1. Introduction

Fourier transform infrared (FTIR) spectroscopy is a well-established scientific method widely used for broadband molecular spectroscopy [1]. The versatility of the technique is much appropriated for the determination of the structure of molecules, the quantitative analysis of complex mixtures, the investigation of dynamic systems, biomedical spectroscopy, hyperspectral imaging, and for the study of many types of interfacial phenomena. The FTIR spectrometer can support both broad spectral coverage and high spectral resolution with a calibrated frequency scale. Traditionally, a Fourier transform spectrometer (FTS) is combined with an incoherent thermal source to cover the mid-infrared (MIR) molecular fingerprint region [2,3]. However, these thermal sources have low spectral brightness and are omnidirectional, making it challenging to achieve long absorption path length for gas-phase spectroscopy; such as with the use of an optical cavity or a multi-pass cell. In addition, a long measurement time is often required to obtain a high signal-to-noise ratio (SNR) from the thermal-source based FTS systems, which is more prominent in high-resolution measurements.

Advances in compact and reliable MIR supercontinuum (SC) sources have circumvented these challenges. Commercial low-noise broadband MIR SC sources with sub-nanosecond pulse durations and low MHz repetition rates are currently available. They have high spatial coherence and a very high spectral brightness, much higher than thermal sources and even synchrotrons [4–6], allowing high sensitive measurements. Furthermore, their high repetition rate can potentially be used in a synchronous demodulation technique to reduce 1/f noise and further improve the detection sensitivity [7].

Combining an FTS with an MIR SC source can lead to fast and sensitive gas sensing for multiple species [8,9]. The high spectral resolution offered by an FTS can improve the distinction of individual molecular absorption spectra in complex gas mixtures, especially for molecular gases with overlapping absorption profiles. The resolution of an FTS increases by extending the optical path difference (OPD) [1]. Obtaining a higher spectral resolution in an FTS inherently results in a lower SNR. The SNR can be improved by longer averaging time to achieve higher detection sensitivity. To reduce the measurement time, the intensity noise can be reduced significantly by implementing a balanced detection scheme. For this, we measure the two out-of-phase outputs of the interferometer by two detectors. The two interferograms are 180 degrees out-of-phase but have an in-phase intensity noise. By subtracting the two outputs, the interferograms are doubled, while the noise is substantially reduced, yielding an improvement in SNR [10,11].

In this contribution, we present a multi-species trace gas sensor employing a broadband MIR SC source combined with a compact, home-built FTS utilizing a balanced detection scheme. We investigate the FTS characteristics in terms of spectral resolution, spectral coverage, detection sensitivity, detection precision, long-term stability, and system linearity. We compare the performance of the FTS-based sensor with our previously developed scanning Grating-based Spectrometer (GS) [7], by determining the concentrations of individual gases in a complex gas mixture. For this, we measure fruit-produced volatiles from fermentation (ethanol, acetaldehyde, ethyl acetate) [12–14], ripening (ethylene) [15], and rotting (acetone, methanol) [16–18], simultaneously with both the FTS and GS. It is important to monitor these processes, e.g. in commercial fruit storage rooms, as fruit can develop physiological or pathological disorders, resulting in spoilage or quality classification downgrading, thereby reducing commercial value. It is estimated that in long-term storage, 25% of all fruits and vegetables are lost, causing major economic losses of €6.1 billion globally [19]. The results demonstrate the potential of both sensors for applications in practical commercial-scale fruit storage facilities with sensitivities at sub-ppmv Hz^-1/2 level.

2. Materials and methods

Figure 1 illustrates the scheme of the experimental setup. We utilized a broadband MIR SC source (NKT Photonics), with a total power of 500 mW, an average power spectral density of 200 µW/nm in the spectral range between 2 to 4 µm and a 2.5 MHz pulse repetition rate. The MIR SC beam is directed into a multi-pass absorption cell (HC30L/M-M02, ∼0.85 L volume, Thorlabs) with a nominal optical path length of 31.2 m, which contains the gas sample. The pressure and the gas flow in the absorption cell are controlled by a pressure meter/controller (EL-PRESS, Bronkhorst) and a flow meter/controller (EL-FLOW Prestige, Bronkhorst), respectively. The transmitted beam of the absorption cell is guided into a compact home-built Michelson interferometer-based Fourier Transform Spectrometer (FTS). In the FTS, the input beam is split into two arms by a beam splitter (BS, BSW711, Thorlabs). Both beams are sent towards two hollow retroreflector mirrors (HRR201-P01, Thorlabs) mounted back-to-back on a 10 cm motorized linear translation stage (DDSM100, Thorlabs). This configuration doubles the FTS’s Optical Path Difference (OPD) compared to a traditional one-fixed-one-movable mirror configuration, resulting in a more compact system. The retroreflector mirrors reflect the beams parallel to the incident ones with a slight horizontal and vertical displacement, avoiding the overlap of the incident and reflected beams. The reflected beams are recombined on the BS so that their transmission and reflection from the BS overlap and create two outputs with out-of-phase interference patterns. These interferograms are recorded separately via two MIR thermo-electrically cooled single-point photodetectors (PVI-4TE-4, VIGO System). For the data handling, the output voltages of the two photodetectors are subtracted using a differential amplifier (SR560, Stanford Research System) in a balanced detection scheme. We used an optical filter (FB3250-500, Thorlabs) to improve the SNR in the desired spectral range. The recorded data were transferred to a computer via a data acquisition card (DAQ, BNC-2110, National Instruments) and processed by a LabVIEW-based program.

