Compressive Sensing with Low-Power Transfer and Accurate Reconstruction of EEG Signals

Tele-monitoring of EEG in WBAN is essential as EEG is the most powerful physiological parameters to diagnose any neurological disorder. Generally, EEG signal needs to record for longer periods which results in a large volume of data leading to huge storage and communication bandwidth requirements in WBAN. Moreover, WBAN sensor nodes are battery operated which consumes lots of energy. The aim of this research is, therefore, low power transmission of EEG signal over WBAN and its accurate reconstruction at the receiver to enable continuous online-monitoring of EEG and real time feedback to the patients from the medical experts. To reduce data rate and consequently reduce power consumption, compressive sensing (CS) may be employed prior to transmission. Nonetheless, for EEG signals, the accuracy of reconstruction of the signal with CS depends on a suitable dictionary in which the signal is sparse. As the EEG signal is not sparse in either time or frequency domain, identifying an appropriate dictionary is paramount. There are a plethora of choices for the dictionary to be used. Wavelet bases are of interest due to the availability of associated systems and methods. However, the attributes of wavelet bases that can lead to good quality of reconstruction are not well understood. For the first time in this study, it is demonstrated that in selecting wavelet dictionaries, the incoherence with the sensing matrix and the number of vanishing moments of the dictionary should be considered at the same time. In this research, a framework is proposed for the selection of an appropriate wavelet dictionary for EEG signal which is used in tandem with sparse binary matrix (SBM) as the sensing matrix and ST-SBL method as the reconstruction algorithm. Beylkin (highly incoherent with SBM and relatively high number of vanishing moments) is identified as the best dictionary to be used amongst the dictionaries are evaluated in this thesis. The power requirements for the proposed framework are also quantified using a power model. The outcomes will assist to realize the computational complexity and online implementation requirements of CS for transmitting EEG in WBAN. The proposed approach facilitates the energy savings budget well into the microwatts range, ensuring
a significant savings of battery life and overall system’s power.
The study is intended to create a strong base for the use of EEG in the high-accuracy and low-power based biomedical applications in WBAN.