Optical gas sensing: a review

TL;DR: In this article, the authors present the basis for each technique, recent developments in methods and performance limitations, and present a performance comparison of different techniques, taking data reported over the preceding decade, and draw conclusions from this benchmarking.

Abstract: The detection and measurement of gas concentrations using the characteristic optical absorption of the gas species is important for both understanding and monitoring a variety of phenomena from industrial processes to environmental change. This study reviews the field, covering several individual gas detection techniques including non-dispersive infrared, spectrophotometry, tunable diode laser spectroscopy and photoacoustic spectroscopy. We present the basis for each technique, recent developments in methods and performance limitations. The technology available to support this field, in terms of key components such as light sources and gas cells, has advanced rapidly in recent years and we discuss these new developments. Finally, we present a performance comparison of different techniques, taking data reported over the preceding decade, and draw conclusions from this benchmarking.

Summary

1 Introduction

• Gas detection has an impact across a wide range of applications.
• Various potential biomarker gases are also under study for use in breath diagnostics, including nitric oxide (NO), ethane, ammonia (NH3), and many more [3] .
• Because the transduction method makes a direct measurement of a molecule’s physical properties (its absorption at a specific wavelength), drift is reduced and, because the incident light intensity can be determined, measurements are self-referenced, making them inherently reliable.
• Gas detection applications can cover a very wide range of gas concentrations.
• Several examples are summarised in Table 1.

2 Basic principles

• Many chemical species exhibit strong absorption in the UV/visible, near infrared or mid infrared regions of the electromagnetic spectrum.
• The absorption lines or bands are specific to each species and this forms the basis for their detection and measurement.
• The availability of high quality sources and detectors, primarily derived from telecommunications applications, can counteract this disadvantage and signal : noise ratios can be relatively high.
• In analytical chemistry and for liquid phase samples, the Beer Lambert Law is typically described using base 10 rather than e, with the result that quoted values of α are 2.3 times smaller, despite having the same apparent units.
• Variations in the linewidth at different pressures must be accounted for, especially in high resolution measurement schemes.

3 Optical gas cells

• Use of long path cells, to increase the magnitude of the signal according to equation (2).
• Use of hollow core optical fibre to form a long, thin gas cell.
• Coupling light into many of these cells is facilitated by the use of lasers, as a significantly greater proportion of the light may be collimated in a narrow beam and / or launched into optical fibre than would be the case for broadband sources.
• Each configuration is now discussed in turn.
• Because gas lines can be inherently narrow, interference fringes can also be a significant performance-limiting factor in broadband absorption measurements [23] .

3.1 Long path gas cells

• Gas cells with the highest sensitivity often employ multipass cells with pathlengths from tens of metres (multipass cells such as those devised by Herriott [24] , White [25] and Chernin [26] ) to km (using so-called cavity-enhanced and ringdown techniques).
• When using a Herriott cell, one might expect the improvement in SNR to scale with the Page 8 pathlength enhancement.
• Various alternative approaches have been demonstrated in recent years; a number are compared in Table 3.
• This was coupled to an FTIR spectrometer and used to measure CO2 at 4.2μm.
• Integrating spheres are formed by taking a material with high diffuse reflectivity (typically barium sulfate, poly[tetrafluoroethylene].

3.2 Gas cells linked via optical fibre

• Common to most is that light is guided in a core region, typically with the core having a higher refractive index than the surrounding cladding, setting up the right conditions for total internal reflection for light incident over a given range of angles.
• The drawing technique provides an excellent surface finish, in particular at the junction between core and cladding, such that scattering losses at this interface are very low, and centering of the core is achieved to a high precision.
• The first is that multiplexing of multiple gas cells allows the relatively high cost of certain light sources (such as tunable diode lasers) to be shared, bringing the cost per sensor head to a level where it can compete with traditional installed sensors.
• Attempts to multiplex more than one sensor head per fibre have included deliberate use of etalons created by gas cells of unequal length in a ladder topology [50] , and frequency shift interferometry [51] , a technique formerly used to interrogate optical fibre Bragg gratings.

3.3 Use of microstructured optical fibre

• Photonic crystal fibre (PCF) was originally developed for photonic and telecommunications applications [53] .
• A periodic refractive index within the fibre (typically generated by an array of holes that pass along the length) forms a photonic crystal, the periodicity of the refractive index in the photonic crystal being analogous to the periodicity of electronic energy levels in more conventional ionic crystals.
• A photonic quantum well is formed that confines the light modes more effectively than in conventional fibre, such that, in principle, photonic crystal fibres of the right lattice spacing and hole size can maintain singlemode operation over almost their entire transmission range [53] .
• Solid core fibres can also be used where the core is small and the guided mode extends into the surrounding voids, ideally with very thin supporting structures [55] in so-called suspended core fibres, in which the level of mode / air overlap is small for gas phase sensing.
• Figure 8 shows examples of microstructured optical fibres used in chemical sensing.

