A low-power wearable embedded system that continuously captures, buffers, processes, and transmits audio context.
Capture PDM audio continuously with hardware-level noise-aware capture.
Core 1 handles real-time audio ingestion with PDM and DMA; Core 2 handles BLE and orchestration.
Maintain a rolling 2-minute buffer in about 8 MB PSRAM and persist selected context to about 16 MB flash.
Support always-on low-power operation through efficient regulation and power-path management.
Transmit selected audio context to a phone or desktop.
Keep RF path layout-constrained and tunable for stability.
Trigger events and expose state without a heavy interface.
| Decision | Reason | Tradeoff | Consequence |
|---|---|---|---|
| Use a dual-core MCU split | Continuous audio ingestion and BLE orchestration compete for timing and power budget. | Firmware architecture becomes more complex. | Real-time capture can be isolated from communication and orchestration tasks. |
| Use rolling context with event persistence | The device should save last N minutes plus future stream without storing every low-value signal region. | Requires correct circular buffer and metadata handling. | Selective storage reduces flash usage while preserving context around meaningful events. |
PDM microphone -> DMA -> circular buffer -> compression/event trigger -> BLE offload
The device maintains a continuous context buffer, saves the last N minutes plus future stream when triggered, and filters silence or low-signal regions at the edge.
This is a full-stack embedded system, not a component demo. The system design is coherent and scalable, while firmware implementation remains future work.