1. Introduction

The development of appropriate gas sensors [1] is necessary in many areas, particularly where low concentrations can cause health or environmental concern. Detection is typically addressed using non-dispersive infrared (NDIR) technique, primarily based on the use of narrow band spectral filters. In combination with signal chopping, a multi-pass optical cell and a gas filter correlation (GFC) technique it is a recognized and reliable technique to detect such gases to very low concentration levels [2]. An alternative optical measuring method is Tuneable Diode Laser Absorption Spectroscopy (TDLAS). However, most mid-infrared lasers operate at cryogenic temperatures to cover the fundamental absorption bands required for ultra-sensitive gas analysis, and near-infrared room-temperature diode lasers give access mainly to significantly weaker overtone bands, requiring extra-long optical paths and very high reflectance multilayer optics [3].

Long term stability or drift has been a major issue for gas sensors irrespective of what technologies or methodologies are used. Zero drift still requires frequent calibrations for these devices to really become unattended systems. Several reasons can be found behind this fact depending on the setup, however, thermo-mechanical instability of the complete system (emitter, optical cavity and sensor) is the main source of signal drift.

Multi-pass optical cavities have been used for gas detection because they provide a many-fold increase in gas sensitivity. Figure 1 illustrates a classic White’s cell arrangement [4] used in NDIR systems. Typically, an infrared emitter is focused or imaged onto a detector after multiple passes through the optical cavity. For a given source size some strategies can be follow to achieve a maximum optical path length within a volume (see for example [5]).

 

Fig. 1 White’s cell multi-pass diagram for NDIR gas monitoring (4 pass in the figure). Blue color is used for chief rays and red color for marginal rays (aperture). The centers of curvature of the objective and field mirrors are indicated as C_O, C_O’ and C_F. Multiple images of the light emitter are projected on the field mirror M_F by the objective mirrors M_O and M_O’. The light beam bounces repeatedly until it is finally projected onto a detector. A band-pass filter, specific of the spectrum target gas, is interposed in the optical path to reduce background noise.

Download Full Size | PDF

Thermal gradients and vibrations affect the thermo-mechanical stability of a multipass optical cavity and it gets worse as the total path length increases. For example, a misalignment of any objective mirror in a White’s cell (see Fig. 1) of only ~20” arc leads to a beam lateral displacement of ~2 mm at detector port for a cavity with a total path length of 10 m. This means, for common detector sizes, losing completely the signal. Common practice to get thermo-mechanical stability is to heat the optical enclosure well above ambient temperature with a thermostatic control.

Sensitivity and global performance is also related to different emission/detection techniques [1]. Thermal stability of emitters is an issue for detectors like PbSe (typically used in industry). Despite its high detectivity (peak D*~2·10¹⁰ cm Hz^1/2 W⁻¹) it is significantly non-linear and spatially non-uniform [6] and it has a temperature coefficient typically larger than 1% K⁻¹ [1, 6]. Therefore, small thermo-mechanical drifts in the optical cavity and the detector platform spoil zero-drift and span calibrations.

Consequently, thermal control is usually applied to the whole system (i.e. the emitter, the optical system and the sensor) in conventional configurations. This is at the expense of high energy consumption and technical complication. One goal of this contribution is to provide an emitter-cavity-sensor design highly robust to thermo-mechanical variations and aging, making practically unnecessary any thermal control.

On the other hand thermal emitters are still unsurpassed in its radiance, compared to other semiconductor sources in the 3-12 μm range; although they are slow and cannot be modulated at frequencies above 10 Hz. Thermopiles detectors have detectivities one or two orders of magnitude lower than known semiconductors. However this lower detectivity can be compensated for the superior radiance of thermal emitters. Compared to other technologies, thermopiles temperature coefficients are one or two orders of magnitude lower (<0.1% K⁻¹). Thermopiles also cover a very wide spectral range and they usually do not need thermal control [7]. Moreover, compared to PbSe photoconductors, thermopiles are highly linear. In principle, the association of a thermal emitter and a thermopile detector can perform well at low frequencies (<10 Hz), based in the fact that thermopiles need no bias and have negligible intrinsic 1/f noise [7].

Any NDIR method needs a reference and a sampling signal to measure transmittance and correlate it to the gas concentration within the optical cavity. Typically, a chopping wheel with filters attached provides an adequate scheme to obtain these two channels, multiplexed in time, and avoid many of the long term thermo-mechanical drift and low frequency noise. Alternatively, multiplexing in space is possible with a dual channel-detector: one channel is filtered with a spectral band pass (specific of the target gas) and the other channel, provided with a near band pass filter where the target gas has no absorption, is therefore used as a reference channel. This is a very common approach with no moving parts, but the stability of this approach depends on the lamp output spectral power stability.

