Low-Power Human-Machine Interfaces: Analysis And Design

Human-Machine Interaction (HMI) systems, once used for clinical applications, have recently reached a broader set of scenarios, such as industrial, gaming, learning, and health tracking thanks to advancements in Digital Signal Processing (DSP) and Machine Learning (ML) techniques. A growing trend is to integrate computational capabilities into wearable devices to reduce power consumption associated with wireless data transfer while providing a natural and unobtrusive way of interaction. However, current platforms can barely cope with the computational complexity introduced by the required feature extraction and classification algorithms without compromising the battery life and the overall intrusiveness of the system. Thus, highly-wearable and real-time HMIs are yet to be introduced. Designing and implementing highly energy-efficient biosignal devices demands a fine-tuning to meet the constraints typically required in everyday scenarios. This thesis work tackles these challenges in specific case studies, devising solutions based on bioelectrical signals, namely EEG and EMG, for advanced hand gesture recognition. The implementation of these systems followed a complete analysis to reduce the overall intrusiveness of the system through sensor design and miniaturization of the hardware implementation. Several solutions have been studied to cope with the computational complexity of the DSP algorithms, including commercial single-core and open-source Parallel Ultra Low Power architectures, that have been selected accordingly also to reduce the overall system power consumption. By further adding energy harvesting techniques combined with the firmware and hardware optimization, the systems achieved self-sustainable operation or a significant boost in battery life. The HMI platforms presented are entirely programmable and provide computational power to satisfy the requirements of the studies applications while employing only a fraction of the CPU resources, giving the perspective of further application more advanced paradigms for the next generation of real-time embedded biosignal processing.