Deep learning-based Human Body Communication baseband transceiver for WBAN IEEE 802.15.6

Abstract

Recently, Wireless Body Area Network (WBAN) has revolutionized e-health-care. WBAN boosts monitoring vital signs utilizing tiny wireless sensors implanted in or around the human body. In February 2012, the IEEE 802.15.6 WBAN standard was released for low-power and short-range communication around the human body. The standard defines one medium access control layer and three different physical layers: narrow band , ultra-wideband, and Human Body Communication (HBC) layers. We are motivated by exploiting the human body as a communication medium. We propose a novel optimized architecture for the HBC baseband transceiver based on deep learning. The receiver utilizes two deep neural networks: one for frame synchronization to recover data and timing precisely and the other for the channel decoder to improve transceiver performance and reduce power consumption. In addition, low-complexity Preamble/SFD generator, Walsh modulation, and FSC spreader modules are proposed to reduce the power consumption while preserving the transceiver performance. Compared with the traditional hard-decision channel decoder, the proposed neural network decoder improves the block error rate by 2 dB. The proposed HBC transceiver supports 1.312 Mbps data rate at 42 MHz clock rate. The transceiver is implemented in RTL and synthesized on 90 nm CMOS technology. It consumes 493 pJ/bit on the receiver side and 105 pJ/bit on the transmitter side.