Digital piano build

[TedWakeshaw]

Hi everyone,
Some time ago, I started building a digital piano. The project is based on a Teensy 4.1 board. The main idea is to create a fully functional, full-size digital piano with velocity sensitive, weighted keys. The Teensy 4.1 will serve as the MIDI controller, and a Raspberry Pi will act as the MIDI synthesizer. Each key will have three sensors.

At the moment, I’m working on a keyboard with single sensors per key, so there’s no dynamic response yet. To obtain multiple inputs, I’m using 74HC166 shift registers connected in series (eventually, there will be 3 rows of these registers).









If anyone is interested, here’s a link to a video showing the build process:



I’ve run into a small issue at this stage. Sometimes the keyboard works fine, but occasionally some keys randomly start repeating by themselves, or pressing one key triggers multiple notes at once. Any ideas what might be causing this? Below is the source code:

Code: [Select]

#define ioSelect 8
#define clockPulse 9
#define dataOut 10

int note;
int noteState[88];
int edgeTestSteps = 300;
int edgeTestStep[88];
int edgeState[88];
int j;

void setup() {
  pinMode(ioSelect, OUTPUT);
  pinMode(clockPulse, OUTPUT);
  pinMode(dataOut, INPUT);
 
}

void pianoKeys()
{
 
digitalWriteFast(ioSelect, 0);
digitalWriteFast(clockPulse, 0);
digitalWriteFast(clockPulse, 1);
digitalWriteFast(ioSelect, 1);
   
  for(j = 0; j < 88; j++)
  {
  delayNanoseconds(200);
   note = 21 + j;
 
    if (digitalReadFast(dataOut))
    {
     if (noteState[j] == 0)
      {
        if (edgeState[j] == 1)
        {
          edgeTestStep[j]++;
        }
        else
        {
          edgeTestStep[j] = 1;
          edgeState[j] = 1;
        }
        if (edgeTestStep[j] > edgeTestSteps) {
          usbMIDI.sendNoteOn(note, 69, 1);
          noteState[j] = 1;
          edgeTestStep[j] = 1;
        }
      }
     
    }
   
    else
    {
     
      if (noteState[j] == 1)
      {
        if (edgeState[j] == 0)
        {
          edgeTestStep[j]++;
        }
        else
        {
          edgeTestStep[j] = 0;
          edgeState[j] = 0;
        }
        if (edgeTestStep[j] > edgeTestSteps) {
          usbMIDI.sendNoteOff(note, 1, 1);
          noteState[j] = 0;
          edgeTestStep[j] = 0;
        }
       
       
      }
     
    }
digitalWriteFast(clockPulse, 0);
delayNanoseconds(10);
digitalWriteFast(clockPulse, 1);
 
   }

}

void loop()
{

 pianoKeys();

 
}

Any comments and suggestions are welcome!

[Phil1977]

Nice project!

Can you post a sketch of the whole button circuit? Of everything that´s behind the shift register, including the power connection of the buttons?

I´m quite sure it´s about some parasitic capacitive or inductive coupling between or towards your buttons. Do you have a scope available to measure the voltage slope when a button is pressed?

[moffy]

If you are using a switch type sensor per key at present, you will need to debounce them, switches don't open or close cleanly, they bounce between open and closed before finally settling. Unfortunately with the hammer mechanism that mechanical/electrical oscillation might be prolonged and give false readings. The judicious use of capacitors per switch usually helps, it would also help with the crosstalk if that is the cause of multiple notes being triggered, but it might also be mechanical vibration being coupled to the other switches when a key is pressed hard.

[PGPG]

My experience is that if any electronic device behaves differently than you expect the most probable source are problem with capacitors.
Looking at your PCB I don't see any blocking capacitors.
Each digital IC should have blocking capacitor (like 100nF) connected between its VCC and GND pin with as short as possible connection. Without them digital circuits can do everything you would never expect from them.

My suggestion - consider making all in SMD technology. Hand soldering SMD (like SO14 and 0603) is even easier than working with THT elements. You need not to revert PCB when soldering.
Well designed PCB (with one layer being whole GND) will be more robust against any disturbance (try to use cellphone near your piano PCBs).
You can design PCBs using KiCad (free PCB design software).

