Acoustic Collar - STM32 Dev and Power Studies

Last updated 8/29/2025

This repo is intended to provide a comprehensive code base and documentation for the acoustic collar project at E4E. The current branches, in descending order of date last updated, are as follows:

• cm4-active: Testing SAI. DMA move to Cortex-M4 core for dual-core wake-up system
• LoRa: LoRa integration for data uplink
• RTC: Real-Time Clock integration for data labelling
• inference (Obsolete): Testing inference
• nothing (Obsolete): Baseline for power tests

Project Details

• Target board: STM32H747I-DISCO (future move intended to smaller board)
• Input: On-board MEMS microphone
• Output: SD card write or LoRa (in progress)
• Power estimates: See blog post

System Outline

• The SAI4/BDMA collect audio from the onboard MEMS microphone and store it in a circular buffer
• Optionally, the SAI1/DMA pipe audio data to the headphones for debugging playback
• The X-CUBE-AI middleware on the Cortex-M7 core processes the audio data into a mel spectrogram and runs inference
• The SDMMC stores the inference result and audio .wav file on local SD card

Development Tools

• STM32CubeIDE
  • Import project from menu
• STM32CubeH7 library and drivers

Next Steps

• Complete Cortex-M4 integration for DSP
• Integrate LoRa (point-to-gateway) for audio uplink
• Build and attach autoencoder for audio compression
• Test inference

Checkout STM32HELP.md for useful environment and debugging tips!