3.3.1 Hollow core waveguides

• Interest has been strong in the use of hollow core waveguides as a long pathlength, low volume gas cell.
• Sub-nanolitre sample volumes have even been demonstrated for gas sensing [63] .
• Following the same conventions as the majority of the literature, the authors term these structures hollow core waveguides (HCWG), in contrast to hollow core photonic bandgap fibres (HC-PBGF) described above.
• Losses can be consequently high, but can generally be accommodated over distances of several metres without significantly affecting sensor performance.
• The attenuation in hollow core waveguides varies with inner core radius a and bend radius r as 1/a 3 and 1/r [67] ; typical figures are shown in Figure 9.

3.3.2 Performance limitations

• The feasibility of using the same design of waveguide for spectral measurements in the UV (180-205 nm) has also been reported [65] .
• Page 15 Charlton et al have coupled the output from a 10.3 μm quantum cascade laser (QCL, see section 6.3) into a 700 μm core diameter capillary of this type, for the purpose of detecting ethyl chloride [66] .
• They used a cell length of 4 m and HgCdTe detector to achieve a detection limit of 0.5ppm, but the noise equivalent absorbance (NEA) was not stated.
• This was reduced for (i) straight fibres rather than bent fibres, (ii) locating the detector as close as possible to the end of the fibre, so as to collect a larger number of modes, and finally by vibrating the fibre at a frequency much greater than the measurement bandwidth, so as to average over a larger number of modes.

3.3.3 Practical matters

• Photonic crystal fibres and other PBGFs can be cleaved using standard equipment and spliced to each other and to conventional fibre [53] .
• By recleaving the joint after splicing, they were able to show the potential for surface tension to create a concave cavity in this case, shown in Figure 11.
• Techniques to improve the sample filling time include increasing the pressure difference across the fibre to drive the gas through, increasing the cell diameter (at the expense of using a larger volume cell) or introducing holes to allow gas flow or diffusion along the waveguide’s length, reducing the effective value of .
• Holes can be drilled to a high tolerance using femtosecond lasers and have the advantage of being convenient to use once in place.
• Slowly pumping the fluid through the fibre during the drilling process also had the benefit of removing debris.

4 Non-dispersive sensors

• Broadband, non-dispersive gas sensing is one of the simplest techniques to construct, and its great commercial significance is perhaps a result of this.
• Similar broadband measurements can be made in the mid infrared (in traditional non-dispersive infrared or NDIR), near infrared and UV / visible regions of the spectrum.
• Sensors can be built that are very compact, with NDIR sensors in recent times being built into a standard form factor consisting of a 16mm long, 20mm diameter cylinder.
• These sensors are low cost, having few components (a simple microbulb light source, gold coated reflective light path and integrated detector containing two or more filtered detection channels).
• These applications include heating, ventilation and air conditioning (HVAC) control, industrial safety especially in the brewing industry (CO2 is an asphyxiant), and capnography (the measurement of time-resolved carbon dioxide concentration in exhaled breath) for patient monitoring for example during anaesthesia.

4.1 Principle of operation

• Typically, emission from a broadband source (such as a microbulb [79] ) is passed through two filters, one covering the whole absorption band of the target gas (in the active channel), and the other covering a neighbouring non-absorbed region (the reference channel).
• For low αℓ, equation (5) is again linear with α, and therefore with gas concentration.
• The reference channel is used to compensate for changes in the emission of the source, which are assumed to affect the reference and active channel wavelengths in equal proportion.
• For NDIR the figure corresponds to a noise equivalent absorbance where the “absorbance” concerned is an integrated effective absorbance over the measurement band.
• The instrument achieves a reported limit of detection of 0.11 ppm CO2 with an optical pathlength of 12.5 cm, operating in the 4.2 μm measurement band.

4.2 Light sources for NDIR

• Here the authors consider broadband sources used in non-dispersive measurements; lasers are considered later in section 6.3.
• For improved modulation frequencies to have the greatest impact, thermal detectors also need to be optimised for higher frequency operation, otherwise semiconductor detectors (eg PbSe), which have bandwidths of kHz or more, must be used.
• The authors can discuss the consequences of these developments for gas detection.
• They offer superior spectral efficiency than thermal sources, with emission better matched to the gas absorption band [ 105 ] .
• Under research are LEDs with shorter wavelength emission; Figure 20 illustrates the latest results, including AlGaN based LEDs with centre wavelengths down to 222 nm [123] , and AlN based LEDs down to 211 nm [124 ] , almost providing sufficient coverage for the entire 200-400 nm region.

