Efficient Autoencoder-Based Human Body Communication Transceiver for WBAN

Abstract

Human Body Communication (HBC), which utilizes the human body as a communication channel, is a promising communication method for wireless body area networks. In this paper, we use the deep-learning based approach to design and implement a new optimized architecture for HBC system with scalable date rates feature. The proposed transceiver is completely implemented using two deep neural networks, one represents the autoencoder for the transmitter and receiver, and the other for frame synchronization. The proposed autoencoder-based HBC improves block error rate by 2 dB compared to the conventional HBC design. In addition, low complexity modules for CRC encoder and decoder, Scrambler and Descrambler, and Preamble/SFD generator are proposed. Implemented under 45nm CMOS technology, the core size of the proposed design is 0.116 mm2, and the estimated power is 1.468 mW with a peak data rate of 5.25 Mbps. The energy efficiency (Eb) of the proposed design is 280 pJ/b that is over 3.5x better than the conventional HBC designs in literature.

Keywords

• Machine learning
• human body communication (HBC)
• WBAN
• IEEE 802.15.6
• deep learning
• low power