 

Fig. 1. The optical setup for the FTS system. SC: supercontinuum source, He-Ne: Helium-Neon laser, MPC: multi-pass cell, M: mirror, BS: beam splitter, RR: retroreflector, TS: translation stage, BB: beam blocker, BPF: band-pass filter, PD: Photodetector, DA: differential amplifier, DAQ: data acquisition card, PC: computer. The SC and He-Ne beams are shown in blue and red, respectively.

Download Full Size | PDF

To calibrate the OPD, a Helium-Neon laser (He-Ne, HNL020L, Thorlabs) beam is directed such that it passes alongside and parallel to the SC beam in the FTS arms. The He-Ne interferogram is also recorded by a Si free-space amplified photodetector (PDA8A2, Thorlabs) and used for frequency calibration. After removing the DC-offset of the He-Ne interferogram, the exact zero-crossing positions in the He-Ne interferogram are identified, using a linear interpolation. Therefore, the OPD step size of the FTS is found based on the wavelength of the He-Ne laser. Another linear interpolation retrieves the SC interferogram values at the zero-crossings of the He-Ne interferogram. In this way, the SC spectrum is calibrated in the frequency domain after a Fourier transformation of the OPD-calibrated SC interferogram [10,20].

Figure 2 shows an overview of the optical setup for the Grating-based Spectrometer (GS). A comprehensive explanation of the GS design and the system characterization can be found elsewhere [7]. Briefly, the MIR SC beam is sent into a multi-pass cell filled with the gas species under study. The output beam is guided toward a diffraction grating mounted on a galvo scanner that is driven by a sinusoidal current. For each grating scan, the diffracted spectrum is recorded in the time domain by an MIR single-point photodetector. The output signal of the detector is transferred to a lock-in amplifier, which is referenced to the (2.5 MHz) repetition rate of the SC source. The scanner position signal and the lock-in amplifier output are transferred to a computer and synchronized for each grating scan to acquire reproducible absorption spectra.

 

Fig. 2. The optical setup for the GS system. SC: supercontinuum source, M: mirror, L: lens, MPC: multi-pass cell, G: grating, GSc: galvo scanner, CM: cylindrical mirror, PD: photodetector, DLA: dual-phase lock-in amplifier, DAQ: data acquisition card, PC: computer.

Download Full Size | PDF

To calibrate the frequency axis of the GS, we separately measured the broadband spectra of different gas species covering the entire bandwidth of the spectrometer. We compared the measured spectra to the corresponding calculated spectra using the HITRAN or PNNL database (depending on the availability of the species) convoluted with a Gaussian instrumental line-shape. Using a 9th order polynomial equation, we fit the frequency scale of the measured spectra to that of the simulated spectra. The retrieved polynomial coefficients are used for frequency calibration of the other measurements. The frequency calibration stays valid, as long as the driving parameters of the galvo scanner are not changed. Therefore, the shot-to-shot calibration of the spectrometer is rather satisfying; however, a day-to-day recalibration might be needed to remove long-term drifts. In addition, environmental temperature fluctuations affect the frequency calibration significantly. To minimize this effect, the optics for the GS are in a thermally isolated and stabilized box; the temperature is regulated and controlled using several resistive heaters and a feedback control loop.

We evaluated and compared the performance of the FTS-based sensor and the GS [7] for multi-species trace gas detection. For this, we performed a series of laboratory-scale measurements on apples to detect their produced volatiles under different atmospheric conditions. Here, we focus on volatile species that are biomarkers for the physiological and pathological status of the fruits in terms of fermentation (ethanol, acetaldehyde, and ethyl acetate), ripening (ethylene), and rotting-damage (methanol, acetone, and ethane). We used the absorbance of these gas biomarkers as references in a fitting routine for retrieving the concentrations of the gas species.

The measurement routine is quite similar for both systems. The wavelength dependent background intensity, $\textrm{\; }{I_0}(\lambda )$, is measured when the corresponding absorption multipass cell is evacuated (<1 mbar pressure). The transmitted intensity ${I_T}(\nu )$ is obtained when the multipass cell is filled with the gas samples at 900 mbar pressure. For both the FTS and GS-based sensor, the background intensity is measured within one minute. To measure the VOCs emitted from the apples, the transmitted intensity is measured over five minutes, and the concentrations of different species are retrieved from the fitting routines. This procedure is performed repeatedly for the entire measurement period.

3. Results and discussion

In this section, we first characterize and evaluate the performance of the developed FTS system. After that, we compare the performance of the FTS and the GS by monitoring the volatiles emitted by the fruit.

