TinyMeasurement · Real-time Acquisition · DMA Double Buffer — Design notes¶

Actual application acquisition chain¶

flowchart TB
    T[Timer callback] --> N[Notify producer]
    N --> S[Read ADXL355]
    S --> B[Lock and write active buffer]
    B --> F{Buffer full?}
    F -->|Yes| R[Mark ready and switch buffer]
    R --> C[Notify consumer to process and output]
    F -->|No| N

A timer callback notifies the producer task. The awakened producer reads ADXL355 acceleration and temperature, writes samples into a buffer, and notifies the consumer. Locks coordinate buffer switching and shared state. Application sampling is not triggered by a DMA-completion interrupt; SPI driver DMA is a separate layer.

Older sketches below remain as design history. Buffering and task separation overlap acquisition and processing, while memory, buses, locks and network resources remain shared. They do not guarantee zero jitter or maximum throughput. Validate actual sample intervals, dropped/overwritten samples, buffer occupancy and transmission time.

源码 / Source

Intuition¶

Why DMA is Fast¶

Normal: CPU reads sensor data byte by byte → executes instructions → occupies CPU cycles.

DMA: Configure DMA once → DMA controller reads SPI/I2C data into memory autonomously → CPU stays free to process data.

Double Buffer (Ping-Pong)¶

Two buckets A and B:

DMA fills A → A full → DMA switches to B → CPU processes A
DMA fills B → B full → DMA switches to A → CPU processes B

Producer and consumer work can overlap. Slow reads or processing can still cause waiting, overwrites or dropped samples.

Data Flow¶

DMA transfer complete interrupt
    → Buffer A full → notify Consumer task
    → DMA auto-switches to Buffer B
    Consumer reads Buffer A and processes
    → Consumer finishes → waits for next interrupt

Notes¶