A mathematically simple hybrid of the unscented Kalman filter and the genetic algorithm (GA) is presented and applied to the non-ideality estimation in sigma-delta modulators. Parameter estimation is a complicated task, especially if a system must be observed continuously and its internal states have to be tracked in addition. A GA is a low-cost method to find an optimal parameter set but if the system is vastly changing, it cannot be applied. In contrast, the basic Kalman filter is an effective state estimator but cannot be used to estimate parameters of a system without complex mathematical extensions. A combination of both techniques can be beneficial to enable fast and especially lowcost on-chip estimation procedures.Introduction: In case the internal parameters of a dynamic system must be observed during runtime, its internal states must be continuously tracked in order to enable parameter estimation. This turns the estimation task into a nonlinear problem because the system parameters become time variant [1]. Such an estimation problem must be solved for many technical applications e.g. in automotive and aeronautics fields.We present an innovative solution to the identification task focused on a modern data conversion system, namely a sigma-delta (ΣΔ) analogue-to-digital data converter (ADC). Such an ADC is used for high-resolution and wide-bandwidth applications and is affected by nonidealities, originating from manufacturing tolerances and run-time effects [2]. As deviations from the ideal behaviour noticeably affect the performance and stability of the converter, it is often beneficial to recalibrate the components of the ADC during operation in order to restore its ideal characteristics [3]. Therefore, the non-ideal parameters in the ΣΔ ADC's filter must be estimated reliably, quickly and precisely.We recently demonstrated that an unscented Kalman filter (UKF) solves the parameter estimation task very well [4]. However, for an application where the estimator must be implemented on-chip, the approach is not applicable because the UKF requires many complex calculations to operate and the size of its digital realisation would make the technique impractical. We also presented a pure genetic algorithm (GA) solution [5] which reduced the complexity of the estimations a lot. However, the system states were not tracked continuously but only an initial value was estimated and a model was simulated from this point on. Hence, the robustness and accuracy were not as good as from the UKF due to missing relationships of the dynamics and parameters.We propose a hybrid of the GA and the UKF for state and parameter estimations. A GA is powerful to heuristically find a set of parameters with extremely low computational effort but it cannot work on a state estimation task. In the proposed hybrid, we can replace the most costly part of the UKF by a simple GA and it is shown that the solution operates quickly and accurately on a complex example.