Highly sensitive gas sensors that are capable to detect molecules even at low ppbv (10⁻⁹ Volume Mixing Ratio, VMR) or sub ppbv concentrations have recently become one of the most in-demand kinds of devices for scientific research and industrial applications. They are routinely used for tracing relevant environmental species [1–3], in medical [4,5] or clean-room diagnostics [6–8], as well as for the monitoring of hazardous or explosive materials [9,10]. Despite this remarkable versatility there are only a few analytical techniques that can offer the required levels of sensitivity; probably the most common among them are gas chromatography, mass spectrometry, and optical absorption spectroscopy.

Gas chromatography is readily used to identify the symptomatic of metabolism related diseases [11,12]. For example, untreated diabetes causes an increased concentration of exhaled acetone, while trace concentration of dimethylsulfide, methyl or ethyl mercaptane often indicate liver impairment. Methylated alkanes and benzene derivatives are known to be frequent markers of lung cancer; however, it is now clear that the effective diagnosis of this disease must take into account specific molecular profiles rather than individual compounds [13]. A multicomponent technique, like gas chromatography, is particularly well suited to accomplish this task, and a number of clinical studies have been reported in this field (see e.g., [14–16]). Besides medical applications, gas chromatographs are also used for environmental analysis [17], nonetheless, in situ measurements are rather seldom, most likely to due to the bulky construction of the chromatograph’s diffusion columns.

Mass spectrometry is often combined with gas chromatography constituting the so-called Gas Chromatography-Mass Spectrometry (GC-MS) technique [18]. In GC-MS a gas chromatograph separates the constituents of the sample that are subsequently ionized and detected by a mass spectrometer—typically a time-of-fight, ion trap or quadrupole (multipole) filter. The sensitivity of modern mass spectrometers is typically high and their mass accuracy parameter, Δ(m/Z)/(m/Z) (m, Z are molecular mass and charge, respectively), can be as small as 10⁻⁶. It is more than sufficient for even most demanding applications of breath analysis and allows to detect a change of mass corresponding to a single hydrogen among molecules composed of several hundred thousands of atoms. Mass spectrometers are also used as stand-alone detectors in environmental or atmospheric studies [19,20].

At present, absorption spectroscopy is probably the most straightforward, robust, and at the same time cost effective and flexible sensing technique for compound analysis. Microscopically, the effect of light absorption consists in the energy transfer from individual photons to atoms or molecules in the medium. This energy is next converted to other forms of energy, for example to heat, leading to a dissipation of the photon flux, but since the absorption spectra are always strictly related to the chemical composition of a material they can be used to identify their constituents. Inversely, if the absorbers and the absorption cross-sections are known it is possible to quantify the molecular concentrations by determining the magnitude of the radiation loss inside the sample cell (single- or multi-pass). One technique, which for this purpose uses a stable optical resonator, is known as Cavity Enhanced Absorption Spectroscopy (CEAS).

Optical resonators (optical cavities) that are suitable for the CEAS are characterized by a high quality factor (Q-factor) >10⁴ ensuring several tens of back-and-forth reflections of light during its decay and effective absorption lengths of tens of kilometres for meter-sized cavities. Thus, CEAS sensors have low detection limits and sensitivities of single pptv (VMR of 10⁻¹²) [21], which however, strongly depend on the reflectance of the cavity mirrors. As a consequence, the performance of these devices is very likely to degenerate in time due to the ageing process of the mirror coatings. Moreover, if the actual reflectance of the cavity mirrors is not precisely known or substantially differs from that assumed, quantitative CEAS results become uncertain. This problem is widely known and recently has been reported by Watt et al., who suggested to perform additional Cavity Ring-Down Spectroscopy (CRDS) measurements in order to calibrate CEAS detectors [22].

