USB MIDI Controller

I use Ableton live a lot to jam and to DJ, and using the mouse to
control all the parameters got in the way quite often. You can’t
select a scene and turn a knob at the same time, switching between
views was tiresome. I first thought about building a midibox, but then decided to build my own midi system
using an AVR microcontroller, which I’m more used to. Blueloop helped me a lot
with the schematics, showing me a nice trick to control LEDs using the
same matrix as for the switches (this is not integrated yet,
though).

The midi controller is quite simple. It features 33 switches and 56
potentiometers. The switches and potentiometers are laid out in a
matrix. You can select a horizontal row by pulling down a port to 0 V,
and then reading the switch values on another port, or by going
through the ADCs in the case of the potentiometers. The actual circuit
for the microcontroller is really really simple. There is an
oscillator and a resistor for the MIDI output. The rest of the I/O
pins go to a pinhead which connects the front panel to the circuit
board.

There is a second circuit board in the midi controller that
converts the MIDI output of the first circuit to USB. It was an old
USB programmer board I built 2 years ago that was lying around. I
directly connected the power of the first circuit board to the USB
connector, and connected the MIDI output to the RX pin of the USB
programmer. I had no 12 Mhz oscillator (the one which I normally use),
so I used a 20 Mhz quartz out of an old network card. The board is
really simple too, and consists of a PDIUSBD11 chip by Phillips, which
gets commands over I2C and outputs USB, and an ATMEGA8
microcontroller. Sadly, the PDIUSBD11 is not manufactured any more,
which is kind of sad, as it is the only USB controller I am aware of
that uses I2C, which spared pins on the microcontroller side. I
implemented the USB MIDI
specification out of the specification, and the controller seems
to work under Linux and Windows (there is a problem with MacOSX, but I
last checked two years ago and haven’t tried/debugged it since
then). I can’t publish the whole code for the usb-midi adapter though,
as most of it (including the USB stack) was written by Till Harbaum.

The matrices are soldered on the front panel, which I made out of a
plexiglas sheet through which I drilled holes for the switches and
pots. I had no real problems with the plexiglas, except that you have
to drill quite slowly through it in order not to split it, and that I
had to remove the melt plastic around the holes using a rasp. Each
switch has a diode at one of its outputs, so that two knobs pressed at
the same time can be distinguished on the controller side. The
potentiometers are connected to VCC one one side, to the select port
on the other side, and the variable output is connected to the ADC
using a diode. When selecting a switch or a pot, the select pin is
set to GND, all the other pins are set to VCC. Because of the diodes,
the pressed switches or the pots which are not selected are always
high, and the diode won’t allow them to influence the signal.

The code on the microcontroller is almost as simple as the circuit
board. A simple ring-buffer interrupt-based UART handling is used to
handle outgoing MIDI data (the midi controller can’t receive MIDI
events).

#define RB_SIZE 16
#define MIDI_BAUD 31250UL    /* 31.25 kbps */

volatile char tx_buf[RB_SIZE];
volatile char tx_wr = 0, tx_rd = 0;
#define TX_INCR(a, b) (((a) + (b)) & (RB_SIZE - 1))
#define UART_TX_OK (UCSRA & _BV(UDRE))
#define TX_FREE (!(TX_INCR(tx_wr, 1) == tx_rd))

SIGNAL(SIG_UART_TRANS) {
  if (tx_rd != tx_wr) {
    UDR = tx_buf[tx_rd];
    tx_rd = TX_INCR(tx_rd, 1);
  }
}

void uart_putc(char c) {
  if (UART_TX_OK && (tx_wr == tx_rd)) {
    UDR = c;
  } else {
    while (TX_INCR(tx_wr, 1) == tx_rd) ;
    tx_buf[tx_wr] = c;
    tx_wr = TX_INCR(tx_wr, 1);
  }
}

void uart_init(void) {
  tx_wr = tx_rd = 0;
  