4.3 Non-dispersive gas cell development

• Over the last decade, the commercial market has become populated with small footprint gas sensors based on the NDIR principle [126,127] .
• Thus, equipment manufacturers need not alter the dimensions of their products when switching to NDIR sensors, and in some cases even the first stage electronics design is also similar.
• Research in this area has concentrated on the following requirements: (i) the need for compact cells, (ii) the need to maximise the proportion of light coupled through the cell to the detector, in order to overcome the detector noise limit, and (iii) in applications requiring very low limits of detection, the need for very long pathlength cells.
• Page 27 Non-dispersive gas sensing has been brought a step closer to dispersive spectrophotometric analysis by the development of IR filter arrays.
• As long as there is a Page 28 temperature difference between the background scene and the target gas, contrast will result in the image.

4.4 Correlation spectroscopy

• With a sufficiently large absorption coefficient in the cell, only light with wavelengths that are not absorbed by the target gas can reach the detector.
• Correlation spectroscopy can therefore provide selectivity to the target gas using relatively simple equipment, using the gas itself to provide a high resolution matched filter.
• Corman et al have also developed a micro miniature correlation system in which the beam separation was less than 2 mm and the total path traversed by the light was of the order of 10-20 mm [140] .
• Indeed, Kebabian et al have found that, in the case of NO2 detection, a temperature-dependent equilibrium between the target gas and its dimer in the reference cell caused thermal drifts in the output at the level of 0.5 ppm K -1 , responsible for the limiting performance of their instrument [142] .
• Use of the reference cell allows real-time subtraction of the background, allowing false colour imaging without the need to acquire a stable, gasfree background image.

5 Spectrophotometry

• Multiple gases may be detected provided that their absorption lines fall within the wavelength range of the spectrometer, and regions can often be found with little or no absorption to act as a reference, in the manner of the reference measurement in NDIR.
• Absorption strengths in the mid IR Page 32 are also sufficiently strong for good quality spectrophotometric measurements of gases to be made.
• These may be fibre optically coupled so that the electronics can be housed separately from the gas cell.
• UV spectral measurements of different gas species frequently overlap; to discriminate individual gases, spectra must be deconvolved, often using a partial least squares method.
• More advanced algorithms have also been developed to enable low concentration measurements to be made over shorter pathlengths typical of industrial applications [151] .

5.1 Spectrophotometer elements

• New designs for compact spectrophotometers have resulted from both advances in manufacturing techniques and from ingenuity on the part of developers.
• A number of groups have developed MEMs-actuated Fabry-Perot filters for use in spectroscopy [154] .
• The team has also extended the wavelength capability to cover the 8-11 μm range [158] .
• High quality laboratory FT spectrometers are typically based on a Michelson interferometer, one arm of which employs a moving mirror, the spectral resolution being inversely related to the optical path difference resulting from mirror displacement.
• A high resolution spectrometer Page 37 has been configured for measurement of ozone with a spectral range of 307–310.5 nm and resolution of 0.0058 nm [166] , fully resolving the gas absorption lines but with no ability to measure other gases.

6 Tunable diode laser spectroscopy

• With tunable diode laser spectroscopy (TDLS), the emission wavelength of a narrow linewidth laser diode is scanned across an individual gas absorption line at very high resolution [ 177 ] .
• Working at such high resolutions therefore gives the following advantages:.
• A high degree of specificity to the target gas, as long as a region can be found for which the narrow individual absorption line of the target has no overlap with lines of potential interferents.
• Fast operation; the wavelength of many diode lasers can be modulated at frequencies up to MHz, however commonly used signal recovery electronics is often limited to around 100 kHz.
• TDLS has been comprehensively reviewed elsewhere [1,178,179] , therefore the purpose of this section is to bring these reviews up to date with recent developments.

6.1 Basic principles of TDLS

• For simple, single pass or folded path gas cells there are two commonly used techniques for TDLS, termed direct spectroscopy or line scanning, and wavelength modulation spectroscopy.
• The latter also has the (usually undesirable) result of increasing the emitted power, known as residual amplitude modulation (RAM), shown in Figure 33(a).
• The method confers two advantages: an improved signal : noise ratio resulting from improved sensitivity to curved spectral features of the appropriate width, and a zero baseline, improving zero stability.
• The typical forms of different demodulated harmonics are shown in Figure 35 for a single gas line.
• It is not the gas absorption wavelength that varies, but the mean emission wavelength of the laser diode, which cannot be determined with sufficient precision to enable wavelength control, other than by reference to the position of the gas line.

6.2 Limitations and improvements to TDLS

• Detection sensitivity in TDLS is often limited by optical interference fringes rather than the theoretical limit given by detector noise [192,193] .
• Gas absorption linewidths are narrow (full width at half maximum of ~5 GHz at atmospheric pressure) and therefore to resolve the gas line requires the use of lasers with narrower emission linewidths; tens of MHz is typical, and more than adequate for pressure broadened gas lines.
• Small Fresnel reflections often persist in the optical path [177] or develop after a period of time in the field [44] .
• Noise ratio for the measurement is to be improved, other sources of noise and uncertainty, such as interference fringes, must be reduced below the level of laser excess noise, also known as If the signal.
• For WMS, many commercial lock-in amplifiers are often limited to a maximum operating frequency of around 100 kHz; note that this determines the maximum frequency of the highest recovered harmonic.

