Part of State-of-charge.
Extended Kalman filter doesn't work well at cold temperatures because of non-linearity of the model.
Beware: Hall effect-based current sensors are prone to DC bias. But even shunt resistance-based sensors can have bias, too. But Kalman filters (both extended and Sigma-point) assume the process and sensor noice with zero mean. Adding a fictitious noise current component to the Sigma-Point Kalman filter state solves this problem.
Estimating State-of-Charge with Kalman filter is not impacted significantly by the Cell capacity fade and the increase of the Cell internal resistance (the symptoms of Cell degradation) because Kalman filter can adapt.
When doing dual or joint estimation of state-of-charge and Cell capacity we can make the Kalman Filter more robust by including a "state-of-charge" output alongside voltage to the model, and using the state-of-charge computed using a very simple cell model (accounting for only Cell open-circuit voltage and Cell internal resistance, but not Cell voltage hysteresis and Cell diffusion voltage) as the "noisy measurement" (with sufficiently big variance). The cell internal resistance in this simple model should come from Estimating cell internal resistance using current jumps. This simple model is guaranteed to output state-of-charge that has physical sense (albeit with big variance because we don't account for hysteresis and diffusion voltage).
This secondary output in the Kalman Filter ensures that the dual or joint estimation doesn't diverge from physically meaningful results.
This secondary model assumes that the OCV relationship doesn't change much as the cell ages (or, it should use different OCVs depending on the previously estimated cell capacity). It can also be tedious to remove current sensor bias consistently, because it also affects the internal resistance estimation.
Alternatives to estimating state-of-charge with a Kalman filter:
‣