What are the common process and measurement noises in INS/GNSS EKF, and how are they chosen?

• A. Process noise accounts for unmodeled dynamics and biases; measurement noise reflects sensor noise; they are chosen from sensor specifications, calibration data, and tuned to balance responsiveness and stability.
• B. Process noise is zero; measurement noise is infinite
• C. Process noise only depends on weather
• D. No such thing as noise

Answer: (A)

In an INS/GNSS EKF, you’re modeling how uncertain the system dynamics are and how noisy the observations are. The process noise represents uncertainties in the state propagation due to unmodeled dynamics and biases. It captures the random walk and drift of inertial sensor biases (gyros and accelerometers) and the small, unpredictable accelerations the vehicle might experience that aren’t captured by the simple motion model. The measurement noise reflects the uncertainty in the observations you fuse in, mainly the GNSS measurements (pseudorange, carrier phase, Doppler) and any other sensors you use to update the state.

How these are chosen comes from practical sensor understanding. The process noise covariance is built from the IMU’s noise characteristics (noise density and bias instability) and from a model of how biases evolve over time, then scaled to reflect how much unmodeled dynamics you expect in your motion. The measurement noise covariance uses the GNSS observation noise typically provided by the receiver (standard deviations for pseudorange and carrier-phase measurements, etc.), and it can be adjusted upward to account for challenging conditions like multipath or outages. In practice, you tune these to make the filter respond adequately to real motion while remaining stable and not overreacting to sensor noise. Weather doesn’t set these noises, and they aren’t zero or infinite; recognizing that noise is always present is key to a robust fusion.