MINIMUM DETECTION LIMITS (MDL)
One can readily enough detect any spectral line or band that is approximately as high in the spectrum as the noise level. The RIAS technique used with QASoft can actually detect lines or bands that are below the noise level. For making a conservative tabulation, however, we assume that bands at their detection limit are as high as the noise. Detectability is increased in direct response to reduction of the noise. The noise level is a function of the source intensity, the number of scans, the spectral resolution, the types of optical filters used, the stability of the optical components, and especially the quality of the detector itself. 20 years ago, in an FT-IR experiment at a folded path of several hundred meters, a signal-to-noise ratio of 2,000 was obtained. (Hanst, et al., 1973). With present day equipment and a 100 meter White cell a signal-to-noise ratio of 10,000 is readily obtained.
There are two ways to lower the detection limit for a given compound: (1) lower the noise level in the spectrum, and (2) increase the size of the absorption features.
The noise level is made lower by choosing a low noise detector and by bringing to that detector a maximum number of photons, while avoiding detector saturation. The best available detectors today are the photoconductors that operate at liquid nitrogen temperature, like mercury-cadmium-telluride. These detectors usually produce a noise level in the spectrum that is about 100 times lower than the noise level produced by detectors that operate at room temperature. The rate at which photons are delivered to the detector is kept high by having a bright source, an efficient Fourier transform spectrometer and an optical system that conserves the energy. In designing systems for air analysis, this energy conservation factor is sometimes not given enough emphasis.
An additional way to lower the noise level in the spectrum is to use longer measurement times and co-add more interferometer scans. It is difficult to make up for lost photons in this way, however, because the lowering of the noise level is not proportional to the extension of the time. The lowering is only proportional to the square root of the time extension. If for some reason the rate of delivery of photons to the detector is reduced by a factor of 10, the noise level in the spectrum will go up by a factor of 10. If one sought to restore the lower noise level by extending the measurement time, a 100-fold extension would be required, changing a 10 minute measuring time to 17 hours.
For maximum spectral detail and maximum line heights the spectrometer should fully resolve the spectral lines. When working at one atmosphere total pressure a resolving power of 0.125 cm-1 is a high as one need go because pressure-broadened line widths are about 0.2 cm-1 (full width, half way up the line). For most instruments, however, going to higher resolution increases the noise in the spectrum at a faster rate than it increases the height of the lines. A compromise in the matter of resolution is required. A value of 0.5 cm-1 seems a good choice.
The main way to increase the size of the spectral features is to lengthen the optical path. The principal limitation on path lengthening is that it should not waste the available photons. There is also an additional problem that when the path gets longer than about 100 meters, absorption by water vapor and carbon dioxide precludes the use of three important bands of air pollutants: the NO2 band near 1600 cm-1, the SO2 band near 1360 cm-1, and the benzene band at 674 cm-1. The importance of these three bands makes it worthwhile to keep the total path at 100 meters or less.
When the optical path is at 100 meters, the resolution is at 0.5 cm -1, and the spectrum noise level is at 10-4 absorbance units, nearly every important gas can be measured at parts-per-billion levels. Minimum detection limits under these conditions were calculated from our atlas of spectra, and are listed in Table 1. An important condition for validity of Table 1 is that the water and carbon dioxide lines must be subtracted from the spectrum without introducing noise. In the past this condition was almost impossible to meet, but now it can be met. The modern software has made it possible to match the line positions and line widths of two spectra with a high degree of precision. The software also has made it possible to create reference spectra that show deviations from the absorbance law that are the same as the deviations in the sample spectrum.
The table of minimum detection limits (MDL's) shows that under the conditions discussed, most compounds are detectable down to the parts-per billion level. Some compounds have MDL'S much below one part-per-billion. Note the extemely low numbers for CF4 and SF6.
Some of the strongest bands of the compounds have not been used in preparing the table because of interferences from water or carbon dioxide. All the carbonyl bands in the region 1800 to 1700 cm-1 have been left out, for example, because of water vapor interference.
In the table a narrow and intense spectral feature has been designated "spike". These features occur in the majority of the spectra and they are the main reason for the specificity and sensitivity of the infrared method of measurement. It is the families of spikes and families of lines that give us confidence that a spectrum can be identified and measured when the maximum absorbance is equal to the peak-to-peak noise level.
It is found that the MDL's in the table are similar to those published 20 years ago by Hanst, et al. (1973). The eason for this is that the calculations then and now were based on similar experimental results obtained with an FT-IR spectrometer system, mercury-cadmium-telluride detector and multiple-pass cell. The main difference between then and now is that formerly we could not properly subtract the interferences but now we can.
The MDL's in our table are some 10 to 100 times lower than the MDL's recently published by Grant, Kagann and McClenny (1992) for an FT-IR system working over an open-air doubled long path totaling 400 meters. The main source of the differences is the two different spectrum noise levels--1.3 x 10-3 absorbance units for Grant, et al., and 1 x 10-4 absorbance units for us. The reason for the two different noise estimates it that they were derived from experience with two radically different optical systems. The simple fact is that our measurement system with the multiple-pass cell is a high throughput (low noise) system while their measurement system with the long path retro-reflector technique is a low throughput (high noise) system.
