Comparison of Mid-IR LED-Based and MEMS Hotplate-Based Photoacoustic CO2 Sensors

The focus of this work is the development of two types of low-cost photoacoustic (PA) CO2 sensors, the first was made from mid-infrared (mid-IR) light emitting diode (LED) and the second was made from micro-electro-mechanical system (MEMS) hotplate. Both sensors use MEMS microphones for the detection of the generated acoustic signal. These sensors have advantages of low power consumption, small size and low cost. Analyses of results from both sensors show that temperature change (within−20°C–50°C range) affects their signal amplitude and sensitivities. Results also show that the absence of buffer gas in the cell reduces the PA signal amplitude. Furthermore, comparison of results from the sensors indicate that the LED sensor is more sensitive by about one order of magnitude while the hotplate sensor, which is less expensive, has a better noise performance. This is the first time that such comparison is done and it is important because it reveals the advantages and limitations of these sensors. The findings is vital for determining which sensor will be suitable for certain applications, considering the cost.

Introduction

The effect of climate change and air pollution on our lives as well as the environment is at a critical level. CO₂ is at the top of the Kyoto protocol [1] list of greenhouse gases (GHGs), therefore its emission must be reduced. Greenhouse gases absorb heat radiation from the earth’s surface, thereby making the heat to stay in the atmosphere and consequently increases the earth’s temperature [2]. Legislations put in place for monitoring and control of the emissions of these GHGs are not kept, partly due to lack of low-cost sensors [3]. This work compares the performances of two low-cost PA CO₂ sensors, one developed using mid-IR LED and the other is a MEMS hotplate-based. These components are used to detect CO₂ because they emit radiation in the mid-IR region where the unique absorption band (4.26 µm) of CO₂ exist. The PA effect was discovered by Alexander Graham Bell in 1880 [4] and his work has been a solid foundation to work like the one presented in this paper. PA effect is a phenomenon that deals with the conversion of light energy into sound energy. Gas phase PA effect results from the non-radiative relaxation of excited gas molecules. This kind of relaxation causes the release of localised heat due to increase in molecules’ kinetic energy, leading to the production of thermal wave and then acoustic wave [5]. Variations of the acoustic wave will be synchronous with the radiation modulation frequency. The generated acoustic signal is detected with a microphone for processing and storage or display. The two major processes involve in PA signal generation are the production of heat and the subsequent generation of PA signal. Both processes depend on the absorption of modulated radiation. Position and time dependent heat density in the gas sample is given by Kreuzer [6]:

(1) $$\begin{array}{r}H(r,t)=\alpha I(r,t)\end{array}$$

where H (r, t) is the heat density and α is the absorption coefficient of the sample. This produced heat generates the acoustic pressure represented by the lossless wave equation [6]:

(2) $$\begin{array}{r}{\mathrm{\nabla }}^{2}p(r,t)-\frac{1}{c^2}\frac{{\mathrm{\partial }}^{2}p(r,t)}{\mathrm{\partial }{t}^2}=\left[\frac{\gamma -1}{c^2}\right]\frac{\mathrm{\partial }H(r,t)}{\mathrm{\partial }t}\end{array}$$

where c is the velocity of sound and $\gamma =c_p/c_v$ is the specific heat ratio or adiabatic constant of the gas. A PA sensor can be designed to operate in resonant or non-resonant mode. In simpler terms, for a mixture of gases with n number of absorbing components in a cell the PA signal is given by Bozoki and Pogany [7]:

(3) $$\begin{array}{r}S=PM(C_{\mathit{c}}\sum _{i=1}^n{\eta }_i{\alpha }_{i}C+B_g)\end{array}$$

where S is the PA signal at the frequency of light modulation obtained from the microphone, P is the exciting light power, M is the sensitivity of the microphone, $C_c$ is the cell constant, ${\alpha }_i$ is the molar optical absorption coefficient of the light-absorbing components at the wavelength of the exciting light, C is the concentration of the light-absorbing species, ${\eta }_i$ is the efficiency of the conversion of absorbed light energy into heat, and $B_g$ is the generation efficiency of background signal. The other part of (3), $PM{B}_g$ is known as the background signal or noise which arises mostly from absorption of the modulated radiation by the cell window or inner walls.