3.1 Spectral resolution

The highest spectral resolution of an FTS is obtained by a maximum OPD. The utmost OPD for this customized FTS configuration is ∼40 cm, yielding a spectral resolution of better than 1 GHz (0.033 cm^-1). However, gaining a high spectral resolution translates into a lower SNR, requiring longer measurement time. To assess the performance of the FTS with high spectral resolution, we measured the absorbance spectrum of 5(±0.1) ppmv methane (diluted in synthetic air, 80% nitrogen and 20% oxygen, Linde Gas) with a 1 GHz spectral resolution and 6 s measurement time (single FTS scan). The averaged absorbance spectrum (250 averages) is shown in Fig. 3 (in red) along with a fitted methane model based on HITRAN database [21] parameters and a Voigt profile (in blue, inverted for clarity). Note that a simple box-car apodization was used for the measured interferograms; and no instrument line shape function was convoluted with the spectrum. The retrieved methane concentration from the fit is 4.98(±0.03) ppmv, in which the error values are the standard deviations of 10 consecutive measurements. The residual of the fit is shown (in green) in the lower panel of Fig. 3. The excellent agreement of the measurement and the model is clear from the flat and featureless residual. The standard deviation of the residual is 0.014.

 

Fig. 3. Left Panel: Measured absorbance spectrum of 5 ppmv methane in synthetic air at 900 mbar (in red), averaged over 250 scans (25 minutes), along with the simulated absorbance model based on HITRAN database (in blue, inverted). The residual of the fit is shown in the lower panel (in green). Right Panel: An enlargement to the methane P5 lines at 2958.3 cm^-1.

Download Full Size | PDF

3.2 Detection sensitivity

As long as the SNR of the retrieved spectrum is dominated by white-noise, the SNR can be increased by averaging over time. We investigated the long-term stability of the system and the optimum averaging time to obtain the minimum detectable concentration. For this, we measured the background spectra (3 GHz resolution) every 1.9 s successively for a period of 12 h, while the multi-pass cell was filled with 100% nitrogen gas (pressure 900 mbar, gas flow 5 l/h). We normalized the obtained spectra to the first measured background spectrum and fitted the methane absorption spectrum model (based on HITRAN database parameters and a Voigt profile) to acquire the Noise Equivalent Concentrations [NECs, Fig. 4 (a)] [7]. The retrieved NECs and their corresponding Allen-Werle plot are shown in Fig. 4. The Allen-Werle plot shows that the FTS has a noise equivalent detection sensitivity of 300 ppbv Hz^-1/2 for methane. Averaging over longer periods improves the detection sensitivity according to τ^-1/2-law of white noise. The absolute minimum detectable methane concentration is 10 ppbv for 15 minutes averaging time.

 

Fig. 4. (a) The methane noise equivalent concentrations (NECs, shown over 6000 seconds), acquired by fitting the methane absorbance spectrum to the noise equivalent absorbance spectra. Noise equivalent absorbance spectra were achieved from a sequence of background spectra measured in pure nitrogen every 1.9 s and normalized to the first measurement. (b) Allen-Werle plot of the NEC of methane (blue curve), and the τ^-1/2-dependency (in red) associated with white noise.

Download Full Size | PDF

The Noise equivalent absorption (NEA) is determined following the equation:

3.3 System linearity

We characterized the linear response of the system by applying various ethane concentrations ranging from 1 to 100 ppmv. The various diluted concentrations were prepared by dynamic mixing of pure nitrogen gas with 100(±2) ppmv ethane from a standard calibrated gas bottle (Linde Gas). For each mixture, 70 spectra with a 3 GHz resolution were obtained and averaged. Figure 5 shows the determined ethane concentrations (from the fitting routine) versus their applied concentration. The high agreement between the calculated and applied concentrations is substantiated by a linear fit to the data, featuring a Pearson’s correlation coefficient of 0.99992.

 

Fig. 5. Linearity response of the system to various applied concentrations of ethane. The uncertainty is based on the standard deviation of the calculated values for 70 measurements.

Download Full Size | PDF

3.4 Performance comparison of the FTS and GS for multi-species detection

We compare the performance of the FTS-based sensor and the GS (grating-based spectrometer) for quantifying gas concentrations in a complex mixture of various gases. In general, retrieving the concentration of an individual gas from a multi-species gas mixture with overlapping absorbance characteristics is not straightforward [23–26]. If the spectral resolution is coarse, as in the case of the GS, the distinction for different gas absorption features is more challenging. Thanks to the high spectral resolution provided by the FTS, this problem is circumvented. We applied a non-negative least square (NNLS) curve fitting [27] to quantify the concentration (contribution) of each gas in the gas mixture. For this, we prepared a gas mixture of 5 ppmv ethane with 25 ppmv ethyl acetate in nitrogen, from 100(±2) ppmv calibrated gas bottles of ethane and ethyl acetate in nitrogen (Linde Gas). We selected these gas species, to cover a broad spectral range with overlapping absorption profiles; while ethane has narrow absorption features, ethyl acetate has a broad absorption profile in this wavelength region.

