atmosphere

Article
A Novel Bandpass Filter for the Analysis of Carbon
Monoxide Using a Non-Dispersive
Infrared Technique
Trieu-Vuong Dinh 1 , Ji-Won Ahn 2 , In-Young Choi 1 and Jo-Chun Kim 1, *
1 Department of Civil and Environmental Engineering, Konkuk University, 120 Neungdong-ro, Gwangjin-Gu,
Seoul 05029, Korea; dinhtrieuvuong@gmail.com (T.-V.D.); force@konkuk.ac.kr (I.-Y.C.)
2 International Climate and Environmental Research Center, Konkuk University, 120 Neungdong-ro,
Gwangjin-Gu, Seoul 05029, Korea; jwahn81@konkuk.ac.kr
* Correspondence: jckim@konkuk.ac.kr; Tel.: +82-2-450-4009; Fax: +82-2-3437-8654




Received: 22 October 2018; Accepted: 12 December 2018; Published: 14 December 2018


Abstract: In this study, two novel narrow bandpass filters (BPF) obtained from the high-resolution
transmission molecular absorption (HITRAN) data for a carbon monoxide (CO) non-dispersive
infrared (NDIR) analyzer were investigated and compared with a commercial BPF (4.64 µm). The new
BPF was made using a two-cavity filter method with different center wavelengths and bandwidths
from the commercial BPF. The wavelengths of the two BPFs were 4.5 µm and 4.65 µm. The gas
emission pattern of a coal-fired power plant was used as a case study. Various concentrations of target
gases were used to theoretically estimate the interference, and to practically determine it. It was found
that although the transmittances of the two new BPFs were lower than that of the commercial BPF,
the signal-to-noise ratio caused by two novel BPFs was approximately 20. In terms of interference
effect, carbon dioxide (CO2 ) was found as a strong interfering gas on the commercial BPF at 4.64 µm
and the new BPF at 4.65 µm. In contrast, the new BPF at 4.5 µm cut off the interference effect of all
target gases. The measurement error of the NDIR analyzer applying the BPF at 4.5 µm was similar to
that of gas filter correlation (GFC) NDIR and was less than 1%. This indicates that the novel BPF at
4.5 µm can be used instead of a GFC for a CO NDIR analyzer, thus overcoming the limitations of
using a GFC.

Keywords: NDIR; bandpass filter; CO analysis; power plant

1. Introduction
Carbon monoxide (CO), which is produced mainly from the combustion process, is a toxic gas
when its concentration is >35 ppm [1]. Besides being a greenhouse gas, CO has an important role in
atmospheric chemistry [2–4]. CO has also been widely considered as a signature means of detecting
fires [5,6]. Therefore, monitoring CO is an important issue. In the monitoring of CO, two main methods
have been used: gas chromatographic technologies (e.g., gas chromatography-flame ionization
detectors, gas chromatography-electron capture detectors, gas chromatography-mercuric oxide
detectors and gas chromatography-mass spectrometry) and spectroscopic technologies (e.g., gas filter
correlation (GFC)-NDIR, Fourier-transform infrared spectroscopy, tunable diode laser spectroscopy,
and resonance fluorescence) [2]. Among these instruments, NDIR has been widely used for field
monitoring of CO because its optics and detecting system are less complicated than that of other
instruments [7]. Furthermore, NDIR technique is also suitable for in situ application because it is often
less bulky than gas chromatographic technologies. In addition, the GFC-NDIR method is well-known
as a standard method for monitoring CO [8].

Atmosphere 2018, 9, 495; doi:10.3390/atmos9120495 www.mdpi.com/journal/atmosphere

Atmosphere 2018, 9, 495 2 of 14

The principle
Atmosphere of PEER
2018, 9, x FOR the NDIR technique
REVIEW is based on the Beer–Lambert law [9], 2 of 14

I = I0 × exp(−kCL) (1)
I = I0 × exp(−kCL) (1)
where II is
where is the
the intensity
intensity of
of light
light after
after penetrating
penetrating the
the target gas, II00 is
target gas, is the
the initial
initial intensity
intensity from
from the
the
infrared
infrared (IR) source, kk is
(IR) source, is the
the absorption
absorption coefficient
coefficient for
for aa specific
specific gas
gas and
and filter combination, C
filter combination, C is
is the
the
concentration the target gas, L
concentration of the target gas, L is the path length between the IR source and detector, and kL is
of is the path length between the IR source and detector, and kL is
defined
defined asas the
the sensitivity
sensitivity of
of the
the sensor
sensor [10].
[10].
Interference
Interference is is one
one of
of the
the main
main issues
issues with
with the
the NDIR
NDIR technique.
technique. Figure
Figure 11 shows
shows the
the absorption
absorption
ratios of several compounds emitted from a coal-fired power
ratios of several compounds emitted from a coal-fired power plant. plant.

Figure 1. Absorption ratios of several gases emitted from a coal-fired power plant observed from the
high-resolution
high-resolution transmission
transmissionmolecular
molecularabsorption
absorption(HITRAN)
(HITRAN)database
database[11].
[11].(The
(Theabsorption
absorptionratio atat
ratio a
spectral line was the fraction of that line’s absorption intensity and the maximum adsorption
a spectral line was the fraction of that line’s absorption intensity and the maximum adsorption intensity
of that gas).
intensity of that gas).

As
As shown
shown inin Figure
Figure 1,
1, the
the spectra
spectra of
of NO,
NO, NO
NO22,, and
and SO overlapped with
SO22 overlapped with H O. Hence,
H22O. Hence, the
the analysis
analysis
of
of these
these gases
gases using
using NDIR
NDIR would
would result
result in
in bias.
bias. Likewise,
Likewise, COCO22 caused
caused an
an interference
interference error
error for
for CO.
CO.
To overcome this interference issue, new developments within the NDIR technique
To overcome this interference issue, new developments within the NDIR technique have evolved. A have evolved.
A review
review ononthe
thedevelopment
developmentofofthe theNDIR
NDIRtechnology
technology can
can be
be found
found elsewhere
elsewhere [12]. The general
[12]. The general
improvement of the CO NDIR analyzer is shown
improvement of the CO NDIR analyzer is shown in Table 1.in Table 1.

1. Recent
Table 1.
Table Recent studies
studies on
on CO
CO non-dispersive
non-dispersive infrared
infrared analysis.
analysis.
Type
Typeof Non-Dispersive
of Non- Application Field Improvement Part Advantages
Infrared (NDIR)
dispersive Application Field Improvement Part Advantages
infrared (NDIR) Coupled with off-axis integrated cavity Fast response due to faster gas
Greenhouse gases
Analyzer [13] output spectroscopy to a Weiss-type equilibration process. Detection limit
from ocean Coupled with off-axis integrated Fast response due to faster gas
Greenhouse gases equilibrator <40 ppt, high precision, stable
Analyzer [13] cavity output spectroscopy to a equilibration process. Detection limit
from ocean Cross-interference correction with
Weiss-type
Gas emission from a equilibrator
multi-bandpass filter. Using fitted <40High
ppt,accuracy
high precision, stable
due to apply fitted
Analyzer [14] Cross-interference correction with
stack interference functions for nonlinear interference functions
absorption compounds
Gas emission from multi-bandpass filter. Using fitted High accuracy due to apply fitted
Analyzer [14]
a stack interference functions for nonlinear
Applied broad-spectral light-emitting interference functions
LED is lower cost and smaller size than
diodes (LED)
Gas emission from absorption in the 3–5 µm wavelength
compounds
Sensor [15] a laser system. Fast response due to
combustion engine as an infrared (IR) source instead of a
Applied broad-spectral light- LED is frequency
high lower costofand
LEDsmaller
(250 Hz)size
Gas emission from laser source
emitting diodes (LED) in the 3–5 µm than a laser system. Fast response
Sensor [15] combustion Noise was removed by digital filter
wavelength as an infrared (IR) due to high frequency of LED (250
Construction
engine Digital filter to remove noise. Applied leaded to high accuracy. Stability in
Sensor [4] source instead of a laser source Hz)
industry sensor networks with Wi-Fi long-term operation due to integrated
Noise
datawas
fromremoved by digital filter
sensor networks
Construction Digital filter to remove noise. leaded to high accuracy. Stability in
Sensor [4]
industry Applied sensor networks with Wi-Fi long-term operation due to
integrated data from sensor networks
The barrier helped to remove thermal
Air monitoring in Uncooled small InSb-based detector
Sensor [16] insulation. Hence, the size of the
vehicles, building with AlInSb barrier. Several single
sensor was reduced. The i-layer
Atmosphere 2018, 9, 495 3 of 14

Table 1. Cont.

