Doctoral Thesis: Low Complexity Quantized Controllers for LTI Systems: peak-to-peak Performance Guarantees

SHARE:

Event Speaker: 

Giancarlo Baldan

Event Location: 

32-D677

Event Date/Time: 

Monday, January 11, 2016 - 10:00am

Thesis committee: 

Professor Munther Dahleh (advisor)

Professor Alexandre Megretski (co-advisor) 

Professor Sanjoy K. Mitter

 

Abstract:

Optimization-based techniques to design controllers for LTI systems have been established for more than twenty years.  In this thesis, we provide a similar set of design tools that can cope with a particular communication constraint in the feedback loop.

In the first part, we propose a novel feedback control scheme for unstable LTI systems performing noise attenuation via a finite-rate digital channel. We introduce the structure of the control system as well as the encoder and decoder used to transmit the required control signals along the digital channel. The performances of the proposed algorithm are then evaluated by providing explicit bounds on the peak-to-peak (ptp) noise attenuation, in regards to the induced ptp gain of the closed loop . This result is obtained by constructing a new class of storage functions that can be employed to verify the validity of a suitable dissipation inequality for the closed loop system. 

In the second part of the thesis, we examine the trade-off between the closed loop performances and the required rate of the channel. While the digital channel imposes some limitations on the achievable induced ptp gain, we show how the performances of the proposed scheme can still approximate those achievable without communication constraints provided that the rate of the channel is large enough. A numerical optimization problem is then devised to design the parameters of the control scheme in order to minimize the strain on the channel while matching some prescribed constraints on the closed loop induced ptp gain.