Abstract:The phase locked stimulus technology has great application prospect in the neural mechanism research and clinical treatment. However, the problem of phase locked between electroencephalogram (EEG) and stimulus need to be solved. Due to the complicated timevarying characteristics of EEG, there is still lack of effective stimulus algorithms that can be used to lock with the EEG phase. Therefore, a phase locked stimulus method for EEG is proposed, which is based on the variational mode decomposition (VMD) and autoregressive (AR) prediction. Firstly, EEG is processed by VMD to obtain multiple eigenmode signals. Then, each eigenmode signal is predicted by the AR model. The predicted values corresponding to all modes are accumulated. Finally, according to the frequency and phase characteristics of the accumulated results, the stimulus is generated, which is phaselocked with EEG. The method is evaluated in the synthesized EEG and 20 subjects (aged 20~36, male 12, female 8) offline resting EEG respectively. Results show that VMDAR can overcome the influence of EEG instability and generate the stimulus with higher phaselocked value (PLV). When the length of prediction time increases from 001 s to 04 s, PLV of opened EEG decreases from 099 to 039, and PLV of closed EEG decreases from 099 to 065. When the length of modeling time increases from 025 s to 25 s, PLV of opened EEG increases from 064 to 083, and PLV of closed EEG increases from 053 to 065. The phase locked performance of VMDAR is superior to the methods of AR and AR based on empirical mode decomposition EMDAR under all test conditions. This method can also be applied to other nonstationary closedloop phaselocked systems. .txt