Type of Non-Dispersive
Application Field Improvement Part Advantages
Infrared (NDIR)
The barrier helped to remove thermal
insulation. Hence, the size of the sensor
Uncooled small InSb-based detector
Air monitoring in was reduced. The i-layer helped to
with AlInSb barrier. Several single
Sensor [16] vehicles, building air improve signal-to-noise S/N ratio,
photodiodes with i-layer thickness
conditioning systems which leads to increase sensitivity.
connected in series
The detector could detect a wide band
for multi-gas detection
One IR source coupled with 4
Multi-detectors for multi-gas detection.
Environment pyroelectric detectors. Compensation
Sensor [17] Effect factors were compensated
monitoring parameter for temperature, humidity,
resulted in high accuracy
and pressure
Help to improve selectivity and
Sensor [12] Fire warning Applied photoacoustic detector
precision

Although the interference effect is a pivotal issue for the NDIR analyzer, development of the CO
NDIR analyzer has not advanced as much as the development of the sensor. The reason is that the GFC
is the best available method to compensate for the interference in a CO NDIR analyzer, especially for an
unknown emission source. However, the GFC method has limitations, such as gas leaks and a limited
correction range [18]. Using a multi-bandpass filter is a good method, but leads to a cost increase,
which is another issue. Therefore, this study was carried out to develop a non-interference bandpass
filter (BPF) based on the high-resolution transmission molecular absorption (HITRAN) database for a
CO NDIR analyzer of a continuous monitoring system. The scope of this study was the gas emission
patterns of a coal-fired power plant.

2. Material and Methods

2.1. Apparatus
A GFC-NDIR analyzer (NMA-N500, Nara Controls Inc., Seoul, Korea) was used and modified for
this study. Information about the analyzer is shown in Table 2.

Table 2. Characteristic of the NDIR analyzer.

Parameter Value Unit

Measurement range 0–200 ppm
Optical path length 8 m
Air flow rate 1 L/min
Gas chamber temperature 45 ◦C
Bandpass filter (BPF) for CO wavelength (half bandwidth) 4.64 (180) µm (nm)
Reference BPF (half bandwidth) 3.95 (90) µm (nm)

Figure 2 depicts the experimental set-up. Standard gases were purchased from Rigas Co., Ltd.,
Daejeon, Korea, which were produced based on the gravimetric method [19] (ISO 6142-1:2015).
Humidity was produced using the bubble method at 25 ◦ C (see Figure 2). The humidity was measured
by a humidity sensor (645, Testo SE & Co. KGaA, Lenzkirch, Germany). Several gas mixtures were
also used to investigate the response of the analyzer. A 25-L Tedlar bag (SKC Inc., Pittsburgh, PA,
USA) was used to store the mixtures. They were re-checked by a multi-gas analyzer (60i, Thermo
Fisher Scientific, Waltham, MA, USA). The experiment was repeated 3 times. The relative standard
deviation of the repeated experiments was less than 1%. The transmittances of BPFs were measured
using Fourier-transform infrared spectroscopy (FTIR) (MB3000, ABB Group, Mannheim, Germany).
Atmosphere 2018, 9, 495 4 of 14
Atmosphere 2018, 9, x FOR PEER REVIEW 4 of 14

Figure
Figure 2. Schematicdiagram
2. Schematic diagramofofexperimental
experimental set-up.
set-up. NDIR:
NDIR: non-dispersive
non-dispersive infrared;
infrared; MFC:MFC:
massmass
flow
flow controller.
controller.

2.2. Investigation of the Non-Interfering Wavelength for CO Gas Emitted from a Coal-Fired Power Plant
2.2. Investigation of the Non-Interfering Wavelength for CO Gas Emitted from a Coal-Fired Power Plant
A case study on the gases emitted from a coal-fired power plant was conducted. Untreated gases
A case study on the gases emitted from a coal-fired power plant was conducted. Untreated gases
emitted from a coal-fired power plant consist typically of 3 to 4% oxygen, 0–20 ppm of CO, 15 to 16%
emitted from a coal-fired power plant consist typically of 3 to 4% oxygen, 0–20 ppm of CO, 15 to 16%
CO2 , 5–7% H2 O (vapor), approximately 73% N2 , 0–1000 ppm of SO2 , and 0–500 ppm of NOx [20,21].
CO2, 5–7% H2O (vapor), approximately 73% N2, 0–1000 ppm of SO2, and 0–500 ppm of NOx [20,21].
Therefore, CO, CO2 , NOx (NO + NO2 ), SO2 , and H2 O (vapor) were used as the target gases in this
Therefore, CO, CO2, NOx (NO + NO2), SO2, and H2O (vapor) were used as the target gases in this
study. N2 O is an interference gas for CO [13,15]. However, the emission of N2 O from the combustion
study. N2O is an interference gas for CO [13,15]. However, the emission of N2O from the combustion
process was not a significant amount [15]. Therefore, the effect of N2 O on CO measurement in this
process was not a significant amount [15]. Therefore, the effect of N2O on CO measurement in this
case was neglected.
case was neglected.
The line-by-line absorption intensity from the HITRAN database [11] (hitran.org; Atomic and
The line-by-line absorption intensity from the HITRAN database [11] (hitran.org; Atomic and
Molecular Physics Division, Harvard-Smithsonian Center for Astrophysics, Cambridge, MA, USA)
Molecular Physics Division, Harvard-Smithsonian Center for Astrophysics, Cambridge, MA, USA)
was used as a reference. Since the main infrared (IR) absorption of CO is approximately 4.5 to 5.0 µm
was used as a reference. Since the main infrared (IR) absorption of CO is approximately 4.5 to 5.0 µm
(see Figure 1), the wavelengths of 2.5 to 10 µm from the HITRAN database were considered. From the
(see Figure 1), the wavelengths of 2.5 to 10 µm from the HITRAN database were considered. From
line-by-line absorption intensity, the maximum value of each gas was determined. Since different gases
the line-by-line absorption intensity, the maximum value of each gas was determined. Since different
demonstrated different absorption levels, the absorption ratio was estimated for easy comparison. For a
gases demonstrated different absorption levels, the absorption ratio was estimated for easy
certain gas, the absorption ratio at a spectral line was the fraction of that line’s absorption intensity and
comparison. For a certain gas, the absorption ratio at a spectral line was the fraction of that line’s
the maximum adsorption intensity of that gas. All absorption ratios which were >0.1% (i.e., 99.9% of
absorption intensity and the maximum adsorption intensity of that gas. All absorption ratios which
absorbance) of all target compounds were scanned and matched to find the non-interfering range. Then,
were >0.1% (i.e., 99.9% of absorbance) of all target compounds were scanned and matched to find the
novel BPFs based on this range were made by the Seoul Precision Optics Co., Ltd., Shiheung, Korea.
non-interfering range. Then, novel BPFs based on this range were made by the Seoul Precision Optics
Generally, the absorbance (A) of a gas can be obtained from the transmittance (T) through the
Co., Ltd., Shiheung, Korea.
following relationship.
Generally, the absorbance (A) of a gas can be obtained from the transmittance (T) through the
T = I/I0 = 10−A (2)
following relationship.
From Equations (1) (see Section 1) and (2), the relationship between gas absorbance and its
T = I/I0 = 10−A (2)
concentration can be derived as
From Equations (1) (see Section 1)Aand = log(2),
(I0the
/I) =relationship
εCL between gas absorbance and (3) its
concentration can be derived as
where ε = log(e) × k = 0.43 k.
Since k is the absorption coefficient forAa=specific
log (I0/I)gas
= εCL
and filter combination [9], the transmittance (3)
and
wherebandwidth
ε = log(e) of
× kthe BPFk.should be considered. Within the spectral range of a certain BPF, there
= 0.43
are many line-by-line spectracoefficient
Since k is the absorption with different absorption
for a specific gas andintensities. Therefore,
filter combination [9],a the
total absorption
transmittance
intensity
and bandwidth of the BPF should be considered. Within the spectral range of a certain BPF, BPF
was calculated. The total absorption intensity of a gas with respect to a certain therewas
are
calculated to compare
many line-by-line the with
spectra effects of interferences
different absorption [22]. The totalTherefore,
intensities. absorption intensity
a total was calculated
absorption intensity
was calculated. The total absorption intensity of a gas with respect to a certain BPF was calculated to
compare the effects of interferences [22]. The total absorption intensity was calculated based on
Atmosphere 2018, 9, 495 5 of 14