Characterisation of Sensors’ Components

SUP01410HR5H-PB MEMS Microphone

The microphone was connected to a circuit designed and placed in the middle of the PA cell’s resonator for the measurement of the frequency response. Signal from the microphone was displayed by the signal analyser and then through the Hewlett-Packard (HP) 7470A Plotter Emulator to the computer for display as presented in Fig. 1.

Fig. 1. SUP01410HR5H-PB MEMS microphone frequency response curve under atmospheric conditions and no light source.

This microphone was purchased from Digi-Key and has the dynamic range of 0–10 kHz. Fig. 1 shows the dominance of 1/f noise in frequencies less than 1 kHz. Beyond this frequency range, the response curve is relatively flat, which is a desired characteristic of good microphones. The ranges shown in this curve were crucially used in the design of the PA cell.

Mid-IR LED (Lms43LED-4m)

To characterise the mid-IR LED, an experiment was set up in which the radiation source and detector combination was Lms43LED-4m and Lms43PD-03-R-PA, respectively. The PD’s wavelength range of 3.0 µm–4.6 µm suitably matches the radiation emission range of the mid-IR LED which is between 3.0 µm and 5.0 µm. These components were bought from Microsensor NT and they represent the mid-IR LED and photodiode (PD). A non-dispersive infrared (NDIR) sensor arrangement was made and CO₂ and N₂ were allowed to flow into the cell. The schematic in Fig. 2a shows the set-up.

Fig. 2. NDIR measurement showing (a) the set-up cell with CO₂ flowing through and the LED radiating light toward the photodetector, (b) Mid-IR LED driving current (labeled a) and PD output (labeled b).

In Fig. 2b, the signal marked “a” is the modulated current driving the LED, while the signal below it marked “b” is the output of the PD, which represents the amount of radiation power that the PD captures from the LED and converts into an electrical signal. Modulation (ON-OFF) of the current driving the LED is at 50% duty cycle, and the maximum value is 800 mA. When the PD receives light from the LED, it produces a current which passes through its resistor–capacitor (RC) connection of the pre-amplifier circuit. This charges the PD circuit capacitor during the ON period of the LED and discharges it during the OFF period, thus producing the type of signal marked “b” in Fig. 2b. This measurement provided more information on the mid-IR LED’s radiation characteristics, which were useful for the sensor design (More information can be found in the design section). The LED has the TO-18 metal casing, which is used as a heatsink, and because the mid-IR LED’s elements are placed deep inside the casing, there is a restriction of radiation dispersion by the cylindrical open end of the casing. Similarly, the PD also has a TO-18 metal casing, which, in addition, carries a compound parabolic collector, which enhances the amount of radiation collected from the source.

EMIRS 200 Micro Hotplate

Again, another NDIR set-up was used to examine the variation of the hotplate’s output power with modulation frequency. The EMIRS 200 micro hotplate packaged in TO-39 compact casing with a reflector was used as the radiation source, while Lead selenide (PbSe) was used as the detector. PbSe is a photoconductive detector whose resistance changes when it is illuminated by IR radiation. Fig. 3 shows the hotplate’s output power as a function of the modulation frequency.

Fig. 3. How the power output of the micro hotplate reduces with an increase in modulation frequency (a) frequencies between 5 and 100 Hz (b) frequency range of 500–4000 Hz.

There was a significant reduction of the hotplate’s output power when the modulation frequency was increased between 2.8–4.0 kHz. This frequency range for measurements in this work was carefully chosen in order to minimize 1/f noise, meaning that the hotplate will deliver lower power for PA signal generation because of the higher frequencies within the range. It is important to state that the mid-IR LED’s modulation in high frequency does not lead to output power loss as observed in the hotplate.