The CRDS technique is principally distinct from CEAS, although both are sometimes classified together as cavity-enhanced techniques [23,24]. This is justified only in the sense that they both are based on high Q-factor optical cavities, but it fails to account for the fundamentally different measurement principles: in CRDS the temporal decay of the optical energy trapped in the resonator is observed, while in CEAS the absolute time-averaged intensity transmitted through it. The primary result of a CRDS measurement, the rate of intra-cavity energy loss, is expressed in terms of the so called ring-down time, i.e., the decay constant of the CRDS signal. Since the difference of ring-down times determined for the empty and absorber filled cavity is directly proportional to the extinction coefficient, α, the CRDS is a self-referencing technique, and contrary to CEAS it is not affected by the accuracy to which the reflectance of the cavity mirrors is known; still it benefits from long absorption lengths provided by high Q-factor cavities. Initially, the CRDS was developed as a reflectometric tool for the characterization of low-loss, highly reflective mirror coatings [25]. Shortly after it was further adapted to measure absorption spectra [26]

Cavity-enhanced methods are appropriate for single- or multi-wavelength analysis of absorbers and recent advancement in the design and manufacturing of Photonic Crystal Fibres (PCFs) and the development of supercontinuum (SC) generation technologies revitalized the interest in multispectral techniques. Today, PCF-based SC light sources are readily used in CEAS [21,22,27–29], but surprisingly, they are not so popular in CRDS. To bridge this gap we have developed a multiwavelength CRDS gas sensor, which to the best of our knowledge for the first time incorporates a miniature PCF supercontinuum laser.

A distinction between monochromatic and broadband CRDS is somewhat arbitrary and usually refers to the optical bandwidth of the incident light. The former typically uses narrow line lasers (preferably continuous wave, CW) that are spectrally and spatially matched to excite only the fundamental mode of the cavity (TEM₀₀). A single-mode exponential decay is strictly monotone, which facilitates the evaluation of decay constants by reducing errors related to the fit, and which has a positive impact on the sensor’s sensitivity. But even in a single-wavelength measurement it is possible to gain knowledge about absorption spectra, yet in a stepwise manner, if the laser’s wavelength can be scanned over a line or a band of interest [26,30–33]. This scheme is called the “wavelength scan method” [24].

The broadband approach employs a different principle. Here, the light source has a sufficiently broad spectrum to cover most of the characteristic spectral features of the investigated absorber. The ring-down events are simultaneously recorded for more than one wavelength using a wavelength and time resolving detector. Thus, broadband sensors are preferable for analytical applications as they provide a fast means to discriminate between different molecular spectra. The tradeoff is the excitation of many cavity modes (longitudinal and transversal) resulting in multi-exponential ring-down decay [34]. Furthermore, interference between modes usually causes characteristic oscillations of CRDS signals [35]; however, there are also adequate methods that allow to reduce the impact of those less desirable side effects of multimode operation of the cavity. We will describe some of them later on.

Broadband techniques can be divided in two main groups. The first one constitutes a direct analogue of the wavelength scan principle with the exception that the selection of wavelengths takes place not on the side of the light source, but on the detector’s side, where the radiation leaking out of the cavity is filtered by a monochromator or a tuneable narrow line interference filter. Many authors have successfully applied this procedure, for example, Crosson et al. [36], who used broadband radiation from a free electron laser (25 nm FWHM centred at 5.38 μm) to measure mid-IR spectra of water vapour at a spectral resolution of 0.03 nm, or Thorpe et al. [37] who investigated gas phase absorption spectra of several molecules (H₂O, O₂, NH₃ and C₂H₂) with a frequency-comb system. A distinctive feature of this device was its exceptionally broad spectral width of nearly 100 nm (at 0.5 nm wavelength resolution). Fourier transform and Fourier transform phase-shift methods constitute a separate subgroup of wavelength selective CRDS. The feasibility of Fourier transform CRDS was demonstrated for the first time by Engeln and Meijer [38] who, using broadband fluorescence radiation from a dye laser (emission spectral width 400 cm⁻¹), measured a rotationally resolved spectrum of molecular oxygen between 760 and 768 nm. Several years later, Hamers et al. [39] used the same absorption band to demonstrate the principle of phase-shift CRDS. In this study the molecules were excited by the spectrally broad white-light from a Xenon arc lamp. However, to improve the contrast of the recorded interferograms a narrowband transmission filter (4 nm FWHM at 764 nm) had to be used, greatly reducing the actual bandwidth covered by the measurement.

