Skip to main content

Sensors, sampling, coordinates, batching, and battery

Android sensors include physical sensors (accelerometer, gyroscope, magnetometer, pressure) and composite/software sensors (gravity, linear acceleration, rotation vector). Availability, range, resolution, power, and quality vary by device.

Discover; never assume​

val manager = context.getSystemService(SensorManager::class.java)
val accelerometer = manager.getDefaultSensor(Sensor.TYPE_ACCELEROMETER)
    ?: return SensorCapability.Unavailable

val metadata = SensorMetadata(
    resolution = accelerometer.resolution,
    maxRange = accelerometer.maximumRange,
    powerMilliAmps = accelerometer.power,
    minDelayMicros = accelerometer.minDelay,
)

Manifest <uses-feature required="false"> can advertise optional capability while allowing installation on unsupported devices.

Sampling is a request​

manager.registerListener(
    listener,
    accelerometer,
    20_000,   // desired sampling period: 50 Hz
    500_000,  // maximum report latency: allow batching
)

Delivery rate may differ. Use SensorEvent.timestamp for actual sample timing; do not assume constant intervals or use callback arrival wall clock. Very high rates can require specific permission and materially increase power/CPU.

Batching lets hardware collect samples and wake the application less often. It increases latency but can save battery. A step-history recorder may batch; an interactive game cannot tolerate the same delay.

Coordinate system​

Most motion sensors use device-natural coordinates:

  • +X toward the device’s right;
  • +Y toward the natural-orientation top;
  • +Z outward from the screen.

Axes do not automatically rotate when display orientation changes, and a tablet’s natural orientation may be landscape. Use display rotation and SensorManager.remapCoordinateSystem when mapping readings to UI coordinates.

Raw acceleration is not movement alone​

Accelerometer includes gravity. For motion acceleration, use the linear-acceleration sensor where available or filter/fuse signals deliberately.

// Simplified low-pass gravity estimate—not production sensor fusion.
gravity[i] = alpha * gravity[i] + (1 - alpha) * event.values[i]
linear[i] = event.values[i] - gravity[i]

Filters introduce lag and depend on sample rate. Orientation from accelerometer/magnetometer is affected by magnetic interference and motion; rotation vector is often a better fused signal.

Lifecycle and threading​

Register only while the owning feature needs samples and unregister symmetrically. Copy values needed beyond callback because event objects/arrays are framework-managed. Move heavy processing off callback threads using a bounded queue and defined overflow strategy.

Accuracy and calibration​

Watch accuracy changes but do not pretend a status value is a calibrated guarantee. Define product thresholds, smoothing, warm-up, outlier handling, and user calibration guidance. Test walking, vehicles, vibration, different mounting orientations, screen rotation, and low-power modes.

Sensor-derived facts are probabilistic

“Phone moved,” “user fell,” or “device faces north” are classification decisions with false positives/negatives. Preserve confidence and design confirmation/escalation appropriate to harm.

Sources​

Exercise​

Design a fall detector: sensor choices, sampling/batching, feature extraction, false-positive confirmation, background lifetime, battery budget, privacy, and emergency failure behavior.