This project captures audio from an I2S microphone using the Raspberry Pi Pico W. It utilizes the RP2040’s programmable I/O (PIO) to implement an I2S receiver and transfers captured audio to memory via DMA. The system is built using Rust in a no_std environment and structured around the async Embassy framework.
- I2S Audio Capture via PIO + DMA
- Async Task Architecture using Embassy
- Double Buffering for continuous data flow
- Planned SD Card Storage of audio as
.wavfiles
At the heart of the system is a custom PIO state machine implementation that samples I2S input and streams it into memory through DMA. The PIO program implements left/right channel synchronization and clock signal decoding using the RP2040’s side-set and loop control features.
The custom PIO program cycles between left and right channel reads based on the LR clock, capturing 24-bit samples per channel:
.side_set 2
set x, 14 side 0b01
left_start:
set x, 23 side 0b00
left_loop:
in pins, 1 side 0b01
jmp x--, left_loop side 0b00
set y, 7 side 0b00
dummy_left:
nop side 0b01
jmp y--, dummy_left side 0b00
set x, 23 side 0b10
right_loop:
in pins, 1 side 0b11
jmp x--, right_loop side 0b10
set y, 7 side 0b10
dummy_right:
nop side 0b11
jmp y--, dummy_right side 0b10
jmp left_start side 0b00This sequence ensures 24 bits per channel are captured with dummy cycles for synchronization and alignment.
The PioI2sMic struct sets up the state machine, configures PIO pins, and establishes shift-in behavior:
ShiftConfigis set to auto-fill 24-bit samples shifted left.- Clock Divider is computed as:
PIO Clock = Sample Rate × 64 (BCLKs/frame) × 2 (phases)The divider adjusts PIO timing to match the I2S BCLK requirement. FifoJoinis configured to RxOnly for unified FIFO access.- DMA Transfers are initiated through
dma_pull()to transfer RX FIFO contents to memory buffers.
Embassy uses a cooperative, single-threaded async runtime tailored for embedded systems. Each #[embassy_executor::task] function is compiled as a state machine that yields execution when awaiting, allowing other tasks to run. The Spawner in main schedules these tasks onto the executor. Under the hood, tasks are polled in a loop, and context switching is minimal—there's no preemption or thread stack switching. This approach makes task execution predictable and efficient, ideal for real-time operations like audio capture and buffer handoff. Because all tasks share the same core and run to completion between await points, timing-sensitive tasks (like DMA completion and buffer queuing) can be precisely coordinated without interrupt-heavy designs.
Captures audio data continuously:
- Waits for DMA to complete
- Swaps the ping-pong buffers
- Sends data through an Embassy channel
Receives buffers and currently logs sample values. It is designed to be replaced with logic for writing to an SD card.
- Filesystem Support (e.g. FatFs or embedded SD drivers)
- Streaming
.wavFile Writes- Handle proper WAV header creation and patching
- Write buffers in real-time
- Error Handling for write failures and card removal
Licensed under either of:
- Apache License, Version 2.0 (LICENSE-APACHE)
- MIT License (LICENSE-MIT)
at your option.
Unless explicitly stated otherwise, any contribution intentionally submitted for inclusion in this project by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.