And local information (you of course like most people here can use china manufacturers).
Typically once per month 'technoservice' have a PCB-day. Orders from such days have special price.
Standard orders - you pay lot (I don't know actual prices) for preparing documentation for your PCB, but next time (even years later) you can order thousands of the same PCB paying only for cm².
Prototype orders - you pay much less for preparing documentation (not sure - may be around 80PLN) but more for cm². Number of the same PCB in prototype order is probably limited but you can make a PCB containing many of your PCBs and it is for them one ordered PCB (times that limited number). They don't keep your documentation - next time you pay once more for it. They simply put many orders together for manufacturing so can't repeat only yours.
Nearest PCB day is 19.09. You pay 1PLN for processing costs for prototype orders.
As I remember all PCB days were always at Thursday. You can try to ask them.

I can't find it at their homepage (may be I'm trying at some old page - its not me who orders our PCBs - I'm little lost in their www) but I have link from mail they send to us:
https://app.getresponse.com/view.html?x=a62b&m=Boydbv&mc=Id&s=BrsZtks&u=MDVgB&z=EQdlwGn&

Edit.
It looks that they simply have not up to date news on their site (tspcb.pl). I have send them a message about it.

[ledtester]

The capacitors recommended by PGPG are also called decoupling/bypass capacitors.

This blog post has a good explanation of why and how to use them:

https://www.vagrearg.org/content/decoupling

and includes this pic of of a suggested method of including them on a breadboard:

[TedWakeshaw]

Thank you all for your responses.

Nice project!

Thank you for your kind words.

Do you have a scope available to measure the voltage slope when a button is pressed?

Unfortunately, I don't have a scope.

If you are using a switch type sensor per key at present, you will need to debounce them, switches don't open or close cleanly, they bounce between open and closed before finally settling.

Actually, I’ve implemented a kind of debounce:

Code: [Select]

if (edgeTestStep[j] > edgeTestSteps) {
          usbMIDI.sendNoteOn(note, 69, 1);
          noteState[j] = 1;
          edgeTestStep[j] = 1;
        }

The MIDI note is not triggered until the same input state is confirmed 300 times in a row. So, it's probably not that. I'm pretty sure bouncing would manifest differently.

@PGPG
Thank you for the detailed response. Those capacitors make a lot of sense. I did some reading on the subject, and I'm almost certain this is the cause, especially since I have 11 shift registers connected in series, and the power for the next one is taken from the previous one. I measured the supply voltage on each shift register, and the voltage drops with each subsequent one. On the last shift register, the voltage drop is significant, and the problems usually occur with the later shift registers. I’ll try running a separate power line with thicker wires and solder a decoupling capacitor to each shift register.

My suggestion - consider making all in SMD technology.

I'm not a fan of SMD, but maybe I’ll come around someday. I guess I’m a bit old-fashioned and conservative

This blog post has a good explanation of why and how to use them:

https://www.vagrearg.org/content/decoupling

Thank you for the link. Very interesting and useful post.

[PGPG]

Unfortunately, I don't have a scope.

Scope is 'must have'. For me it was so 'must have' that to have it I have designed and build one in 1982 (you can find its description here: https://archive.org/details/Re0184OCR/Re_04_84_OCR/page/n5/mode/2up ).
Nowadays, for not a lot of money, you can have scope that I couldn't even dream those time. If not 'true scope' than at least USB scope.

I measured the supply voltage on each shift register, and the voltage drops with each subsequent one. On the last shift register, the voltage drop is significant, and the problems usually occur with the later shift registers.

Voltage drop that you see is not the main problem. Problem are short current pulses (ns range times) taken by ICs when they switch their state (you can't see them with voltmeter). This current pulses generate short voltage drop at supply. But not the wire resistance is the main factor but the wire inductance (the longer distance from supply (supply certainly have capacitor at its output) the higher inductance).
I'm sure that adding capacitors to each IC without changing wires will solve your problem. But of course if you see too big voltage drop at wire it is worth to change them.

