In this work, a novel approach to the usage of an autocorrelation the SNR while preserving exactly the frequencies and linefunction in order to improve signal-to-noise ratio (SNR) is presented. widths of the signal spectral components. The function also This method avoids the usual problems entailed by standard autocorallows signal weighting based on T 2 which can be useful relation function-based approaches to nonstationary signals such as when processing data in successive steps. The latter feature NMR signals. The Cadzow autocorrelation matrix approach to tranmakes a new multistep processing procedure viable by splitsient data is often not suitable for time-domain signal analysis; in ting the autocorrelated data fitting into sequential sessions fact, it does not maintain the Hankel structure of the prediction on which partial parameter retrieval is performed. matrix, which is mandatory for many linear-prediction (LP) applications. The approach presented here conserves the Hankel structure THEORY of the prediction matrix and, moreover, does not change the frequency and linewidth parameters of the signal components. Furthermore, the proposed autocorrelation-like function permits a weighting Modified Autocorrelation-Like Function of the individual components according to their T 2 decay constant. When the ergodic theorem holds, for instance, when a signal This property opens new possibilities for retrieving signal parameters is strictly stationary and the ensemble of its random realizations by LP procedures. These new procedures are applied to simulated does not contain any strictly stationary subensembles occurring 2D signals and 1D NMR measurements of phosphorus metabolites in frog muscle.