Sensors based on tunable diode laser absorption spectroscopy (TDLAS) have been widely applied to measure gas parameters such as concentration and pressure for its noncontact nature, high sensitivity, stability, and selectivity, and fast response. Direct absorption spectroscopy (DAS) is among the most extensively utilized measuring methods in TDLAS [1], [2], [3], [4], [5], [6], [7]. DAS can be performed using simple and affordable equipment that can analyze gas parameters directly from the measurement signal; however, it is vulnerable to interference. In complex environments, such as combustion environments where DAS is used to measure gas properties, the measurement signal is unstable owing to flame fluctuations. Consequently, multiple spectral lines in the selected wave number range may be mixed because of pressure broadening in the high-pressure environment created by combustion, thereby rendering it impossible directly obtain parameters of gas properties from the measured signal [8]. Different data processing algorithms are used to solve such problems in DAS [9], [10], [11], [12], [13]. This study developed a fast and accurate baseline extraction algorithm with DAS to directly obtain accurate absorption signals.

DAS is used to analyze the transmittance and shape of selected absorption spectral lines by tuning the laser frequency and obtain important information such as the spectral line strength using which broadening coefficient, gas temperature, concentration, and pressure values can be obtained. To obtain the gas parameters from the absorption signal of the DAS, a reference signal without absorption is required as the baseline, which is a signal that is not absorbed by other gases during the propagation of the laser along its beam path [14], [15], [16]. DAS exhibits a good non-absorption signal for single spectral line measurements at normal temperature and pressure. However, spectral overlapping occurs in case the spectral lines are densely distributed in the researched wavenumber range or there is an increase in pressure. A non-absorbing baseline is sometimes not available, and thus the absorption peaks of the target gas are affected by the neighboring gas, resulting in inaccurate fitting of the spectral line shape and affecting the calculation of the gas parameters. Thus, several gas components must be detected simultaneously in the laser scanning range; however, it is difficult to obtain the parameters of different gas components with a single laser, and past experiments have used more than one laser, which incurred increased cost [17], [18], [19], [20].

In DAS measurements, the baseline is obtained by fitting a low-order polynomial to the non-absorbing region of the absorbing signal, and then the integrated absorption area is defined by line fitting to determine the gas parameters. This is suitable for completely isolated absorption features; however, once the features are blended, the non-absorbing area used to fit the baseline becomes minimal, which results in inaccurate baseline fitting results. Xie et al. [21] proposed an ultra-low sampling high accuracy TDLAS temperature measurement method based on artificial neural network, wherein the DAS data was used as the input of the training set of the artificial neural network with high accuracy and insensitivity to noise; however, only a single training parameter of temperature could be input. Wang et al. [6] introduced a fast uncalibrated wavelength modulation spectroscopy algorithm based on even harmonics, only requiring the extraction of harmonics’ peaks and offers great advantages at high temperature and/or high pressure. However, this process is more complex than the DAS method and requires a considerable quantity of algebraic operations. Weisberger et al. [8] also used DAS to obtain gas parameters using baseline fitting and compared the results with the WMS method, but they did not apply it to high and low pressure cases, with tests conducted only at normal pressure.

The baseline extraction method proposed in this study can obtain accurate baselines even when multiple absorption peaks of different gas components are overlapped. The points used to fit the baseline were estimated by coupling the measurement data with the simulation to obtain the transmission fraction of the peak data between the absorption features, and the analog signal used to couple the measurement data in this method was continuously iterated. Therefore, the baseline fitting was more flexible, functional, and accurate for the gas parameter measurements. Moreover, the method is also suitable for the case of absorption feature isolation.