I'm not a fan of SMD

For prototype like assembling THT is better, but when you design PCB for your device than SMD is much better. Try once and you will not get back to using THT in design. I have changed from THT to SMD around 1990.

[Phil1977]

It would really help if you would sketch your button circuit out. It helps you to understand what's happening and it helps us to explain.

Adding capacitors is one important step and may be enough to get rid of the worst problems. But if you want the piano to flawlessly detect the key inputs then you really should think about reasonable dimensioning of caps and resistors. A standard pull-up or pull-down resistor for example may have a huge tolerance on a breadboard test design but that doesn't mean any arbitrary value will be suited for reliable operation.

I think you really can get a lot of useful information in this forum, and especially hands-on-projects are usually really well supported. But you need to give some starting points like all the connections of the key contacts to power rails and shift registers.

[ledtester]

It would really help if you would sketch your button circuit out. It helps you to understand what's happening and it helps us to explain.

From the video it looks like a standard daisy-chain of 74hc166 shift registers. Each protoboard consists of three '166 chips and it looks like there are four such boards. There are 10K pull-ups (or pull-downs) on the shift register inputs (at 19:22 in the video.)

At 25:47 you can see how the keys work:

https://youtu.be/jvl6SZLRJpM?t=25m47s

When a key is depressed it lifts a wire off of a rod that runs the length of the keyboard. The rod is probably grounded and so the 10K pull-up presents a HIGH signal to a shift register input.

The signals running between the protoboards would have to be: power, ground, SH/LD, CLK and SER/QH.

[dietert1]

A nice build! As a boy i made something similar for an analog synthesizer together with a friend. We wanted to be a band.
For sensing the dynamics one would need analog position sensing for each key. Something cheap like an IR sensor. The shift registers would get replaced by multiplexers, probably with some buffer amplifier to drive the cables.

Regards, Dieter

[nctnico]

Nice to see people are still interested in this kind of projects. I've built something similar nearly 30 years ago    But I used a matrix to scan the keys. Much less electronics and wiring.
Velocity sensing is typically done by measuring the time between opening and closing the contacts on each key (each key has a normally open and normally closed contact).

[pcprogrammer]

For sensing the dynamics one would need analog position sensing for each key.

For sensing the velocity there is no real need for analog signals. Just setup a scanning matrix with two bars per section and check per key if it is on the lower bar or the upper bar or in between. When it leaves the bottom bar start a count, or read a timer value and when it touches the top bar stop the count or read the timer value again and subtract the two to get the elapsed time. Scale it to get a proper velocity value.

The attached image shows how this is done in the Siel Opera 6. There it is scanned with an old eight bit microcontroller, so with a modern day MCU it should be a piece of cake.

Another more advanced option to also get polyphonic after touch as well as the velocity an analog hall sensor can be used. Stick a magnet underneath each key and position analog hall effect sensors underneath. Scan the analog signals with a MCU and set values for upper, lower and after touch positions. When a value passes a set value take action to measure the velocity or the after touch.

[ledtester]

The OP experiments with measuring velocity via the two switch contact approach in this video:

https://youtu.be/WKrjwQOasvg

and another person has a demo of the Hall effect sensor approach here:

https://youtu.be/tgWXtYCHDI4

[tooki]

You also might want to take a look at how Yamaha did the sensing in most of their AvantGrand digital pianos (which have a practically unmodified full piano action, just without strings). They use optical fibers to implement remote optical density sensing without adding bulk to the action. (The fibers shine light through a gradient on a moving flap, giving continuous position information, rather than just on-off sensing of multiple positions.) Apparently the newest models now use magnetic sensing, but I don't know whether it's binary or continuous.

(A friend of mine, who is an accomplished recreational pianist, bought one (at his home in USA) after spending over three hours playing them at the big music store here in Zurich. The salesman was apparently delighted to have such an enthusiast, even knowing that it would not result in a sale.)

