<title>Hybrid multivariable controller architecture</title>

Abstract

We present a multivariable controller architecture that is a hybrid combination of a classically designed controller and an observer-based controller. The design process starts with a classical multivariable feedback controller, designed by any convenient method, such as sequential SISO loop closing. After designing the classical controller, an observer-based modern controller is designed to be stable in parallel combination with the classical controller. The hybrid configuration is realized by introducing an additional feedback path between the two feedback controllers, to subtract the effects of the classical controller from the observer-state estimate. All of the controller gains are re-tuned to improve a variety of performance measures. The additional feedback path does not increase the number of states in the controller but allows significantly higher gains to be used in the observer-based controller, resulting in better isolation from input disturbances. A six-input, nine-output lightweight space structure (LSS) provides a working example. The classical controller was implemented as six 40th-order SISO feedback controllers, at a sample rate of 20 kHz, closed in parallel around the six main mount struts, achieving very good isolation across the struts. A 240th-order observer-based modern controller, also at a 20 kHz sample rate, was designed to work with the classical closed loops and has been implemented in the hybrid configuration described. This non-square modern controller uses feedback signals from three non-collocated sensors, in addition to the six used by the classical SISO controllers, and improves isolation by about 5 dB in the most critical regions of the controller bandwidth.