i-Sense (South Korea)

Full-duplex (FD) zero-intermediate-frequency (zero-IF) transceivers are attractive for the Internet of Things (IoT) because they reduce cost, power consumption, and hardware complexity. However, their practicality is limited by severe self-interference (SI) dominated by DC offset, frequency-selective in-phase/quadrature (IQ) imbalance, and power-amplifier (PA) nonlinearities that vary over time with thermal effects and protection logic. Conventional cancellers that rely on uniform, time-domain models perform poorly under these frequency-selective and state-dependent distortions. This paper proposes a hardware-aware, frequency-domain self-interference cancellation (SIC) framework that (i) analytically characterizes zero-IF distortion spectra to enable selective per-subcarrier basis allocation, and (ii) introduces lightweight state-adaptive processing using DC-magnitude detection to switch coefficient sets in real time. A USRP X310 prototype demonstrates mean SIC of 35.5 dB across both stable and dynamic PA states at 15 and 21 dBm transmit power, achieved with only 2.57 basis functions per subcarrier on average. Compared with polynomial, parallel Hammerstein, and neural-network baselines, the proposed design achieves superior cancellation with orders-of-magnitude lower complexity. These results establish frequency-domain, hardware-aware SIC as a practical and scalable solution for resource-constrained IoT devices operating in dynamic wireless environments.