I2S input -> Codec -> ESP Now

[dodds-cc]

Overview

I’m trying to evaluate low latency audio streaming using ESP32 NOW. I'd like to experiment with tweaking the codec choice and setting to achieve the lowest high quality stream over ESP NOW. Some progress has been made from the excellent library and examples, but I’m struggling with this particular pattern



ESP32 A = EXTERNAL AUDIO SOURCE -> I2S OUT

ESP32 B = I2S IN -> FORMAT CONVERTER -> ENCODER -> ESP NOW TX
ESP32 C = ESP NOW RX -> DECODER -> I2S OUT

A -> B -> C -> External DAC

Eventually, the I2S IN will be supplied by another board , but for the sake of learning this library and how it handles the signals, I have bought another ESP32 S3, which simply runs the following

HLS-> I2S example: here. So this code is running on ESP32 A

I am using a PCM5102 DAC, detailed here.



A few positive observations so far

I have verified that the DAC is working by plugging directly into the ESP32 A. I hear clean stereo audio.


I have also successfully run this example which encodes a generated sine tone into SBC on ESP32 B and sends it over ESP NOW. This broadcast is received on ESP32 C fine and plays through the DAC as expected. Excellent. Promising stuff.



Issues receiving an I2S input stream

However, I’m now trying to modify the example to instead take an I2S stream input from ESP32 A rather than the generated sine tone. I currently try to take the incoming i2s input, convert the format to 32khz, mono, 16bit and pass into ESP NOW just like the example above.

I currently can just about make out the correct signal in the ESP32 C output to the DAC, but it is 99% noise. This suggests to me that something is right at least.


ESP32 A is running this example
ESP32 C is running this example


ESP32 B is running the following code


/**
 * @file i2s-converter-encoder-espnow.ino
 * @author Phil Schatzmann's example code + Thomas Dodds' edits
 * @brief Receive i2s audio, convert it and transmit over ESP Now
 * @version 0.1
 * @date 10-12-2025
 *
 */

#include "AudioTools.h"
#include "AudioTools/Communication/ESPNowStream.h"
#include "AudioTools/AudioCodecs/CodecSBC.h"

I2SStream i2sIn;
AudioInfo i2sInFormat(44100, 2, 16);
AudioInfo sbcFormat(32000, 1, 16);

ESPNowStream now;
SBCEncoder sbc;
EncodedAudioStream encoder(&now, &sbc); // encode and write to ESP-now
FormatConverterStream conv(i2sIn);
StreamCopy copier(encoder, conv); 

const char *peers[] = {"0C:B8:15:C3:40:68"};

void setup() {
  Serial.begin(115200);
  AudioToolsLogger.begin(Serial, AudioToolsLogLevel::Debug);

  auto cfg = now.defaultConfig();
  cfg.mac_address = "94:E6:86:04:89:44";
  now.begin(cfg);
  now.addPeers(peers);


  // I2S Input
  Serial.println("starting I2S...");
  auto config_in = i2sIn.defaultConfig(RX_MODE);
  config_in.copyFrom(i2sInFormat); 
  config_in.is_master = false;
  config_in.pin_bck = 18;
  config_in.pin_ws = 17;
  config_in.pin_data = 16;
  config_in.use_apll = true;

  i2sIn.begin(config_in);

  // Start format convertor
  conv.begin(i2sInFormat, sbcFormat);

  // start encoder
  encoder.begin(sbcFormat);
  
  
  Serial.println("Sender started...");
}

void loop() { 
  copier.copy();
}

Note: I have change the debug levels on the various boards to rule out the performance issues introduced with lots of logging. This is not the issue here.

I have searched for similar discussions and info on the wiki, but not found anything that has been trying to do quite what I have in mind.



Any advice from anyone would be greatly appreciated!

[nielsnl68]

LOL, we are want to do almost the same thing, here.
Altrough i'm using the M5stack's Atom EchoS3R for the communication. In the end i would also like to use the SBC-Encoder/-decoder with my setup.

From what i can see in the ESPNowStream source code is that there is no control over from where an audio stream is coming from. All that is stored in on buffer. this will not work with SBC. Somehow we should split-up the incoming streams in to there own pipelines but i have no clue how we could catch those within the main program.
When you have an idea @pschatzmann i love to implement that into the ESPNowStream class. I have already looked into the Inputmixer to merge the streams and that could work for me, but having the option to have for example the SBC-decoder working on every stream is still something i have no clue how that should work.

At the moment i'm reworking the ESPNowStream so it allows to also broadcast the audio instead of using multicast as it is now. And i would love to add an unicast option as well. Working on the receiving side is next on my agenda.

@dodds-cc lets work together to make an good solution for both of us.

[dodds-cc]

Yes, I have tried this arrangement in the past, not having any luck with it. I since introduced a converter too, but again, not working well.

[nielsnl68]

are you sure data is indeed send over ESP_NOW?

[dodds-cc]

yes, I observe incoming data and can hear something. A very distorted signal that contains maybe 1% of the signal.

[nielsnl68]

okay, that sounds good.
Did you also try without the SBC encoder?

[dodds-cc]

I have tried sending over raw PCM data as per this example, which works: https://github.com/pschatzmann/arduino-audio-tools/tree/main/examples/examples-communication/esp-now/pcm

But again, this is a generated signal, not real-time. What I'm looking to achieve is to replace the generated signal with an I2S input stream, whilst applying any encoder/decoder (SBC initially), to increase the audio quality.

Just checking you've also seen these useful blog posts @nielsnl68?

https://www.pschatzmann.ch/home/2022/04/27/low-latency-streaming-of-audio-data-using-esp-now/
and
https://www.pschatzmann.ch/home/2022/04/28/arduino-low-latency-streaming-of-audio-data-codecs/

These two get us to where we are right now.

[dodds-cc]

@pschatzmann Could you advise the correct pattern to take an I2S input on physical pins, pass it through the SBC encoder and then onto ESP NOW for transmission?

I've included some example code above where I am also converting the incoming I2S stream to 32000khz, mono 16bit to make its the same as your SBC codec sinetone example. Am I missing a ring buffer or queue somewhere? Do I need to wait until I have a certain amount of raw I2S data before I try to encode and send it over?

[pschatzmann]

No need to redesign anything: Just use your own receive callback to implement the mixing!

[pschatzmann]

I don't see any reason why the input from another audio source should not work:

• First test with a simple sketch if input is working correctly!
• If this is the case you can try to use some other codecs: L8 might be a good starting point.