Carbon dioxide (CO2) is an important component of the Earth’s atmosphere that sustains life on our planet via photosynthesis and heat preservation. However, the increase of CO2 concentration in recent decades is considered as one of the main factors responsible for global warming, resulting in the famous greenhouse effect. Moreover, such negative influence is emerging with the massive emissions from extensive human activities in industrial, agricultural, and ecological fields. Given a report from the Intergovernmental Panel on Climate Change (IPCC), CO2 concentration has increased by about 35% since pre-industrial times at an average rate of 1–3 ppm/yr [ppm (part per million)/year]^[1]. High-sensitivity detection of CO2 concentration is essential to identify such a subtle increase over a long time span for atmospheric monitoring.

In recent years, CO2 detection technologies have been extensively developed, among which tunable diode laser absorption spectroscopy (TDLAS) has become the preferred strategy due to its excellent selectivity, high sensitivity, robust anti-interference ability, and long service life. In a TDLAS-based CO2 sensor, laser sources operating at 1572 nm, 2004 nm, 2.7 µm, and 4.3 µm are usually employed to probe the CO2 absorption lines. The advantages of near-infrared (NIR) lasers at 1572 nm are that they have a low cost and are easy to use, but their line strength is very weak. In contrast, the absorption line strength at the mid-infrared (MIR) region is significantly stronger, for example, the line strength of 4.3 µm is stronger than that at 1572 nm by 105 times, which is beneficial for high-sensitivity detection. However, lasers and photodetectors at 4.3 µm are expensive and require a cooling environment for optimal signal-to-noise ratio (SNR). The 2004 nm laser, due to its balanced performance on the economy and absorption line strength, has been widely applied to TDLAS systems for CO2 detection. Generally, the TDLAS mainly includes direct absorption spectroscopy (DAS)^[2] and wavelength modulation spectroscopy (WMS)^[3–5]. Compared with the DAS technique, the WMS method enables the detection of weak absorption by virtue of its high SNR. In a WMS-based system, a high-frequency modulation signal is added to the scanning ramp to bring the detection to a higher frequency range. Absorption-induced harmonic signals are measured by a lock-in amplifier to indicate the gas concentration. Thus, the low-frequency noise (e.g.,  1/f noise) is eliminated, and white noise (e.g.,  thermal noise and shot noise) is substantially reduced. As a result, the gas detection limit and sensitivity are both improved. Meanwhile, calibration and calibration-free techniques were also developed and applied to the WMS system to improve its stability^[6-8]. To further improve the detection sensitivity, multi-pass cells were proposed to increase the absorption path length^[9]. White cells and Herriott cells can extend the optical path length to tens of hundreds of meters^[10-12]. Cavity ringdown spectroscopy (CRDS) is another technique for atmospheric CO2 monitoring^[13,14], which can further improve the sensitivity by extending the absorption path length to the kilometer scale. As early as 2008, Picarro developed a CRDS-based analyzer that already had the capabilities of measuring atmospheric levels of CH4, CO2, and H2O^[15].

When greenhouse gases are mentioned, most people focus on the detection of carbon dioxide in the atmosphere. However, 70% of the Earth’s surface is covered by oceans, and approximately one-third of the CO2 emitted by humans is absorbed by bodies of water. Therefore, monitoring the dissolved CO2 in water is equally important for global climate research. More and more researchers have started to focus on this field in recent years^[16-19]. However, the volume of long-optical-path cells in these studies is always as high as several hundred milliliters, whether it is WMS systems or CRDS instruments. Limited by degassing efficiency, the response speed is a common problem of such sensors. Photoacoustic spectroscopy (PAS)^[20,21] is a rapidly developing technique that greatly decreases the dependence on the optical absorption path length. This allows us to design miniaturized resonant cells for low gas consumption measurement^[22], meanwhile achieving high sensitivity by high-power light sources^[23], high-sensitivity acoustic transducers, and specially designed resonant PA cells^[24-27]. Utilizing the characteristic of low gas consumption, several PAS sensors have been reported for dissolved CO2 detection^[28-30]; as a result, the response speed is significantly improved by virtue of sub-milliliter gas consumption. However, the detection sensitivity of CO2 is generally limited to the level of ppm (part per million) or above due to the low optical power of the 2004 nm laser. To increase the photoacoustic excitation power, we proposed a thulium-doped fiber (TDF) amplifier-enhanced strategy in this study. The TDF was used to amplify the 2004 nm laser power in advance and then to excite the photoacoustic signal. Thus, low gas consumption and high sensitivity were both realized for trace CO2 detection based on the enhanced PAS technique.