6.3.1 Solid state, monolithic devices

• In this section the authors consider laser diodes and their mid IR counterparts (which are not strictly diodes but can be operated in a similar manner).
• Figure 38 shows examples of the resulting spectra for acetylene.
• For the spectra to remain stable these laser diodes must be made with reduced gain in the cavity [207] .
• Decoupling the emission wavelength from the bandgap has allowed QCLs to be developed over a very wide range of wavelengths from the mid IR to the terahertz (THz) region, however their use for gas sensing in the THz region has been limited [211] .
• Interband cascade lasers (ICLs) cover an important gap in wavelength coverage between diode lasers and QCLs with a standard operating range between 2.5 and 4.4 μm.

6.3.2 Widely tunable laser systems

• External cavity lasers have long offered narrow linewidth, singlemode operation with tuning over a wide range [216] .
• These are suitable for use in external cavity lasers with wide tuning that cover a range of different gas species [221] , and turnkey systems have been commercialised by several companies [222] .
• These systems therefore have the potential to replace FTIR spectrometers for certain applications requiring a reduced wavelength range.
• The key to this process is an optical crystal with a high nonlinear coefficient, which is transparent at the wavelengths concerned, such as lithium niobate (LiNbO3).
• For efficient nonlinear conversion, conservation of momentum also requires phase matching of the pump, signal and idler beams.

6.3.3 Associated light sources

• The following light sources offer similar advantages and employ pump lasers in their generation, so they are included here.
• Supercontinua have been generated in standard telecoms fibre [238,239] and optical fibre tapers [240] , however the use of femtosecond pulsed pumps with specially tailored PCF generally offers the highest conversion efficiency [240,241] .
• Page 49 A final development is the use of frequency combs for gas spectroscopy.
• Originally developed for precision time and frequency metrology, frequency combs employ femtosecond (fs) pulsed lasers to generate a comb of precisely evenly spaced wavelength emission modes over a wide spectral range, with a spacing uniformity that can be better than a few parts in 10 17 [244] .
• Compared with a conventional FTIR employing an incoherent source, measurements have higher spectral brightness and a better signal to noise ratio, which in turn enables shorter averaging times and faster FTIR scanning.

6.4 Standoff detection of backscattered light

• Lasers provide the opportunity for novel detection geometries including remote standoff detection of gases, typically based on tunable diode lasers operating over distances of 10 m or more [186,254,255] .
• Iseki et al have modelled their system’s reflectivity and noise performance, relating this to a minimum detectable gas concentration signals in ppm.m for different circumstances of target Page 51 reflectivity and distance [254] .
• They demonstrated gas leak detection with leak rates as low as 0.01 litre min -1 and concentrations of around 100 ppm.m.
• For this reason, the correspondence between the standoff measurement and that of a point sampling instrument is surprisingly good, as Figure 45 shows.
• Higher power lasers can provide greater resolution and / or refresh rates, and the use of the mid IR again improves detection limits.

6.5 Light scattering in TDLS

• The advantages are very simple alignment and improved stability to temperature changes.
• In reflective cells, the optical configuration is similar to that found in the standoff laser pointers of the previous section, albeit with a fixed, known pathlength.
• A group at Lund University have used TDLS to explore gas concentrations held within porous and scattering media including pharmaceutical samples [261] and biological tissue [262] .
• When such materials are used in high resolution spectroscopy, the well-known, structured interference fringes of conventional gas cell configurations are disrupted, interference effects still dominate.
• They can be particularly troublesome for diffusely reflecting materials.

7 Cavity enhanced techniques

• Significant recent activity in this area merits its inclusion as a separate section.
• General reviews of cavity-enhanced techniques and their relative performance have been produced by Paldus and Kachanov [265] , and Foltynowicz et al [266] .
• In each of the techniques described here, extremely long pathlength cells are constructed using high reflectivity (R > 99.99%) mirrors, which can be achieved using dielectric coatings.
• Figure 46 illustrates the variation in R and F with wavelength for a typical high finesse cavity in the near IR.
• The different techniques described below adopt different approaches to address these issues.

7.1 Cavity ringdown spectroscopy

• Cavity ringdown spectroscopy (CRDS) was pioneered by O’Keefe and Deacon [ 267] , building on a technique previously used to measure mirror reflectivity.
• The following source modulation techniques have been used in CRDS [47] : (a) A CW source may be switched off rapidly (directly or using an external modulator), after which the detector observes smooth exponential decay.
• By using laser pulses shorter than the cavity round trip time, the problem of coupling to specific modes within the cavity is avoided and the emission wavelength may be scanned over an absorption feature.
• This approach ensured that mode locking was achieved under the same conditions as ringdown measurement, being matched in both wavelength and beam position within the cavity during a ringdown event.
• Use of the Levenberg – Marquardt fitting algorithm can be computationally intensive, Page 58 limiting the speed of the technique [265] .

