Skip to content

0002. Multi-rate estimation pipeline driven by the IMU FIFO

  • Status: Accepted
  • Date: phase 1 (recorded on 2026-10-27)

Context

The gyroscope has to drive the propagation and the rate loop at 960 Hz. The accelerometer is noisy and vibrates, so it does not need to be used at 960 Hz. The magnetometer is slow, and it is read through the sensor hub. Timestamps from separate reads would add jitter.

Decision

  • The LSM6DSV16X FIFO holds gyro, acc, magnetometer and (later) barometer samples in the order they were taken. One DMA burst per interrupt reads them all.
  • est_pipeline.c handles the samples:
    • propagation every gyro step (960 Hz);
    • accelerometer update every 4 steps (240 Hz), on a rotation-compensated window mean;
    • magnetometer update when a new sample arrives, with its age compensated.
  • The sample period is measured against the MCU crystal, not assumed.

Consequences

  • One time base for every sensor, with no per-sensor timestamps.
  • The rate loop hook runs right after the propagation, before the corrections, which gives the lowest latency.
  • The pipeline is pure C, so the same code runs in the PC simulation, in the host tests and on the board.
  • Cost: the INT1 level on an edge-triggered EXTI needs the "read again at once" recovery (see Troubleshooting).

Alternatives considered

  • One update per sample at 960 Hz. It costs CPU, and it feeds vibration into the attitude.
  • Separate tasks per sensor. They need locks and timestamps, and they bring jitter.