On Optimal Design Of ERP-Based Brain-Computer Interface

Event-related potential (ERP)-based brain-computer interface (BCI) is an important category of BCI implementations. It has drawn wide attention since the P300 speller paradigm was introduced by Farwell and Donchin in 1988. Based on the ERP-BCI paradigm, diverse applications have been established. The main goal of this paper is to investigate the factors that influence the design of an ERP-BCI paradigm, and improve its perfor-mance through a thorough consideration and optimization.The principle, signal processing approaches, and performance evaluation criteria are reviewed, and the design of ERP-BCI is considered from several aspects, i.e., the visual design, the stimulus code, and the stimulus timeline. These factors are clearly defined and based on which the following optimizations are carried out.Regarding the visual design, the paper focuses on a factor that has important impact on the ERP characteristics and distributions, the stimulus type. The stimulus types are categorized on two ways. According to the image properties that are changed such as color, position, size, texture, and sharpness, the stimulus types could be flash, translation, rotation, zooming, pattern changing, and sharpening. According to the way the status is changed, there are impulse mode and step mode. The comparision results indicate that except P300, motion visual evoked potentials could be evoked by the motion related types, and late positive components could be elicited by sharpening type which requires more mental effort. Choosing the subject specific stimulus type could effectively improve the performance.The stimulus code and timeline control are critical factors that determine the clas-sification efficiency of ERP-BCI system. The paper points out that another important parameter, the item number, has not caught deserved attention. The item number is deter-mined directly by the task requirement, and has great impact on the optimization of other factors. The cube stimulus code is proposed. Under the embedding cube code design framework, the cube code is optimized for given item numbers by a cross-cube predic-tion approach through limited training. The arbitrary dimension embedding cube code consistently outperforms the classical row-column code. The optimization reveals how the code influences the ERP characteristics. The optimized stimulus code is especially valuable for practical ERP-BCI applications.One important target population of BCI is the people with motor impairments. For the people who lose the eye movement control ability, developing the gaze independent BCI paradigm is of special importance. The paper introduces the oddball visual search task, and propose a novel independent BCI paradigm. The paradigm has two advantages. On the one hand, it limits the items in a small visual angle, so the high spatial acuity of the central vision can be exploited. On the other hand, due to that the visual search task requires more mental task, the elicitation of the P300 as well as the late positive components (LPC) are enhanced. The experiments show that under the gaze fixation condition, the classification accuracies achieve 87.8% and 84.1%, for random position and fixed position mode respectively. The written symbol rate (WSR) criterion, which takes the error correction into consideration, is 1.38 symbols/min. The results outperform other gaze independent BCI paradigms that are reported by the literature.