English: (A) Naive Bayes models were trained separately on LFP and electroencephalogram (EEG) features, and the predicted kinematics were obtained through decision fusion and posterior probability from Bayes classifiers. (B) The DNN, consisting of LSTM and a convolutional layer, used MWP features in multiscale to predict movement. (C) TCN was used as the decoder for predicting kinematics. (D) The LFP-based decoder used LSTM, as illustrated in the gray-shaded block.