Nowadays, Brain Computer Interface has an important role in the life quality of parallelized people. However, this technique is mainly affected by the quality of the recorded signal in each trial. This problem could be solved by rejecting low-quality trials. But developing the processing based on the recorded signal from the brain, which is a mixture of the target signal plus noise and artifact, would not be favorable in situations that all trials have low quality. This paper solves this problem by presenting a new fast algorithm for separating recorded source signals. Results indicate the improvement in classification accuracy of the proposed method compared with the classification accuracy of processing on the recorded mixture signal.