Wong et al. [8] propose the use of a spatial dual-channel and dual-detector setup where both detectors see the same spectral band pass but each channel has a different path length. A short path now acts as a reference channel and a long path acts as the sampling channel, showing higher absorption. If we define the optical response as the ratio of the signals provided by these two channels, then the response is independent of emitter flux. This fact allows temporal modulation of the source. In the work of Wong et al, the sampling channel is approximately two times longer than the reference channel, and the detectors are far away from each other. Increasing the system sensitivity would lead to inappropriate long cavities.

We also propose a dual channel strategy with a common spectral window, but a modified optical design that sets a short channel (reference) and an arbitrary long channel (sampling), folded within the same multi-pass optical cavity. In particular, we introduce simple modifications to convert a single-path White’s cavity into a multi-path cell with independent path lengths. It shows an increased robustness to misalignment. Experimental results demonstrate that this new design, in combination with thermal emitters and thermopile detectors, provide stable sub-ppm gas monitoring (applied to carbon monoxide). No thermal control is applied to any of the emitter, cavity or detector components. Low frequency source modulation (~1Hz) is enough to reduce ambient or background 1/f noise influence.

2. Enhanced optical cavity design and procedure

Figure 2 shows the basic arrangement of the optical cavity design. The key points are: 1) a dual-channel cavity with different path lengths that use the same spectral window for auto-reference, and 2) a Köhler illumination system at the detectors.

 

Fig. 2 Description of the dual-channel optical cavity proposed. Dashed blue line represents the path of the reference channel, formed with the field mirror associated to the objective mirrors M_OR and M_OR’. The field mirror and the other pair M_OS-M_OS’ configures the sampling channel with independent path length (beam not drawn). All the mirrors are concave spherical mirrors and they have the same radius of curvature which is also equal to the length of the cavity.

Download Full Size | PDF

In order to set multiple paths in a single cavity, we replicate the common pair of objective spherical mirrors, based in the White cell basic design [4], in multiple pairs. Every objective mirror pair sets an independent path. Thus an array of conjugated mirror pairs allows setting multiple channels with different paths. In Fig. 2 we show an array of objective mirrors valid for two independent channels.

As the light enters the system it overfills the objective mirrors (see Fig. 2). Only the light reaching M_OR and M_OS directly from the source get trapped in the optical cavity. The light incident on M_OR’ and M_OS’ directly from the source is reflected away and lost. In the dual channel configuration of Fig. 2 one channel corresponds to the conjugated pair M_OR-M_OR’ (reference) and the other channel is set with the conjugated pair M_OS-M_OS’ (sampling). The field mirror is common to both channels. In Fig. 2 we illustrate the chief ray trajectories for only one channel (reference). The light coming from every channel is finally collected by a lens and projected to the corresponding detector surfaces planes. This lens also provides aperture tolerance as explained later on. To increase total path length for a given channel it is sufficient to get the centers of curvature of the conjugated objective mirrors closer to each other (e.g. C_OR and C_OR’ in Fig. 2).

The light spots distribution is explained in Fig. 3 for the sampling channel. The field mirror, M_F, receives repeated images of the input port (IN) until finally light exits through the output port (OUT). M_F can be filled with multiple light-spots contained within rectangular cells as drawn in Fig. 3(a) with dashed lines. These cells are coupled to the location of the centers of curvature of objective mirrors (C_O and C_O’). The closer they are to each other the greater the number of spots or the total light path. Basically, light bounces repeatedly in the order described in Fig. 3 where spots distribute symmetrically to C_O and C_O’ in alternative way. The total number of passes is N = 4m+2 (for reference channel m = 1 as illustrated in Fig. 2). At the output port a collecting lens projects the image of objective mirror M_OS onto a detector surface (similarly for reference channel with M_OR). Thus, the total path length for sampling or reference channels can be set differently.

 

Fig. 3 Optical association of the mirrors at the cavity. (a) Spot diagram formed over the field mirror by the sampling channel. Light coming from input port (IN) goes to objective mirror M_OS and it is then focused back onto the field mirror M_F at position 1. From there light is reflected to the conjugated mirror M_OS’ and then back to hit mirror M_F at position 1’. The series follows the order: IN-M_OS-1-M_OS’-1’-M_OS-2-M_OS’-2’-…-M_OS-m-M_OS’-m’-M_OS-OUT. Where m is an arbitrary number, limited by M_F size. (b) The objective mirrors are grouped in conjugated couples, one pair for reference (M_OR-M_OR’) and one pair for sampling channel (M_OS-M_OS’). Only one mirror of every pair is finally imaged on an independent detector.