based on Equation (4). The average transmittance was used because the transmittance displayed in the
line-by-line spectra varied within the BPF spectral range. Therefore, using the average transmittance
was better than using the maximum transmittance.
n
Total absorption intensity = ∑ ai × Ngas × T 0 × L (4)
i =1

n
where ∑ ai is the integration of all line-by-line absorption intensities from HITRAN database of each
i =1
gas from line 1 to line n within each BPF spectral range (cm−1 /molecule.cm−2 ), Ngas is the number of
molecules of that gas (i.e., the number of molecules was converted from the concentration because the
absorption intensity from HITRAN was calculated based on molecule) (molecule), T’ is the average
transmittance of the BPF (unitless) and L is the optical pathlength of the gas chamber (cm).
On the other hand, the signal-to-noise ratio (S/N), which is a more useful variable than only noise,
is also an important factor in determining the quality of an instrument [23,24]. S/N should be higher
than 3 in order to detect a signal [24]. The S/N depends on many factors. In terms of the BPF, the S/N
caused by the BPF is proportional to the detectivity of the detector, which is presented through the
following equation [23,25]
D ∗ × Pin
S/N = p (5)
Ad ∆ f
where S/N is the signal-to-noise ratio, D* is the detectivity of the detector, Ad is the active area of the
detector, Pin is the IR intensity reaching the detector, and ∆f is the bandwidth.
Since the same detector and IR source were always used for all experiments, the S/N fraction of
the 2 BPFs could be derived as follows:
s
S/NBPF1 ∆ f BPF2 T
= × 1 (6)
S/NBPF2 ∆ f BPF1 T2

where ∆fBPF1 and ∆fBPF2 are the bandwidths of bandpass filter 1 (BPF1 ) and bandpass filter 2 (BPF2 ),
respectively and T1 and T2 are the average transmittances of BPF1 and BPF2 , respectively. The average
transmittance was employed instead of the radiant input because an identical IR source was used to
supply the same IR radiant. In this equation, the bandwidth is the most important factor because the
bandwidths of new BPFs were much lower than that of the commercial one while their transmittances
were not much different.
For a direct current (DC) signal, S/N can be practically observed from a variety of signals as
follows [24].
S
S/N = (7)
δS
where S is the mean of signal and σS is the signal standard deviation.
Accordingly, the sensitivity, theoretical and experimental interference, and theoretical and
experimental S/N of the 2 new BPFs and a commercial BPF were evaluated and compared.

2.3. Production of Novel BPFs

The new BPFs were made based on the results from Section 2.2. The novelty of these BPFs was
their center wavelength and bandwidth. In general, BPFs at 4.64 µm or 4.75 µm have been widely
used for CO analysis using the NDIR techniques, whereas the new BPFs had wavelengths of 4.5 and
4.65 µm. In addition, the bandwidths of new BPFs were narrower than those of the commercial BPFs.
The optical properties of the new BPFs will be introduced in more detail in Section 3.1. Two-cavity
filter method was used to make the new BPFs with Germanium substrate and SiO coating material.
Atmosphere 2018, 9, 495 6 of 14

2.4. Investigation of the Effect of Interfering Gas on the Novel BPFs

A commercial BPF at 4.64 µm (180 nm) (NBPF 4.64; Seoul Precision Optics Co., Ltd., Shiheung,
Korea) was selected to compare with the new BPFs. The description of the novel BPF will be introduced
in more details in Section 3.1. Each BPF was applied to the NDIR analyzer to investigate its performance.
Various concentrations of target gases were introduced into the analyzer to investigate the interference.
The N2 gas (99.999%, Dong A Co., Ltd., Gyeonggi, Korea) and CO standard gases at 10, 50, 100,
and 298 ppm (Rigas Co., Ltd., Daejeon, Korea) were used to calibrate the analyzer. Based on the
literature [20,21], the NO, NO2 , and SO2 levels were varied from 1 to 1000 ppm (Rigas Co., Ltd.,
Daejeon, Korea), and CO2 levels were varied from 1 to 50% (Rigas Co., Ltd., Daejeon, Korea) to
investigate the interference effect. In general, a cooler or a membrane is usually used to remove the
H2 O vapor before introducing the gas stream into the analyzer. Hence, it was presumed that the inlet
air of the analyzer would be dry. However, these instruments could not completely remove all H2 O
vapor. Accordingly, we assumed that the inlet H2 O vapor concentration of the analyzer was less than
25,800 ppm at 25 ± 2 ◦ C. The H2 O vapor levels, which were directly introduced into the analyzer, were
varied from 1400 to 23,000 at 25 ◦ C in this study.
Several gas mixtures were also used to investigate the performance of the analyzer with the
original structure (i.e., GFC-NDIR), the novel BPF, and the BPF at 4.64 µm. Based on the Korean
emission regulations for a new coal-fired power plant, the SO2 and NOx concentrations should be
≤50 ppm [26]. Moreover, CO and CO2 concentrations were selected based on the literature [20,21]
because there were no emission regulations for these gases. Consequently, gas mixtures consisting
of CO (10, 100, and 200 ppm), NO (50 ppm), CO2 (15%), NO2 (50 ppm), SO2 (50 ppm), and H2 O
(~14,360 ppm) were generated and introduced into the analyzer to determine the measurement errors.
The measurement errors were obtained from the relative percent difference between the standard gas
and the analytical results. NO, CO, NO2 , and SO2 were diluted from 2000 ppm of the standard gases
(Rigas Co., Ltd., Daejeon, Korea). CO2 was diluted from 50% of the standard gas (Rigas Co., Ltd.,
Daejeon, Korea). H2 O was diluted from 80% of relative humidity. N2 was used as the balancing gas.
A 25-L Tedlar bag was used to prepare gas mixtures. N2 was introduced into the Tedlar bag with the
flow rate controlled by a mass flow controller (M3030VA, Line Tech Ltd., Co., Daejeon, Korea). Standard
gas samples were introduced into the Teblar bag with a 500 mL glass syringe (500MAR-LL-GT, SGE
Analytical Science Pty. Ltd., Ringwood, Australia). Each gas was introduced into the NDIR analyzer
for 15 min. Results from the first 5 min were ignored due to equilibrium concentration in the analyzer.
The values from the last 10 min were collected. Each experiment was repeated 3 times. The signal was
recorded every second. The relative absorbance of each gas was practically calculated based on the log
of the detector signal from the reference channel to the detector signal from the CO channel.