An additional advantage for us it that we calculate for only a 100 meter path in air. This lets us consider some important absorption bands that are not accessed in a 400 meter path. This includes strong bands of NO2, SO2 and benzene. Finally, we have said that we can recognize a line or band when its height is equal to the peak-to-peak noise level in the spectrum, while they assumed their detectable band height to be 3.3 times greater than the noise.
Measurements made with the RIAS technique have confirmed many of the detection limits in the table and have even shown that some can be surpassed. These measurements were done by combining the database spectra, appropriately reduced, with noise spectra and then running RIAS.
TABLE 1. MINIMUM DETECTION LIMITS WHEN MEASURING IMPURITIES IN AIR, for compounds in the Infrared Analysis, Inc. atlas of digitized quantitative reference spectra.
M.D.L. is the Minimum Detection Limit in parts-per-billion (10-9 atmospheres), when resolution is 0.5 cm-1, optical path is 100 meters and spectrum noise level is at 10-4 absorbance units (logarithm, base 10).
Compound (cm -1) (PPB) Remarks
Acetaldehyde 2900-2600 6.0 Broad band with some fine structure;
structure; open region.
Acetic acid 642 1.0 Spike.
Acetone 1218 4.0 Broad band with some structure.
Aceto nitrile 1080-1010 30.0 Array of lines.
Acetyl chloride 1130-1090 1.0 Band with spikes.
Acetylene 739 0.15 Spike.
Acrolein 959 5.0 Spike.
Acrylic acid 640 0.5 Spike.
Acrylo nitrile 954 3.0 Spike.
Ammonia 970-920 1.0 Two bundles of lines.
Analine 782 10.0 Spike.
Arsine 2126 2.0 Spike.
Benzene 674 0.6 Sharp line; CO 2 must
Boron trichloride 970-940 0.2 Strong band.
Bromoform 1150 2.0 Smooth band.
Bromo methane 1030-900 15.0 Array of lines.
1,3 Butadiene 908 1.0 Spike.
Butane 2968 1.5 C-H band; has a spike.
2-Butanone 1171 3.0 Band about 40 cm -1 wide.
Carbon dioxide 2363 0.4 M.D.L. is 0.4 PPB if no
other CO 2 is present.
In air, the minimum
detectable change in CO 2
would be about 50 PPB.
Carbon monoxide 2200-2100 2.0 Array of lines.
Carbon tetra-chloride 800-790 0.2 Very strong band.
Carbonyl sulfide 2080-2030 0.3 Broad band.
Chloro benzene 741 0.8 Spike.
Chloro difluoro 809 0.5 Spike.
Chloro ethane 995-950 9.0 Band with P-Q-R structure.
Chloroform 780-760 0.5 Wide band, but strong.
Chloro methane 732 5.0 Spike.
Chloro trifluoro 1107 0.7 Spike.
Crotonaldehyde 2900-2600 10.0 Broad bands in unique position.
Cyclo hexane 2934 0.5 Spike in C-H band.
Cyclo hexene 2936 1.0 Spike in C-H band; also
spikes in region 1160 to 650.
Cyclo pentene 698 1.2 Spike.
Cyclo propane 3101 1.0 Sharp spike in C-H band.
1,2 Dibromo ethane 1195-1180 2.5 Band with characteristic shape.
m-Dichloro benzene 800-770 1.5 Band with spike.
o-Dichloro benzene 749 1.5 Spike.
p-Dichloro benzene 820 1.5 Spike.
Dichloro difluoro 1161 0.2 Strong spike.
1,1 Dichloro ethane 1060 3.0 Spike.
1,2 Dichloro ethane 740-710 2.0 Broad band.
cis-1,2 Dichloro ethylene 696 0.6 Spike.
Dichloro methane 770-740 1.3 Band with some structure.
1,2 Dichloro 1250-800 0.7 Group of strong bands.
Diethyl amine 1130-1160 3.0 Broad band.
Diethyl ether 1150-1120 1.0 Smooth band.
Di-isopropyl ether 1400-1000 2.0 Group of bands.
Dimethyl ether 1179 2.5 Band with P-Q-R structure.
1,1 Dimethyl hydrazine 2814, 2775 4.0 Spikes.
Dimethyl sulfide 2980-2900 8.0 Structure in C-H band.
Dinitrogen 1240, 740 1.0 Two strong bands.
Ethane 3050-2880 1.5 Structure in C-H band.
Ethanol 1090-1020 4.0 Broad band with some structure.
Ethyl acetate 1270-1225 0.5 Broad band.
Ethyl acrylate 1210-1180 0.5 Broad band.
Ethyl benzene 698 4.0 Spike.
Ethyl formate 1210-1150 0.5 Broad band with some structure
Ethylene 950 1.0 Spike.
