The MLP-based and TDNN-based models were trained for AF recognition. (The output layer of example covers a total temporal context of 9 frames).