Multiresonant fiber acoustic sensors with stabilized triple-phase demodulation for large dynamic ranges

Current extrinsic fiber acoustic sensors face two primary challenges. The first lies in the time-varying initial phase of the interference signal, which arises from environmental vibrations, temperature fluctuations, and variations in laser linewidth. These effects collectively cause significant drift in the interferometric operating point, introducing strong 1/f noise that degrades the signal-to-noise ratio (SNR), particularly in the low-frequency range. This drift also leads to unstable interference fringes, making the output unsuitable for direct audio waveform reconstruction. To address this, it is essential to stabilize the interferometric operating point at a specific phase—typically the quadrature point. Existing solutions commonly rely on monitoring the interference spectrum using an optical spectrometer to dynamically tune the laser wavelength, or scanning and locking onto the quadrature point by identifying fringe midpoints. However, these methods all require a pre-calibration step before each measurement, as interference fringes tend to drift over time, limiting practicality and real-time deployment.

A collaborative team led by Professor Wenming Zhang and Associate Professor Lei Shao from Shanghai Jiao Tong University reports the development of an extrinsic fiber acoustic sensor featuring a stabilized triple-phase demodulation (STPD) method. This approach effectively suppresses 1/f noise and achieves a flat, low noise floor of 173 μrad/Hz without requiring any pre-calibration of the interference spectrum. The STPD method enables simultaneous phase demodulation and drift-free control, allowing for direct audio signal output in a compact, simplified, and cost-effective system (Figure 1). In addition, the team introduces a labyrinth-inspired multiresonant membrane that incorporates three cascaded piston-like resonant modes at 1500 Hz, 6500 Hz, and 9400 Hz, respectively (Figure 2). This structural innovation not only ensures high sensitivity across a wide acoustic spectrum, but also supports a remarkable linear dynamic range of 115.72 dB. Combining these two strategies, the proposed sensor achieves a broad frequency response from 20 Hz to 15 kHz, a maximum phase sensitivity of 1.02 rad/Pa (−119.83 dB ref: 1 rad/μPa), and a minimum detectable pressure (MDP) of 81.84 μPa/Hz at its first-order resonance (Figure 3).

To validate the real-world sensing capabilities of the proposed stabilized triple-phase demodulation (STPD) architecture, the research team conducted three representative application experiments. First, in ecosystem monitoring (Figure 4), the sensor was deployed in an outdoor woodland environment, where it successfully captured weak sounds such as bird tweets and cat meows, demonstrating its high sensitivity in non-contact and low-amplitude scenarios. Second, leveraging the real-time phase output enabled by STPD, a Transformer-based speech recognition system was developed, which significantly reduced the word error rate and confirmed the sensor’s potential for stable and reliable human–machine interaction (Figure 5). Finally, in high sound pressure environments (Figure 6), the sensor exhibited a highly linear response to both piling impacts and balloon bursts, closely matching the waveforms recorded by a commercial sound level meter.

This work presents a fiber acoustic sensor that integrates a broadband multiresonant sensing membrane with a stabilized triple-phase demodulation (STPD) method, which enables automatic stabilization of the interferometric operating point without the need for pre-calibration. The STPD approach is implemented via a miniaturized circuit board capable of simultaneous interference locking and phase demodulation, effectively suppressing 1/f noise and achieving an ultra-low phase noise floor of 173 μrad/Hz. As a result, the sensor demonstrates broadband acoustic spectrum coverage from 20 Hz to 15 kHz, a large linear dynamic range of at least 115.72 dB, a maximum phase sensitivity of 1.02 rad/Pa (−119.83 dB ref: 1 rad/μPa), and a minimum detectable pressure (MDP) of 81.84 μPa/Hz at its first-order resonance. The sensor also exhibits outstanding sensitivity and MDP performance in the low-frequency range. Experimental evaluations further confirm the sensor’s excellent acoustic detection capabilities across a wide range of sound pressure conditions, including ecosystem sound monitoring, machine-learning-assisted speech recognition, and high-intensity shock detection. These findings suggest that the proposed multiresonant extrinsic fiber acoustic sensor with STPD holds strong promise as a robust, low-power, and lightweight solution for advanced voice activity detection and intelligent voice interaction systems.