Wire-Free Wearables: Adaptive Low-Power Transceivers for Human Body Communication

Overview

Wearable and implantable sensing is limited by its radio, not by its sensor. On the body, conventional RF links suffer shadowing and multipath, radiate energy into the surrounding room, and end up dominating the power budget of a node whose battery has to be small enough to wear. Human Body Communication (HBC) removes that radio: the signal is coupled into conductive tissue through skin electrodes and travels along the wearer, cutting transmit energy by orders of magnitude while keeping the data confined to the person carrying it.

This project is the analog / mixed-signal integrated-circuit thread of the CCSL HBC programme. It is not a single demonstrator but a deliberate sequence of four taped-out generations in TSMC 65 nm CMOS — HBC-I through HBC-IV — each closing one specific gap between what HBC promises on paper and what a wearable can actually be built from: transmitter energy, interference robustness, throughput, and finally full transceiver integration with on-chip processing and sensing on a single die.

That silicon then becomes a platform for applied work: privacy-preserving in-car driver state monitoring, non-invasive hydration status detection that reuses the communication channel itself as a biomarker, and NABDH — a wireless Holter for continuous cardiac monitoring, awarded a Gold Medal at the International Exhibition of Inventions of Geneva 2026.

System Model — the body as a channel

In capacitive-coupled HBC the transmitter drives a signal electrode against the body while the return path closes through the environmental ground. The body behaves as a lumped, largely capacitive network whose transfer function depends on electrode geometry, posture and skin state. Two operating regimes matter to the circuit designer, and they call for different front-ends:

Four silicon generations, one architectural line

Proposed Technique — four silicon generations

Each chip is a full tapeout in TSMC 65 nm, and each answers the question the previous one left open. Below, one card per generation: the problem it targets, the circuit idea, and what the silicon does.

HBC-I: Fully digital transmitter with a pre-trained autoencoder encoder

HBC links are power-asymmetric: the body-worn transmitter runs from a coin cell, while its receiver — built into a car, a bed rail, a piece of clinical equipment — is mains-powered. HBC-I exploits that asymmetry. The learned symbol-to-waveform mapping of an autoencoder is trained offline and hardcoded in silicon with resource-shared weights, removing memory-access overhead entirely, while the matching decoder stays fully reconfigurable on an FPGA-based standalone receiver. All the machine-learning adaptability lives where the energy is.

• Silicon: TSMC 65 nm CMOS · 0.01 mm² · 0.7 V supply · fully digital implementation flow.
• Two modes without any hardware change — AE-OOK (narrowband, programmable carrier) and AE-Only (broadband baseband).
• Data rate 2.3 kbps – 1 Mbps; TX core power 65 nW – 1.4 µW; energy efficiency 1.3 – 31.4 pJ/bit.
• 33 % and 74 % longer communication range than conventional OOK and AE-Only respectively, measured end-to-end.
• Practical EQS-band noise surveys across real environments yield carrier-selection guidelines and data-rate-to-carrier ratio limits for reliable OOK demodulation.

Proposed system architecture for a wearable HBC TX and standalone RX

System-level validation: (a) measurement test setup with human subject, and (b) BER performance at different distances comparing AE-Only, AE-OOK, and OOK-Only transmission schemes.	Fabricated chip: a) packaged die, b) unpackaged die micrograph

HBC-II: Interference-aware adaptive-frequency transceiver with an on-chip MCU

Below 10 MHz the EQS channel is quiet in the radiative sense but crowded in the conducted sense. Rather than buying robustness with conservative carrier-to-data-rate ratios, HBC-II measures the channel and moves. A tiny on-chip microcontroller performs real-time channel assessment, drives adaptive frequency hopping as interference-driven channel selection, and interfaces sensors directly — no external host anywhere in the control loop.

• RecSilicon: 65 nm CMOS · 0.117 mm² · complete analog front-end integrated with the digital baseband.
• eive chain: preamplifier → envelope detector → post-amplifier → comparator → 3-bit ADC.
• Data rate 22 – 360 kbps; energy efficiency 62.2 – 79.5 pJ/bit.
• Robustness: −33 dB signal-to-interference ratio tolerated at BER = 10⁻³.
• 40× BER improvement over single-frequency operation at 180 cm body distance — the full length of an on-body link.

HBC-III: Digital body-resonance platform with an embedded microcontroller

HBC-III moves the architecture out of the EQS band. It keeps the digital-first flow and the MCU-centred control of HBC-II — OOK framing, programmable data rate, direct sensor interfacing — but drops the autoencoder path and the ExG front-end, and retargets the entire design to high-frequency operation with an embedded microcontroller clocked up to 100 MHz. Its purpose is to establish body-resonance operation as a practical high-throughput mode for HBC rather than a laboratory curiosity, and to give the group a digital platform on which the framing, CRC and data-recovery blocks are brought up before they are hardened alongside full-custom analog in HBC-IV. Role: bridge generation — from sub-10 MHz EQS narrowband links to Mb/s-class body-resonance links. Embedded MCU operating up to 100 MHz, opening the throughput headroom that streaming physiological signals require. Status: characterisation in progress; measured results and figures will be added to this page as they are released.