The measured absorbance spectrum (in black) and the calculated fits for each gas species obtained by the FTS-based sensor (500 averages in 15 min, 3 GHz spectral resolution) and the GS (3000 averages in 5 min, 75 GHz spectral resolution) are shown in Figs. 6(a) and 6(b), respectively. The measured transmission spectra of the sample gases were normalized to the background transmission spectra measured in pure nitrogen at the same pressure and averaging time for each spectrometer. The absorbance spectra were calculated from these normalized spectra, respectively. The reference spectra used for the FTS were based on the PNNL database [2]. The reference spectra for the GS were pre-measured with the same spectrometer using calibrated gas mixtures. A 3^rd order and a 2^nd order polynomial baselines were added as additional references for the FTS and the GS fits, respectively. These baselines were used to compensate the spectral drifts during the measurements. The measured absorbance by the FTS (in black) is in excellent agreement with the corresponding fit model (in red), yielding almost a flat residual (in green). However, for the GS some discrepancies are observed between the measured absorbance and the reference spectra, resulting in a less uniform residual. The standard deviation of the residual of the FTS and GS are 0.011 and 0.0073, respectively. The retrieved concentrations by the fit for the FTS are 4.77(±0.05), 25.6(±0.3) ppmv for ethane and ethyl acetate, respectively. The error is the standard deviation of 10 consecutive measurements. These values are within the error margin of the gas mixture, due to the precision of the calibrated gas mixture and the mass flow controllers (∼2%). The retrieved concentrations by the fit for the GS are 4.42(±0.07) and 28.78(±0.06) ppmv for ethane and ethyl acetate, respectively, which show slight systematic deviations from the expected values.

 

Fig. 6. The measured absorbance and the corresponding fits, obtained by (a) the FTS and (b) the GS.

Download Full Size | PDF

The SNR for a single measurement of the transmitted spectrum of the FTS (between 2800-3300 cm^-1) and GS (∼2700-3200 cm^-1) when their absorption cell filled with a non-absorbing gas (i.e. nitrogen) are ∼6.8 and ∼99.5, respectively. These values obtained from the standard deviation of normalization of two successive spectra (SNR = 1/σ). Although the GS system as compared to the FTS reaches a higher SNR value in a shorter measurement time, its relative accuracy is less than the FTS. This effect is mostly due to the less precise frequency calibration of the GS, as a well-calibrated spectrum is essential for the NNLS fitting. The FTS uses the He-Ne laser wavelength as the reference, providing well-calibrated spectra in the frequency domain. For the GS, the grating mounted on a galvo scanner is driven via a sinusoidal wave (input), and the spectrum is obtained by synchronizing the galvo scanner (output) position signal and the recorded signal from the photodetector. However, due to the lack of a perfect reproducibility of the spectrum, well-calibrated spectra (such as the one for the FTS system) in this infrared wavelength region are difficult to achieve. Therefore, the calibration mismatch for the GS yields a more structured residual of the fit and, as a result, less relatively accurate concentrations are retrieved. Although GS cannot resolve the closely spaced absorption bands of ethane and given that there is a very strong overlap in absorption spectra of the gasses, the selectivity of the NNLS fitting routine is still found to be satisfactory in distinguishing ethane from ethyl acetate.

3.5 Sensor comparison by monitoring apples under fermentation and normal atmosphere

Under normal aerobic conditions, biological tissue respires and thereby provides energy to the cells [28]. If aerobic respiration is impeded, alcoholic fermentation will provide the cells with energy without the involvement of oxygen, albeit with low efficiency. In commercial fruit storage rooms, the oxygen concentration is lowered, to typically 1%, to decrease the respiration rate of the fruit, thereby extending storage time. However, too low oxygen levels induce fermentation, causing the fruit to deteriorate quickly. It is therefore of the utmost importance to control the atmosphere around stored fruit in commercial storage rooms. To test whether our systems can detect fruit fermentation at an early stage, we induce fermentation in apples by applying a nitrogen atmosphere (for 17 hours). Then, we replace the nitrogen gas by normal air and observed the response of the apples in the presence of oxygen.

The measurements were performed with the FTS-based and GS sensor systems simultaneously, each with their own SC source. The two multi-pass cells were connected in series, sharing the same gas-handling system. Details of the gas-handling system are described elsewhere [29]. Water vapor is a common spectroscopically interfering species, especially for multi-species detection, therefore, a water trap was integrated into the gas flow before the sampling cells [29]. Off-the-shelf biological apples (∼2 kg, Royal Gala) were stored at 23 °C in a water-locked polypropylene container (65 L). The storage container was connected to the gas-handling system in a closed-loop configuration; the emitted volatiles from the fruit were transported with the gas flow (using a pump) from the container via the water trap to the absorption cells and back to the storage container. The gas-handling system was controlled via a LabVIEW program. By recycling the gas, we mimic the accumulation process of emitted volatiles in storage rooms.

