Balancing Benefits and Risks of Embedding System Knowledge in Linear Active Disturbance Rejection Control

There are multiple ways to incorporate model information into active disturbance rejection control (ADRC) to facilitate a more reliable and efficient design. However, if this knowledge is misused, it can undermine the benefits of informed ADRC synthesis. In this paper, we provide design recommendations to guide ADRC users in selecting the control structure based on model availability, observer speed, and system characteristics. These guidelines stem from a comparative analysis of three linear ADRC variants, each utilizing a priori plant knowledge differently: the classical form, which relies only on basic plant information; the generalized version, which allows broader model incorporation; and a third approach, proposed by the authors, which modifies the generalized framework. While the latter two share the same assumed model, they differ in how it is integrated into the control architecture: the former embeds the linear drift directly into the state matrix, altering both observer and controller gains, whereas the proposed approach introduces it via a regression-based formulation that preserves the original gain structure of classical ADRC. A transfer function-based analysis shows that, even with identical models, the integration method yields distinct closed-loop properties. In particular, the proposed approach outperforms the generalized ADRC variant in cases where observer bandwidth is limited relative to plant dynamics. In this work, we focus on benchmark second-order systems, which strike a balance between generality and tractability in control design. The theoretical insights are supported by simulations and validated experimentally.