7.2 Cavity-enhanced absorption spectroscopy

• In principle, what distinguishes cavity ringdown spectroscopy from cavity-enhanced absorption spectroscopy (CEAS) is that CRDS measurements are made in the time domain, with the measured absorption being a direct function of ringdown times or phase shifts.
• A large number of closely spaced cavity modes is excited; mode fluctuations dictate that the larger the number that can be averaged, the better the noise characteristics.
• Also making use of time domain information, phase shift measurements have been used to calibrate the effective pathlength or ringdown time [305] .
• Obtaining a good calibration of optical pathlength can be an issue for broadband CEAS, as noted by Langridge et al [312] , because the beam from these sources is less well-collimated than a laser beam, resulting in geometric light loss from the cavity and excitation of cavity modes with higher losses.
• Broadband NO2 measurement has been further developed by Kebabian et al [314] using PS-CEAS in a method they refer to as cavity attenuated phase shift spectroscopy (CAPS).

7.3 High finesse optical fibre cavities

• There has been great interest in the use of gas cells in which the high finesse cavity is created within optical fibre, either using high reflectivity elements such as fibre Bragg gratings or by looping the fibre to form a ring resonator [47] .
• Stewart et al developed the first fibre loop cavity for gas detection coupled to a micro-optic gas cell with a pathlength of 5cm [325] .
• Because of relatively high losses in the loop (compared to free space CRDS), an erbium doped fibre amplifier (EDFA) was used in combination with a variable attenuator, the latter providing fine adjustment of the loop gain so as to control the ringdown time.
• In another study by von Lerber and Sigrist [327] , the high finesse fibre cavities were sensitive to bending losses in the fibre, confirming one of the practical difficulties in stabilising the system over time.
• Another problem is the high level of ASE noise introduced by the EDFA, causing accuracy problems similar to those seen in free space systems.

8 Photoacoustic detection

• Photoacoustic sensors differ in the way that the absorbed light is detected.
• In photoacoustic sensors, the absorbed light is measured directly, as illustrated in Figure 53.
• The photoacoustic signal, resulting from integration of the energy absorbed during each modulation cycle, is also proportional to 1/f, so the general result is no net improvement in signal to noise ratio.
• Acoustically resonant systems have been used extensively to amplify the acoustic signal; Figure 54 shows a number of different geometries in use.
• This provided an almost perfect reference, to correct for matrix interfering chemical species as well as environmental changes and acoustic noise that could also affect the measurement.

9 Comparison of different gas detection techniques

• Within this article, the authors have attempted to compare performance using the common metrics of NEA and αmin.
• Methods to reduce such deterioration include active temperature stabilisation of the instrument [346] and frequent re-zeroing the instrument with clean air (sometimes known as “zero air”) [228] , all of which comes at a cost to instrument complexity.
• After some optimum averaging time, drift effects can be seen as a rise in variance as averaging time is increased.
• It may be that the best instruments have been designed to avoid interference fringes through a combination of balanced detection and thermal management.
• For the short optical pathlengths used, PAS shows very good performance compared with other techniques; this may well be due to the pathlength invariance of equation (12) already discussed.

10 Discussion and conclusions

• One of the technical compromises in optical gas sensing is to work at very high resolution with limited wavelength coverage, or to make measurement over a wide range of wavelengths but with a worse signal to noise ratio.
• The same is true of review articles; in an article of this breadth the authors cannot cover any one area with sufficiently high resolution to do it justice, and neither can they report on every clever advance that has been reported in the field.
• As mid IR lasers develop further, the authors may also see consolidation in this field.
• Their small size has advantages for system response time and is potentially an advantage in itself for example in portable equipment.
• To conclude, optical gas sensing remains an important field that complements other gas detection technologies.

Figures

Figure 16. Emission parameters for a 1.7×1.7 mm thermal emitter (Patinor 15-22). Taken from [90].

Figure 3. Absorption spectrum for 100% methane at atmospheric pressure at 1.651 μm, calculated from Hitran [20] . The feature actually consists of four main underlying absorption lines that cannot be resolved at atmospheric pressure.

Figure 7. Techniques providing interaction between light guided within an optical fibre and an external measurand: (a) light coupled out of the fibre and back in; evanescent fields in (b) a tapered region, (c) an etched region, (d) a side-polished region or (e) using long period gratings to couple light in and out of cladding modes; (f) diffusion of some gases (eg H2) into the fibre core. Modified from [47].

Figure 38. MUMAS spectra recorded around 1.529 μm for acetylene at a pressure of 10 Torr. The

Table 6. Summary of techniques used to improve output coupling from mid IR LEDs.

Figure 34. Example of a dual gas measurement of methane and ethane using a wavelength scan from a single DFB laser. Taken from reference [182].

