Skip to main content
Open Access Publications from the University of California


UCLA Electronic Theses and Dissertations bannerUCLA

Disturbance Rejection and System Identification Using Multi-Channel Adaptive Filtering and Receding-Horizon Control


The research presented in this dissertation explores the application of a receding-horizon adaptive controller to increase the disturbance rejection bandwidth far beyond the capabilities of a classical LTI controller. The receding-horizon controller is applied to two laboratory experiments to analyze and test various aspects of the controller. In the laser beam steering experiment a complex broadband disturbance is significantly reduced and the laser spot is held tightly on the target. Control penalty parameters are analyzed and frequency weighting is implemented in the single-input-single-output (SISO) experiment.

In the magnetic bearing experiment a steel shaft is magnetically levitated and rotated to speeds up to 7200 rpm. The emphasis in this experiment is to test the receding- horizon controller’s ability to reject time-varying disturbances and to operate in a multi-input-multi-output (MIMO) configuration. Additionally, the forgetting factor parameter, unique to the recursive least-squares (RLS) algorithm employed by the controller, is examined and analyzed in the context of disturbances with varying statistics. Furthermore,

in a collaboration project, a combination of the receding-horizon adaptive controller and a peak resonator controller is applied in a parallel structure to reduce a complex disturbance. Independence of the controllers is demonstrated and it is shown that the parallel combination is capable of reducing the disturbance to a greater degree than either controller applied individually.

Finally, many advanced controllers including the receding-horizon adaptive controller require an accurate plant model in order to operate effectively. Currently, this plant model must be generated by a one-time, offline system identification procedure in which the system is driven with a white noise sequence with no disturbance present. The collected input/output data is batch-processed at the end of the data acquisition period to generate a plant model which can then be used by the adaptive controller. Because this scenario is not always feasible in practice, a novel system identification procedure in which plant models can be generated in real-time and in the presence of a disturbance is analyzed. The online system identification process successfully demonstrates a real-time identification of a complex plant in the presence of a broadband disturbance. The resulting plant model that is generated is comparable in accuracy to the state-of-the-art subspace identification methods.

Main Content
For improved accessibility of PDF content, download the file to your device.
Current View