Download Full Size | PDF

The final collecting lens at the output provides a stable irradiance at detectors thanks to a Köhler illumination configuration (see Fig. 4 ). The emitter (real or a projected image) is finally imaged at the exit port where the lens is located. For every channel, this lens projects an image of the corresponding objective mirror (M_OR or M_OS) onto the surface of a corresponding detector (see also Fig. 2 and Fig. 3(b)). The aperture of the lens should be greater than the image of the emitter to effectively capture spot displacements. Therefore, this setup is expected to yield robustness to thermo-mechanical drifts and vibrations by providing aperture tolerance to spot displacements. Thermo-mechanical drift may come from subtle misalignment of objective mirrors. The mesh distribution of spot cells in the field mirror (dashed lines in Fig. 3a) depends on the precise location of the centers of curvature of objective mirrors and the distance between them. Thus, the displacement of C_O or C_O’ will provoke a displacement of every light spot proportional to the spot number m. It is simple to see that the final spot displacement is proportional to the total beam path. Therefore, longer optical paths lead to more unstable systems where the role of the collecting lens should be fundamental.

 

Fig. 4 Köhler illumination configuration applied at detection for a conventional White’s cell. A simple lens, situated at the exit spot, images the surface of the objective mirror onto the surface of detector. This provides tolerance to beam displacements as indicated by dashed lines. In the figure we draw only the final output of a hypothetical misalignment of any of the objective mirrors.

Download Full Size | PDF

Based on Beer-Lambert law of absorption, the detected signals ratio can be related to gas concentration in the following way (in absence of signal offsets):

M_S and M_R are the measured signals at each detector, from the sampling and reference channel respectively. C_S0 and C_R0 are system constants for every channel; they only depend on the transmission of optical elements, geometrical parameters and the detectors itself. In turn, C₀ is defined as C₀ = C_S0/C_R0 and it is independent of the source intensity or the radiance of the emitter. N = N_S–N_R are the different number of paths within the cavity between the sample and the reference channel. r is the radius of curvature of the mirrors and K is a sensitivity factor, it depends on the particular gas and it is a constant only for monochromatic light. The term KC is the standard absorption coefficient (cm⁻¹) and it is assumed to be the same for every channel as long as the spectral band pass filter is the same for each detector. C is the gas concentration. Finally, the linear approximation in Eq. (1) should be considered only in the limit of very low gas concentration. In this case, C is directly proportional to the ratio M_S/M_R or M_SR.

In order to obtain for each channel a signal free of offsets and drifts (thermal background, operational offsets, thermal drifts, etc…), light emission was modulated at a given frequency and each channel signal post-processed in the computer, where a band pass filter, centered at the modulation frequency, was applied.

Although it is possible to use expression (1) for every sampled temporal point of these processed signals, the system parameter C₀ can be slightly time dependent if the response time of each detector is dissimilar. To cancel this periodic residual effect we calculated the standard deviation of the M_S and M_R signals within an integer number of periods to finally estimate M_SR.

3. Experimental results

To test the previous analysis an experimental arrangement was set for the detection of CO. The dual path optical cavity of Fig. 2 was set with spherical mirrors of radius r = 30.5 cm. A gas chamber was filled and a single optical CaF₂ window allowed for IR light input and output. The reference channel was set to m=1, and m=7 for the sampling channel, therefore N = 4(7–1) = 24, which gives a total path length difference of N × r = 732 cm. Simulation with HITRAN database at 25 °C for this path difference and the detectors filter bandwidth gives a variation in M_SR of –0.1% ppm⁻¹ when CO concentration is low (≤30 ppm).

A micro-machined thermal emitter (Heimann HLS-EMIRS reflector type) is projected at the input port and modulated at ~0.5 Hz (voltage sinus function) reaching a maximum temperature of ~450 °C. A CaF₂ lens is situated at the exit port and projects an image of objective mirrors M_OR and M_OS onto the thermopile detectors at a distance of s’ = 3 cm. The two detectors (Heimann HTS-Q21) are square thermopiles chips of ~2 mm size separated 1 mm side by side. They are mounted in a single compact housing. This arrangement minimizes thermal gradients and it is very convenient to image the objective mirrors of our optical cell on them. The detectors have attached a band pass filter intended for CO detection (λ_p = 4.64 μm HWFM = 0.18 μm) and they have a nominal normalized detectivity of D* = 2.7·10⁸ W⁻¹ cm Hz^1/2. Minimum detector signal noise is ~38 nV (intrinsic Johnson limit).

