Finite sample properties of system identification with quantized output data

Erik Weyer, Sangho Ko, Marco C. Campi

Abstract: In this paper, we consider system identification with quantized output data. We show that by applying an LSCR (leave-out sign-dominant correlation regions) algorithm guaranteed non-asymptotic confidence sets for the system parameters can be obtained. The results and the confidence sets are valid for any finite number of data points, and there are no assumptions on the noise. Simulation examples are provided which illustrate the usefulness of the algorithm and which also point to open research problems in quantizer and input signal design.