Figure 33. Conventional configuration for TDLS employing a laser diode, showing detected signals as a function of laser drive current for (a) direct scanning of the gas line, or (b) 2 nd harmonic wavelength modulation spectroscopy (2f WMS). The effects of residual amplitude modulation can be seen in the

Figure 20. Recent spectral emission results of UV LED research, covering centre wavelengths from 211 nm to 282 nm: (a) AlGaN based LEDs, taken from [123], (b) an AlN based LED, taken from [124].

Figure 46. Variation in reflectivity R and finesse F for a typical high finesse cavity in the near IR, illustrating the effect of using dichroic mirrors on the wavelength range of operation. Taken from [246].

Figure 10. Construction of a low volume, fast (6 s) response gas cell using a coiled hollow core waveguide. (a) Schematic diagram, (b) photograph of inlet manifold, (c) close-up of inlet manifold showing closely spaced, 50 μm diameter drilled holes. Taken from [67].

Figure 42. Schematic diagram of a difference frequency generation (DFG) source using periodically poled lithium niobate (PPLN). After [229].

Figure 50. Off-axis integrated cavity output spectroscopy (OA-ICOS) as implemented by Baer et al, taken from [296]. The etalon was used for relative wavelength measurements during a scan by the laser diode.

Figure 26. A gas correlation system based on the complementary source modulation technique. After Dakin et al [138].

Figure 25. Modulation of the absorption within the gas reference cell allows a correlation system to have matched reference and active beams. After [136].

Table 1. Examples of applications for methane detection, illustrating the need for gas measurement over different concentration ranges. Not all applications currently employ optical techniques.

Figure 4. Early Herriott cell, using smoke to visualise the reflected beams between concave mirrors. Taken from [27].

Figure 24. Example of correlation spectroscopy using a UV LED as a source, illustrating spectra for source emission, target gas absorption, and received light at the two detectors. Taken from [135].

Figure 35. Generation of different harmonic signals in wavelength modulation spectroscopy (WMS). (a) DC scan through a single gas line, background-corrected to remove the residual amplitude modulation for clarity. (b) Form of 1f, 2f and 3f harmonics as a function of position in the current scan, displaced for clarity. The RAM creates a small but measurable DC offset on the 1f signal.

Figure 60. Summarised performance of different categories of gas detection techniques as a function of optical pathlength: (a) NEA, (b) αmin.

Figure 17.Example IR emitter fabricated from SiO2–on–Si (silica on insulator, SOI) MEMS technology. (a) Mask layout, (b) visible emission recorded at 1000°C, taken from [97].

Figure 31. Micro-optical bench (14 mm long) incorporating free space micro optics, a tunable FabryPerot filter and fibre optic coupling. Taken from [156].

Full text

Page 1
Optical gas sensing: a review
Jane Hodgkinson* and Ralph P Tatam
Department of Engineering Photonics, School of Engineering, Cranfield University, Cranfield,
Bedfordshire, MK43 0AL, UK.
j.hodgkinson@cranfield.ac.uk
Abstract
The detection and measurement of gas concentrations using the characteristic optical absorption of
the gas species is important for both understanding and monitoring a variety of phenomena from
industrial processes to environmental change. This article reviews the field, covering several individual
gas detection techniques including non-dispersive infrared (NDIR), spectrophotometry, tunable diode
laser spectroscopy and photoacoustic spectroscopy. We present the basis for each technique, recent
developments in methods and performance limitations. The technology available to support this field,
in terms of key components such as light sources and gas cells, has advanced rapidly in recent years
and we discuss these new developments. Finally, we present a performance comparison of different
techniques, taking data reported over the preceding decade, and draw conclusions from this
benchmarking.
1 Introduction
Gas detection has an impact across a wide range of applications. Early markets have included the
process and petrochemical industries, where sensors are used to ensure safety (eg via detection of
toxic or flammable gases), monitor feedstocks and measure key species in products and processes,
some of which can be rapidly changing
[
1
]
. Use of high sensitivity gas detectors is widespread in
atmospheric science, where they are used to measure and understand the profile and pathways of
different gas species including greenhouse gases
[
2
]
. Various potential biomarker gases are also under
study for use in breath diagnostics, including nitric oxide (NO), ethane, ammonia (NH
3
), and many
more
[
3
]
.
Quantitative detection of gases is traditionally dominated by laboratory analytical equipment such as
gas chromatographs, with sampling that precludes real-time data
[3]
, or small ultra-low-cost devices
such as pellistors, semiconductor gas sensors or electrochemical devices. Pellistors are robust
devices that respond to combustion on a catalyst bead
[
4
]
; they perform well in detecting flammable
gases close to the lower explosive limit, however suffer from zero drift at parts per million (ppm) levels.
Semiconductor gas sensors can be highly sensitive at the low ppm level
[
5
]
, however these also suffer
from drift and cross-respond to other gases and changing humidity levels. Electrochemical gas
Measurement Science and Technology, Volume 24, Number 1, 2013, Paper number: 012004