A second group of broadband cavity-based methods comprise simultaneous time-wavelength techniques that are sometimes called two-dimensional [24]. Conceptually, they are very simple: broadband incident light initiates multi-wavelength ring-down, and a suitable array of photodetectors registers time-dependent, spectrally resolved intensity changes of light leaking out of the cavity—the challenge here is purely technical and mainly related to the detector. Namely, currently there are not so many detectors that would provide time resolution and wavelength discrimination simultaneously. Photomultipliers or photodiodes are sufficiently fast to measure CRDS signals of approximately μs duration, but currently their number available in an array is limited. Additionally, the area of these detectors is typically too large (∼0.5 mm × 0.5 mm) to provide a satisfactory wavelength resolution when they are attached to commercially available spectrographs with linear dispersions matched to commonly used array detectors. The elements of CMOS sensor arrays are small, but as typical day-time image sensors they are not well suited for low-intensity light applications. Charge Coupled Device (CCD) arrays are sensitive and have small pixel sizes (∼10 × 10 μm), but since they provide at best μs readout times of individual pixels (ms readout times of a full chip), they are too slow to capture CRDS signals if operated in one of the conventional exposure/readout modes.

The first CRDS technique that successfully coped with these issues—Ring-Down Spectral Photography—incorporated a streak camera principle [40,41]. The cavity outgoing light was spectrally resolved along one coordinate of the CCD detector by a grating spectrograph, and additionally deflected by a rotating mirror, scanning the dispersed light along the other coordinate such that its x- and y- directions corresponded to time and wavelength coordinates, and “spectral snap-shots” of the ring-down decay were taken in each detector row. The same approach was more recently adopted in frequency-comb based CRDS as an alternative method of obtaining molecular spectra [37].

Ball et al. [42,43] simplified the spectral photography setup by replacing the rotating mirror with a specially modified CCD camera. This device embodied an optically opaque mask situated right in front of the detector array. Its hidden part served as photo charge storage and was not exposed to the incident light. A few uppermost, uncovered pixel rows were used for actually generating photocharge, which was sequentially shifted by from the illuminated to the dark region of the detector. In each so called “clocking” event the charge was transferred to the next empty hidden pixel row, encoding the temporal evolution of the signal by the amount of charge contained in each row. At the end of this sequence the total charge was read row by row via the shift-register, and the CCD array was charge-cleaned for the next registration. The charge transfer method was not explicitly fast, and, as admitted by the authors, provided typical time resolution of 5 μs, only a factor 10 shorter than the average duration of CRDS signals [42].

Cavity Ring-Down Spectrography (CRD-Spectrography) bases on a simpler principle and does not require any custom-made detectors. Here, the multi-wavelength ring-down is reconstructed from a set of cavity transmitted spectra registered as independent exposures at successive, discrete delays with respect to the injected light pulse. This however requires that for every such exposure a new cavity decay has to be initiated. In addition, spectra corresponding to a particular delay are typically averaged over many laser shots to reduce noise. This procedure can be automated and is not labour-intensive, but it is inherently sensitive to any pulse energy drift of the light source, particularly if the time required for the spectrum acquisition is longer than the characteristic drift time; however, the number of time-sampled spectra and consequently the duration of the measurement is directly controllable by the number of steps in the delay sequence. The first reported CRD-Spectrography measurements on a real world sample—NO₂– gas have been performed by Czyżewski et al. [44]. Their spectrograph operated on amplified fluorescence of stilbene 3 dye delivering 15 nm FWHM spectral bandwidth between 410−440 nm. This broadband radiation was passed trough the ring-down cavity and dispersed in wavelength by a spectrograph. A time evolution of the spectra was reconstructed from successive acquisitions using an intensifier fitted CCD camera (ICCD) that registered ring-down events at well defined delays with respect to the laser pulse. The sensitivity of this setup was rather modest ∼10⁻⁵ cm⁻¹ and, as admitted by the authors, was most likely attributable to relatively low reflectivity of cavity mirrors; between 99.0% and 99.8% in selected wavelength range.