The PAS technique has many unique advantages compared to traditional DAS techniques^[20], among which are photoacoustic strength scales with optical excitation power as described in the following equations: Spas=kαcLPQfV,(1)where α is the absorption coefficient, c is the concentration, L is the absorption path length involved in the PA cell, P is the optical power, Q is the quality factor, f is the operating frequency, V is the PA cell volume, and k is a constant describing other system parameters. Obviously, the most straightforward method to enhance photoacoustic strength is using a high-power light source to excite the photoacoustic signal. Therefore, in this work, we attempt to propose a strategy to first amplify the power of the 2004 nm laser and then perform PAS-based CO2 detection.

As an excellent gain medium, thulium-doped silica fiber has been widely used to construct fiber lasers and amplifiers operating at the 1.9–2.1 µm wavelength range. Simplified energy level bands are displayed in Fig. 1(a); the thulium ions can be pumped by 790 and 1550–1910 nm, respectively, and emit the laser at the 2 µm band. Even some people also use 1180 nm for pumping based on excited state absorption (ESA) as shown in Fig. 1(b); however, ground state absorption (GSA) is the most frequently utilized. Due to the high absorption of thulium ions at 790 nm, TDF lasers and amplifiers are usually pumped with a 790 nm light source. However, the nonlinear effects and optical efficiency would become increasingly difficult to balance by geometric optimization. As a result, the slope efficiency is usually less than 60%^[32]. Compared with the 790 nm pump, pumping in the wavelength range of 1550–1910 nm has several advantages as follows. First, the absorption cross-section of thulium is significantly higher compared with 790 nm, leading to more efficient energy transfer and higher pump absorption efficiency. This results in lower threshold power for the onset of 2 µm generation and potentially higher output power. Additionally, longer pump wavelengths in the NIR region experience lower dispersion and attenuation in optical fiber, reducing signal degradation and enabling longer fiber lengths for efficient power scaling. Moreover, the high-power pump laser at the NIR is very easy to obtain and has a very low cost because of the rapid development of the telecommunications industry. So, we chose an in-band-pump strategy in this study, in which a 1567 nm laser was used to pump the TDF to amplify the 2004 nm laser for PAS CO2 detection.

The TDF amplifier enhanced photoacoustic spectroscopy (TDFE-PAS) system is depicted in Fig. 2, and a commercial data acquisition (DAQ) card USB-6361 was employed for signal generation, data collection, and processing. A nanoplus 2004 nm distributed feedback (DFB) laser was used to probe the CO2 absorption lines at the 2 µm band, providing an optical power of 5 mW before amplification. A sawtooth wave at a frequency of 0.1 Hz and a sine wave at a frequency of 1555 Hz were generated and added by the DAQ card to drive the 2004 nm DFB laser for wavelength scanning and modulation, respectively. The laser temperature was controlled by LDC501. A 5 m TDF was used as the gain medium, which has an absorbance of (340±50) dB/m at 1560 nm. The TDF was forward pumped by a 1567 nm laser via a customized 1567/2004 wavelength division multiplexer (WDM). The 1567 nm laser has the ability to provide an adjustable pump power ranging from 0 to 10 W. An optical isolator (ISO) was inserted between the 2004 nm DFB laser and WDM to prevent the backscatter inference on the seed laser. The ISO, WDM, and TDF were fiber-fused together with a fusion loss of ≤0.02dB. The amplified seed laser was introduced to a differential PA cell via a fiber collimator. The fiber collimator was customized to provide a laser beam with a diameter of <1mm over 10 mm working distance so that the laser could not touch the inner wall of the PA cell. The PA cell, fabricated with 316 L stainless steel, has a resonant frequency of ∼3.11kHz as displayed in Fig. 2, whose details have been reported in Ref. [29]. The absorption-induced photoacoustic signal was measured by two microphones (Mic1 and Mic2) and amplified 50 times, respectively; after a 0-dB-gain differential circuit, the photoacoustic signal was collected by the DAQ card for data processing. A self-developed digital quadrature lock-in amplifier (DLIA) was used to extract the second-harmonic signal for CO2 concentration inversion. The bandwidth of the 8-order lowpass filter of the DLIA was set to 1 Hz to reduce noise interference. Sample CO2 used in the experiments was provided by a gas mixing system, including two flowmeters, pure nitrogen, 1% CO2, and a mixing chamber. The full range of a 500 mL/min flowmeter was connected to pure nitrogen, and another full range of a 20 mL/min flowmeter was connected to the 1% CO2 so that low-concentration CO2 samples could be generated.

