powered by:
MagicWare, s.r.o.

Criteria for system identification with quantized data and the optimal quantization schemes

Author:Tsumura Koji, The University of Tokyo, Japan
Topic:1.1 Modelling, Identification & Signal Processing
Session:Filtering and Estimation
Keywords: identification, quantization, least-squares method, maximum likelihood principle

Abstract

In this paper, we first examine several criteria for systemidentification with quantized output data and show that the ordinaryparameter estimator for quantization-free case is still reasonableaccording to those criteria. Then, we give the optimal quantizationschemes for minimizing the estimation errors under a constraint on thenumber of the quantized subsections of the output signals or theexpectation of the optimal code length when the quantized data isencoded.