  /* set baudrate */
  UBRRL = (((F_CPU) / (16 * MIDI_BAUD)) - 1);
  /* enable TX, interrupts */
  UCSRB = _BV(TXEN) | _BV(TXCIE);

  sei();
}

The static buffer “tx_buf” is used to buffer outgoing serial
data. “tx_wr” and “tx_rd” are used as indices into the ring
buffer. “TX_INCR” handles the indices when a character is added to the
ring buffer. “TX_FREE” checks if there is free space in the ring
buffer. “UART_TX_OK” checks if the serial interface is ready to send
another character. “SIGNAL(SIG_UART_TRANS)” is the interrupt routine
that is called when a character has successfully been sent. It checks
if the ring buffer is empty, and sends the next character if it is
not. “uart_putc” is the API to the UART code, and directly sends the
character if the serial interface is ready and the ring buffer is
empty. If not, it waits until there is space in the ring buffer, and
adds the new character to the ring buffer. “uart_init” handles the
initialization of the serial interface, sets the correctly baudrate,
and enables TX and the TX_TRANS interrupt. Finally, interrupts on the
AVR are enabled.

void io_init(void) {
  /* disable watchdog */
  WDTCR = 0;

  /* DD = 1, output */
  /* PORT = 1 wenn DD = 0, dann pull-up */
  /* portd = output */
  DDRD = 0XFF;
  
  /* enable ADC */
  DDRA = 0;
  PORTA = 0xFF;
  ADMUX = _BV(ADLAR);
  ADCSRA = _BV(ADEN)|_BV(ADPS2)|_BV(ADPS1)|_BV(ADPS0);
  // ADCSRA = _BV(ADEN);

  uart_init();
}

The routine “io_init” initializes the Atmel, first disabling the
watchdog, then setting PORTD to output (which is not really necessary,
as the status of the port is set by the switch polling routine). The
ADC is enabled by setting the PORTA to input, enabling pullup
resistors, and enabling the ADC in a 10 bit mode, and using AVCC as
reference voltage. Finally the UART is initialized.

/*** MIDI zeugs ***/

#define MIDI_NOTE_ON     0x9

#define MIDI_NOTE_OFF    0x8

#define MIDI_AFTERTOUCH  0xA

#define MIDI_CC          0xB

#define MIDI_PROG_CHANGE 0xC

#define MIDI_CHAN_PRESS  0xD

#define MIDI_PITCH_WHEEL 0xE
#define MIDI_STATUS_TYPE(c) (((c) & 0xF0) >> 4)

#define MIDI_STATUS_CHAN(c) ((c) & 0xF)
static inline midi_is_status(char c) {

  return (c & 0x80);

}
void midi_send_note_on(unsigned char channel, unsigned char note, unsigned char velocity) {

  uart_putc((MIDI_NOTE_ON 
The midi controller sends three types of midi messages. When a switch

is pressed, it sends a note on message, when a switch is release, it

sends a note off message, and when a pot is modified it sends a

control channel message. All three messages are three bytes long and

sent directly to the uart using "uart_putc".
#define LED_BUTTON_PORT     D

#define PORTLED_BUTTON_PORT PORTD

#define DDRLED_BUTTON_PORT  DDRD

#define PINLED_BUTTON_PORT  PIND

#define SET_PIN_OUTPUT(port, pin) { DDR ## port |= _BV(pin); }

#define SET_PIN_INPUT(port, pin) { DDR ## port &= ~(_BV(pin)); }

#define SET_PIN_INPUT_PULLUP(port, pin) { SET_PIN_INPUT(port, pin); PORT ## port |= _BV(pin); }

#define SET_PIN_LOW(port, pin) { PORT ## port &= ~(_BV(pin)); }

#define SET_PIN_HIGH(port, pin) { PORT ## port |= _BV(pin); }

#define PIN_VALUE(port, pin) ((PIN ## port & _BV(pin)) ? 1 : 0)
#define NUM_BUTTONS 30
#define BUTTON_STATUS(i) (button_status[(i)] & 0xF)

#define SET_BUTTON_STATUS(i, val) { button_status[(i)] = button_status[(i)] & 0xF0 | (val); }

#define LED_STATUS(i) ((button_status[(i)] >> 4) & 0xF)

#define SET_LED_STATUS(i, val) { button_status[(i)] = button_status[(i)] & 0xF | ((val)  1 (bounce?) -> 2

   wenn knopf auf 1 dann nachricht senden (note on), wieder auf 0 (note off)

