Fast auditory evoked potential extraction with real-time singular spectrum analysis

The fast and real-time extraction of precise auditory evoked potential (AEP) is important for the diagnosis of auditory diseases as well as for the brain–computer interface. For fast AEP extraction, the effect of various noises such as motion artefact, eye blink, or power line noise should be minimised in the AEP recording. The existing fast AEP extraction methods that use the Kalman filter or wavelet transform have limitations owing to parametric arbitrariness. In this study, the singular spectrum analysis (SSA) in the time domain was adopted and optimised for AEP extraction with various types of noises. Moreover, the hardware architecture for an optimised SSA was implemented in an FPGA to realise real-time operation. The results show that the optimised SSA method can reduce the stimulus repetition by 61.2% compared with the conventional ensemble averaging and obtain the maximum similarity to the original AEP signal of 83.2%. The designed hardware is favourable for wearable BCI applications in terms of hardware complexity and required clock frequency.