Abstract: Measures should be taken to overcome the time-variability of underwater acoustic channel, which will degrade the performance of underwater communication systems. Per-survivor processing is an approximation of maximum likelihood sequence estimation, and it provides a solution to carry out channel estimation and information detection at the same time without using training data. An algorithm is proposed in this paper that the channel estimations are updated depending on the received data. By comparing the current costs in different survivor paths, the variation of channel is measured, hence whether to update the channel estimation or not is decided. Simulation result shows that the per-survivor processing algorithm based on data-depended updating of channel estimation is a good solution to underwater communication system in time-varying channels.