Time-frequency pattern wake-up detector for low-power always-on sensing of acoustic events

In acoustic event detection scenarios, average power consumed for continuous monitoring can be lowered by an low-power always-on sensor interface, waking-up power-hungry signal sampling and processing circuitry only upon detection of the specific audio-pattern. We research on power consumption of such wake-up sensing (WUS) interface consisting of a multichannel programmable analog filtering bank, signal detectors, and a digital wake-up pattern detector circuit. Focusing on the pattern detector, this paper investigates a design performing wake-up upon matching a sequence of its multi-channel inputs against a digitally preset template of time-frequency states. We compare two hardware implementations of a proposed design: one exploiting a dedicated programmable sequential logic chip, and the other on a low-power 16-bit microcontroller (MCU). In most cases, experimental results generally demonstrate lower energy expenditure obtained by the MCU. Only 330 nW is required for always-on listening, and 1.27-1.71 μJ is spent for analysis of a 3-channel sequence consisting of 4 successive states, each lasting for 200-500 ms. Depending on the application scenario, proposed WUS interface enables for extension of sensor node's battery-lifetime up to 7.4 times, compared to continuous sampling and processing.