Two-dimensional CRDS setups as they were briefly described above and in general the majority of two-dimensional setups use dye-lasers to generate spectrally broad light for the cavity excitation. They are capable to deliver sub-MW optical powers typically required in broadband measurements due to the low absolute transmission of optical cavities having effective transmission (input and output mirrors) coefficients T∼10⁻⁶ or less. Lack of radiation sources being: (i) spectrally broad and (ii) rich in spectral power density (optical power per unit wavelength) has been limiting multiwavelenght CRDS for a long lime. The situation has changed with the discovery of SC light which is generated when the spectrum of a laser pulse of typically picosecond or femtosecond duration is broadened due to its nonlinear interaction with the medium in which it propagates. The short duration of the pulse ensures that MW optical powers can be generated at moderate energies of several μJ or even nJ. Not surprising is therefore that the first SC-CRDS sensor entirely based on the short pump pulse SC technology. The sensor was constructed by the group of Stelmaszczyk et al. and since then the technique is known as a Supercontinuum CRDS (SC-CRDS) [45].

In the innovative approach of Stelmaszczyk et al. [45] the SC was obtained via multiple filamentation of the beam [47] by transmitting a train of fs laser pulses (800 nm wavelength) through a block made of fused silica. This type of material is highly non-linear yielding efficient SC generation [46]. When injected into an optical cavity the SC initiates multi-wavelength light ring-down characterized by wavelength dependent decay constants. Exactly this principle was employed by Stelmaszczyk et al. to measure the wavelength dependent reflectance of UV CRDS mirrors. Yet, these measurements were not strictly two-dimensional basing on the wavelength-scan principle. Therefore, the group aimed at demonstrating full two-dimensional capabilities of CRDS in a subsequent experiment by measuring the absorption spectrum (423−455 nm) of NO₂ gas [48]. The setup implemented in this case based on a CRD-Spectrography scheme and reached a detection limit of an absorption coefficient α∼6.2 × 10⁻⁸ cm⁻¹, corresponding to 5 ppbv of NO₂ when taking into account the average absorption cross-section of this molecule in the respective wavelength range.

More recently, the Supercontinuum Cavity Ring-Down Spectrography (SC-CRD-Spectrography) has been proposed as a broadband diagnostic tool for highly reflective mirror coatings [49,50]. Although this idea is essentially not new and, as mentioned above, has been already demonstrated by earlier investigations of Stelmaszczyk et al. [45,48], the new aspect of the work [49,50] is the SC light source itself, because the authors for the first time use SC radiation generated in a PCF. Nonetheless, it has never been proved whether or not such PCF-based supercontinuum sources are also eligible in ring-down setups that are designed to detect molecular absorption. First measurements with the system described below clearly confirmed such possibility.

The VIS absorption (610–730 nm) of atmospheric air at normal pressure was investigated with the SC-CRDS sensor of Figure 1. Before each experiment run the cavity was evacuated and reference decay constants were determined. The resonator was next filled with dust free air initially purged through a PTFE Teflon® syringe filter (GE Osmonics Labstore, pore size 0.45 μm) to remove bigger particles and minimize losses related to Mie scattering. The gas inside the cavity consisted of a mixture of dry atmospheric air (lab air) and humid air, which was generated by bubbling the dry air in a glass vessel containing distilled water. Depending on the rate, normally 2−6 litres per minute, the relative humidity (RH) of the humid air was typically between 50−80%. The RH of the dry air was systematically lower, in the range of 20−40%, and was mainly dependent on meteorological conditions. The dry and humid air were mixed at known partial pressures the sum of which was corresponding to the ambient pressure. Additionally, the actual value of RH inside the cavity was verified before each series of experiments with the precision of ±2% absolute using the thermohygrometer (ATP Messtechnik GmbH P470). Direct contact of the gas with stainless-steel cavity walls provided a good thermal equilibrium with the ambient temperature of 22 °C. A special precaution was taken to assure a quasi-adiabatic expansion of the air, which otherwise would lead to water condensation.

At atmospheric pressure the fraction of H₂O molecules that is adsorbed on the cavity walls is typically negligibly small. Thus, it is reasonable to estimate the actual concentration of gas-phase water dipoles(N_H2O) from the physical parameters of the inlet air and the formula: (4) N H 2 O = p * H 2 O p atm V k B T ( R H dry     p dry + R H humid     p humid ) .

