This paper presents an off-line algorithm for dy-namical time delay recovery for which the whole observation block is used. The time offset varies over the observation interval following a random walk model. The proposed synchronizer applies to data-aided (DA), non-data-aided (NDA) and code-aided (CA) modes. Theoretical performance of the off-line technique is derived and compared to simulation results. The Bayesian Cramer-Rao Bound (BCRB) is also evaluated for DA, NDA and CA modes and for both the off-line and on-line scenarios. Simulation results show the improvement brought by the off-line and the CA schemes. The presented algorithm outperforms the conventional on-line estimator, which only takes into account the current and previous observations, and its MSE approaches the BCRB.