monitoring audio input and send to PT8211

[baritonomarchetto]

Hello, I have written a simple sketch intended to monitor a analog signal in the audio frequency range and output to a I2S DAC (PT8211).

In this phase my goal is to see the input signal outputted as-it-is.

I am using the square wave generated by my oscilloscope as input (1KHz, 3V amplitude).

The sketch works with two (main) flaws:
1 - the out wave frequency is too low (approx 300 Hz).
2 - the DAC output level is too low (0.3V).

Could anyone help me improve the sketch by solving these two issue someway? :)

(the hardware is a Pi Pico on latest Pico core)

#include <I2S.h>

I2S i2s(OUTPUT);

#define ADC_PIN A0
#define I2S_BCLK 2
#define I2S_WS   3
#define I2S_DOUT 4

void setup() {
  i2s.setBCLK(I2S_BCLK); // Note: LRCLK (WS)= BCLK + 1
  i2s.setDOUT(I2S_DOUT);
  i2s.setBitsPerSample(16);
  i2s.setFrequency(44100);
  i2s.setLSBJFormat();
  i2s.begin();
}

void loop() {
  int value = analogRead(ADC_PIN); //12 bit -> 0–4095
  int16_t dacValue = (value - 2048)<<4;//12 bit signed -> 16 bit unsigned
  i2s.write16(dacValue, dacValue); // LEFT, RIGHT
}

Thanks in advance!

[earlephilhower]

The PT8211 doesn't follow I2S standards and is a Japanese format DAC, so be sure to set the LSBJ mode on the I2S device. That'll give 2x the output volume since LSBJ is shifted from I2S.

Look at ADCInput for a way to get the ADC sampled periodically at the same HZ you're running your I2S. Your sketch will not accurately repetitively sample the ADC at the right rate since it's uncontrolled.

[baritonomarchetto]

Thanks for pointing me in the right direction Earle!

The sketch works way better now!!

It's still not perfect, but a big step in the right direction.

Now I have to test it with a real audio signal :)

May I ask: what is the "better" frequency to set I2S communication here? I set 44100, but what is the limit (and drawback) increasing it?

Is there a better strategy to make the output wave closer to the one inputted? I will use also analog reads in the final sketch and I am concerned about how analog reads will load the overall quality...

(thanks!!)

#include <I2S.h>
#include <ADCInput.h>

#define ADC_PIN A0
#define I2S_BCLK 2
#define I2S_WS   3
#define I2S_DOUT 4

I2S i2s(OUTPUT);
ADCInput adc(ADC_PIN);

void setup() {
  i2s.setBCLK(I2S_BCLK); // Note: LRCLK (WS)= BCLK + 1
  i2s.setDOUT(I2S_DOUT);
  i2s.setBitsPerSample(16);
  i2s.setLSBJFormat();
  i2s.begin(44100);
  adc.begin(44100);
}

void loop() {
  int value = adc.read(); //12 bit -> 0–4095
  int16_t dacValue = (value - 2048)<<4;//12 bit signed -> 16 bit unsigned
  i2s.write16(dacValue, dacValue); // LEFT, RIGHT
}

[earlephilhower]

You might want to do a

void loop() {
  while (adc.available()) {
    ..your read/scale/write...
  }
}

ADCInput will deliver a packet of some number of samples all at once. You'll go from 0 to (say) 128 samples available since it works with DMA buffers (no CPU). What you have will work, but will be a little slower than processing all samples at once w/the while

With onboard ADC having only ~9 bits of useful data and not great linearity, decreasing I2S speed to something like 16khz or even 8khz (if your I2S DAC supports that low) may be just fine given all the sampling noise anyway.

Increasing I2S frequency means you need to get more samples there faster. The PIO HW can do 192kHZ, but there's absolutely no reason to do so with this data and it makes it more likely you'll have hiccups in the stream should USB packets come in, etc.

[Andy2No]

@baritonomarchetto

I'm not sure the dacValue calculation is doing what you want. Suppose the read value is 2047, so value - 2048 is -1 (all 1s, because adding 1 then gives you zero, ignoring the overflow). Shifting that left 4 bits gives you -16 (4 zeros at the end, so adding 16 gives you zero with overflow). That means the smallest value that would be considered just below zero becomes almost the maximum output value - mostly 1s, except in the lsb end.

Maybe try just:

void loop() {
  uint16_t value = adc.read(); //12 bit -> 0–4095 (cast it to unsigned 16 bit)
  uint16_t dacValue = value<<4;//12 bit unsigned -> 16 bit unsigned
  i2s.write16(dacValue, dacValue); // LEFT, RIGHT
}

The problem would be more obvious with a non-square waveform, e.g. a sine wave.

[earlephilhower]

I think he does need that 2048 bias because the ADC is a 0...1 unsigned value while I2S is a -1...1 signed one. He's picking 2048 as the 0-point since it's the midpoint (modulo epsilon).

So he needs to map 0 => -32767, and 4095 => +32767

Mathing it out from his original code

#include <stdio.h>
#include <stdint.h>

int16_t calc(int32_t value) {
  int16_t dacValue = (value - 2048)<<4;//12 bit signed -> 16 bit unsigned
  return dacValue;
}

int main(int argc, char **argv) {
    printf("0: %04x\n", calc(0) & 0xffff);
    printf("1: %04x\n", calc(1) & 0xffff);
    printf("2: %04x\n", calc(2) & 0xffff);
    printf("4094: %04x\n", calc(4094) & 0xffff);
    printf("4095: %04x\n", calc(4095) & 0xffff);
    return 0;
}

Gives

0: 8000
1: 8010
2: 8020
4094: 7fe0
4095: 7ff0

0x8000 == -32768
0x7ff0 == 32752

Which is good enough for this case...

[earlephilhower]

(but I now see the comment is actually backwards... // 12 bit *UN*signed -> 16 bit *SIGNED*...but the logic is sound)

[baritonomarchetto]

Thanks guys for the help and sorry for the backward comment (and general inaccuracies: I am not a coder in real life, but try to do my best, especially when asking for help!)

I will test a new version of the sketch with "while" loop and lower samples as suggested by Earle and report back as soon as I will have my hardware in front of me (likely tomorrow morning).

Lurking the discussion forum a little bit more, I found a topic where a user adopted ADCInput callbacks for a task close to this. Would the use of I2S and ADCInput callbacks be a good idea here?

[earlephilhower]

For this simple of an application, it's probably not worth the effort. Callbacks run at IRQ level so they can't do many things safely.