Comparison of discrete subsample time delay estimation methods applied to narrowband signals

The problem of estimating the subsample time delay between two noise-corrupted narrowband signals is considered. Significant improvements to two previously proposed estimators, the MXS and MSX, are made. The performance of the revised estimators is assessed through simulation over a range from -10 to 20 dB, with four well known estimators, two of which are asymptotically maximum likelihood (the direct correlator and the average square difference function) and the other two are windowed versions of them (the generalized cross-correlator and generalized least squares). Also, the estimators are applied to ultrasonic experimental data at three different SNRs. At low SNR, the MXS/MSX estimators are shown to exhibit a significant reduction in the variance of the estimates.