Error estimates for results of nonstationary noise analysis derived with linear least squares methods

Nonstationary noise analysis of electrophysiological data is applied to the estimation of the single-channel current, i, and the number of active channels, N(C), whenever they cannot be determined directly due to limited resolution. Using least squares methods, the accuracy of estimating i and N(C) chiefly depends on the statistical error of the ensemble variance. It is shown that if the correlation among the binned data points is taken into account correctly, the variability of i and N(C) can be remarkably reduced and exact confidence limits of the parameters can be calculated. Least-squares methods are introduced which consider the measured error-covariance matrix of the binned variance in a model-independent fashion. Employing Monte Carlo methods, it is demonstrated that both the error predictions and the confidence limits are correct. The method is used to investigate the performance of nonstationary noise analysis at low channel open-probabilities. The application of the approach to simulated data as well as to experimental, i.e. non-ideal, data is discussed.