PWMAudio multiple devices (specifically 2 Stereo devices)

[crp500]

The new PWMAudio/AudioBufferManager libraries look very interesting and the 2 examples work great. Well done Earle.

The documentation suggested that i could create multiple PWMAudio devices so I tried to make 2 Stereo devices using the PlayStereo example. One device on pins 0,1 and one device on pins 2,3.

I basically just duplicated the relevant calls to the PWMAudio lib that were in PlayStereo.

The result was that i only got audio out of one set of pins. Whichever set I started last with the begin() call.

I tried a few things like dropping the sample rate to 32000, trying with and without larger buffers, even overclocking and trying different pin combinations with varying degrees of success. The best i did was with low sample rate and overclocking. In that case i got sound out of both sets of pins but the set i started first sounded bit crushed or had extra harmonics added.

I want to get this working so that i can try to make 4 (or event better 8) wavetable oscillator voices for a eurorack module from 2 or 4 stereo PWMAudio devices. At the moment it works great with 2 voices (1 stereo device) but no more.

I feel like i have done something wrong in setting this up.
What is the correct way to setup multiple PWMAudio devices?

My patched version of the .PlayStereo looks like this

/*
This example plays a tune that alternates between left and right channels using a simple sine wave.

Released to the public domain by Earle F. Philhower, III earlephilhower@yahoo.com
*/

#include <PWMAudio.h>

PWMAudio pwm(0, true); // GP0 = left, GP1 = right
//add 2nd device
PWMAudio pwm1(2, true); // GP2 = left, GP3 = right

const int freq = 48000; // Output frequency for PWM

int16_t al = 0;
int16_t ar = 0;

const int notes[] = { 784, 880, 698, 349, 523 };
const int dly[] = { 400, 500, 700, 500, 1000 };
const int noteCnt = sizeof(notes) / sizeof(notes[0]);

int freqL = 1;
int freqR = 1;

double sineTable[128]; // Precompute sine wave in 128 steps

unsigned int cnt = 0;
//add 2nd counter
unsigned int cnt1 = 0;

void cb() {
while (pwm.availableForWrite()) {
double now = ((double)cnt) / (double)freq;
int fl = freqL << 7; // Prescale by 128 to avoid FP math later on
int fr = freqR << 7; // Prescale by 128 to avoid FP math later on
pwm.write((uint16_t)(al * sineTable[(int)(now * fl) & 127]));
pwm.write((uint16_t)(ar * sineTable[(int)(now * fr) & 127]));
cnt++;
}
}

//add 2nd callback
void cb1() {
while (pwm1.availableForWrite()) {
double now = ((double)cnt1) / (double)freq;
int fl = freqL << 7; // Prescale by 128 to avoid FP math later on
int fr = freqR << 7; // Prescale by 128 to avoid FP math later on
pwm1.write((uint16_t)(al * sineTable[(int)(now * fl) & 127]));
pwm1.write((uint16_t)(ar * sineTable[(int)(now * fr) & 127]));
cnt1++;
}
}

void setup() {
// Set up sine table for waveform generation
for (int i = 0; i < 128; i++) {
sineTable[i] = sin(i * 2.0 * 3.14159 / 128.0);
}
pwm.setBuffers(4, 32); // Give larger buffers since we're are 48khz sample rate
pwm.onTransmit(cb);
pwm.begin(freq);

//init 2nd device
pwm1.setBuffers(4, 32); // Give larger buffers since we're are 48khz sample rate
pwm1.onTransmit(cb1);
pwm1.begin(freq);
}

void loop() {
delay(1000);

// Send it out on the LHS
ar = 0;
for (int i = 0; i < noteCnt; i++) {
freqL = notes[i];
al = 5000;
delay(dly[i]);
}
al = 0;
delay(1000);

// Hear reply on RHS
for (int i = 0; i < noteCnt; i++) {
freqR = notes[i] / 4;
ar = 15000;
delay(dly[i]);
}
ar = 0;
delay(3000);
}

[earlephilhower]

Interesting use case! I'm not so sure PWM is good enough quality for anything as "pro-level" as you're thinking of. There's no oversampling, and at 48khz I think there's only ~10bits of precision available (due to PWM hardware frequency limitations).

Anyway, there may be a limitation in the PWM hardware module with respect to DMA requests, or I may be miscalculating some internal HW slice addresses. The PWM HW is kind of non-orthogonal due to being built of only 8 slices for ~30 IOs. I'll take a look at the datasheet and see. FWIW, yhe AudioBufferManager has all of its own state inside the instantiated object, so there's no bottleneck there that I can see.

[crp500]

Thanks Earle
I am also going to try your i2s library when some stereo decoders show up from china. Interesting to see the difference.

I think there is a small bug in your playStereo example.
You cast the PWM.write sample value to uint16_t when it looks like the pwmAudio library function for write is expecting an int16_t
I think that means you are changing all the negative values in your wavetable to positive.
When i looked on an oscilloscope i noticed that the sine wave was cut in half...only positive values.
I changed your code from this:
pwm.write((uint16_t)(al * sineTable[(int)(now * fl) & 127]));
pwm.write((uint16_t)(ar * sineTable[(int)(now * fr) & 127]));
to this
pwm.write((int16_t)(al * sineTable[(int)(now * fl) & 127]));
pwm.write((int16_t)(ar * sineTable[(int)(now * fr) & 127]));

And then i got a full sine wave on the oscilliscope.

[earlephilhower]

I changed your code from this:
pwm.write((uint16_t)(al * sineTable[(int)(now * fl) & 127]));
pwm.write((uint16_t)(ar * sineTable[(int)(now * fr) & 127]));
to this
pwm.write((int16_t)(al * sineTable[(int)(now * fl) & 127]));
pwm.write((int16_t)(ar * sineTable[(int)(now * fr) & 127]));
And then i got a full sine wave on the oscilliscope.

Would you like to do a PR to fix the example?

[crp500]

ok, i think i have created a PR.

[earlephilhower]

:) Almost. You need to go to https://github.com/earlephilhower/arduino-pico and click the New-PR option. You've got a private branch w/the fixes but it's not been submitted here. Thx!