Spatial-coded Visual Brain-Computer Interface Towards Natural Interaction

Visual brain-computer interfaces(BCIs),popular for their high performances,have successfully demonstrated their potential to be an efficient human-machine interaction channel.Nowadays,how to help visual BCIs find their place in daily life has become an important issue in the field.The spatial-coding method,which encodes targets with their spatial locations relative to the stimulus,has regained attention due to its feasibility in improving the user experience.Unlike the popular time-coding and frequency-coding methods which are developed based on the communication principle,the underlying principle supporting such a spatial-coding method is the neural mechanism of the spatial information processing.Therefore,the spatial-coding method is able to encode multiple targets with one stimulus and separate the targets from the stimulus at the same time.These characteristics enable the spatial-coding method to implement BCI applications with lower visual load and more flexible design,suggesting its potentials to make visual brain-computer interaction more "natural".This thesis investigated spatial-coding methods from three aspects: the explorations on the spatial-coded characteristics of typical visual evoked potentials(VEPs),explorations of the spatial information decoding methods,and the design and implementation of practical BCI systems.First,based on two typical VEPs: steady-state visual evoked potential(SSVEP)and motion-onset visual evoked potential(m VEP),this thesis investigated the neural representations of spatial information and the feasibility of real-time spatial information decoding.While previous spatial-coded BCI studies only utilized spatial direction information,this thesis considered visual eccentricity information as an encoding dimension for the first time,and achieved the simultaneous decoding of the direction and eccentricity.The joint decoding of eccentricity and direction information is expected to substantially increase the number of targets,by making better use of the visual spatial information.Secondly,according to the spatial characteristics of SSVEP and m VEP,this thesis proposed a series of spatial information decoding methods,from the template matching to individual-template canonical correlation analysis(ITCCA),amplitude-phase canonical correlation analysis(APCCA),and inverted encoding model(IEM).These methods reflected different levels of the information decoding from separating the responses with classifiers to modeling responses based on the neural mechanism.This thesis also proposed a multi-stimuli spatial information decoding method to enhance the practical performance of spatial-coded BCIs.Finally,this thesis proposed a series of single-stimulus,multi-target BCIs,where the reduction of stimuli provided greater flexibility in user interface design.This thesis also proposed a "dial on paper" BCI system which is able to recognize hand-written characters,providing new solutions for the classical BCI spellers.While previous visual BCI studies have mainly focused on the time-coding and frequency-coding methods,this thesis presents the first systematic study of the spatialcoding method.On the one hand,this thesis expands the dimension of spatial-coding and demonstrates the potentials of the spatial encoding method towards natural braincomputer interaction.On the other hand,this thesis also provides examples of the spatial information decoding method and the paradigm design of a practical spatial-coded BCI system.Altogether,by exploring the characteristics of the spatial-coding method,this thesis provides a new solution to achieve the natural interaction of visual BCI systems.