Figure 7 illustrates the gas concentrations of the emitted volatiles for the two sensors when the apples were in the nitrogen atmosphere for 17 hours (left sub-panels). Each data point represents the averaged concentration of five independent two-minute-measurements for the FTS and five independent one-minute-measurements for the GS. The error bars are calculated based on the standard deviation of the five measurements. Note that the concentration of the emitted gases varies over time due to change of biological behavior of the apples, which also affects the error bars. The data show ethanol, a well-known fermentation indicator produced in the absence of oxygen [12]. A continuous increase in accumulated ethanol is observed by both sensors, up to 10 ppmv. At 17 hr. the nitrogen gas is replaced by normal air thereby removing all the accumulated gases. The accumulation of ethanol in atmospheric conditions is then followed from 21-30 hr. Although the fermentation is stopped/strongly reduced, ethanol release is observed due to the presence of liquid ethanol in the apple tissue. At ∼22h this release stopped, since the concentration is not increasing further. Acetaldehyde, a precursor of ethanol in the fermentation cycle, was found at levels of 2 ppmv by the FTS sensor (GS sensor 750 ppbv). Acetaldehyde is more volatile than ethanol. Under post anaerobic conditions, a part of the present ethanol is oxidized back to acetaldehyde, which explains the relative high emission rate of the acetaldehyde between 21 and 30 h [30]. Ethylene is a gaseous ripening hormone [15], which can only be produced in the presence of oxygen. Here we see low ethylene level under nitrogen conditions, while as the respiration process is restored the ripening process starts again. As we expected, the ethylene level reaches ∼2.5 ppmv at 30 hr. (FTS), this reaches up to ∼4 ppmv measured by the GS. Methanol, acetone and ethyl acetate reach levels of ∼1 ppmv. Within the FTS, the sensitivity is too low to determine proper concentration levels of these volatiles. Within the GS sensor, they can be properly quantified in the aerobic atmosphere.

 

Fig. 7. Volatile emission from apples under nitrogen atmosphere (left sub-panel, for 17 hours). At 17 hr. the nitrogen gas is replaced by normal atmospheric air (right sub-panel. Emissions in normal air are followed from 21-30 hr. Panel (a) shows the measurements by the FTS-based sensor and Panel (b) the GS sensor. The error bars in the data clearly indicate the detection sensitivity for the different gases.

Download Full Size | PDF

As supplementary information, numerical simulations are shown in Appendix 5.1, demonstrating the resilience of the fitting error to changes in the relative concentrations of overlapping species. Moreover, the actual measured spectra in real-life measurements related to Fig. 7 are presented in Appendix 5.2.

3.6 Comparing the FTS-based and the GS sensor

As shown above, both the FTS-based and the GS sensor demonstrate successful and rather satisfactory results in the detection of fruits gaseous biomarkers. The concentration values measured by both sensors were relatively close to each other, and their variations exhibit similar trends. However, we can also observe discrepancies in the results. As shown in Fig. 6, the measured and the fitted spectra are better matched for the FTS-based sensor compared to the GS sensor. Therefore, the retrieved concentrations are expected to be closer to the actual values for the FTS-based sensor.

Another apparent distinction is that the GS sensor has higher SNR at the spectra as compared to the FTS. Consequently, the error bars for the determined gas concentrations are smaller for GS, and we had to average the FTS data for a longer period: 10 minutes compared to 5 minutes for the GS data.

Within the GS, the reference spectra for each gas are obtained experimentally, using calibrated gas mixtures. These reference spectra are sensitive to misalignments of the SC-beam within the sensor system. Misalignments lead to alteration of the spectral resolution and changes in the frequency calibration. Therefore, the reference spectra are less valid if the optical alignment is varied substantially. On the other hand, the FTS is more robust and less sensitive to optical misalignments. While a change in FTS alignment can reduce the SNR, the spectral resolution and frequency calibration remain unaffected, as both SC and He-Ne beams share the same set of mirrors and experience the same alignment change. In addition, the reference spectra for the FTS are available from reference databases, such as HITRAN and PNNL, with more than four hundred gas species available.

The mid-infrared photodetectors in the FTS receive the full power of the light source at all times, so power attenuation methods may be required to avoid detector saturation. The best practical solution is to use a band-pass filter to achieve a higher SNR in the desired spectral range, removing the undesired out-of-the-band power and preventing the photodetectors from saturation. In contrast, the photodetector in GS receives only a small fraction of the power from the light source in the time domain during the grating scan. Therefore, the highest SNR for each wavelength can be achieved without any power attenuation.

As was mentioned earlier, FTS has a slower scan speed compared to the GS. A single spectrum with 3 GHz spectral resolution is measured in 1.9 s using the FTS. A higher spectral resolution demands a more extended OPD requiring a longer measurement time. Meanwhile, the GS sensor provides fast measurements as short as 100 ms for a single spectrum. However, its spectral resolution is 75 GHz and cannot be dynamically changed. In addition, the FTS can cover the entire bandwidth of the SC source, while GS coverage is limited to a wavelength coverage of 1000 cm^-1.

4. Conclusion

We have developed a multi-species trace gas sensor by combining a broadband mid-infrared supercontinuum source with a multi-pass absorption cell and a compact, home-built Fourier Transform Spectrometer. The FTS uses a balanced detection scheme to improve the SNR and the detection sensitivity. We achieve detection sensitivities on the order of 300 ppbv Hz^-1/2 for methane, and similarly a few hundred ppbv Hz^-1/2 for various hydrocarbons, alcohols, and aldehydes. We also evaluated the linearity, precision and multi-species detection capability of the FTS-based sensor. We utilized the FTS system alongside a previously developed scanning grating spectrometer (GS) and simultaneously measured the volatile species produced by fruit under various atmospheric conditions. Both sensors successfully determined fruit-produced ethanol, acetaldehyde, ethyl acetate, ethylene, acetone and methanol, indicators of fermentation, ripening, and rotting of plant tissue. The FTS-based sensor shows better relative accuracy; in concentrations retrieval they showed lower precision. The GS sensor provides a higher precision but lower relative accuracy. We believe that the main contribution to the lower accuracy of the GS is the lower reproducibility of the measured spectrum as well as long term drifts of the system, which could be improved in the future. Both sensors are suitable for various gas sensing applications, and their selection depends on the requirements of a specific application.

