1. Preprocessing. 2. Calculating Fisher ratios over discreted channel-frequency-time bins, using the training data. 3. Based on the Fisher ratios, selecting optimal time sections, selecting channels, and designing Mu and Beta passband filters in each section. 4. Doing multi-class CSP (with one-against-rest method) on each Mu or Beta band in each time section. 5. Combining all CSP features and doing classification with support vector machines. Note all parameters were fine-tuned with cross-validation. By the way, in reference to our research group, could you add another researcher, Dr Yuanqin Li, to the member list, please? I forgot to include him last time, but he has indeed done some important contributions to the work. So the details of our research group shall read Cuntai Guan, Haihong Zhang, Yuanqin Li {ctguan, hhzhang, yqli2}@i2r.a-star.edu.sg Neural Signal Processing Lab Institute for Infocomm Research, Singapore