To remove the bad influence of residual normal moveout upon seismic data res-olution, we put forward a method which takes the following steps: Derive the model trace from stacked seismic traces by doing forward and backward predictions.

0, while encoding the residual coefficient of one block, some context-based VLC tables are set in terms of pre-prediction in order to adapt to different probability distributions of non-zero coefficients.

Our results show that the min squared error of m steps backward linear prediction for a steady and random signal equal to the energy impelled by the m steps auto regressive model.

Least-squares lattice(LSL)algorithm is presented which has a reduced computational Complexity and makes use of both a priori and a posteriori prediction errors, prior to RLS algorithm for adaptive filtering, LSL algorithm requires only O(L) operations, moreover, LSL gives an optimal filter for the given data.