Time Series Reconstruction from Quantized Measurements

This paper describes a Quantization Regression (QR) algorithm which generates a nonlinear estimate of an autoregressive time series from quantized measurements. Its purpose is to retrieve the underlying information from quantized signals such as those from the analogue to digital converter of a plant instrument. The reconstructed signals have uses in data-centric applications such as controller performance assessment and system identification. The algorithm based upon Ziskand and Hertz is a combination of the ‘Gaussian Fit’ scheme of Curry with expectation-maximization (EM) algorithm of Dempster et al. The performance of Quantization Regression algorithm is compared with two other methods in fitting of an autoregressive time series for the reconstruction of a quantized signal.