Research On P300-Speller Technique For Adaptive Optimization Of Visual Stimulus Onset Asynchrony (SOA)

As a classical brain-computer interface(BCI)paradigm based on event-related potentials(ERP),P300-Speller has been reported in many clinical trials.The performance of the P300-Speller is mainly dependent on the stimulus coding efficiency and electroencephalography(EEG)decoding efficiency.Stimulus onset asynchrony(SOA)is an important factor affecting stimulus encoding efficiency,but there are few researches on SOA.Therefore,this article explores the optimization strategy of P300-Speller system from SOA perspective to develop a new technology path.In this paper,we proposed a P300-Speller optimization strategy based on dynamic stoppingstrategy and adaptive SOA adjustment,so that the system can automatically adjust the instruction coding speed according to the real-time operation of individuals.The online experiment results of 18 subjects showed that the ERP classifier and the dynamic stop algorithm established at 200 ms SOA can maintain the recognition performance under various SOA conditions ranging from 50 ms to 300 ms.The system can then reduce the SOA from an initial 200 ms to an average of about 98.5 ms while maintaining letter output accuracy.The average theoretical information transfer rate(ITR)was significantly improved from 42.4 bit/min to 85 bit/min(the maximum was232 bit/min).These results demonstrate that the adaptive dynamic adjustment strategy established in this paper can optimize the SOA personalized settings of the subjects and effectively improve the system performance of P300-Speller.This article furtherdiscussed the recognition effect of the ERP classifier based on convolutional neural network(CNN)for dynamic SOA system from the perspective of decoding efficiency,and analyzed the identification performance and migration effect of CNN model in the face of individual differences and SOA differences.The results showed that the CNN model based on the one-dimensional convolution kernel has good generalization ability and easy migration,and the cross-individual letter recognition rate reaches about 80%.Moreover,this model can realize model migration across individual and cross-SOA conditions on the data volume less than 10 letters,and the letter recognition rate can be improved to more than 90%.In summary,the work in this paper showed the feasibility of improving the efficiency of P300-Speller system through SOA optimization,which provides a certain technical basis for the development of efficient and stable P300-Speller system.