Shift I2S input data by 1 bit

[earlephilhower]

Fixes #2037

[frohro]

@earlephilhower I am happy to report you fixed the issue perfectly according to my test. Just in case I made a mistake here is what I did. There are parts of this I have never done before.

1. I changed the platformio.ini to include this line: platform_packages =
   framework-arduinopico@https://github.com/earlephilhower/arduino-pico.git#i2sin1
2. I connected my pico w and built and uploaded with the -> button in VScode PlatformIO. It took a long time reconfiguring the PlatformIO framework but finally finished.
3. Then I changed my code so it was as we thought it should be, shifting 8 bits for 32 bit encoded packets or 0 bits if I am using 24 bit encoded packets. That now looks like this:

void BufferFiller::fillBuffer()
{
    while (!mutex_try_enter(&my_mutex, &mutex_save))
    {
        // Mutex is locked, so wait here.
    }
    char *buffer = queue.getNextBufferAndUpdate(true);
    mutex_exit(&my_mutex);
    if (buffer != nullptr)
    {
        static int32_t r, l, packet_number = 0;
        uint32_t bufferIndex = 4;
        while (bufferIndex < BUFFER_SIZE - 4) // Leave space for the packet number
        {
            i2s.read32(&l, &r);
            if (BITS_PER_SAMPLE_SENT == 24)
            {
                // l = l << 1;
                // r = r << 1;
                l &= 0xFFFFFF;                       // Keep only the lower 24 bits
                r &= 0xFFFFFF;                       // Keep only the lower 24 bits
                memcpy(buffer + bufferIndex, &l, 3); // Copy only the lower 3 bytes
                bufferIndex += 3;
                memcpy(buffer + bufferIndex, &r, 3); // Copy only the lower 3 bytes
                bufferIndex += 3;
            }
            else if(BITS_PER_SAMPLE_SENT == 32)
            {
                // l = l << 9;
                // r = r << 9;
                l = l << 8; // Keep only the lower 24 bits
                r = r << 8; // Keep only the lower 24 bits
                memcpy(buffer + bufferIndex, &l, sizeof(int32_t)); // Copy only the lower 3 bytes
                bufferIndex += sizeof(int32_t);
                memcpy(buffer + bufferIndex, &r, sizeof(int32_t)); // Copy only the lower 3 bytes
                bufferIndex += sizeof(int32_t);
            }
        }
        memcpy(buffer, &packet_number, sizeof(int32_t));  
        packet_number++;
        Serial.printf("Filled %d\n", packet_number);
    }
}

4. I sent it via UDP to my Linux box, received and plotted it there using a python script that was working just fine with the extra bit shift and old framework. The plot looks like this:
   
   and this is what I was expecting, as I have a sinewave at 1 kHz and 2 V amplitude feeding both left and right channels.
   That was the test.

I would appreciate your advice about when I should change frameworks, and what I should change to.

Thank you so much for fixing this bug for me!!

Rob

[earlephilhower]

Thanks for the update! With the waveform captured it was pretty obvious what was going on.

You should probably use the git version (I think you need to just drop the #xxx) until the next release to get this minor patch.