Photoacoustic Multigas Sensor via CNN-Based Mode Division Multiplexing

Photoacoustic multigas detection has important applications in fields such as environmental monitoring and fault diagnosis of power equipment. However, existing detection technologies all have inherent limitations: time division multiplexing (TDM) cannot realize simultaneous multigas detection; frequency division multiplexing (FDM) faces challenges such as complex frequency resource allocation and increased system costs; mode division multiplexing (MDM) technology enables simultaneous same-frequency detection, but it relies on a restrictive assumption that different gas signals exhibit distinct waveform widths and are linearly superimposed, which limits its universality. This paper proposes a convolutional neural network-based mode division multiplexing (CNN-MDM) technology. This technology is not only applicable to various different waveform characteristics, but is also capable of accommodating both linear and nonlinear superposition scenarios, thereby providing a universal MDM solution. The core of this scheme lies in assigning unique waveform characteristics to different gases at the modulation stage, and then identifying and separating the mixed signals through a CNN. Experiments were conducted using the decomposition components of SF₆, namely H₂S and CO, as the detection targets. The results demonstrate that the signals separated by the neural network exhibit a strong linear relationship with gas concentrations, with linear fit R² values of 0.996 and 0.995 for CO and H₂S, respectively, and detection limits of 50 and 426 ppb. This CNN-MDM scheme provides a universal framework for simultaneous same-frequency multigas detection and can be extended to photoacoustic systems with mixed modulation modes as well as tunable diode laser absorption spectroscopy (TDLAS).