The paper presents a novel semi-blind equalization method for time-varying channel. The complex exponential basis expansion model is introduced to describe the time-varying channel taps. The new method adopts the iterative-Newton based equalization algorithm which combines the independent component analysis criterion with soft-decision directed criterion to remove the inter-symbol interference. What’s more, the method reduces the iterative computation burden by making use of the complex exponential basis expansion model structure. A small number of training symbols are utilized to provide a rough initial estimate of the equalizer’s weight vector and decide if the iteration should go on. Compared with existing ones, the proposed algorithm leads to better equalization result and appears more robust to noise. Simulation results prove the efficiency of it.