Design of the Sensors

The geometry of a PA cell determines the sensitivity of the PA sensor. Understanding pressure distribution in the cell is crucial to PA sensor optimization; this enables the achievement of optimal radiation excitation, PA signal generation, and microphone transduction. Different PA cell geometries have been demonstrated; they include T-cell [8], simple cylindrical cell, Helmholtz resonator, multi-pass cell [5], differential cell [7], and cylinder with two buffer volumes [9], [10]. The PA cell designed in this work follows the resonator (cylinder) with buffer volumes attached to each end. The length and radius of the resonator are 4 cm and 0.7 cm, respectively. Radiation spread from the LED and hotplate was used to derive the resonator diameter. This was done to ensure that there is sufficient passage of radiation into the cell and the amount of radiation hitting the surface of the is at minimum. The two identical buffer volumes are 2 cm long and 4.8 cm in diameter. These buffer volumes are a quarter wavelength of the resonance frequency, and at that optimum length, there is maximum destructive interference of the standing acoustic noise coming into the buffers, hence they filter the acoustic signal. Diameter of the of the buffer volumes was calculated from the condition that it needs to be at least three times the resonator’s diameter, described by $d_{buffer}\ge 3d_{resonator}$. The PA cell configuration and simulated cell are given in Fig. 4.

Fig. 4. The PA cell (a) its configuration showing the parts and microphone placed at the middle of the resonator (b) FEM simulated cell showing regions of high (red) and low (blue) acoustic signal. This was used to verify that the middle of the resonator is the antinode.

The use of modulation frequencies below 1 kHz was avoided because of the significant influence of 1/f in that region. Modulation frequency within the 1–5 kHz interval has been recommended for this type of PA system. This means that the resonator length will be in the range 3.4 cm–17 cm, calculated from this relationship given in (4):

(4) $$\begin{array}{r}{f}_n=\frac{nc}{2L}\end{array}$$

where f is the modulation frequency, n is the number of longitudinal modes, c is the speed of sound, and L is the resonator length. Using 343 m/s as the speed of sound in air, the fundamental longitudinal resonance frequency is 3543.14 Hz when the effective length of the resonator was taken into consideration. The inner walls of the stainless-steel PA cell were coated with gold to reduce the effect of background signal caused by cell wall radiation absorption. At about 42% CO₂ concentration and normal atmospheric conditions, a resonance profile was plotted from measurement results around the resonance frequency, and the quality factor Q computed is 88. The Q is the PA cell’s amplification factor and it shows the relationship between the resonance frequency and bandwidth of the cell. Zinc selenide (ZnSe) windows with antireflection coatings were employed. Transmittance of these windows were plotted and presented in Fig. 5.

Fig. 5. Transmittance of antireflection coated ZnSe optical window. The inset shows more clearly the uncoated and coated transmittances within the wavelength range of 4.1–4.3 μm.

These infrared windows through which the radiation enters and leaves the PA cell also serve as filters, allowing the intensity radiation around the peak absorption wavelength of CO₂. Antireflection-coated Infrared windows offer the advantage of increased transmittance over uncoated ones; however, the value is still less than 100%.

Experimental Set-Up and Procedures

Fig. 6 is the block diagram for the practical measurement set-up used in this work. Two different electronics circuits were built for modulating the LED’s and hotplate’s driving currents based on the resonance frequency of the cell. These circuits also perform resonance tracking functions because of the variation of resonance frequency with temperature and gas composition or concentration. The modulated radiation was passed through the PA cell, which also has the gas mixture flowing through it. PA signal was generated at each measurement CO₂ concentration, then amplified by the microphone amplifier circuit before being passed to the lock-in electronics. At the lock-in stage, PA signal was extracted at the resonance frequency from the noisy background signals. Hewlett Packard (HP) 3561A Dynamic Signal Analyzer does the lock-in and display of the PA signal. Another HP device, the 7470A Plotter Emulator, connects the signal analyzer to a computer for PA signal capture and display or storage.

Fig. 6. Stages of the generation and detection of PA signal shown as a block diagram of the practical measurement set-up (a) The constructed PA cell with inlet/outlet pipes connected and microphone cable shielded against interference (b).

Gas flow through the cell was achieved by using mass flow controllers connected to the MKS Multi-Gas Controller 647B. Flow rates in standard cubic centimeters per minute were set in calculated proportions for the target and buffer gases in order to produce desired gas concentrations. Temperature analysis was done using measurements obtained from the WEISS WKL34 environmental chamber. Inside the chamber, specific temperatures and humidity were set for measurements to be carried out. Monitoring of the frequency, modulated current of radiation sources, and the output signal from the microphone was done on the oscilloscope.

