NOTES¶

Producer-Consumer — Ring Buffer Decouples Acquisition from Processing

Default architecture. The Producer writes samples to a ring buffer; the Consumer reads and processes them periodically. When acquisition outruns processing, data queues up.

Intuition¶

Ring Buffer = Bicycle Chain¶

You (Producer) keep hanging items on the chain. Someone else (Consumer) takes items off to process. The chain is the ring buffer.

Key points:

Producer never stops — even if Consumer hasn't finished the last batch
Consumer wakes periodically (default 50 ms) and processes the N most recent samples
If Consumer falls behind, new data overwrites old data (Producer never blocks)

Buffer Size Trade-off¶

Larger buffer → more history → better burst tolerance → more memory
Smaller buffer → less memory → drops data faster when consumer lags

Default: 512 samples × 28 bytes = ~14 KB. Reasonable on ESP32-S3.

Data Flow¶

Producer task (high priority)
    each tick: read ADXL355 → write ring buffer
        ─────────────────
Consumer task (every 50 ms)
    read last N samples → threshold detection → feature extraction → output

Notes¶

When to use Producer-Consumer

Default is safe. Change only if you need sub-50 ms processing intervals.

Consumer processing time must be < fill time

If Consumer takes longer than Producer takes to fill the buffer, samples drop continuously. Fix: - Reduce CONSUMER_PROCESS_SAMPLE_COUNT - Switch to DMA + Dual Core architecture