3. Results and Discussion

3.1. Non-Interfering Wavelengths for CO Gas Emitted From a Coal-Fired Power Plant
From the HITRAN, the non-interfering wavelengths for CO in the presence of CO2 , NOx , SO2 , and
H2 O were found to be in the range of 4.76–4.78 (bandwidth = 20 nm), 4.55–4.75 (bandwidth = 200 nm),
and 4.46–4.54 µm (bandwidth = 80 nm). Due to technical limitations, a new BPF at 4.5 µm (i.e., BPF_1)
with a half bandwidth of 50 nm and a new BPF at 4.65 µm (i.e., BPF_2) with a half bandwidth of 60 nm
were manufactured. In general, a BPF filter consists a sort of thin-film layers. The bandwidth of the
BPF depends on the number of layers. The more thin-film layers there are, the narrower the bandwidth
of the BPF is. However, the transmittance of the BPF is also influenced by the BPF thickness and the
tolerance of the coating process [27]. Therefore, it is a challenge to reduce the bandwidth as much as
the values above while maintaining a high transmittance. The transmittance should be high enough to
retain a good S/N ratio for the analyzer. The commercial BPF at 4.64 µm with a half bandwidth of
180 nm was labelled as BPF_3. The transmittances of three BPFs are presented in Figure 3.
Atmosphere 2018, 9, 495 7 of 14
Atmosphere 2018, 9, x FOR PEER REVIEW 7 of 14

Figure 3.
3. Spectra of target
Spectra band
of target passpass
band filtersfilters
(BPF_1, BPF_2,BPF_2,
(BPF_1, and BPF_3) observed
and BPF_3) from thefrom
observed Fourier-
the
transform infrared spectroscopy.
Fourier-transform infrared spectroscopy.

As shown
As shown in inFigure
Figure3,3,the the full
full bandwidth
bandwidth at transmittance
at transmittance ≥0.1%
≥0.1% of theoftwo
thenew
two BPFs
new was
BPFsmuch
was
much narrower than that of BPF_3. The full bandwidths of the BPF_1,
narrower than that of BPF_3. The full bandwidths of the BPF_1, BPF_2, and BPF_3 were 141, 262, and BPF_2, and BPF_3 were
141,
567 nm,262,respectively.
and 567 nm,Table respectively.
3 shows Table 3 shows
a summary a summary
of the of thefor
specifications specifications
each BPF. The formaximum
each BPF.
The
transmittances of the BPF_1 and BPF_2 did not differ greatly from that of BPF_3. Theofaverage
maximum transmittances of the BPF_1 and BPF_2 did not differ greatly from that BPF_3.
The average transmittances of BPF_1, BPF_2, and BPF_3 were 26.8%,
transmittances of BPF_1, BPF_2, and BPF_3 were 26.8%, 24.6%, and 27%, respectively. With the 24.6%, and 27%, respectively.
With
optical the optical
path path
length of 8length
m, theofvalues
8 m, the
for values for sensitivity
sensitivity kL of BPF_1, kLBPF_2,
of BPF_1,andBPF_2, and BPF_3
BPF_3 were 5.45 × were
10−16,
5.45 × 10 −16 , 1.82 × 10− 15 , and 2.19 × 10 −15 (1/molecule·cm−2 ) respectively. This indicates that
1.82 × 10 , and 2.19 × 10 (1/molecule·cm ) respectively. This indicates that the sensitivity of BPF_3
−15 −15 −2

the
wassensitivity
four timesofhigher
BPF_3thanwas thatfour oftimes higher
BPF_1 andthan that ofhigher
1.2 times BPF_1thanand that
1.2 times higher
of BPF_2. than thatthe
However, of
BPF_2. However,
theoretical fractiontheoftheoretical
S/NBPF_3 tofraction
S/NBPF_1of S/Nonly
was to S/N
BPF_30.50, and thatwas
BPF_1 only
of S/N 0.50,
BPF_3 to and
S/NBPF_2 of S/N
thatwas only to
0.74.
BPF_3
S/N
TheBPF_2 was only 0.74. The theoretical
theoretical S/NBPF_3 was not much different S/N BPF_3 was not much different from the new BPFs
from the new BPFs because its average transmittances because
its
wereaverage
similar transmittances wereFrom
to that of others. similar to that of
Equation (7),others. From Equation
the practical (7), theBPF_2,
S/N of BPF_1, andS/N
practical of BPF_1,
BPF_3 were
BPF_2, and BPF_3 were calculated as 20.0, 19.6, and 49.7, respectively. In the
calculated as 20.0, 19.6, and 49.7, respectively. In the experimental results, the S/N caused by BPF_3experimental results, the
S/N caused by BPF_3 was approximately 2.5 times higher than that caused
was approximately 2.5 times higher than that caused by BPF_1 and BPF_2 due to their relatively by BPF_1 and BPF_2 due
to theirvalues
lower relatively
for lower
maximum values for maximumHowever,
transmittance. transmittance.
these However,
S/N values these
wereS/N values
much werethan
higher much 3.
higher than 3. Therefore, these new
Therefore, these new BPFs can be applicable [24]. BPFs can be applicable [24].

Table 3.
Table Specification of
3. Specification of BPFs
BPFs using
using in
in this
this research.
research.
Center
Center Half
Half Band FullFull
Band Bandwidth
Bandwidth ** Maximum
Maximum Average
Average
BPFBPF Wavelength (µm) Width (nm) (nm) Transmittance (%) Transmittance (%)
Wavelength (µm) Width (nm) (nm) Transmittance (%) Transmittance (%)
BPF_1
BPF_1 4.54.5 5050 141141 70.1
70.1 26.8
26.8
BPF_2 4.65 60 262 75.4 24.6
BPF_2
BPF_3 4.65
4.64 60180 262567 75.4
83.8 24.6
27
BPF_3 4.64 180 567 83.8 27
Note: * Full bandwidth from 0.1% of transmittance.
Note: * Full bandwidth from 0.1% of transmittance.
Although BPF_1 and BPF_2 were made with very narrow half bandwidths, their real full
Although
bandwidth wasBPF_1 andthe
still over BPF_2
rangewere made from
estimated with HITRAN.
very narrow half the
Hence, bandwidths,
theoreticaltheir real full
and practical
bandwidth was
interference still over
should the range estimated
be investigated. from HITRAN.
The theoretical Hence,isthediscussed
interference theoretical
in and
thispractical
section.
interference
The practicalshould be investigated.
interference The theoretical
will be discussed in Sectioninterference is discussed
3.2. Based on in this
the absorption section.from
intensity The
practical interference
HITRAN, willlength
the optical path be discussed in Section
of the analyzer, 3.2. Based
the average on the absorption
transmittance intensity
of BPFs, and from
the various
HITRAN, the optical
concentrations path
of target length
gases, theoftotal
the absorption
analyzer, the averageoftransmittance
intensity each gas was of BPFs, andThese
evaluated. the various
values
concentrations
are depicted in of target
Figure 4. gases, the total absorption intensity of each gas was evaluated. These values
are depicted in Figure 4.
Atmosphere 2018, 9, 495 8 of 14
Atmosphere 2018, 9, x FOR PEER REVIEW 8 of 14
Atmosphere 2018, 9, x FOR PEER REVIEW 8 of 14

