Time-Efficient SNR Optimization of WMS-Based Gas Sensor Using a Genetic Algorithm

Abstract

This paper presents the description of the wavelength modulation spectroscopy (WMS) experiment, the parameters of which were established by use of the Artificial Intelligence (AI) algorithm. As a result, a significant improvement in the signal power to noise power ratio (SNR) was achieved, ranging from 1.6 to 6.5 times, depending on the harmonic. Typically, optimizing the operation conditions of WMS-based gas sensors is based on long-term simulations, complex mathematical model analysis, and iterative experimental trials. An innovative approach based on a biological-inspired genetic algorithm (GA) and custom-made electronics for laser control is proposed. The experimental setup was equipped with a 31.23 m Heriott multipass cell, software lock-in, and algorithms to control the modulation process of the quantum cascade laser (QCL) operating in the long-wavelength-infrared (LWIR) spectral range. The research results show that the applied evolutionary approach can efficiently and precisely explore a wide range of WMS parameter combinations, enabling researchers to dramatically reduce the time needed to identify optimal settings. It took only 300 s to test approximately 1.39 × 10³² combinations of parameters for key system components. Moreover, because the system is able to check all possible component settings, it is possible to unquestionably determine the operating conditions of WMS-based gas sensors for which the limit of detection (LOD) is the most favorable.

Keywords: LWIR; SNR optimization; WMS; artificial intelligence; genetic algorithm; laser absorption spectroscopy; methane sensor.

Figure 1

(a) Schematic of the CH₄ detection system; (b) HITRAN simulation results (1.8 ppm CH₄ absorption spectrum at temperature 296 K, pressure of 100 mbar, pathlength 31.23 m with air absorption bands of 7% H₂O and 322 ppb N₂O).

Figure 2

Measured first four WMS harmonics signals with SNR value.

Figure 3

The flowchart of the optimization algorithm.

Figure 4

Mean scores versus generation for odd harmonics with various selection functions.

Figure 5

Mean scores versus generation for even harmonics with various selection functions.

Figure 6

Box plots of best individuals for GA optimization. Central mark indicates the median, and the bottom and top edges of the box indicate first and third quartile. The whiskers extend to the most extreme data points, and the outliers are plotted individually using the ‘+’ marker symbol.