Comparison of the decoding effects of different LFP signals in continuous reconstruction of unidimensional limb trajectory

Continuous decoding of gait through EEG signals is one of the important applications of brain-computer interface tech-nology, and it is also of great significance to applications such as smart prostheses. The local field potential (LFP) signal has become the mainstream research direction because of its important motional information contained. To achieve good effects on decoding motional signals of the rat's knee, we applied Short-Time Fourier Transform (STFT) feature extraction method to reduce data di-mensions instead of decoding raw LFP signals by SVM. To further improve the reconstruction effect and explore decoding mecha-nism, processed LFP signals are divided into different frequency sections, and the significance of different signals for the recon-struction of motion trajectory is verified. We chose the correlation coefficient (r) between predicted and real signal as a standard, and the result shows that raw LFP data presents r of 0.01208, while using of STFT improve it to 0.33779. By using finer frequency band signals, we found that low frequency delta shows r of 0.44215 while high frequency high gamma shows r of 0.19299 In conclusion, a time-space feature extraction can improve decoding effect lower limb motions, and the lowest frequency data shows better per-formance than higher frequency LFP data.