Figure 4. Total absorption intensity of CO, NO, NO2, SO2, CO2, and H2O. The concentration of CO
Figure
Figure 4.4. Total
Totalabsorption
absorptionintensity
intensityofofCO,
CO,NO,NO,NO
NO 2 ,2,SO
SO2 2, ,CO
CO22,, and
and HH22O.
O. The
The concentration
concentration ofof CO
CO
was from 1 to 298 ppm; that of NO, NO2, and SO2 was from 1 to 1000 ppm; that of CO2 was from
was
was from
from 11to
to298
298ppm;
ppm; that
thatof
ofNO,
NO, NO
NO22, and SO22 was from 1 to 1000 ppm; that of CO22 was from
, and SO was from 1 to 1000 ppm; that of CO was from
10,000 to 500,000 ppm; and that of H2O was from 1431 ppm to 22,972 ppm at 25 °C, ◦ C,1 1atm.
10,000
10,000to
to500,000
500,000ppm;
ppm;and
andthat
thatofofHH22O
O was
was from
from1431
1431ppm ppmto to22,972
22,972ppm
ppmat at25
25 °C, 1atm.
atm.
As shown in Figure 4, NO, NO2, and SO2 did not reveal significant effects on any BPFs. H2O had
As
Asshown
shown in
inFigure
Figure 4,4,NO, NO
NO22,,and
andSO
NO,concentrations SO22ofdid
did not
not reveal
reveal significant
significant effects on any BPFs.
BPFs.H H22O
O had
a slight effect on BPF_2 at low CO. CO 2 showed a strongeffects onCO
effect on anymeasurement. had
aa slight
slight effect
effect onon BPF_2
BPF_2 at low
low concentrations
atbetween
concentrations of CO. CO showed aa strong
CO22 showed strong effect onon CO measurement.
measurement.
This cross-interference CO and COof 2 isCO.
well-known [13–15]. Theeffect
CO2 gas CO demonstrated a
This cross-interference between CO and
andCO 22 is
iswell-known [13–15]. The
The CO gas demonstrated a
This cross-interference
strong effect on BPF_3between
because COthe linear CO well-known
regression of CO2 was [13–15].
the closest to22that
CO gasofdemonstrated
CO (i.e., the a
strong
strong effect on onBPF_3 because thethe
linear regression of CO of22COwas the closest to that
to of COof(i.e., the slope
slopeeffect
between BPF_3
two because
curves was 2.3 × linear
10−4). Inregression
contrast, CO 2 wasless
showed theeffect
closest
on BPF_1thatand CO (i.e.,
BPF_2. Atthe
between two curves was 2.3 × 10 −4 ). In contrast, CO showed less effect on BPF_1 and BPF_2. At 50%
50%between
slope CO2, the twoslope
curvesbetween
was 2.3two
× 10curves
−4). In contrast, 2
at BPF_1 COwas aboutless
2 showed 0.074 and
effect onthat
BPF_1at and
BPF_2 was At
BPF_2.
CO
50% 2 , the
COslope
approximately between
2, the slope
twoindicates
0.017. between
This curves
twoatthatBPF_1
curvesBPF_1was about 0.074
at reduced
BPF_1 and
more about
was that
interference at BPF_2
0.074 effect was the
and than
that approximately
at other
BPF_2two was
0.017.
BPFs. This indicates that BPF_1 reduced more interference effect than the
approximately 0.017. This indicates that BPF_1 reduced more interference effect than the other two other two BPFs.
BPFs. The interference
The interference ratios
ratioscaused
causedby bymixing
mixinggasesgases were
were also estimated.The
also estimated. Theconcentration
concentration ofof target
target
gases
gases in the mixture was described in Section 2.3. The interference ratios are depicted in Figure 5.5.
in the mixture was described in Section 2.3. The interference ratios
The interference ratios caused by mixing gases were also estimated. The concentration of target are depicted in Figure
gases in the mixture was described in Section 2.3. The interference ratios are depicted in Figure 5.

Figure
Figure 5. Interference
5. Interference ratios
ratios of target
of target gases
gases with
with respect
respect to different
to different mixtures.
mixtures. (a) (a) 10 ppm
10 ppm of CO;
of CO; 50
50 ppm
ppm of NO, NO , and SO ; 15% of CO ; and 14,360 ppm of H O. (b) 100 ppm
of NO, NO2 , and SO2 ; 15% of CO2 ; and 14,360 ppm of H2 O. (b) 100 ppm of CO; 50 ppm of NO, NO2 ,
2 2 2 2 of CO; 50 ppm of NO,
Figure
and SO2,25.
NO andInterference
; 15% SOof2; CO
15%2 ;of ratios
CO14,360
and of target
2; and ppm gases
14,360 ofppm with
of
H2 O. H2O.
(c) respect
200toppm
(c)ppm
200 ofdifferent
CO;of 50 mixtures.
CO;ppm
50 ppm of(a)
of NO, NO,
NO102NO
,ppm of
SOCO;
2, and
and SO 2; 50
2 ; 15%
ppm
of 15%
COof of CO ; and 14,360 ppm of H O.
NO, NO2, and SO2; 15% of CO2; and 14,360 ppm of H2O. (b) 100 ppm of CO; 50 ppm of NO,
2 ; and 14,360 ppm of H2 O.
2 2

NO2, and SO2; 15% of CO2; and 14,360 ppm of H2O. (c) 200 ppm of CO; 50 ppm of NO, NO2, and SO2;
15% of CO2; and 14,360 ppm of H2O.
Atmosphere 2018, 9, 495 9 of 14
Atmosphere 2018, 9, x FOR PEER REVIEW 9 of 14

As shown
As shown ininFigure
Figure5, 5,
at 10
at ppm of CO,
10 ppm of CO
CO,2 revealed significant
CO2 revealed interference
significant on all BPFs.
interference However,
on all BPFs.
the lowest effect appeared on BPF_1. On the other
However, the lowest effect appeared on BPF_1. On the other hand, hand, H 2 O showed significant effects on
H2O showed significant effects on BPF_2.
The higher
BPF_2. the COthe
The higher concentration, the lower
CO concentration, the effect
the lower of COof
the effect 2 . CO
At 2100
. Atppm of CO,
100 ppm CO2CO
of CO, denoted
2 denotedthe
lowest
the effect
lowest on BPF_1,
effect followed
on BPF_1, followedby BPF_2
by BPF_2andand
BPF_3. At At
BPF_3. 200200ppm ppmof CO, COCO
of CO, 2 showed
2 showedless effect
less on
effect
both of BPF_1 and BPF_2. In all cases,
on both of BPF_1 and BPF_2. In all cases, CO CO demonstrated significant effects on
2 2 demonstrated significant effects on BPF_3. BPF_3. In terms
terms ofof
HH22O,
O,significant
significanteffects
effectson
onallallBPFs
BPFscould
could not
not bebe found
found atat high
high CO CO concentrations.
concentrations. Although
Although the the full
full
bandwidth was
bandwidth was large,
large, BPF_1
BPF_1 compensated
compensated moremore for for the
the interference effect of CO22 and and HH22OO than
than the
the
other two
other two BPFs.
BPFs.

