Meaning
Numerical instability in state estimation algorithms arises when the estimated error covariance matrix becomes unrealistically large due to unmodeled system dynamics or rounding errors. This covariance explosion can occur in Kalman filters when the model assumes a system is more predictable than it actually is, leading the filter to ignore real-time sensor measurements. If the filter relies too heavily on its own incorrect state propagation, the variance values grow unbounded.
Numerical Origin
Divergence in sequential estimation begins when the gain matrix approaches zero while the state transition matrix continues to amplify state values. During a covariance explosion, the algorithm progressively loses its ability to correct for small cumulative errors. This process occurs because the updating step does not receive sufficient correction from the measurement residue, causing the error bounds to expand.
Filter Instability
Disconnect between the actual state and the estimate causes complete failure. The algorithm ceases to provide useful feedback, producing highly erratic capacity calculations.
Mitigation Strategy
Sourcing and purchasing decisions for microcontrollers rely on algorithms that incorporate protective bounds. Preventing this growth requires the implementation of covariance reset techniques or the addition of fictitious process noise to force the filter to trust incoming voltage readings. These constraints keep the error covariance matrix bounded, ensuring long-term tracking stability.