Ethylene oxide 3066 0.5 Spike.
Ethyl vinyl ether 1230-1200 2.0 Wide band with some structure.
Fluoro benzene 754 0.5 Spike.
Formaldehyde 2711, 2779 3.0 Two sharp spikes.
Formic acid 1105 0.8 Spike.
Furan 744 0.2 Spike.
n-Hexane 2970 1.0 C-H band not specific; total
hydrocarbon is often quoted
in hexane equivalents.
Hydrazine 1000-890 4.0 Band with complex fine structure
Hydrogen bromide 2700-2400 8.0 Array of lines.
Hydrogen chloride 3050-2700 1.5 Array of lines.
Hydrogen cyanide 712 0.4 Q branch.
Hydrogen fluoride 4200-3700 1.0 Array of lines.
Hydrogen sulfide 1300-1200 400 Many lines, but extremely weak.
Isobutane 2967 1.0 Spike in C-H band.
Isobutanol 1060-1030 3.0 Broad band.
Isobutylene 890 1.5 Spike.
Isoprene 900 1.5 Two distinctive spikes near 900.
Isopropanol 1420-900 7.0 Group of bands.
Mesitylene 836 3.0 Spike.
Methane 3018 2.0 Spike and fine structure in C-H band.
Methanol 1033 1.5 Spike.
Methyl acetate 1265-1230 1.0 Broad band with some structure.
Methyl acrylate 1220-1180 1.0 Broad band with some structure.
Methyl amine 820-720 2.5 Band with structure.
2-Methyl-2-butene 2969 5.0 C-H band not specific; use spike
at 890 for identification.
3-Methyl-1-butene 2970 2.0 C-H band not specific; use spike
at 913 for identification.
Methyl formate 1210 1.5 Spike.
Methyl methacrylate 1215-1150 1.2 Two-hump broad band.
Methyl nitrite 860-790 2.0 Band with structure.
2-Methyl pentane 2967 1.0 C-H band not specific.
3-Methyl pentane 2967 1.0 C-H band not specific.
2-Methyl-1-pentene 2970 2.5 C-H band not specific; use feature
at 892 for identification.
2-Methyl-2-pentene 2973 2.0 C-H band not specific.
4-Methyl-2-pentene 2967 1.5 C-H band not specific; use feature
at 721 for identification.
Methyl vinyl ether 1400-800 1.0 Five different spikes.
Methyl vinyl ketone 744 8.0 Spike.
Nitric acid 896, 879 2.0 Spikes.
Nitric oxide 1920-1870 4.0 Array of lines.
Nitro benzene 701 1.0 Spike.
Nitro ethane 900-850 10.0 Band with P-Q-R shape.
Nitro methane 656 3.0 Spike.
Nitrogen dioxide 1600 2.0 Several bundles of lines.
Nitrous acid 852, 790 0.5 Q branches.
Nitrous oxide 2210 1.0 Array of lines.
Ozone 1054 2.0 Band with structure.
n-Pentane 2966 1.5 Difficult to distinguish
from other large alkanes.
1-Pentene 2968 2.0 C-H band strongest but
not specific; use feature
at 915 for identification
2-Pentene 2970 2.0 C-H band strongest but
not specific; use feature
at 961 for identification
Phosgene 864-830 0.3 Strong band with structure.
Phosphine 992 10.0 Spike.
Phosphorus 515-495 3.5 Smooth band.
Propane 2968 1.0 Spike in C-H band.
Propionaldehyde 2850-2650 8.0 Broad bands in the open.
Propionic acid 1144 2.0 Band 30 cm -1 wide.
Propylene 913 2.0 Spike.
Propylene oxide 3050 3.0 Spike.
Styrene 695 2.0 Spike.
Sulfur dioxide 1361 2.0 Spike; water must be
Sulfur hexafluoride 950-940 0.04 Extremely strong band.
1,1,1,2 Tetrachloro ethane 980-700 1.3 Group of bands.
Tetrachloro ethylene 930-900 0.7 Smooth band.
Tetrafluoro methane 1283 0.03 Extremely strong spike.
Toluene 729, 694 2.0 Spikes.
1,1,1 Trichloro ethane 735-715 0.5 Smooth band.
1,1,2 Trichloro ethane 760-720 0.8 Broad band.
Trichloro ethylene 783 1.4 Spike.
Trichloro fluoro methane 855-840 0.2 Very strong band.
Trichlorotrifluoro ethane 1240-800 0.6 Group of strong bands.
Vinyl acetate 1240-1210 0.7 Broad band.
Vinyl chloride 942, 896 2.0 Two spikes.
Vinylidene chloride 869 1.0 Spike.
Water 1700-1400 5.0 M.D.L. is 5 PPB if no other
water is present. In humid air,
the minimum detectable change in
water content would be 1000 PPB.
m-Xylene 768 2.0 Spike.
o-Xylene 741 1.0 Spike.
p-Xylene 795 2.0 Spike.