It takes into account an ideal gas. In Equation 4 p^*_H2O and p_atm denote the saturation pressure of water vapour [66] and the atmospheric pressure, respectively, V is the volume of the cavity, k_B is the Boltzmann constant, T is the temperature and the RH and p correspond to the relative humidity and the pressure of the dry and humid air. The result of such theoretical estimation is shown in the Figure 6. In the presented case the partial pressure of humid air is 500 hPa. The left panel corresponds to the number concentration of water molecules, while the right one to their corresponding volume mixing ratio. Both quantities are plotted as a function of the RH of the dry and humid air. According to the calculation the partial pressure of water in the cavity does not exceed 17 hPa (1.6 % VMR) and for the ambient temperature of 22 °C is well below the saturation pressure of 25.5 hPa [66]. Indeed, virtually all water molecules inside the cavity are in the gas phase.

Figure 7(c) shows a spectrum as it was typically obtained in experiments. It covers two stretching-bending overtones of H₂O (polyads 4ν + δ and 4ν) and two vibronic transitions of the weak spin-forbidden b-X band of O₂ (b¹Σg⁺(ν″ = 1)←X³Σg⁻(ν′ = 0) and b¹Σg⁺(ν″ = 2)←X³Σg⁻(ν′ = 0)). Figure 7(b) and 7(a) present, respectively, the line strengths and simulated absorption spectrum of air for the corresponding wavelength range. For the simulation the Hitran 2000 data base [67] was used and 21% VMR of oxygen and the 53% RH were assumed the latter corresponding to the value measured in the cavity.

Since the ratios of line strengths of O₂/H₂O do not differ by more than 5–8% absolute, we can state relatively good agreement between measurements and simulations. Nonetheless, it is sensible to ask why it is not possible to obtain accurate quantitative information in our measurements even though the absorption cross-sections were precisely known. It turns out that the problem is more general and relates to the wavelength resolution of the measurement, which in our case was not sufficient to resolve narrow line resonances of absorbing species, because typical line widths of, O₂ and H₂O at normal pressure (1,013 hPa) are 7 GHz (0.01 nm) and 4 GHz (0.005 nm), respectively, far beyond the resolving capabilities even of the 1,800 grooves/nm grating. Such narrowband resonances tend to saturate over long absorption paths encountered in the cavity, and since even a weak saturation suffices to alter the spectra, they cannot be represented by a simple convolution of absorption cross-sections with the instrument slit function. The net result is that true molecular concentrations are hardly derivable unless special retrieval algorithms are used [43]. Such advanced data treatment is, however, beyond the scope of the present work, which aimed at demonstrating applicability of PCFs in CRD-Spectrography.

Finally, we point out that the magnitude of energy dissipation in the cavity that is not related with absorption losses or transmission losses through the mirrors further confirms a good quantitative performance of the sensor. It can be determined in spectral windows that are free from molecular resonances of O₂ and H₂O; in our case 610−625 nm and 670−680 nm. The wavelength averaged rate of the energy loss determined in this way is 7.1 ± 1.2 × 10⁻⁸ cm⁻¹. The Rayleigh scattering theory yields a similar value of 5.9 × 10⁻⁸ cm⁻¹ and the small discrepancy of about 1 × 10⁻⁸ cm⁻¹ is likely caused by Mie scattering on residual dust particles. In any case, the overwhelming part of the background signal must have been caused by Rayleigh scattering.

We have demonstrated the suitability of a compact PCF based SC sources for the simultaneous CRDS measurement of multiple absorbing components in a broad spectral region. This simple approach offers a range of several potential applications such as humidity control, monitoring of air quality in tunnels and chimneys, detection of hazardous substances in airports etc., where a fast and reliable sensor to detect changes in concentration is needed.

The stray light coming from internal reflections inside the spectrograph can considerably affect the signal, causing multi-exponential decays at the very beginning of the transients, and it also limits the dynamic range of the detector due to the initial high intensities. Thus, one should consider using gated devices to control the exposure time of the sensor, besides filters to select the spectral range of interest and reduce problems related to stray-light reflection that are common for grating spectrometers. Additionally, the reflection coefficient curve of the SC-CRDS mirrors should be possibly flat to achieve similar dynamic ranges at all wavelengths of interest.

Although the maximum resolution of the instrument is approximately one order of magnitude lower than would be necessary to resolve the absorption lines of typical atmospheric relevant molecular absorbers, it is not a limiting factor for their identification. However, accurate quantitative information on their concentration can be obtained only when special retrieval algorithms are applied, which produce a linearized absorption cross-sections throughout the fit of the natural logarithm of the time dependent cavity transmission spectra convolved with the resolution of the instrument as described in [43].