Appendix

5.1 Simulation of retrieving concentration of fruit-produced volatiles

Here we provide numerical simulations to demonstrate the resilience of the fitting routine to changes in the relative concentrations of overlapping species. We used the same absorbance reference spectra of GS and FTS for the simulations as we used for the actual fit in section 3. All the absorbance reference spectra were generated for 1 ppmv concentration. The water spectrum was also added to the references since water vapor presents in the real-life measurements. To simulate the measured absorbance spectrum including target gases with different concentrations, we multiplied each reference spectrum with a specific factor and summed them up. Different random noise levels for the FTS and GS, with the same standard deviation as what we have had in the actual measurements, were also added to the simulated measured spectra.

Similar to the real measurements presented in section 3, we simulated five measurements with different concentrations for various species and obtained the mean value and standard deviation (error bar in the plot) of the retrieved concentrations from the fits. To evaluate the fermentation process, we increased the concentration of ethanol from 1 to 20 ppmv (1 ppmv step), and acetaldehyde by a lower rate from 150 ppbv to 3 ppmv (150 ppbv step). We assumed fixed concentrations of 1 ppmv and 500 ppbv for ethylene and methanol, respectively. The results for the FTS and GS are shown in Fig. 8. It is obvious that due to a better SNR, the error bars of the retrieved concentrations by the GS are smaller compared to the FTS.

 

Fig. 8. A simulation of fermentation process for the FTS (left panel) and the GS (right panel) by increasing ethanol from 1 to 20 ppmv, acetaldehyde from 150 ppbv to 3 ppmv, and keeping the other gas species at fixes concentrations (methanol at 500 ppbv and ethylene at 1 ppmv). The real SNR of the spectrometers were considered in the simulations.

Download Full Size | PDF

For evaluating the ripening process, we increased the concentration of ethylene from 1 to 20 ppmv and ethylene from 10 to 30 ppmv, while we kept the other volatiles at constant concentrations (ethanol at 10 ppmv, and acetaldehyde at 5 ppmv). The results are represented in Fig. 9.

 

Fig. 9. A simulation of the ripening process for the FTS (left panel) and the GS (right panel) by increasing ethylene from 1 to 20 ppmv and keeping the other gas species at fixes concentrations (methanol at 1 ppmv, ethanol at 10 ppmv, and acetaldehyde at 5 ppmv). The real SNR of the spectrometers were considered in the simulations.

Download Full Size | PDF

The assessment of the rotting process was performed by increasing methanol from 1 to 20 ppmv and keeping the other volatiles at fixed concentrations (ethanol at 10 ppmv, methanol at 1 ppmv, and acetaldehyde at 5 ppmv), Fig. 10.

 

Fig. 10. A simulation of the rotting process for the FTS (left panel) and the GS (right panel) by increasing methanol from 1 to 20 ppmv, ethylene from 10 to 30 ppmv and keeping the other gas species at fixes concentrations (ethanol at 10 ppmv and acetaldehyde at 5 ppmv). The real SNR of the spectrometers were considered in the simulations.

Download Full Size | PDF

For the last evaluation, we repeat the first simulation but implement the same SNR for FTS as for GS. The results indicate that both sensors provide almost the same results (Fig. 11).

 

Fig. 11. A simulation of fermentation process for the FTS (left panel) and the GS (right panel) by increasing ethanol from 1 to 20 ppmv, acetaldehyde from 150 ppbv to 3 ppmv, and keeping the other gas species at fixes concentrations (methanol at 500 ppbv and ethylene at 1 ppmv). The same SNR was considered for the simulations from the FTS and GS.

Download Full Size | PDF

In short, the simulations demonstrate the resilience of the fitting routine to changes in the relative concentrations of overlapping species. The error for the concentrations obtained via the FTS based sensor is larger compared to the ones obtained by the GS based sensor. However, when the SNR for both spectrometers is the same, the retrieved concentrations and their error values are almost the same.

5.2 Actual measured spectra in real-life measurements

To represent more information regarding Fig. 7, we show the absorbance spectrum, the total fit and their residual obtained by the FTS (Fig. 12) and the GS (Fig. 13) for the first measurement (top panel) and the last measurement after 17 h (bottom panel).

 

Fig. 12. The absorbance spectrum, the total fit and their residual obtained by the FTS for the first (top panel) and last measurement after 17 h (bottom panel).

Download Full Size | PDF

 

Fig. 13. The absorbance spectrum, the total fit and their residual obtained by the GS for the first (top panel) and last measurement after 17 h (bottom panel).

Download Full Size | PDF

The obtained concentration for different gas species from the total fit of Fig. 12 are listed in Table 1.