The readout of the thermopiles was performed using a high resolution data logger (PICO Technologies ADC24). It provides a ± 0.6 μV signal resolution for an integrating time of 0.66 s. Irradiated thermopiles gave maximum AC signals of ~150 μV and final reading was averaged over a ~76 s interval. These parameters predict a resolution of ± 5·10⁻⁴ for the optical response M_SR or a CO ppm equivalent resolution of ~ ± 0.5 ppm.

We perform a preliminary test to check that the optical sensor is not sensitive to the IR source variations. We work in DC mode and subtracted offsets (dark signal) for each channel. The results are shown in Fig. 5 . We may appreciate that the optical sensor output is sensitive to gas concentration variations but it is not sensitive to strong source radiation instabilities.

 

Fig. 5 Optical sensor response working in DC mode for gas injections and source bias variations. Above: reference signal (blue line) and sampling signal (red line). Below: optical sensor output (signals ratio) variations.

Download Full Size | PDF

Finally we operated the sensor as indicated in section 2. Small CO gas injections in the chamber were carried out every hour approximately; then clean air was circulated inside the chamber to remove CO in several steps. An electrochemical sensor (EL-USB-CO from Lascar Electronics) was introduced in the chamber for cross checking CO concentration. This experiment is illustrated in Fig. 6 , it shows the optical response (equivalent to M_SR in our analysis) and the electrochemical detector response expressed in ppm. The observed optical resolution is in the order of ± 0.5 ppm which is in agreement with the expectative. A good correlation is observed between our system and the tracking detector. We should point out that a somewhat poor performance was observed for the electrochemical sensor below ~3 ppm.

 

Fig. 6 Response step changes in CO concentration from 0 to 30 ppm. Above: Optical sensor. Below: electrochemical sensor for tracking.

Download Full Size | PDF

A similar test was done to test zero-drift over a day-long period (see Fig. 7 ). From the fourth hour on the concentration was kept constant. It is shown that the response stability stays nearly within the resolution range for about 20 hours (~ ± 0.4 ppm at 76 s integration time). The lab temperature varied ~2 °C (Fig. 7 below). No special thermal stabilization was established in this experiment. We only used a laminate shield covering the thermopiles package to reduce air currents around the detectors.

 

Fig. 7 Above: Optical sensor response to CO concentration variations (~0-20 ppm peak to peak) and drift (from ~4 h to 24 h). 7.3 m path difference and no thermal stabilization, integration time = 76 s. Below: ambient temperature measured at thermopile substrate.

Download Full Size | PDF

According to specifications, thermopile intrinsic noise is ±0.038 μV Hz^-1/2 and the amplifier noise is ±0.5 μV Hz^-1/2 (13 times larger). Thus, we believe there is room for improvement based on optimized electronics. This point of view is supported by the work of Foote [5] on thermopile detection for example. In the limit of ideal amplifier performance we could achieve a concentration resolution of ~ ±0.04 ppm (value established as an asymptotic result considering the present set-up). Notice that for a multiple channel setup within the same volume the SNR decreases as the number of channels increases. Alternatively, we can increase the emitter temperature to obtain a 4 times larger radiance, and also increasing the light source size may improve the flux at detector a factor of ~10. The combination of both aspects (electronics and optical setup) would lead theoretically to CO concentration resolution below ~ ±10 ppb in the current setup. Of course, other sources of noise or interference enter to play. Anyhow, it should be analyzed to which extent thermal control can be reduced to minimize drifts within this theoretical resolution. For example, a thermal conductive sink around the thermopile package would reduce further thermal gradients within detector enclosure and probably would help to further reduce drift.

4. Conclusions

We demonstrate the potential for high resolution and stable gas detection based on a new optical cavity design. A White cell array sharing the same cavity is the basis for multichannel and auto-referenced gas detection within the same spectral window. For carbon monoxide, we obtained sub-ppm sensitivity and stability without thermal control. The optical cavity design was presented for a single gas sensor but the general idea can be extended for multiple gas detection within the same cavity, just by increasing the number of conjugated objective mirrors within the same cavity, to get more independent path optical channels.