Results and Discussions

PA Signal Variation with Increase in LED Current

Further analysis on the mid-IR LED was carried out. CO₂ was made to flow through the PA cell at 100% concentration from the mass flow controllers, and the modulated current at 50% duty cycle was increased in steps of 40 mA while the PA signal produced at each measurement step was recorded. Fig. 7 is the graph that was plotted from the results obtained.

Fig. 7. Variation of PA signal with the mid-IR LED driving current. The PA signal increases proportionally with increasing driving current.

The plot shows that the relationship between PA Signal and LED driving current is virtually linear. This is because the PA signal is linearly dependent on the output power of the LED [11], which in turn is directly proportional to the current driving it (P = IV).

PA Signal Variation with CO2 Concentration

This section presents results from the two sensors: Mid-IR LED and MEMS hotplate PA CO₂ sensors. Measurement procedures for both sensors were the same. Results were obtained by passing radiation from the infrared sources through the ZnSe windows into the cell containing proportions of CO₂ and the non-absorbing buffer gas, N₂. Fig. 8 represents the plots of the results from the sensors.

Fig. 8. The mid-IR LED-based sensor’s variation of PA signal with respect to CO₂ concentration increase at room temperature (20°C) (a). Change of PA signal with respect to CO₂ concentration increase at room temperature for the hotplate sensor (b).

PA signal generated in both sensors varies linearly with CO₂ concentration, having R² values of 98.82% and 98.91% for the Mid-IR LED sensor and hotplate sensor, respectively. From the graphs in Fig. 8, it is clear that the mid-IR LED sensor’s signal amplitude is higher by one order of magnitude in comparison to that of the micro hotplate sensor. The PA signal of the LED sensor increases for every 1% increase in concentration of CO₂ by 65.3 µV while that of the hotplate sensor increases by 4.6 µV, that is another one order of magnitude advantage in terms of sensor sensitivity.

Temperature Effect on the PA Signal of the Two Sensors

Both sensors show remarkable repeatability and approximately the same percentage loss of signal amplitude from temperatures of −20°C to 50°C. Temperature is one of the most varying environmental factors which affects the practical application of PA sensors. Therefore, the effect that temperature has on the LED and hotplate sensors was investigated. PA signal was measured in the temperature range of −20°C−50°C for increasing CO₂ concentrations. Results from the two sensors are given in Fig. 9.

Fig. 9. Temperature effect on PA signal from −20°C to 50°C (a) Results from Mid-IR LED sensor (b) MEMS hotplate sensor results.

These results show that PA signal decreases with increasing temperature, which agrees with previous works [9], [12]. At 25% concentration of CO₂ in the cell, experimental results obtained were used to plot the graphs shown in Fig. 10. For the procedure, CO₂ was kept at a constant concentration while the temperature was varied, and the PA signal was recorded.

Fig. 10. Variation of PA signal with change in temperature at 25% fixed concentration of CO₂ in the cell (a) Mid-IR LED sensor (b) MEMS hotplate sensor.

At 25% of the proportion of CO₂ in the PA cell there was PA signal amplitude reduction of 45.5 µV for every 1°C increase in temperature, which resulted in approximately 76% reduction from −20°C to 50°C in the LED Sensor. Similarly, the hotplate sensor had approximately 77% loss of PA signal in the same temperature range at the rate of 4.4 µV/°C. For 10°C rise in temperature from 10°C to room temperature (20°C), the signal attenuated by about 17.7% and 15.3% in the LED sensor and Hotplate sensor, respectively. Measurements at all temperatures were done with resonance tracking to ensure that the temperature effect observed was not due to the variation of resonance frequency.

Effect of Temperature on the Sensors’ Sensitivities and Microphone Sensitivity

The sensitivity of the microphone was provided by the manufacturer in the data sheet, while the sensor’s sensitivities are the gradient calculated from the graphs of PA signal vs. CO₂ concentration in the cell. Computed gradients of the graphs in Fig. 9 were plotted as the dependent variable on temperature within the measurement range and given in Fig. 11.

Fig. 11. Variation of the CO₂ sensors’ sensitivities with temperature. These graphs represent the sensitivity-temperature relationships for: (a) the Mid-IR LED sensor (b) the MEMS hotplate sensor.

