Abstract

Brain-Computer Interface (BCI) system has become a topic of study, with numerous interests from the researchers, over recent decades, especially after the elaboration of the machine intelligence methods‎. Application of BCI systems has been increasingly expanded from rehabilitation of the people with mental traumas such as cerebral stroke ‎to diagnosis of psychological and neural disorders. Extraction informative contents of brain activities and relating them to the important brain functions such as memory mechanism, the function of the motor and sensory units of the brain along with their interconnections for the perception purposes, are still unrecognised and considered as the research gaps. The appearance of virtual reality-based systems, although it helped the researchers to obtain a better understanding of the brain function, failed to be adopted to the real-world scenarios. Augmented Reality (AR) has been introduced as an adaptation of virtual reality to respond to such an important research question of the real-world scenarios. However, the vital bottleneck in the development of such the AR-based systems is the extraction of the informative contents of brain signals reflected by ElectroEncephaloGram (EEG). In these applications, the EEG signals are synchronously recorded from different areas of scalp, EEG channel, and the temporal variation of all the channels provide a temporal-spatial demonstration, which carries indicative information of the brain activities. The extraction of this information is, however, a very complicated task, especially considering that another dimension of spectral contents of each channel is also associated with the analysis. Recent developments in machine learning methods opened the way to learn the signal characteristics of extracting medical information from EEG. Thereby, in this project proposal, we aim at the following goals:
• To automatically extract indicative metrics to score the capability of different brain parts, including memory, visual, auditory, and motor cortex, based on applying advanced machine learning methods to AR-assisted multi-channel EEG signals.
• To explore the deviation of the metric from their normal ranges at the presence of different mental disorders such as memory malfunction. Metrics.
• To interpret brain commands from EEG signals using an AR-assisted system.
• To extract informative contents of multi-channel EEG signal related to the memory action, emotion, visual and auditory action, sleep phases, dreams, and sensory actions, using an AR-assisted system.
• Interpretations and deployment of neuro-feedback in an AR-assisted system for different BCI applications such as diagnosis, patient monitoring and rehabilitation.

Moreover, when the number of channels increases, the computational throughput performance degrades. Due to the rapid advancement of the device fabrication, state-of-the-art microelectrode array can provide more than ten thousand recording channels simultaneously and provides simultaneous stimulations. Although this advancement enables encouraging prospective to BCI, it also demands innovation from the computational compartments that can catch up with the significant bandwidth from the recording front-ends. Field programmable gate arrays (FPGAs) provide massively parallel computing resources and are suitable for real-time and high-performance applications. It can give substantial flexibilities for real-time multi-channel recording systems, in which parameters, such as the dimension of neuronal spikes and data word length, need to be tunable and adaptable for various circumstances. Compared to application-specific integrated circuits (ASIC), system-on-chip (SoC), and programmable processor implementations, FPGAs provide high performance, good scalability, and flexibility at the same time. These are crucial criteria for multi-channel neuronal signal recording systems. Therefore, we will also aim to develop an FPGA-based system that integrates efficient spike analysis algorithms into silicon devices to enable a scalable system with reliable performance.