Table 1. The retrieved concentrations for different gas species from the total fit of Fig. 12 for the FTS.

View Table | View all tables in this article

The obtained concentration for different gas species from the total fit of Fig. 13 are listed in Table 2.

Table 2. The retrieved concentrations for different gas species from the total fit of Fig. 13 for the GS

View Table | View all tables in this article

For the first measurement, the water absorbance peaks are clearly observed while the other gas species do not exist or they are less than the detection limit. For the last measurement, the absorbance peaks of other species such as ethanol are clearly observed.

Funding

Interreg (363); Nederlandse Organisatie voor Wetenschappelijk Onderzoek (14709); H2020 Industrial Leadership (732968).

Acknowledgments

The authors would like to thank NKT Photonics for providing the mid-infrared supercontinuum light sources.

Disclosures

The authors declare no conflicts of interest.

References

1. P. R. Griffiths, J. A. De Haseth, and J. D. Winefordner, Fourier Transform Infrared Spectrometry (Wiley, 2007).

2. S. W. Sharpe, T. J. Johnson, R. L. Sams, P. M. Chu, G. C. Rhoderick, and P. A. Johnson, “Gas-phase databases for quantitative infrared spectroscopy,” Appl. Spectrosc. 58(12), 1452–1461 (2004). [CrossRef]  

3. D. W. Griffith, N. Deutscher, C. Caldow, G. Kettlewell, M. Riggenbach, and S. Hammer, “A Fourier transform infrared trace gas and isotope analyser for atmospheric applications,” (2012).

4. C. R. Petersen, U. Møller, I. Kubat, B. Zhou, S. Dupont, J. Ramsay, T. Benson, S. Sujecki, N. Abdel-Moneim, Z. Tang, D. Furniss, A. Seddon, and O. Bang, “Mid-infrared supercontinuum covering the 1.4–13.3 μm molecular fingerprint region using ultra-high NA chalcogenide step-index fibre,” Nat. Photonics 8(11), 830–834 (2014). [CrossRef]  

5. T. Cheng, K. Nagasaka, T. H. Tuan, X. Xue, M. Matsumoto, H. Tezuka, T. Suzuki, and Y. Ohishi, “Mid-infrared supercontinuum generation spanning 2.0 to 15.1 μm in a chalcogenide step-index fiber,” Opt. Lett. 41(9), 2117–2120 (2016). [CrossRef]  

6. C. R. Petersen, P. M. Moselund, L. Huot, L. Hooper, and O. Bang, “Towards a table-top synchrotron based on supercontinuum generation,” Infrared Phys. Technol. 91, 182–186 (2018). [CrossRef]  

7. K. E. Jahromi, M. Nematollahi, Q. Pan, M. A. Abbas, S. M. Cristescu, F. J. Harren, and A. Khodabakhsh, “Sensitive multi-species trace gas sensor based on a high repetition rate mid-infrared supercontinuum source,” Opt. Express 28(18), 26091–26101 (2020). [CrossRef]  

8. V. V. Goncharov and G. E. Hall, “Supercontinuum Fourier transform spectrometry with balanced detection on a single photodiode,” J. Chem. Phys. 145(8), 084201 (2016). [CrossRef]  

9. I. Zorin, J. Kilgus, K. Duswald, B. Lendl, B. Heise, and M. Brandstetter, “Sensitivity-Enhanced Fourier Transform Mid-Infrared Spectroscopy Using a Supercontinuum Laser Source,” Appl. Spectrosc. 74(4), 485–493 (2020). [CrossRef]  

10. A. Foltynowicz, T. Ban, P. Masłowski, F. Adler, and J. Ye, “Quantum-Noise-Limited Optical Frequency Comb Spectroscopy,” Phys. Rev. Lett. 107(23), 233002 (2011). [CrossRef]  

11. A. Khodabakhsh, V. Ramaiah-Badarla, L. Rutkowski, A. C. Johansson, K. F. Lee, J. Jiang, C. Mohr, M. E. Fermann, and A. Foltynowicz, “Fourier transform and Vernier spectroscopy using an optical frequency comb at 3–5.4 μm,” Opt. Lett. 41(11), 2541–2544 (2016). [CrossRef]  

12. E. Boamfa, M. Steeghs, S. Cristescu, and F. Harren, “Trace gas detection from fermentation processes in apples; an intercomparison study between proton-transfer-reaction mass spectrometry and laser photoacoustics,” Int. J. Mass Spectrom. 239(2-3), 193–201 (2004). [CrossRef]  

13. D. Ke, E. Yahia, M. Mateos, and A. A. Kader, “Ethanolic Fermentation of `Bartlett’ Pears as Influenced by Ripening Stage and Atmospheric Composition,” J. Am. Soc. Hortic. Sci. 119(5), 976–982 (1994). [CrossRef]  

14. F. R. Nunes, C. A. Steffens, A. S. Heinzen, C. Soethe, M. A. Moreira, and C. V. T. d. Amarante, “Ethanol vapor treatment of ‘Laetitia’ plums stored under modified atmosphere,” Revista Brasileira de Fruticultura41(5), (Scielo, 2019). [CrossRef]  