[Postal2]

In the Soviet set for children "musical keyboard Start-9069" (which I have), the pressing force is registered by means of 5 contacts that close in sequence when the key is moved.
https://rw6ase.narod.ru/index1/konstr/rk_komplek/start9069.html

[dietert1]

Just saw linear hall sensor TI DRV5053 at Digikey, € 40.53 for 100 pieces. In addition one needs the magnets. A cheap and pretty robust solution.

Regards, Dieter

[PCB.Wiz]

You also might want to take a look at how Yamaha did the sensing in most of their AvantGrand digital pianos (which have a practically unmodified full piano action, just without strings). They use optical fibers to implement remote optical density sensing without adding bulk to the action. (The fibers shine light through a gradient on a moving flap, giving continuous position information, rather than just on-off sensing of multiple positions.) Apparently the newest models now use magnetic sensing, but I don't know whether it's binary or continuous.

There are increasing numbers of interesting opto sensors too

This one from Vishay has 2 emitters and 4 PhotoTX to give a mix of linear and digital.



and this one from Nisshinbo has reflective with lens, so is almost linear 0-4mm



Other lens-reflective PhotoTX models are ITR1502SR40A/TR8 Everlight Elec and Sharp GP2S700HCP
Vishay also have a TCND5000, that is lens reflective, using PIN photodiode to remove PhotoTX  HFE variations.

[pcprogrammer]

A possible solution for using the hall effects is to make a PCB per octave that fits underneath the front of the keys. No need for analog multiplexers when a cheap MCU like for instance the STM32F103C8T6 is used. With the two ADC's it can read two keys at the same time and multiplex them over 6 channels each. With a sample rate of 100kSa/s per key it should be fast enough to get usable data for velocity and after touch.

Then connect the boards in series via the available UARTS to send out the MIDI data. Have all but the last one send at the highest baud rate it can handle, which is many times the MIDI baud rate, and have the last one send it out at the normal MIDI baud rate.

Needs some priority system on the data between the boards to not be swamped by after touch data, but that is not that hard to implement.

Use a couple of pins to configure the octave the board is used for and another one to indicate it being the last one. This way the software can be the same for all the boards.

[TedWakeshaw]

Wow, I didn’t expect so much interest and so many posts on my thread! It looks like this forum has an amazing community.

As I’m reading through your replies, I realize that even though I’ve been into electronics as a hobby for over 20 years, I feel like a total noob compared to you all!

Scope is 'must have'

I know, I’ve been thinking about getting a scope for almost 10 years now, but somehow I’ve managed to get by without one so far.

For me it was so 'must have' that to have it I have designed and build one in 1982

That’s incredible! I’m impressed. I wasn’t even born yet at that time!

Voltage drop that you see is not the main problem. Problem are short current pulses (ns range times) taken by ICs when they switch their state (you can't see them with voltmeter).

Yeah, I get it. What I meant is that it seems like those short current pulses you're talking about might have a bigger impact when there’s a voltage drop. At least that’s how I understand it.

As for SMD components, I’ve got hundreds, maybe even thousands of THT parts at home because I always end up buying some extras whenever I make a purchase. And guess how many SMD components I have? Exactly zero. This reminds me of the time I had to repair my mother-in-law’s plasma TV and had to drive across the whole city (Lodz) just to get one SMD resistor

It would really help if you would sketch your button circuit out. It helps you to understand what's happening and it helps us to explain.

Like ledtester said, the aluminum rod is grounded, and wires are constantly in contact with it. When a key is depressed, the wire is lifted from the rod. All the shift register inputs are connected to Vcc through 10k resistors. There are four PCBs—three of them have three shift registers each, and the last one has two.

In the future, I plan to add two more rows of shift registers to measure the speed of the keys, like in this video:

The OP experiments with measuring velocity via the two switch contact approach in this video:

https://youtu.be/WKrjwQOasvg

I chose this method of connecting multiple keys instead of a matrix for easier programming and indexing of the keys. And since I managed to buy 40 shift registers for less than $40, I figured it was worth it. Initially, I considered optical sensors, but I went with the mechanical solution because it’s cheaper. The great thing about DIY projects is that you can always upgrade them later on!

