one for the ADC resolution: #8.