Algorithm Parameter
An optimization technique adjusts the statistical uncertainty matrix within a state estimation filter to match changing system dynamics. Through process noise covariance scaling, tracking algorithms can dynamically alter their responsiveness to sensor measurements versus internal prediction models. This scaling prevents the filter from diverging when the physical system undergoes rapid transitions that exceed the baseline process model.
The scale factor is calculated continuously based on the residual errors of the filter.
Filtering Optimization
Inertial navigation and sensor fusion systems experience performance degradation when the assumption of constant process noise is violated by unpredictable maneuver changes. If the assumed noise covariance is set too low, the filter will ignore actual sensor trends and lag behind the true state of the system. Implementing process noise covariance scaling allows the filter to increase the process noise parameter during highly dynamic phases.
This adjustment forces the filter to place more weight on incoming sensor data, which improves tracking accuracy.
Adaptive Tuning
The computation of the scaling factor relies on monitoring the innovation sequence, which represents the difference between predicted and measured states. When the innovation sequence consistently exceeds the expected boundary, the process noise covariance scaling algorithm increases the covariance values to expand the filter’s acceptance limits. Conversely, during steady-state conditions, the scaling factor is reduced to allow the filter to smooth out sensor noise and provide a stable output.
This adaptive mechanism is particularly useful in environments where the physical platform transitions between periods of high activity and complete rest.
System Verification
Software simulations must be conducted to verify that the scaling algorithm remains stable under extreme noise conditions. If the scaling factor increases without bound, the filter can become overly sensitive to high-frequency measurement noise. Technicians run test scenarios with simulated sensor failures to define the safety limits for the scaling factor.
These limits prevent the system from accepting corrupt data during sensor anomalies.
