Deep Learning-Based Bioimpedance Spectroscopy Using Start of Frame Delimiter in Human Body Communications

Bioimpedance spectroscopy (BIS) is a frequency-based technique which has been extensively used in medical and agricultural applications. BIS involves estimation of the power spectral density (PSD), to determine the frequency response of biological material modelled as an impedance. This PSD is then used to determine other biological parameters of interest. A deep-learning-based BIS technique is proposed for body area networks (BAN) using the IEEE 802.15.6 Human Body Communications (HBC) specification. The proposed technique uses the start of frame delimiter (SFD) from the HBC physical layer (PHY) frame to perform BIS. A deep-learning-based strain gauge was developed based on this SFD-BIS framework as a proof of concept. Results show an average RMSE of around 3.87 N and an average absolute error of 1.20 N for all noise variations explored. The relatively low error showed the feasibility of this approach validating this proof of concept. This has great potential when applied to cyber-physical systems.