Estimation of single event-related potential with adaptive median filtering based method

Event-related potentials(ERP) is of great importance in medical research and clinical diagnosis. Estimation of single event related potential (sERP) is the goal of the ERP processing and is very difficult because of strong background noise. A new technique, adaptive median filtering based method for estimation of sERP, is presented. It can be seen that the waveform of the background noise resembles pulses in appearance. The pulse noise can be effectively removed by the median filtering. The estimation process is performed in two stages; first, the zero crossing method was used to measure pulse widths of the noises in each recorded signal, and the length of the window of median filtering was set adaptively for each recorded signal based on the maximum of all measured widths; then, the data from the odd-ball trials were processed by the median filtering and sequentially by the low pass filtering. The experimental results showed that the performance of sERP estimation was excellent. The method proposed is suitable for removing pulse noise whether it is nonstationary-colored non-Gaussion-ditributed or not. So it is robust enough to deal with sERP processing.