3.2. Effect
3.2. EffectofofTarget
Target Gases
Gases on
on the
the Novel
Novel BPFs
BPFs
Various concentrations
Various concentrations of of aa single
single target
target gas
gas were
were used
used to
to determine
determine thethe real
real interference
interference effect
effect
on new
on new BPFs,
BPFs, as
as well
well asas the
the commercial
commercialBPF. BPF. The
The CO
CO calibration
calibrationcurves
curvesarearepresented
presentedin inFigure
Figure6.6.
Relative absorbances were calculated from the log of ratio values between detector
Relative absorbances were calculated from the log of ratio values between detector signals from the signals from the
reference channel (i.e., reference BPF at 3.95 µm) and detector signals of the CO measurement
reference channel (i.e., reference BPF at 3.95 µm) and detector signals of the CO measurement channel channel
(i.e., BPF
(i.e., BPFofofCO
COgas).
gas).ItItwas
wasfound
foundthat
thatallall BPFs
BPFs showed
showed good
good performance
performance in terms
in terms of calibration
of calibration (r2
(r 2 > 0.99).
> 0.99).

Figure
Figure6.
6.CO
COcalibration
calibrationcurves
curvesassociated
associatedwith
withBPF_1,
BPF_1, BPF_2,
BPF_2, and
and BPF_3.
BPF_3.

The SO
The SO22, ,NO
NO2,2 ,and
andNO NOgas gasconcentrations
concentrationswere werevaried
variedat atthe
thefollowing
followingvalues:
values:1,1,10, 10,50,50,100,
100,200,
200,
300, 400, 500, 600,
300, 600, 700,
700,800,800,900
900and and1000
1000ppm.ppm.The TheCO CO2 concentrations
2 concentrations werewere varied
varied as 1%,
as 2%,
1%, 3%,
2%, 4%,
3%,
6%, 6%,
4%, 10%,10%,12%,12%,23%23% andand 50%.50%. TheTheH2 OH(vapor)
2O (vapor)waswasvaried as 5%,
varied 10%,10%,
as 5%, 20%,20%, 30%,30%, 40%,40%,50%,50%,60%,60%,70%
and and
80% 80% of relative humidity at 25 at◦ C (i.e., from 1400 to 23,000 ppm). The interference of target gases
70% of relative humidity 25 °C (i.e., from 1400 to 23,000 ppm). The interference of target
on each
gases on BPF
eachisBPF depicted in Figures
is depicted 7–9. 7–9.
in Figures
Asshown
As shownininFigureFigure7,7,BPF_1BPF_1was was notnot significantly
significantly affected
affected by by
NO, NO,NONO 2, or 2 ,SO
or2.SO 2 . This
This patternpattern
was
was similar
similar to thatto of
that
theoftheoretical
the theoretical estimation.
estimation. On theOnother
the other
hand, hand,
H2O H and
2 O and
CO 2 CO
revealed
2 revealed
effectseffects
at lowat
low ppm
ppm concentrations
concentrations (i.e.,(i.e.,
10 ppm)10 ppm) of CO of CO in the
in the theoretical
theoretical estimation,but
estimation, butthe theexperimental
experimentalresults results
did not
did not show
show these
these effects
effects ofof these
these gases.
gases.
As shown
As shownin inFigure
Figure8,8,NO, NO,NO NO , and
2, 2and H2HO2 O
diddid
notnot demonstrate
demonstrate interference
interference effects
effects on BPF_2.
on BPF_2. In
In contrast,
contrast, a slightly
a slightly negative
negative effecteffect
caused caused
by SO by SO2found.
2 was was found. Particularly,
Particularly, CO2 revealed CO2 revealed strong
strong effects
effects
on on BPF_2.
BPF_2. The effectThe effect
of COof 2 wasCO2 stronger
was stronger on BPF_2
on BPF_2 thanthan on on BPF_1
BPF_1 because
because BPF_2had
BPF_2 hadaalarger
larger
bandwidth than BPF_1, and the total CO absorption intensity in
bandwidth than BPF_1, and the total CO absorption intensity in this band was higher than that of this band was higher than that of
BPF_1.BPF_3
BPF_1. BPF_3was wasalsoalsostrongly
stronglyaffected
affectedby byCO CO 2 (Figure
2 (Figure 9).9).
AtAt 50%50%of of
CO CO 2 , the
2, the error
error concentration
concentration of
of CO
CO was was about
about 50 50
ppm.ppm. The The large
large bandwidth
bandwidth ofofBPF_3
BPF_3brought
broughtaboutaboutthe thehighhighabsorption
absorptionof ofCO CO22. .
However, the
However, the effect
effect of of CO22 on on BPF_3
BPF_3 was was weaker
weaker thanthan that
that onon BPF_2
BPF_2 because
because the the COCO absorption
absorption
intensity of
intensity of BPF_3
BPF_3 was was thethe highest
highest (i.e.,
(i.e., largest
largest full
full bandwidth)
bandwidth) based based onon HITRAN
HITRAN data. data. This
This pattern
pattern
was aa little
was little different
different fromfrom thethe theoretical
theoretical evaluation,
evaluation, which showed a stronger effect of CO22 on on BPF_3
BPF_3
than BPF_2. This difference
than differencemight mightbe bea aresult
resultofof
their different
their different transmittance
transmittance values.
values. TheThe
maximummaximum and
average
and transmittance
average transmittance of BPF_3
of BPF_3were were
higherhigher
than those
than of BPF_2.
those Moreover,
of BPF_2. due to the
Moreover, duemulti-pathway
to the multi-
gas chamber
pathway of the analyzer,
gas chamber this gapthis
of the analyzer, increased many times
gap increased manyaftertimes the IR beam
after the IR had beam reached to the
had reached
detector.
to When When
the detector. the same theIR source
same and detector
IR source were used,
and detector wereIRused,intensity which reached
IR intensity to the detector
which reached to the
detector with BPF_3 would be higher than that with BPF_2. Based on Beer–Lambert law, the detector
with BPF_2 would be more sensitive to CO and CO2 gases than that with BPF_3. This resulted in the
Atmosphere 2018, 9, 495 10 of 14

with BPF_3 would be higher than that with BPF_2. Based on Beer–Lambert law, the detector with
BPF_2 would be more sensitive to CO and CO2 gases than that with BPF_3. This resulted in the increase
of the effect of the CO2 . In addition, the errors in the HITRAN data and the observed bandwidth of
BPF by the Fourier-transform infrared spectroscopy (FITR) might also cause this bias.
An example of the analyzer signal with respect to the interference of CO2 is presented in Figure 10.
Linear regression of these analyzer signals was also estimated to compare their patterns. As shown
in the figure, the analyzer signals with respect to BPF_2 and BPF_3 denoted an increase pattern
(i.e., positive slope, r2 = 0.7224 and 0.2938 respectively). In addition, the response of the analyzer with
Atmosphere
Atmosphere 2018,
2018, 9,
9, xx FOR
FOR PEER
PEER REVIEW
REVIEW 10
10 of
of 14
14
respect to BPF_2 for CO2 was very clear compared to that of BPF_3 (i.e., slope = 2 × 10−4 > 4 × 10−6 ).
In terms of
increase of using
the effectBPF_1, the signal
of the was
CO22. In not influenced
addition, by in
the errors COthe
2 (i.e., the slope
HITRAN was
data almost
and zero and
the observed
2
r << 1).
bandwidth of BPF by the Fourier-transform infrared spectroscopy (FITR) might also cause this bias.