*/
#define BUTTON_ROWS 5

unsigned char button_status[BUTTON_ROWS * 8];
void init_buttons(void) {

  int i, j;

  for (i = 0; i 

The button polling routines are a bit longer. The routine

"ask_buttons" polls through all the buttons, and calls the routine

"ask_button" to get the status of a single switch. The result of the

polling is stored in the array button_status. A simple state machine

is used to handle press/release status. "ask_button" sets the

corresponding select pin low, enables the input pin and enables the

internal pullup resistor. The value is then read and handled. After

all buttons have been read, midi_buttons is called, which reads the

switch status in the "button_status" array, and sends the

corresponding midi message.
#define POTI_SEL_PORT        B

#define PORTPOTI_SEL_PORT PORTB

#define DDRPOTI_SEL_PORT  DDRB

#define NUM_POTIS 64

unsigned char poti_status[NUM_POTIS];
void init_potis(void) {

  int i, j;

  for (i = 0; i  MAX_THRESHOLD) return 0x7F;

  uint32_t bla = (uint32_t)((uint32_t)adc - (uint32_t)MIN_THRESHOLD);

  bla *= 0x7F;

  bla /= 0xDFF0;

  return bla;

}
void ask_poti(unsigned char sel_pin, unsigned char adc) {

  SET_PIN_LOW(POTI_SEL_PORT, sel_pin);

  _delay_us(50);

  ADMUX = _BV(ADLAR) | adc;

  ADCSRA |= _BV(ADSC);

  loop_until_bit_is_set(ADCSRA, ADIF);

  loop_until_bit_is_clear(ADCSRA, ADSC);

  last_adc = ADC;

  SET_PIN_HIGH(POTI_SEL_PORT, sel_pin);

  unsigned char val = convert_to_cc(last_adc);

  unsigned char poti = sel_pin + 8 * adc;

  unsigned char old = poti_status[poti] & 0x7f;

  if ((val != old) && ((val == 00) || (val == 0x7F) || (abs(val - old) > 1)))

    poti_status[poti] = 0x80 | val;

}
void ask_potis(void) {

  int i, j;
  DDRPOTI_SEL_PORT = 0xFF;

  PORTPOTI_SEL_PORT = 0xFF;
  for (i = 0; i 
The code to read the potentiometers is pretty similar, except that the

routine "ask_poti" reads the value using the ADC. The diodes on the

variable pins of the potentiometers have a voltage drop of 0.6V, which

we account for by subtracting MIN_THRESHOLD from the read ADC

value. The value is then scaled to 0-127, and sent if it differs more

than one from the previous value. The value is stored into the array

"poti_status". Finally, the routine "midi_potis" sends the value of

the potis which have changed (signalled by the flag 0x80).
/*** Main ***/

int main(void) {

  io_init();
  init_buttons();

  init_potis();
  DDRPOTI_SEL_PORT = 0xFF;

  PORTPOTI_SEL_PORT = 0xFF;

  DDRA = 0x00;

  PORTA = 0xFF;

  ADMUX = 0;
  int i;

  for (i =0 ; i 
The main routine initializes the buttons, potis, I/O, waits for 100

iterations of the main loop to let the pots and switch bounce around,

and then loops asking for switch and pot status, and sending the

corresponding MIDI messages. You can find the code for the project

here
The main problem with the midi controller is that I used "cheap"

pots (though it was quite expensive to buy 56), and some of them have

short circuits or jump around or don't work at all. Another problem is

that I probably will have to convert the linear output of the pots to

something more approaching logarithmic scale because the resolution in

the high-end is too bad. Finally, another problem with the pots is

that the value sometimes oscillate due to environment noise or so, and

that using live in record mode tends to overwrite some parameter

automation envelopes. The next version will probably use less pots,

maybe some rotary encoders, though I have to find somewhere were I can

get them without being too expensive.
All in all, it was a lot of work to build the midi controller, and

you can buy the same stuff for around 120 EUR from Behringer. The next

version will have some features that are not easily found. I

absolutely want to be able to record "automation macros" directly on

the hardware. Press "record pattern A", then move some pots and knobs

for a few bars, stop the recording, and then be able to play in back

"in sync". The next midi controller I am building will be a bad-ass

step sequencer with those macros features and other mad stuff. I'll

post about it when it's ready and working.