[nctnico]

For long term, an aluminium rod is not the best solution due to corrosion. Try to find a stainless steel rod and use stainless steel springs. Stainless steel wire for making springs is widely available as well. As stainless steel has some nickel in it, I expect this to last much longer compared to aluminium. However, using something like a micro-switch (without a click) would be more durable as the contacts are made from the proper materials.

[Phil1977]

Thanks for the diagrams!

For the very beginning I´d add two components per key:



the 1k-resistor in the switch line reduces current peaks when the key is released. The capacitor should average out all bouncing <1ms while still keeping the latency for detection of a pressed key (open contact) <10ms.

It´s just the first mod I would try - I´m sure others have other maybe better ideas. But it definitely should improve the reliability of the inputs.

And as already mentioned: You should definitely put decoupling caps to the power rails. At least per shift register IC you should spend 1 cap, e.g. 100nF.

[RoGeorge]

Nice idea for the key contacts! 

The same idea can be used to read the keys velocity, by adding a second contact for each key.  The second switch is identical with the first contact, just that the tip must be bent a little more, such that the height (distance) between first and second contact opening is the same in all keys, so one switch will open a little later.  How much later, is dictated by the key velocity.

By measuring (in software) the time between the first switch and the second switch opening, the velocity/force of the keypress can be measured (to control the note volume).

[BrianHG]

If you plan on keeping the aluminum rod and spring contacts, google:

gold plating aluminum kit

Plating both you springs and aluminum rod with a good thick layer will keep the contacts corrosion free.
This will improve the aging consistency of your hardware switch solution.

However, this most likely wont ultimately solve your signal bounce issue.  Though, it may be possible to solve this with software if your switch contacts and signal ground are clean throughout.

As others have pointed out, a scope is a must to inspect exactly what is going on.

Even a sub 100$ toy oscilloscope (not recommended) would still at least show you some details in the low frequency domain of a mechanical switch.


As others have pointed out, if you were to go to hall-effect magnetic sensors with a small magnet on each key, there would be 0 contacts to ever break down or wear out.  Also, if you replaced the cmos digital serializers with something like an MCU with multiple ADC inputs reading each sensor, you could then determine velocity and depth of each key press if you wanted to engineer such a capable keyboard.

[PGPG]

it seems like those short current pulses you're talking about might have a bigger impact when there’s a voltage drop.

Where you have smaller voltage you have also longer distance from supply. I think that this distance in sense of wire inductance have bigger influence than DC voltage being smaller.
You use HC serie. 2V is enough for them to work.

I’ve got hundreds, maybe even thousands of THT parts at home

In 70s and 80s I was collecting elements as it was not possible to buy what you need but you could only buy what just was at the moment accessible. Having a big set of elements was the only way to go. All my designs were made to use only elements I just have because in other way it could took me years to get the part I need. Even my oscilloscope was designed only with elements I had bought in past not knowing what for they will be used. This way I had a big collection of elements I have never used.
About 10 years ago I throw away most of my old THT elements to make room in my matchbox cabinets for modern elements.

For the very beginning I´d add two components per key:

If 10k would be at switch side of 1k you will be driving IC input with true 0 and not 1/11 of VCC.
Then you could replace 1k with 100k and 47n with 470p to make both state changes with almost the same time constant.

[TedWakeshaw]

For long term, an aluminium rod is not the best solution due to corrosion. Try to find a stainless steel rod and use stainless steel springs.

I'm aware of that. The springs are made of stainless steel. I was thinking about sanding down the rod with sandpaper and coating it with graphite paint, but I'm not sure if it will stick.

For the very beginning I´d add two components per key:

I'm very grateful for your suggestions.

The same idea can be used to read the keys velocity, by adding a second contact for each key.

The wire I used is quite flexible, and it would be difficult to maintain even spacing between them. I have a different idea for measuring velocity.

If 10k would be at switch side of 1k you will be driving IC input with true 0 and not 1/11 of VCC.
Then you could replace 1k with 100k and 47n with 470p to make both state changes with almost the same time constant.

I wish I could understand what you wrote.