- Semiautomated calibration. Ideally once the hardware is correctly installed, it can be calibrated by playing all of the notes in order. It is expected this will require an attached PC running specialized software.
- It must be possible to make this work with just the extracted action of a grand piano. Mechanics are outside of the scope of what this project intends to worry about, but it would be ideal if a MIDI controller housing could be made for a piano's action, such that the action could be transferred between controller housing and piano with ease.
- Support for multiple common 1/4" TRS sustain pedals, or some other sensors that would be suitable for measuring the state of pedals still in their original state.
- The boards should be designed such that bad firmware can't wreck anything. I feel like this is important for open hardware type projects, where there is the possibility that folks might start fiddling with their firmware without a full understanding of the hardware.
- Integration with a Raspberry Pi 4 Compute Module capable of running Pianoteq so that a standalone piano can be produced.
- Alternatively, try to figure out how to get some open source synth software running on some sort of microcontroller, such that it can take MIDI over I2C and generate audio.
- Design/integration of DACs and audio amplifiers into that synth board, so that it can drive speakers directly.
The hardware/software system is based around the Vishay CNY70 optical reflectance sensor. It consists of an infrared LED, and a phototransistor sensitive to the LED, in a convenient package. It can measure distances in the (roughly!) 0-10mm range.
In the full version of this conversion, 88 of those sensors would be placed under the keys to detect note-off time & velocity, and 88 of them placed at the end of the hammers' travel, so as to measure note-on timing and velocity. This would prevent the "loud note" problem that troubles some hybrid pianos. Additionally, information from the key sensors may be used to signal to the synthesizer that the damper has been raised (permitting resonsnace) before the hammer strikes. Sample based synths could also use that information to preload samples.
(During the initial design of this I completely forgot that pianos usually don't dampen the highest strings. If you want to leave the top 22 notes undampened, you could eliminate one of the ADC boards. Or you could convert a >88 key piano.)
That's 176 sensors in a piano. 22 are monitored by each of these boards. The current through each one is converted to a voltage by a transimpedance amplifier, and monitored by an ADC channel of an STM32H743. Hopefully the STM32 will be able to process all 22 signals at between 10kHz and 20kHz. I expect to apply some digital low pass filtering.
Once the processor has determined that a note on, or off, is occurring and measured the velocity, it is sent out over a RS485 serial link. Use of RS485 should allow us to spread these boards out through the structure of the controller, while maintaining relatively high speed serial communications. The RS485 link takes place over two twisted pairs of an ethernet cable. (NOTE we are NOT using ethernet, just taking advantage of the large supply of high quality premade cabling and jacks.) Having two twisted pairs permits full duplex communication.
The same cable also carries ~5V power. Each board includes a single switching regulator to bring the 5V down to 3.3V. The analog voltage rails are produced by linear regulators, to reduce noise.
The board can be populated with optional components to make it into a USB device. This isn't expected to be necessary, but may be more convenient during the sensor calibration process. Two additional UARTs are brought out to 0.100" header pins. Just in case.
Programming is performed over the JTAG header, intended for use with the ST-Link/v2 programmer.
There is space for four LEDs, and a number of otherwise-unused pins are brought out to sizable test points.
Each sensor board will consist of nothing but some CNY70s, and a 0.100" header. 2*N pins will connect to the phototransistors, and come back to the ADC board via twisted pair. The last two correspond to a chain of all of the LEDs on the board. The LEDs should be chained together in rather long strings. Likely 22 sensors per string.
The LED power board is designed (as of rev3) to use an off the shelf 48V power supply, and provide a constant ~20mA to 8 different strings of LEDs. Each string is intended to have 22 CNY70 LEDs on it.
We derive from the 48V a ~12V power rail for the rest of the analog electronics on the board. The design as it stands should produce constant currents which are extremely stable from time scales of a year, down to 100kHz. The deterioration of the LEDs themselves is expected to be the main source of change in the optical output.
Rev3 of the board adds a microcontroller which can monitor the LED strings individually to detect errors, and also turn all the LEDs off by shutting down the precision voltage reference.
Intended to use a STM32H743 like the ADC board.
8 RS485 serial links, along with 5V distribution to the ADC boards. Per-board high-side MOSFET control of power, so that boards can be brought up in sequence once power has stabilized. Also allows them to be power cycled in event of error, or to reflash them.
USB B device link, as the main/intended link to the host computer.
Four I2C controllers are brought out to three separate Qwiic headers, so that a variety of common I/O devices can be attached without requiring special development. (MIDI DIN, pedals, led power control, RTC clock, ethernet/bluetooth, blinky RGB LEDs, etc)
Firmware is expected to process the fairly raw velocity messages from the ADC boards, appyling any calibration, and sending them out over any of the many MIDI links available to it (serial, USB, ethernet, maybe even bluetooth).
A low powered STM32G0 a board with Qwiic headers, and 4 1/4" TRS jacks for standard sustain pedals. 12 ADC channels go to every pin of the TRS jacks in hopes that we can automatically determine the behavior of the attached pdals.
Like the pedal board, the MIDI board is designed as a Qwiic device. It also uses an STM32G0 to act as an I2C<->UART bridge, and has the appropriate MIDI isolation circuitry.
To facilitate development/testing of the ADC boards without the main board, a "link board" has been designed. It is a full duplex RS485 board with power injection comparable to what the main board will provide.
As an alternative to using the main board, a PC with enough link boards to talk to all of the ADC boards simultaneously could skip out on the use of a main board, in exchange for running more complicated software locally.
See the wiki Assembly Notes for this info.