Page 2
sensors can be relatively specific to individual gases and sensitive at ppm or ppb levels
[
6
]
, however
they have limited lifetimes and also suffer from some known cross-response issues, eg to humidity.
In contrast, gas sensors based on optical absorption offer fast responses (time constants below 1s are
possible), minimal drift and high gas specificity, with zero cross-response to other gases as long as
their design is carefully considered. Measurements can be made in real time and in situ without
disturbing the gas sample, which can be important in process control
[
7
]
. Because the transduction
method makes a direct measurement of a molecule’s physical properties (its absorption at a specific
wavelength), drift is reduced and, because the incident light intensity can be determined,
measurements are self-referenced, making them inherently reliable. In this way, optical gas sensing
fills an important gap between lower cost sensors with inferior performance and high end laboratory
equipment.
Table 1. Examples of applications for methane detection, illustrating the need for gas measurement
over different concentration ranges. Not all applications currently employ optical techniques.
Application
Significant issues
Required
concentration range
Example
ref
Process control: gas quality, ie
measurement of natural gas
composition for regulation,
metering and custody transfer
Accuracy to “fiscal standards”
(0.1%)
70-100 %vol
[8]
Safety: purging gas pipes to
avoid explosions and ensure
pilot lights remain burning.
Accuracy eg to ±5 %vol at
50 %vol
1-100 %vol
[9]
Process control: monitoring
combustion processes
Accuracy in a wide range of
temperature and pressures
0.1-100 %vol
[10]
Safety: quantification of gas
leaks with respect to the lower
explosive limit (LEL) of 4.9 %vol
Accurate at action points eg
20 %LEL (1 %vol) for
evacuation of buildings
0.1 – 5 %vol
[11] [12]
Safety: location of gas leaks,
often outdoors
Reliable zero
Limit of detection approaching
1 ppm
1-10,000 ppm
[12] [13]
Process / environment:
quantification of residual
methane in flares, for carbon
trading
Repeatability 100 ppb
Background methane 1.8 ppm
(higher when close to sources)
100 ppb – 1,000 ppm
(plus background
level of 1.8ppm)
[14]
Environmental modelling:
measurement of the methane
background of 1.8 ppm
Comparison with historic data.
Accuracy of 0.1-5% of reading
required.
30 ppb – 3 ppm
(plus background
level of 1.8ppm)
[15]
Environmental modelling:
methane flux measurement by
eddy covariance technique
Correlation with local
atmospheric eddy currents at
data rates >10Hz
5 ppb – 25 ppm
(plus background
level of 1.8 ppm)
[16]

Page 3
Gas detection applications can cover a very wide range of gas concentrations. The concentration is
typically expressed as a proportion in air (or some other matrix) by volume. Since most gases at
standard temperature and pressure behave as ideal gases to a high degree, this is also equal (or
almost equal) to the molar concentration in the matrix. To put gas concentrations into context, we can
take the example of one gas species (methane) with a variety of applications, each demanding
measurement over a different concentration range. Several examples are summarised in Table 1.
Concentrations are expressed as %vol (% by volume), ppm (parts per million by volume;1 part in 10
6
),
ppb (parts per billion by volume; 1 part in 10
9
) or ppt (parts per trillion by volume, 1 part in 10
12
).
In this review, we discuss commonly used techniques in gas sensing based on measurement of
optical absorption at specific wavelengths. These are non-dispersive gas sensing including non-
dispersive infra-red (NDIR), spectrophotometry, tunable diode laser spectroscopy (TDLS) and
photoacoustic spectroscopy (PAS). Recent developments in the techniques themselves and in
important key system components (such as sources) are considered. Finally, we have completed a
survey of recent published results for the detection of a number of gas species and summarised these
in Table 8 at the end of the article. The list of measurands covers ammonia, benzene, carbon dioxide,
carbon monoxide, ethane, formaldehyde, hydrogen sulfide, methane, nitric oxide, nitrous oxide,
nitrogen dioxide, sulfur dioxide and water vapour. From this survey we have been able to draw cross-
comparisons between different approaches.
2 Basic principles
The fundamentals of molecular absorption spectroscopy and associated instrumental techniques have
been discussed widely elsewhere
[
17
,
18
]
. Many chemical species exhibit strong absorption in the
UV/visible, near infrared or mid infrared regions of the electromagnetic spectrum. The absorption lines
or bands are specific to each species and this forms the basis for their detection and measurement.
Absorption spectra in the different spectral regions have different characteristics, as shown in Table 2.
In the so-called fingerprint region of the infra-red, gas phase absorption spectra exhibit narrow lines as
a result of molecular vibrations at discrete energy levels. These can be measured at high resolution,
resolving the line, or at lower resolution, measuring the absorption band. Near IR spectra are typically
overtones of fundamental vibrations in the mid IR and hence can be significantly weaker (eg around
100 times weaker for methane). However, the availability of high quality sources and detectors,
primarily derived from telecommunications applications, can counteract this disadvantage and signal :
noise ratios can be relatively high.
Table 2. Origin of absorption spectra in different regions of the electromagnetic spectrum.
Spectral region
Cause of absorption
UV (200-400nm)
Electronic transitions
Near IR (700nm – 2.5μm)
Molecular vibration & rotation, 1
st
harmonic
Mid IR (2.5μm – 14μm)
Molecular vibration & rotation, fundamental