The carbon dioxide molecule has a series of absorption lines at 2 µm as shown in Fig. 3(a); the absorbance of 400 ppm CO2 is simulated based on the HITRAN database from 2000 to 2005 nm. Considering that the water molecule is the primary interference within this wavelength range, 4% water vapor is simulated in the same figure for reference. In addition, we also carried out experiments to investigate which absorption line was optimal for enhancing photoacoustic detection in terms of pump efficiency. In the experiment, 1% CO2 was introduced to the PA cell. The temperatures of the 2004 nm laser were stabilized successively at 26.5°C, 29°C, 31.4°C, 34°C, 36.7°C, and 39.5°C by LDC501, making the central wavelengths of the seed laser sequentially positioned at 2002, 2002.5, 2003, 2003.5, 2004, and 2004.5 nm, respectively. Thus, the absorption-induced photoacoustic signal was measured at different absorption lines. Meanwhile, the pump power of the 1567 nm laser was tuned from 0 to 4 W to explore the enhancing effect of different pump powers on the photoacoustic signal. As a demo in Fig. 3(b), the photoacoustic signal was significantly enhanced as the pump power increased. The relationship between the second-harmonic amplitude and pump power for different absorption lines was displayed in Fig. 3(c); it was indicated that the enhancement was nonlinear, which may come from the nonlinear response of the TDFA module to the 1567 nm pump laser. When the pump power was set between 1.2 and 3 W, the second-harmonic amplitude was approximately linear-enhanced with the pump power, and then saturation was observed. A linear fit was applied to the measured data in the shadow region, and the 2003.5 nm absorption line achieved the most efficient enhancement with a slope of 265μV/W. Fortunately, the 2003.5 nm wavelength was an ideal absorption line without water vapor interference, which has been chosen for CO2 detection in many studies. Therefore, the seed laser was stabilized at 2003.5 nm in the following experiment to study the performance of the TDFE-PAS sensor. The pump power of the 1567 nm laser was fixed at 3 W.

The TDFE-PAS sensor was calibrated by measuring different CO2 mixtures with known concentrations. A series of CO2 samples were produced based on the gas mixing system by diluting the 1% CO2 with pure nitrogen, giving 12 groups of concentration from 12 ppm to 10,000 ppm. Several representative second-harmonic signals in the low concentration region were plotted in Fig. 4, and the photoacoustic amplitude for the concentration of 500 ppm was measured to be 314 µV. In a control group, the 2004 nm seed laser was directly connected to the PA cell without any amplification as displayed in Fig. 2. The same CO2 sample of 500 ppm was measured as shown in the inset of Fig. 4, and the photoacoustic amplitude was only 3.1 µV. So, the photoacoustic strength was enhanced by over 101 times by virtue of the TDF-enhanced strategy. Considering that the photoacoustic strength scales with the optical excitation power, it could be indicated that the effective power of the 2004 nm laser after TDFA was estimated to be approximately 500 mW. The measured second-harmonic amplitudes were plotted in Figs. 5(a) and 5(b) as a function of CO2 concentrations. A linear fit was applied to the experimental data, giving an R2 of 0.9997 in a wide concentration range with a slope of 0.6μV/ppm. To reduce the fitted error in the low concentration range, the linear fit was re-performed in concentrations below 500 ppm, and the linearity was improved to R2=0.9999 with a slope of 0.63μV/ppm. If such segmented fitting was used, the measurement error would be controlled under 5% in the full concentration range except for the two lowest concentrations. When generating the CO2 samples under 20 ppm concentration, the gas mixing error was introduced limited by the accuracy of flowmeters. The slope efficiency of the low concentration range was a little bit higher than that of the wide concentration range, and such a difference may come from the saturation effect of high-concentration absorption considering that the PA cell has an effective absorption length of 70 mm^[29]. Accidental measurement errors also play a role in the difference. To evaluate the detection limit, the 1σ noise in pure nitrogen was incorporated into the slope of 0.63μV/ppm, and a minimum detection limit (MDL) of 190 ppb (part per billion) was achieved, which equals a normalized noise equivalent absorption (NNEA) coefficient of 1.3×10−8cm−1WHz−1/2 considering the amplified 2004 nm laser of 500 mW.

