AI-Optimized Wake-Up Radio Systems for 6G IoT

Abstract

Battery life is the hard constraint for massive Internet-of-Things (IoT) at 6G scale. Wake-up radios (WuRs)—ultra-low-power auxiliary receivers that “listen” for short wake-up signals while the main transceiver sleeps—offer orders-of-magnitude energy savings, but suffer from false wake-ups, long tail latency under bursty traffic, and sensitivity/coverage limits. This paper proposes an end-to-end AI-optimized WuR stack that combines (i) a TinyML classifier embedded in the WuR path to suppress false triggers and adapt detection thresholds, and (ii) a reinforcement-learning (RL) scheduler at the base station or gateway that co-optimizes wake-up signaling with 3GPP NR DRX/C-DRX timers. The proposed method, tested using a trace-driven simulator calibrated with published WuR power/latency figures, the approach reduces node-average energy by 41–72% versus strong baselines, while meeting 99% latency targets and cutting false wake-ups by >80%.