15. F. B. Abeles, P. W. Morgan, and M. E. Saltveit, Ethylene in Plant Biology (Elsevier Science, 2012).

16. Q. Liu, X. Meng, Y. Li, C.-N. Zhao, G.-Y. Tang, and H.-B. Li, “Antibacterial and Antifungal Activities of Spices,” Int. J. Mol. Sci. 18(6), 1283 (2017). [CrossRef]  

17. S. Konstantinou, G. S. Karaoglanidis, G. A. Bardas, I. S. Minas, E. Doukas, and A. N. Markoglou, “Postharvest Fruit Rots of Apple in Greece: Pathogen Incidence and Relationships Between Fruit Quality Parameters, Cultivar Susceptibility, and Patulin Production,” Plant Dis. 95(6), 666–672 (2011). [CrossRef]  

18. A. Altemimi, N. Lakhssassi, A. Baharlouei, D. G. Watson, and D. A. Lightfoot, “Phytochemicals: Extraction, Isolation, and Identification of Bioactive Compounds from Plant Extracts,” Plants 6(4), 42 (2017). [CrossRef]  

19. J. Gustavsson, C. Cederberg, U. Sonesson, R. Van Otterdijk, and A. Meybeck, “Global food losses and food waste,” (FAO Rome, 2011).

20. V. Saptari, Fourier Transform Spectroscopy Instrumentation Engineering (SPIE Optical Engineering Press, 2004).

21. L. S. Rothman, I. E. Gordon, Y. Babikov, A. Barbe, D. Chris Benner, P. F. Bernath, M. Birk, L. Bizzocchi, V. Boudon, L. R. Brown, A. Campargue, K. Chance, E. A. Cohen, L. H. Coudert, V. M. Devi, B. J. Drouin, A. Fayt, J. M. Flaud, R. R. Gamache, J. J. Harrison, J. M. Hartmann, C. Hill, J. T. Hodges, D. Jacquemart, A. Jolly, J. Lamouroux, R. J. Le Roy, G. Li, D. A. Long, O. M. Lyulin, C. J. Mackie, S. T. Massie, S. Mikhailenko, H. S. P. Müller, O. V. Naumenko, A. V. Nikitin, J. Orphal, V. Perevalov, A. Perrin, E. R. Polovtseva, C. Richard, M. A. H. Smith, E. Starikova, K. Sung, S. Tashkun, J. Tennyson, G. C. Toon, V. G. Tyuterev, and G. Wagner, “The HITRAN2012 molecular spectroscopic database,” J. Quant. Spectrosc. Radiat. Transfer 130, 4–50 (2013). [CrossRef]  

22. N. R. Newbury, I. Coddington, and W. Swann, “Sensitivity of coherent dual-comb spectroscopy,” Opt. Express 18(8), 7929–7945 (2010). [CrossRef]  

23. O. Kara, F. Sweeney, M. Rutkauskas, C. Farrell, C. G. Leburn, and D. T. Reid, “Open-path multi-species remote sensing with a broadband optical parametric oscillator,” Opt. Express 27(15), 21358–21366 (2019). [CrossRef]  

24. M. Rutkauskas, M. Asenov, S. Ramamoorthy, and D. T. Reid, “Autonomous multi-species environmental gas sensing using drone-based Fourier-transform infrared spectroscopy,” Opt. Express 27(7), 9578–9587 (2019). [CrossRef]  

25. Z. E. Loparo, E. Ninnemann, Q. Ru, K. L. Vodopyanov, and S. S. Vasu, “Broadband mid-infrared optical parametric oscillator for dynamic high-temperature multi-species measurements in reacting systems,” Opt. Lett. 45(2), 491–494 (2020). [CrossRef]  

26. K. Johnson, P. Castro-Marin, O. Kara, C. Farrell, and D. T. Reid, “High resolution ZrF4-fiber-delivered multi-species infrared spectroscopy,” OSA Continuum 3(12), 3595–3603 (2020). [CrossRef]  

27. K. E. Jahromi, Q. Pan, L. Høgstedt, S. M. Friis, A. Khodabakhsh, P. M. Moselund, and F. J. Harren, “Mid-infrared supercontinuum-based upconversion detection for trace gas sensing,” Opt. Express 27(17), 24469–24480 (2019). [CrossRef]  

28. R. F. Evert, P. H. Raven, and S. E. Eichhorn, Biology of Plants (W. H. Freeman, 2012).

29. K. Eslami Jahromi, Q. Pan, A. Khodabakhsh, C. Sikkens, P. Assman, S. M. Cristescu, P. M. Moselund, M. Janssens, B. E. Verlinden, and F. J. M. Harren, “A Broadband Mid-Infrared Trace Gas Sensor Using Supercontinuum Light Source: Applications for Real-Time Quality Control for Fruit Storage,” Sensors 19(10), 2334 (2019). [CrossRef]  

30. E. I. Boamfa, P. C. Ram, M. B. Jackson, J. Reuss, and F. J. M. Harren, “Dynamic Aspects of Alcoholic Fermentation of Rice Seedlings in Response to Anaerobiosis and to Complete Submergence: Relationship to Submergence Tolerance,” Ann. Bot. 91(2), 279–290 (2003). [CrossRef]