Impact of Structured High-Frequency Disturbances on Linear Identification of Lateral Vehicle Dynamics.

This paper investigates the influence of weak but structured high-frequency disturbances on the linear system identification of lateral vehicle dynamics using experimental measurement data. The analyzed dataset originates from previously conducted driver-in-the-loop experiments involving free-driving and slalom maneuvers. Frequency-domain analysis confirms that the dominant vehicle dynamics are concentrated below approximately 2-3 Hz, while a weak but persistent narrow-band disturbance around 10 Hz is consistently present in the steering signal. To investigate the influence of this disturbance on identification, different disturbance-handling strategies are compared, including notch filtering, low-pass filtering, ARX, IV-ARX, and ARMAX model structures. The comparison considers prediction performance, model complexity, identified dynamics, and robustness under different measurement conditions. The results show that increasing the deterministic model order is generally less effective than either targeted preprocessing or explicit noise modeling. When the disturbance is spectrally well separated from the relevant vehicle dynamics, notch filtering combined with a low-order ARX model provides the most effective solution. If preprocessing is not possible, ARMAX models achieve comparable performance by representing part of the disturbance through the noise model. IV-ARX models are used as a benchmark to verify that the main conclusions remain valid under possible closed-loop bias. Based on these findings, a practical engineering workflow is proposed for selecting appropriate disturbance-handling strategies according to the spectral characteristics of the measured signals. The proposed methodology provides guidance for robust control-oriented identification of lateral vehicle dynamics using realistic measurement data.