Page 4
Optical gas detection using absorption spectroscopy is based on application of the Beer Lambert
Law
[18]
;
0
()expII


(1)
Where I is the light transmitted through the gas cell, I
o
is the light incident on the gas cell, α is the
absorption coefficient of the sample (typically with units of cm
-1
) and ℓ is the cell’s optical pathlength
(typically with units of cm). The absorption coefficient α is the product of the gas concentration (for
example in atm – the partial pressure in atmospheres) and the specific absorptivity of the gas ε (for
example in cm
-1
atm
-1
).
Notes
[18]
 In analytical chemistry and for liquid phase samples, the Beer Lambert Law is typically described
using base 10 rather than e, with the result that quoted values of α are 2.3 times smaller, despite
having the same apparent units. In this review, in line with most gas sensing, we use base e
throughout.
 The Beer-Lambert Law applies for monochromatic radiation; when using light sources that are
broader than absorption lines, the width of the source must be accounted for.
 The law also assumes that there are no chemical changes in the sample – at high concentrations,
dimer formation can alter spectra, but this is a minor effect for most gases at standard temperature
and pressure.
For low αℓ, equation (1) is conveniently linear with α, as follows:
0
I
I



(2)
where ΔI = I
0
-I and ΔI / I
0
is the absorbance, which is unitless but often described in “absorbance
units” (AU).
Limits of detection can be quantified as the noise equivalent absorbance (NEA, in AU) or the minimum
detectable absorption coefficient (α
min
, in cm
-1
), allowing instrumental techniques to be compared
without reference to the specific target gas. For estimates of noise and uncertainty throughout this
article, we use the convention that ΔI is the root mean squared (RMS) value of intensity variations
(1σ). For example, an NEA of 10
-6
implies that for RMS changes in received light intensity at the level
of 1 part in 10
6
, the signal : noise ratio (SNR) is unity. For many instruments, white noise dominates
and therefore the SNR also depends on the measurement bandwidth Δf, as SNR  Δf
-1/2
. When
operating in this domain it is therefore important for practitioners to also record the value of the
measurement integration time t used to obtain a certain noise limit, and / or to quote limits in units of
Hz
-1/2
or cm
-1
Hz
-1/2
, normalising to a 1 Hz measurement bandwidth. As the precise conversion
between Δf and t is often system – specific
[18]
, we have simply quoted authors’ own estimates for
either, or both, in this article.
Measurement of the level of absorbed light in the sample is actually proportional to the number density
N of target molecules in the sample. To convert to more typical units of ppm by volume or %volume at
different temperatures T and pressures P, adjustments must be made using the ideal gas equation
PV=Nk
B
T, where V is the volume of a closed cell, k
B
is the Boltzmann constant and N is the number of
molecules in the cell.

Page 5
An absorption spectrum is a plot of α or ε as a function of wavelength (eg in µm) or its reciprocal,
wavenumber (in cm
-1
). Public-domain quantified spectra are available from the US National Institute of
Standards and Technology (NIST)
[
19
]
, Pacific Northwest National Laboratory (PNNL)
[19]
, and may be
calculated using information in the Hitran database
[
20
]
. Typical absorption spectra are shown in Figure
1, for a series of gases in the mid IR, and in Figure 2, for a single gas (methane) plotted at higher
resolution.
Figure 1. Absorption spectra for 5 gases in the mid IR region of the spectrum (all at 100% vol), taken
from the PNNL database
[19]
.
Figure 2. Expanded view of methane spectrum in the mid IR, from 3 – 3.6µm, also taken from the
PNNL database
[19]
.
At atmospheric pressure, a single gas line has a pressure broadened Lorentzian profile
[18]
, such that
 




   


2
2
0
mol
CS
(3)
Where C
mol
is the gas concentration in units of molecules cm
-3
, S is the line intensity
(cm
-1
/molecule cm
-2
), γ is the line halfwidth at half maximum (HWHM, cm
-1
), ν is the wavenumber
(cm
-1
) and ν
0
is the position of the line centre. Figure 3 shows an example absorption line profile for
methane at 1.651 μm at atmospheric pressure. Variations in the linewidth at different pressures must
be accounted for, especially in high resolution measurement schemes. The linewidth can also be
wavelength / μm
3.0
3.1
3.2
3.3
3.4
3.5
0
20
40
3.6
absorption coefficient / cm
-1
4
6
8
10
12
14
16
wavelength / μm
2
0
20
40
60
80
absorption coefficient / cm
-1
carbon monoxide
carbon dioxide
water
ammonia
methane