ESP32 I2S Data Stream Problem

[OllieK]

I am using an oscilloscope to investigate the problems I have when driving a 16x16 WS2812B LED array with ESP32 Pico D4 using I2S interface. Occasionally I am seeing a gap in the beginning and end of the color bit stream. All the bits are there, but it seems that the I2S DMA goes to sleep for 4..6 us without an explanation. Obviously the LED array will fail with the extra gaps in the stream.

First the application sends 256 zero bytes to reset all the LEDs before starting with the color information. My conclusion is that the problem is in the ESP32 hardware or IDF libraries.

In the test setup, the software sets the zero bytes and constant color bytes in a buffer and sends them out once per second with the following statement

Code: Select all

esp_err_t err = i2s_write(I2S_NUM, colorBuffer, bufferSize, &byteCount, 10);

The I2S is initialized with

Code: Select all

    i2s_config_t i2s_config = {
        .mode = I2S_MODE_MASTER | I2S_MODE_TX,
        .sample_rate = BIT_RATE / SAMPLE_SIZE2,               // = 69444
        .bits_per_sample = I2S_BITS_PER_SAMPLE_16BIT,         // = 16
        .channel_format = I2S_CHANNEL_FMT_RIGHT_LEFT,           
        .communication_format = I2S_COMM_FORMAT_I2S,
        .dma_buf_count = 2 + LEDBUFFSIZE / MAX_DMA_BUFFER,    // min = 2, max = 128
        .dma_buf_len = MAX_DMA_BUFFER,
        .use_apll = false,
        .intr_alloc_flags = ESP_INTR_FLAG_LEVEL1};
    esp_err_t stat1 = i2s_driver_install(I2S_NUM, &i2s_config, 0, NULL);
    
    i2s_pin_config_t pin_config = {
        .bck_io_num = I2S_BCK_IO,
        .ws_io_num = I2S_WS_IO,
        .data_out_num = I2S_DO_IO,
        .data_in_num = I2S_DI_IO};
    esp_err_t stat2 = i2s_set_pin(I2S_NUM, &pin_config);

It is possible that I am overlooking something obvious, but the variations in the scope picture are difficult to explain. The scope picture shows that the WS clock works perfectly OK and indicates the left and right channels.

If the gaps would be there all the time with the constant data, then the problem could be in my code or ESP IDF 4.1, but now it probably is in the hardware.

[ESP_Sprite]

Could be that the driver services some of the longer requests using interrupts. If so, you may be seeing an instance of interrupt latency that happens to be somewhat longer than your DMA buffer. You can try increasing MAX_DMA_BUFFER, that might solve the issue.

[OllieK]

The system constant MAX_DMA_BUFFER (1024) is the maximum value for i2s_config.dma_buf_len. It cannot be increased. My application allocates buffers and DMA for 256 LEDs, but in the debugging tests I did use only 3 LEDs. In the background, there are no major interrupts.

If I cannot solve this, I need to try SPI or RMT for the WS2812B LEDs. Do you have a recommendation?