These graphs show that the sensors’ sensitivities decrease with temperature. Nevertheless, the sensitivities are insignificantly affected by temperature since the rate of reduction in sensitivities with temperature differs by two orders of magnitude from the LED sensor’s sensitivity (2 × 10⁻⁴ mV/%/°C) and from the sensitivity (10⁻⁵ mV/%/°C) of the hotplate sensor. Szakall et al. [12] clearly demonstrated that the dependence of sensor and microphone sensitivities on temperature are coincident and, therefore, concluded that the sensor sensitivity dependence on temperature resulted from the dependence of the microphone sensitivity on temperature.

Effect of Background Signal on the PA Signal

Background signal in this type of PA system is the signal generated when the cell windows and inner walls absorb the modulated radiation. It is essential to minimize this noise signal because it is at the same frequency as the main PA signal, which is why both the windows and inner walls of the cell were coated. The effect of the background signal on the sensors is presented in Figs. 12a and 12b.

Fig. 12. PA signal amplitude at 87.5% CO₂, pure nitrogen and radiation source turned off (a) for the LED sensor (b) Hotplate sensor results.

These graphs show that the effect of the background signal (red graph) was more pronounced in the LED sensor because the difference between the measured signals for pure nitrogen and when the LED was switched off is higher compared to what was seen in the hotplate sensor results. Those measured signals are much closer for the hotplate sensor, indicating that it has better noise performance and great potential in gas sensing if its sensitivity can be improved. Table I shows the average values of the PA signal and the uncertainties involved for those three cases.

Table I. Background Signal Analysis Using PA Signal Amplitude when Radiation Source was Switched Off, Gas in Cell was Purely Nitrogen and CO₂ was at 87.5% Concentration

Mode of measurement	Hotplate sensor		LED sensor
	Average acoustic signal (µV)	Uncertainty (µV)	Average acoustic signal (µV)	Uncertainty (µV)
Light source switched off	110.2	2.7	110.2	2.7
100% N2	140.8	6.0	1248.9	61
87.5% CO2	524.2	4.4	6602.8	57

Effect of the Absence of Buffer Gas (N2) in the PA Cell

In terms of the effect of buffer gas, at 0% nitrogen in the cell, the reduction in signal amplitude from the highest signal at 87.5% ${\text{CO}}_2$ concentration and the signal with no buffer gas (100% ${\text{CO}}_2$) were 31% for the hotplate sensor and 16.1% for the LED sensor, see Fig. 13. This indicates that the hotplate sensor was more affected by the absence of nitrogen buffer gas by approximately 2 times.

Fig. 13. Variation of PA signal with increase in CO₂ concentration, showing the effect of the absence of N₂, at 100% CO₂ in the cell (a). Graphs of PA signal variation with CO₂ concentration showing how buffer gas absence in the cell affects the sensor output (b).

One of the main benefits derived from buffer gases is that they dilute the target gas, thereby enhancing the collisional excitation of molecules and, consequently, the PA signal generation. In addition, it helps to convey acoustic signal to the microphone. It can be deduced from the graph that the absence of ${\text{N}}_2$ in the cell at 100% ${\text{CO}}_2$, hampered some of the acoustic signal from reaching the microphone. That is why signal reduction is observed at 100% ${\text{CO}}_2$.

Conclusion

The use of PA effect to develop a mid-IR LED-based and a MEMS hotplate-based CO₂ sensors have been presented in this work. The performance of the sensors was compared under varying temperatures between −20°C and 50°C. At 20°C, the peak measured signal for the LED and hotplate sensors are 6577.5 µV and 525.0 µV, respectively. This implies that the mid-IR LED sensor’s PA signal amplitude is greater by one order of magnitude. Both sensors showed very similar reduction in the magnitude of the PA signal from −20°C to 50°C. It was found that PA signal generated by the hotplate sensor was more affected when the sensor was operated in pure CO₂, without the N₂ buffer gas in the cell. This buffer gas absence effect was about two times less in the LED sensor. Again, the mid-IR LED sensor has a higher sensitivity by one order of magnitude. However, the MEMS hotplate sensor offers better advantages over the mid-IR LED sensor in terms of power consumption, cost, and coherent background noise [13]. The maximum driving current for the hotplate is 86 mA, in contrast with the mid-IR LED, which uses a maximum operation current of 800 mA. From the foregoing, it can be seen that both sensors have great market potential and would greatly contribute to the fight against air pollution and global warming.