Piano World Home Page Forums Our Most Popular Forums Digital Pianos - Electronic Pianos - Synths & Keyboards Cybrid (CyberGene's hybrid MIDI controller) is open-source!

Not obvious.

Hi Cybergene: important and very needed advices on the amount of work that will be spent in a cybrid DYI; this is a project not for the faint of heart... I run a academic research lab in neuroscience, and probably half of my team's time is spent on customization of open-source software and hardware for our own needs. I am well too familiar with these kind of projects and even so, I have second thoughts about attempting a cybrid solution or going the easy way and get a NU1x (N1x cost is unreasonable for me at this moment). Because used grand actions in good conditions are scarce, I am considering an upright-cybrid: it is far easier to find a somehow recent and decent upright and end up with a customizable midi controller for way less money than a NU1x (or get a NU1x, and have the cybrid components at hand and install them into the Yamaha action when the original electronics start to break)...

Cybergene: do you have any indication that the cybrid will not work as well on an upright action? There is far less space to install the optical sensors in a way in which the trimpots are easy accessible to screwdrive adjustment. If below the hammer rail, then all the logic needs to be reversed (from detecting approximation to the sensor to detect distance to the sensor, which makes the system less sensitive at the end of the hammer travel where it matters the most); above the spring rail the dampers need to go most probably... Any thoughts on this?

On an upright piano action it will be tricky to find where to put the boards and the sensors. On a grand piano the shank bases, where I measure velocity, are the uppermost part of the action and at the same time has shortest travel for being closest to the pin. However on an upright the same point is buried inside...

Although it’s a pretty successful project, I look at it more as a personal achievement and fulfillment. In retrospect, and despite it being very expressive and well-working, especially taking in mind how cheap it was, I’d probably purchase the N1X again. And I previously said the opposite 😀 That if I had finished it before I purchased the Yamaha, I wouldn’t have purchased the Yamaha. There’s still something in a mature product as the Yamaha that makes me feel I have a real instrument that I never stopped loving and is still a dream come true. Whereas the Cybrid is a quirky thing that makes me feel good more because I proved something to myself rather than because it is good in itself (and it is). Blame it also on my OCD-ish nature that likes commercial and finished products rather than custom/modded solutions. I have friends who are on the opposite and their first task with a new car is “upgrading” filters, bumpers, radio and whatnot. We’re all different

I have the NU1 myself, bought new in 2015. I opened it. I read the service manual. It is extremely well built. Ok, mine is not very old, but I believe it will never "start to break".

You might be able to find an old grand in UK, where they are a little more common than rest of Europe. Too bad for you that you are not in USA where they are extremely common. If you go the UK route, be mindful of Brexit, which will make it extremely more hard to get it shipped after Brexit is completed (end of 2020, IIRC).

Let's be realistic. Everything eventually breaks. Everything.

I only just recently came across your work, and I'm tentatively considering making a derivative. I'm also a software engineer, but I've got a background in embedded development, and electronics. I was going to wait to have some kind of results before I bothered saying anything, but I felt bad that you hadn't gotten the response you'd hoped for. I know how that goes. I'm very impressed with the Cybrid, it seems to work extremely well, and I'm quite envious.

(Massive) changes I'm likely to implement if I attempt to build one:

The CNY70 sensor board has electrically "long" wires back to the note module board, and the output of it is a rather high impedance signal. It's clearly working well enough, but my inclination is to add a tiny LDO linear regulator on every board, a capacitor, and an op amp as a transimpedance amplifier. That'll give you a low impedance signal out from the board, and be more resistant to any noise on the power lines. That's probably <=$0.50 of parts per key, at the necessary quantities. One of the first things I'll do is produce a new PCB that can either be wired up identically to CyberGene's, or support these alterations. (That'll give me a nice platform with which to experiment on the sensors, while figuring out the next steps.)

I'm curious exactly what trimpot CyberGene used, and where he got so many of them for less than the stated total budget. Brand new appropriate looking ones would run me $300 for the necessary quantity(!!!). To that end, I'm tempted to just have 88 dsPICs, one for every key, watching the sensor via their ADC. It'd cost (me) less, be more compact, and be significantly more flexible than comparators. They can all be tied back to a central slightly beefier microcontroller via a CAN bus. That would only need two wires (plus power) between note boards (or their equivalent) and the main board.

If I make it anywhere with thid, I'll pop back up and say so. I'm also happy to provide what electrical expertise I've got to anyone thinking about this.

Finally, if anyone has any suggestions for acquiring a grand action in Colorado besides "craigslist", I'd appreciate it.

Nice to e-meet you Jay. I'm Software Engineer in Colorado too. Impressed by your suggestions to improve Cybrid. Regarding the ADC's, I think CG tried or considered them too, but they were too slow for the purpose. We need around 1ms (better if lower) response time, which means way much faster if polling. Now sure how fast ADC can be these days (last time I used them seriously -- i.e. not raspberry Pi -- it was 20 years ago!)