Even a ppb-level sensitivity of CO2 detection has already been achieved in DAS systems based on multi-pass cells^[33]; however, such a sensitivity of 190 ppb is still very competitive for a PAS-based sensor. This is because, whether it is at 2 µm or 4.3 µm, the optical power of lasers is generally less than 10 mW, which is very limited for photoacoustic excitation. The laser at 1572 nm can be amplified by erbium-doped fibers^[34,35]; however, the absorption line strength at 1572 nm is as weak as 1.72×10−23cm−1/(molec·cm−2). It is also very difficult to generate a strong photoacoustic signal even with a fiber amplifier. As summarized in Table 1, only an MDL is better than that achieved in this paper, in which a 30 W high-power MIR black-body emitter was utilized to excite the photoacoustic signal. The wavelength band centered at 4.3 µm was filtered to probe the CO2 absorption line at the MIR region. Due to the wide bandwidth of the filter, other gas components whose absorption lines also located in the same bandwidth may bring cross interference to CO2 detection. As a result, the CO2 sensor based on such light sources has poor selectivity compared with the laser source. In addition, limited by the beam quality, it is very difficult to design a resonant PA cell to adapt to such light sources. So, the PAS sensors reported in Ref. [36] operated in a non-resonant mode. Therefore, the TDF-enhanced strategy proposed in this paper is still very competitive. It has better detection selectivity and beam quality. Various resonant PA cells can be designed to further enhance the photoacoustic signal. Moreover, the result achieved in this paper is just a preliminary verification of the TDFE-PAS method, and it still has great potential for further improvement. For example, we can develop a TDF laser operating at 2003.5 nm and place the resonant PA cell inside the laser cavity to utilize the high power within the cavity^[37,38]. Combined with the Q-switching technique^[39], a high-power laser pulse ranging from watts to tens of watts would be generated for photoacoustic excitation. Similarly to PAS, photothermal spectroscopy (PTS) is another indirect spectroscopy technology having high sensitivity by virtue of the phase demodulation technique. Moreover, the TDFA pumped 2004 nm laser is also preferred for PTS to achieve high-sensitivity CO2 detection^[40]. In addition, light-induced thermoelectric spectroscopy (LITES) is also a good choice, which has rapidly developed in recent years. It can combine the advantages of long-path absorption and low-noise photothermal detection together to achieve an ultra-high sensitivity^[41].

Thulium-based lasers have been widely used for biological applications as well as cutting, engraving, and welding plastics. However, using TDF to amplify the 2004 nm DFB laser for photoacoustic enhancement of CO2 detection has now been reported in this paper. In conclusion, a novel TDFE-PAS sensor has been developed for the first time, to the best of our knowledge, to successfully enhance the photoacoustic signal over 101 times in this study. It provides a new direction and more potential to researchers for high-sensitivity detection of CO2 gas based on the PAS method. Verified by experiments, the TDFE-PAS sensor achieves a very good linearity of 0.9997 in a wide concentration range of 0–10,000 ppm. Meanwhile, an MDL of 190 ppb is realized with a response time of 10 s. Future work will involve optimizing the insertion loss and TDF length to further increase the pump efficiency, constructing the Q-switched TDF laser operating at 2003.5 nm. Ultimately, we aim to improve the detection limit to ppb or even sub-ppb level for applications requiring ultra-highly sensitive CO2 detection.