Figure 7. Interference
Figure 7. effect of
Interference effect of NO,
NO, NO
NO222,, SO
SO222,, CO
CO222,, and
and H
H222O
Oon
on CO
CO with
with respect
respect to
to BPF_1.
BPF_1. The
The error
error
bar
bar in
in the figure is
the figure is the
the standard
standard deviation
deviation ofof the
the repeated
repeated experiment.
experiment.

Figure 8. Interference
Figure 8. effect of
Interference effect of NO,
NO, NO
NO222,, SO
SO222,, CO
CO222,, and
and H
H222O
Oon
on CO
CO with
with respect
respect to
to BPF_2.
BPF_2. The
The error
error
bar in the figure is the standard deviation of the repeated experiment.
bar in the figure is the standard deviation of the repeated experiment.

An example of the analyzer signal with respect to the interference of CO22 is presented in Figure
10. Linear regression of these analyzer signals was also estimated to compare their patterns. As shown
in the figure, the analyzer signals with respect to BPF_2 and BPF_3 denoted an increase pattern (i.e.,
positive slope, r22 = 0.7224 and 0.2938 respectively). In addition, the response of the analyzer with
Atmosphere 2018, 9, 495 11 of 14
Atmosphere 2018, 9, x FOR PEER REVIEW 11 of 14
Atmosphere 2018, 9, x FOR PEER REVIEW 11 of 14

Figure 9. Interference effect of NO, NO2, SO2, CO2, and H2O on CO with respect to BPF_3. The error
effect of
Figure 9. Interference effect of NO,
NO, NO
NO22, SO
SO22, CO22,, and
and H22O
O on
on CO
CO with
with respect to BPF_3. The error
bar in the figure is the standard deviation of the repeated experiment.
figure is
bar in the figure is the
the standard
standard deviation
deviation of
of the
the repeated
repeated experiment.
experiment.

Figure 10. Non-dispersive infrared (NDIR) analyzer signal associated with different BPFs and
CO 2 concentrations.
Figure 10. Non-dispersive infrared (NDIR) analyzer signal associated with different BPFs and CO2
Figure 10. Non-dispersive infrared (NDIR) analyzer signal associated with different BPFs and CO2
concentrations.
To investigate the performance of theconcentrations.
analyzer employing these BPFs, different gas mixtures
basedTooninvestigate
the emission
thepatterns of a coal-fired
performance power plant
of the analyzer were introduced
employing these BPFs,into the NDIR
different gas analyzer.
mixtures
To investigate the performance of the analyzer employing these BPFs, different gas mixtures
Table 4 shows the measurement errors with respect to the mixing gases.
based on the emission patterns of a coal-fired power plant were introduced into the NDIR analyzer.
based on the emission patterns of a coal-fired power plant were introduced into the NDIR analyzer.
Table 4 shows the measurement errors with respect to the mixing gases.
Table 4 shows the measurement errors with respect to the mixing gases.
Atmosphere 2018, 9, 495 12 of 14

Table 4. Comparison between standard gas concentrations and analytical results for CO gas associated
with different BPFs.

GFC-NDIR BPF_1 BPF_2 BPF_3

CO Std. Mean of Mean of Mean of Mean of
(ppm) Mea. Mea. Mea. Mea.
Analytical Analytical Analytical Analytical
Error (%) Error (%) Error (%) Error (%)
Values (ppm) Values (ppm) Values (ppm) Values (ppm)
10 10.04 0.40 10.08 0.83 283.74 2737 18.87 88.7
100 100.22 0.22 100.22 0.22 362.93 263 105.94 5.94
200 200.17 0.08 199.84 0.08 505.79 153 205.08 2.54
Note: Measurement error (Mea. error) is obtained from the relative percent difference between the CO concentration
of standard and analytical results. CO std. is concentration of CO standard gas.

As shown in Table 4, the measurement errors associated with BPF_2 were very high due to the
strong effect of CO2 . In the case of the BPF_3, the error was very high (88.7%) at 10 ppm of CO,
but it was lower at 100 and 200 ppm of CO. However, the error was still higher than 2%. Therefore,
BPF_3 could not be employed itself for the NDIR analyzer. The commercial BPF_3 coupled with GFC
helped to remove the interference of CO2. The measurement errors of the GFC-NDIR were all less than
1%. This pattern was similar to that of BPF_1. Although there was an error at 10 ppm of CO when
BPF_1 was applied for the NDIR analyzer, the measurement error was still less than 1%. This result
was comparable to the results of other studies. Sun et al. (2013) developed a multi-NDIR analyzer
using multiple BPFs coupled with a cross-interference algorithm to compensate for the interference.
It was reported that the measurement error of CO was less than 1% with an analytical range of
0–850 ppm [14]. In another research, a combination of array filters and detectors was applied to the
multi-gas NDIR sensor, in which the measurement error of CO was less than 1.5% with an analytical
range of 0–1000 ppm [6].
These results indicate that the new BPF_1 can be applied to the CO NDIR analyzer instead of a
GFC to compensate for the interference effect, especially for CO2 . BPF_1 has the added advantage of
having no leak problems. In addition, our test results showed that BPF_1 had a good performance with
up to 50% CO2 and up to 1000 ppm NOx and SO2 . Therefore, it can be applied to monitor CO emission
from a power plant as well as the combustion process. Although the new BPF could overcome some
disadvantage of the GFC, the center wavelength of the new BPF was different to that of the reference
BPF (i.e., 3.95 µm). Consequently, drift due to aging of IR source or detector and various of temperature
would still occur [28]. Based on Planck’s law, spectral intensity at a certain wavelength depends on
temperature. When temperature varied, the IR intensity penetrated the reference BPF and target gas
BPF would change, which resulted in changing the ratio between reference channel and measurement
channel at the same target gas concentration and pressure. Hence, the base line of the analyzer would
shift, which leads to increase the analyzer bias [26]. Accordingly, the effect of temperature as well as a
new reference method on the analyzer should be investigated in the future.

4. Conclusions
A HITRAN database was used to investigate the non-interference wavelength for CO analysis
using NDIR technology. The non-interference wavelength was determined with respect to the gas
emission patterns of a coal-fired power plant. As a result, two new BPFs at 4.5 µm (with a half
bandwidth of 50 nm) and 4.65 µm (with half bandwidth of 60 nm) were developed. The S/N ratios of
the NDIR analyzer when the new BPFs were applied were approximately 20. In terms of interference
compensation, the BPF at 4.5 µm was found to remove the effect of CO2 up to 50% and that of NOx and
SO2 up to 1000 ppm. The measurement errors of the NDIR coupled with BPF_1 were similar to those
of a GFC-NDIR analyzer, showing less than 1%. Therefore, BPF_1 can be used instead of the GFC for a
NDIR analyzer to monitor gas emission from any combustion process for which the main components
include NOx , CO2 , CO, SO2 , O2 and H2 O. In terms of BPF_2, since its full bandwidth was still large,
a narrower bandwidth of BPF_2 should be produced, and its performance should be investigated
with interference gases in the further work. Although the price of the filter was high, its price will
Atmosphere 2018, 9, 495 13 of 14

decrease in the end due to the increase of demand and the improvement of technology. The effect of
temperature and new reference method should be investigated in future work. The investigation of
BPFs for other compounds which typically require the use of a GFC for the NDIR analyzer would be a
new direction in this NDIR field.