For getting the action, craigslist is a good start, you can get excellent deals if you are patient, especially around May when people sell their houses and need to get rid of stuff quickly or pay penalties to their buyers. I posted my whole thought process (including volume ordering of electronic parts) in this other thread: http://forum.pianoworld.com/ubbthreads.php/topics/3031436/re-cybrid-kickstarter.html#Post3031436

Other options are repair shops of which there are more than I know, but I can send you the list of the ones I know, if you'd like. Those tend to charge more, not necessarily for more value. Perhaps post-pandemic we could join forces in this project?

Absolutely. But reading between the lines of the OP, it sounded like he intended to buy the piano and expect it broken in a couple of years.

When I mention "break" I am thinking in 10-15 years timeframe. And that is for me the appeal of a DYI project: I can change sensors, move from the teensy board to the whatever-arduino-clone will become the defacto norm, alter the code to adapt to a new pedal, change VST or whatever generates the sound, change speakers to resonators to whatever, etc... But the action and cabinet stays the same for dozens of years (well... maybe I don't have many dozens of years left in me...). If I have an OCD component (which I don't think so...) is about the upweight of the digitals. I am unable to not feel it, and I am constantly feeling like fighting the urge of the keys to return to rest, while an acoustic feels like having two resting positions: one stable at the stop and one dynamic at the bottom, but both confortable.

To me, as a long time guitarrist, this is like the feeling of steel strings: they are natural in a electric guitar (slender neck, different string attack, curved fretboard for string bending) but become a nuisance in an acoustic that is basically a nylon design with steel strings slapped on it... My fingers never felt confortable despite the many many hours playing steel strung guitars. A steel string acoustic guitar is convenient because it generates more volume, the same way as a non-escaping action is more convenient for the companies because it's way simpler to create a midi value (the note is held for as long as the sensor is pressed, and this principle remains the same from the most basic action to the GFIII).

So, I rather invest in a NU1x than on anything that costs more than 50% of it. And this gives me few quality options. Hence the charm of cybrid...

Well, certainly Cybergene can better reply to the exact supplier, but china suppliers are cheaper than digikey. Take a look at https://www.lcsc.com/ for example. Or take the alibaba gamble where the same components are 5-10 times cheaper.

Go fund me?

Hi!


I was just talking to someone the other day about how much ADCs and DACs have improved in the last 20 years; it's really driven a lot of our technological improvements over that time.

I'm specifically looking at the dsPIC33CK256MP508 family of parts, which has multiple built-in 12-bit ADCs that can each run at 3.5Msps. 4x oversampling takes us to 14-bit resolution and still leaves us with 875ksps. It also has some built-in op amps, which would make it very convenient to construct an analog differentiator to feed one of the ADCs (derivative of distance being speed, of course, which is what we actually want to measure), if the signal from the CNY70 is sufficiently linear. So one ADC could constantly sample the distance value, and then when the key stick/hammer/whatever crosses the appropriate distance, switch over and take a reading of the speed.

Note that those dsPICs can run at 100MHz, and I'm suggesting one on every single key. This plan involves throwing ~9GHz of CPU at the task. Should be plenty of oomph to watch all the keys with ADCs.


Yeah, I'm starting with craiglist. I've got an email out to someone on there about a grand. I don't have the energy or heart to deal with dissecting a piano while the previous owner watches, so I just offered to cover disposal costs in exchange for the action. We'll see. I suspect they've got hopes of someone wanting to refurbish it. ("works great, just needs a new soundboard!")


I'll get back to you on that if craigslist and near-free hasn't worked out by the time I have a smidgen of working electronics.


I'm definitely interested in any co-conspirators or assistance; it keeps me motivated to actually finish projects. And my wife already made some threatening noises should I acquire a piano action and then just leave it laying around the house to gather dust...

Cool! Exactly the type of crowd I was expecting to contribute and improve 🤝

Sounds nice! Never thought of it but sounds like a very good improvement.

Here's the part: 3296W(64W) 10K DIV

As you see the prices are in US$, I ordered 270 of them and the price for such a quantity was $0.18 per trimpot, or a total of less than $50. I believe this Bulgarian store works with Farnell.

This is exactly one of my initial ideas, however around that time (and still to this date) I only know the word "PIC" but nothing more and my initial research in the vast sea of specs and programming guides left me rather scared that it would be too difficult for me I'm very far from low-level or embedded programming. (As you see, I even implemented my code for the Cybrid in the Arduino compatible high-level paradigm rather than direct low-level C-routines which would be probably the better way). So, I'm a rather high-level developer and am very inflexible in that regard... However I believe this will be the proper solution since separate ADC-s on a group of notes will be very fast and reliable, will eliminate the need for trimpots and will allow for automatic (or semi-automatic) velocity calibration, compared to the mundane and tedious task of manual trimpot calibration. I'd be very glad if experienced guys can do this massive improvement. It will change entirely the scanning algorithm too, because currently I rely on timed measurements across the entire keyboard which is ultimately slow. Having this happening locally per note, or per note group is the way to go 👍🏻

Del Vento, when I said ADC is slow, I meant using 1 or 2 global ADC-s as in the Teensy that will serially scan through all notes.

