High-resolution multipath time delay estimation for broad-band random signals

The resolution limits to multipath time delay estimation for broadband random signals are examined. First, appropriate Cramer-Rao lower bounds (CRLBs) are derived to establish baseline performance for unbiased estimation. The bounds are then compared to computer simulation results. For the two-path case, a maximum-likelihood estimator is implemented, while for the three-path case, the modified forward-backward linear prediction algorithm developed for high-resolution frequency estimation is used. It is shown that both of these techniques can achieve performance close to the appropriate CRLB; in particular, reliable multipath estimation can be achieved even when the multipath time delays are much closer than could be resolved by conventional autocorrelation processing. A criterion is developed for predicting when the bound is no longer reachable. It is shown that even relatively poor a priori knowledge of the attenuation can significantly help the estimate of time delay for small delays.>