5. Patents
Korean patents 10-0897279 and 10-2018-0126957 resulted from the work reported in
this manuscript.

Author Contributions: J.-C.K. and T.-V.D. conceptualized the work. Data curation was performed by J.-W.A. and
I.-Y.C., The methodology was created by T.-V.D., T.-V.D. wrote the original draft, and review and editing was done
by J.-C.K.
Funding: This research was supported by the R&D Center for Green Patrol Technologies through the R&D for
Global Top Environmental Technologies funded by the Ministry of Environment (MOE), Republic of Korea. This
work is supported in part by the MOE under the Graduate School Specialized in Climate Change program and
the Pride Leading Group 2015 of Konkuk University.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Pohanish, R.P. Sittig’s Hanbook of Toxic and Hazardous Chemicals and Carcinogens, 6th ed.; Elsevier Inc.: Oxford,
UK, 2012; ISBN 978-1-4377-7869-4.
2. Novelli, P.C. CO in the atmosphere: Measurement techniques and related issues. Chemosphere 1999, 1, 115–126.
[CrossRef]
3. Tanimoto, H.; Sawa, Y.; Matsueda, H.; Yonemura, S.; Wada, A.; Mukai, H.; Wang, T.; Poon, S.; Wong, A.;
Lee, G.; et al. Evaluation of standards and methods for continuous measurements of carbon monoxide at
ground-based sites in Asia. Pap. Meteorol. Geophys. 2007, 58, 85–93. [CrossRef]
4. Yang, J.; Zhou, J.; Lv, Z.; Wei, W.; Song, H. A Real-Time Monitoring System of Industry Carbon Monoxide
Based on Wireless Sensor Networks. Sensors 2015, 15, 29535–29546. [CrossRef]
5. Chen, T.; Su, G.; Yuan, H. In situ gas filter correlation: Photoacoustic CO detection method for fire warning.
Sens. Actuators B Chem. 2005. [CrossRef]
6. Fonollosa, J.; Solórzano, A.; Jiménez-Soto, J.M.; Oller-Moreno, S.; Marco, S. Gas Sensor Array for Reliable
Fire Detection. Procedia Eng. 2016, 168, 444–447. [CrossRef]
7. Worthington, B. 60 years of continuous improvement in NDIR gas analyzers. In Proceedings of the
50th Annual ISA Analysis Division Symposium—50 Years of Analytical Solutions, Houston, TX, USA,
10–14 April 2005; pp. 95–107.
8. ISO 12039:2001. Stationary Source Emissions—Determination of Carbon Monoxide, Carbon Dioxide and
Oxygen—Performance Characteristics and Calibration of Automated Measuring Systems; ISO: Geneva, Switzerland,
2001.
9. Lee, R.; Kester, W. Complete Gas Sensor Circuit Using Nondispersive Infrared (NDIR). Analog Dialogue 2016,
50, 1–9.
10. Snakenborg, D.; Mogensen, K.B. Optimization of signal-to-noise ration in absorbance detection by integration
of microoptical components. In Proceedings of the Seventh International Conference on Miniaturized
Chemical and Biochemical Analysis Systems, Squaw Valley, CA, USA, 5–9 October 2003.
11. Harvard-Smithsonian Center for Astrophysics. High-Resolution Transmission Molecular Absorption Database;
Harvard-Smithsonian Center for Astrophysics: Cambridge, MA, USA, 2018.
12. Dinh, T.-V.; Choi, I.-Y.; Son, Y.-S.; Kim, J.-C. A review on non-dispersive infrared gas sensors: Improvement
of sensor detection limit and interference correction. Sens. Actuators B Chem. 2016, 231, 529–538. [CrossRef]
13. Arévalo-Martínez, D.L.; Beyer, M.; Krumbholz, M.; Piller, I.; Kock, A.; Steinhoff, T.; Körtzinger, A.;
Bange, H.W. A new method for continuous measurements of oceanic and atmospheric N2 O, CO and
CO2 : Performance of off-axis integrated cavity output spectroscopy (OA-ICOS) coupled to non-dispersive
infrared detection (NDIR). Ocean Sci. 2013, 9, 1071–1087. [CrossRef]
Atmosphere 2018, 9, 495 14 of 14

14. Sun, Y.W.; Liu, C.; Chan, K.L.; Xie, P.H.; Liu, W.Q.; Zeng, Y.; Wang, S.M.; Huang, S.H.; Chen, J.; Wang, Y.P.; et al.
Stack emission monitoring using non-dispersive infrared spectroscopy with an optimized nonlinear
absorption cross interference correction algorithm. Atmos. Meas. Tech. 2013, 6, 1993–2005. [CrossRef]
15. Thurmond, K.; Loparo, Z.; Partridge, W.; Vasu, S.S. A Light-Emitting Diode-(LED-) Based Absorption Sensor
for Simultaneous Detection of Carbon Monoxide and Carbon Dioxide. Appl. Spectrosc. 2016, 70, 962–971.
[CrossRef] [PubMed]
16. Camargo, E.G.; Tokuo, S.; Goto, H.; Kuze, N. Nondispersive Infrared Gas Sensor Using InSb-Based
Photovoltaic-Type Infrared Sensor. Sens. Mater. 2014, 26, 253–262.
17. Tan, Q.; Tang, L.; Yang, M.; Xue, C.; Zhang, W.; Liu, J.; Xiong, J. Three-gas detection system with IR optical
sensor based on NDIR technology. Opt. Lasers Eng. 2015, 74, 103–108. [CrossRef]
18. Dinh, T.-V.; Ahn, J.-W.; Choi, I.-Y.; Song, K.-Y.; Chung, C.-H.; Kim, J.-C. Limitations of gas filter correlation:
A case study on carbon monoxide non-dispersive infrared analyzer. Sens. Actuators B Chem. 2017,
243, 684–689. [CrossRef]
19. ISO 6142-1:2015. Gas Analysis—Preparation of Calibration Gas Mixtures—Part 1: Gravimetric Method for Class I
Mixtures; ISO: Geneva, Switzerland, 2015.
20. McLarnon, C.R.; Steen, D. Combined SO2 , NOx, PM, and Hg removal from coal fired boilers. In Proceedings
of the Joint EPRI-DOE-EPA Combined Utility Air Pollutant Control Symposium, The MEGA Symposium,
EPRI-DOE-EPA, Washington, DC, USA, 19–22 May 2003.
21. Granite, E.J.; King, W.P.; Stanko, D.C.; Pennline, H.W. Implications of mercury interactions with band-gap
semiconductor oxides. Main Group Chem. 2008, 7, 227–237. [CrossRef]
22. Zhao, Y.; Pan, Y.; Rutherford, J.; Mitloehner, F.M. Estimation of the interference in Multi-Gas measurements
using infrared photoacoustic analyzers. Atmosphere 2012, 3, 246–265. [CrossRef]
23. Jones, R.C. Performance of Detectors for Visible and Infrared Radiation. In Advances in Electronics and Electron
Physics; Marton, L., Ed.; Academic Press: Cambridge, MA, USA, 1953; Volume 5, pp. 1–96, ISBN 0065-2539.
24. Douglas, A.; Skoog, F.; James Holler, S.R.C. Principle of Instrumental Analysis; Cengage: Boston, MA, USA,
2007; ISBN 978-8-57-811079-6.
25. Gonz´alez, F.J. Noise measurements on optical detectors. Rev. Mex. Fis. 2006, 52, 550–554.
26. Korea MOE. National Emission Standard for Coal-Fired Power Plant; Korea MOE: Sejong, Korea, 2017.
27. MacLeod, H.A. Thin-Film Optical Filter, 4th ed.; CRC Press Taylor & Francis Group: Boca Raton, FL, USA,
2010; ISBN 978-1-42-007302-7.
28. Wong, J.Y.; Schell, M. Zero drift NDIR gas sensors. Sens. Rev. 2011, 31, 70–77. [CrossRef]

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).