Bidirectional I2S

[relic-se]

New class, audiobusio.I2S, which can be used to both record and playback audio over a unified I2S bus (single pair of bit clock and word select pins).

Notes:

• This functionality currently relies on PIO and is only functional on RP2xxx boards. I am not familiar with the architecture of other boards which implement the audiobusio module: Atmel SAMD, Espressif ESP32, or i.MX RT1020. Support for these boards (if possible) would likely have to come in later updates.
• The interface is mostly synonymous with audiobusio.I2SOut except that data pin direction needs to be specified in keyword, ie: data_out=..., data_in=....
• Keyword arguments are available to specify sample_rate, bits_per_sample, channel_count, etc, but are only necessary if utilizing input. Otherwise, the bus will use the parameters of the sample argument. If the bus is bidirectional and a sample is played that does not match the input parameters, a relevant ValueError will be thrown.
• The audio dma system has been modified to allow bidirectional or input-only operation.
• Testing has been conducted with a Pimoroni Pico Plus 2, an ICS43434 I2S microphone and a PCM5102A I2S DAC.

Comments:

• I worry about delays potentially caused by waiting on the input buffer during audiobusio_i2s_get_buffer.
• I'm calling RUN_BACKGROUND_TASKS when using the bus as an input only while waiting for the incoming buffer. When the bus is operating bidirectionally, it waits with a sample-length sleep loop. There is likely a better solution out there.
• I've created a separate audiobusio.I2SIn that I was testing with initially. I plan on removing this class before marking the PR as ready.

Todo:

• Support mono input with stereo output.
• Fix mismatched buffer sizes.
• (optional) Add record method (or similar) to get current input buffer. Could be useful for monitoring input level.

[relic-se]

I've located a new bug. When using a buffer size on audiomixer.Mixer or an audio effect that is less than that of audiobusio.I2S, the audio distorts significantly.

import board
import audiobusio
import audiomixer

codec = audiobusio.I2S(
    bit_clock=board.GP0,
    word_select=board.GP1,
    data_in=board.GP2,
    data_out=board.GP3,
    sample_rate=22050,
    buffer_size=1024,
)

mixer = audiomixer.Mixer(
    voice_count=1,
    sample_rate=22050,
    buffer_size=512,
)

codec.play(mixer)
mixer.play(codec)

[gamblor21]

Was this the PR you were looking at some help to get past some technical hurdles? I think I may have time to help if you wanted me to poke at this as well. I will have to get some hardware that can do I2S input still.

[relic-se]

Was this the PR you were looking at some help to get past some technical hurdles? I think I may have time to help if you wanted me to poke at this as well. I will have to get some hardware that can do I2S input still.

There have been a couple of changes to audio dma recently. I likely need to revise a lot of this before pushing for a review. I also think the recent changes to garbage collection may improve this feature and was likely why the performance was unacceptable in the first place.

As for compatible hardware, there are a few options:

• I've been using the WM8960 breakout by SparkFun, but the Adafruit Voice Bonnet would be a good option paired with either a Pico to Pi HAT converter or Waveshare RP2040-PiZero.
• The simplest option would be to use an I2S microphone and share the same bus as your DAC.
• I'd like to eventually make a breakout for the TLV320AIC310x, since I think that would be a good candidate and similar to the one on the Fruit Jam.