The outstanding bit error rate (BER) performance of turbo codes (iterative decoding of parallel concatenated convolutional codes with interleavers) has been reported, but the computational complexity and the decoding delay have been problems in many ways. This paper presents a method can be used to reduce the decoding computational complesity. The received signal to noise ration (SNR) is varying with the condition of channel. When SNR is relatively high, the decoding computational complexity can be reduced without degradatin of BER performance. To reduce the decoding complexity we choose the method to remove the learning sequence of Viterbi sliding window method which is widely used for preventing pipeling delays. That is, using adaptive sliding window method, we can reduce about 18% of the decoding complexity without degrading BER performance by applying to CDMA cellular system and fading channel. And the modification of Log-MAP algorithm, which can be calculated without look-up table, is also considered.