BTW, using these 25 turn trimpots for full 5V voltage dividers turned out to be rather silly because the difference between the last two measurement points is (if I'm not mistaken, need to check again) less than 100mV and so even with this fine trimpot it's not very efficient. I thought as a small improvement for a future revision (for the trimpot solution) I will not divide the entire 5V range but will limit it for the entire group in a smaller region, so that the trimpots have broader range.

Those same trimpots on DigiKey here in America cost $4.38 each.

For the 270 quantity you bought ...
At DigiKey the price drops to "only" $3.01 each.
At Mouser they're $2.01 each.
At Arrow only $1.44
At Newark only $0.81
At TTI only $0.78 (I've never before heard of TTI).
The $0.18 price you paid seems out of reach here.

😱 I have always been under the impression that for some reason you can find the cheapest stuff in the USA, especially when talking about electronics. Pretty weird...

I found them as the knockoff brand "Bonens" on Alibaba for only $0.10 in 500 quantity, or $0.25 in 50 quantity.
https://www.alibaba.com/product-det...2700.details.maylikeexp.3.45b83562bReqOC
https://www.alibaba.com/product-det...2700.details.maylikeexp.9.45b83562bReqOC

Still ... if JayKominek could build this pot-free, that would be ideal.

Pianoteq are running a video contest that's open for a few more days.

Just saying

(Would need to be re-shot with pianoteq)

Very interesting! As I said it's been a long time, but happy to take a look and see if I can contribute (more likely to sw rather than hw) if you have something even half-finished to post.

Which one?


For the craigslist option, the disposal fee isn't very different from the transportation fee. If you want to pay that, get it delivered to your garage and do the dissecting in private, so you get the cabinet parts too!

I'm sure your wife (like mine ) will be delighted by the dust and noise in the garage not to mention the nice gift of a very large, non-round, cast-iron-but-gold-colored ring you will gift her (I'm talking about the plate)

Joke asides, building the cabinet in other ways, it's too much work and cost. The most difficult part of the job will be the plate removal, which definitely requires two people (Resources on Arapaho will take it off your hands once you put it in your car and drive it there). The second most difficult one will be cutting the tail (assuming you want it short like the N1x). Not the cutting per-se which is easy with a tree-pruning saw (either chainsaw or reciprocating saw), but the "smooth" final cut before the reassembly. You will need a table saw (or a friend with that) for getting that part of the job done.

Alternatively, I am sure some piano moving companies will be happy to remove the plate for you, for a small fee. I can give you a referral of at least one who should. The problem is that you need to remove the strings FIRST and that is tedious and time consuming, so that will be a higher fee (or a second trip, they will not wait for you to spend 2-3 hours unwinding the strings).

Happy to talk more about this if you decide to go with this route.



Everything sounds very good!

PS: where are you? I'm in Boulder (you can PM me or simply not respond if you prefer to keep some privacy)

Hi. My non-expert 2 cents opinion about hardware improvements. I don't think that PICs or any other type of microprocessor with ADCs is the best way to go. We must not forget the intrinsic limitations of MIDI (Note-on, note-off transmission of values in the range 0-127, at very low speed). That's all. There is no need to full nanosecond resolution of key or hammer instantaneous position. ADCs are needed for continuous detection, but that is not necessary for speed of note-on, note off. Cybergene solution is EXTREMELY elegant and efficient and solves a myriad of timing issues. It is the same sensor that makes multiple readings from each hammer and timestamps each crossing of 3 sets of voltage thresholds. The simple method of sequential reading of keys looks very non-sophisticated, but it is still faster than the rate of MIDI transmission, so there is no need to be even faster. The use of set voltage points for binarization is also non sophisticated compared to MHz ADCs, but it is already capable of detecting displacement speeds in excess of the 0-127 range, so there is also no need of higher resolution of hammer or key speed (we can discuss the merits of MIDI 2.0, but 127 values at the present speed rate do not seem to cause a limitation in sound quality). ADCs still require the use of optical sensors (with a rate of operation slower than ADC) or better yet 2-axis hall-effect magnetic sensors. All of these require intra and intersensor calibration if many are read by the same ADC; and then computation of speed is needed from the ADCs values... Again: I don't think that it is possible to beat the efficacy and simplicity of CG solution. It can be replicated by other strategies, but there really no need for that.

What I think is needed (and Jay already mentioned important possible improvements) is:
- more stable voltage calibration of the optical sensors (I am limited in electronics, but Jay suggestion of LDOs is probably the best solution);
- include a 4th reading per key to allow for key-off speed; this would require 4 trimpots per key instead of 3;
- in alternative to the above, making the voltage thresholds software based instead of hardware based: this would permit to increase or decrease the number of readings per sensor without changing PCBs. I really don't know if this is possible at all; more expert people (Jay?) can comment on this.