System Identification via Data of Finite Wordlengths

Abstract

For system identification of continuous-time plants, we commonly use digital data converted from analog signals. Since digital variables are of finite wordlengths, they contain quantization errors. Such errors may deteriorate identification accuracy significantly, especially in the case of mechanical systems where large movement and precise positioning are required. To overcome this problem, the present paper proposes to estimate the quantization errors and thus true analog sampled signals to improve identification accuracy. Estimation of quantization errors and system identification are carried out simultaneously.