Journal of Quantitative Spectroscopy & Radiative Transfer 112 (2011) 1420–1428

Contents lists available at ScienceDirect

Journal of Quantitative Spectroscopy &

Radiative Transfer
journal homepage: www.elsevier.com/locate/jqsrt

60-GHz oxygen band: Precise experimental profiles and extended

absorption modeling in a wide temperature range
D.S. Makarov a,, M.Yu. Tretyakov a, P.W. Rosenkranz b
a
IAP RAS, Uljanova str. 46, 603950 Nijniy Novgorod, Russia
b
Massachusetts Institute of Technology, Cambridge, MA 02139, USA

a r t i c l e in f o abstract

Article history: The 60-GHz band of atmospheric oxygen was studied in the temperature range of  281
Received 13 January 2011 to þ 60 1C at atmospheric pressure by means of a resonator spectrometer with
Received in revised form absorption-variation sensitivity of 0.002 dB/km. The experimental data obtained have
24 February 2011
sufficient signal-to-noise ratio to take second-order mixing into account, increasing the
Accepted 26 February 2011
accuracy of the millimeter-wave propagation model (MPM). A refined set of mixing
Available online 3 March 2011
coefficients for the model is derived from the new data and presented. The fidelity of
Keywords: the new model to the spectrometer data is generally better than 2% between 54 and
Microwave spectroscopy 65 GHz.
Collisional coupling
& 2011 Elsevier Ltd. All rights reserved.
Absorption model
Atmosphere
Molecular oxygen

1. Introduction reasons only odd rotational quantum numbers N are allowed

for the principal isotopologue, 16O2. Since for the oxygen
Accurate information on the absorption of radiation in molecule spin S¼1, possible values for the total angular
the atmosphere is in demand for atmospheric applica- momentum quantum number J are Nþ1, N, N1. The
tions, wireless communications, remote sensing, etc. The selection rule DJ ¼ 7 1 leaves only two series of lines:
main atmospheric absorbers of microwave radiation are J 0 ’J ¼ N’N þ 1 (usually denoted as Nþ lines) and
water vapor and molecular oxygen. In the millimeter- J 0 ’J ¼ N’N1 (Nlines). Also, the diagonal matrix ele-
wave range, widely used for the aforementioned applica- ments of the magnetic dipole are nonzero, which leads to a
tions, molecular oxygen has a strong band between non-resonant absorption due to collisional reorientation
50 and 70 GHz and a single spectral line at 118.75 GHz. (classical approach to this phenomenon is given in [1]).
The absorption band near 60 GHz and the line at Other transitions, with nonzero DN, appear at submillimeter
118.75 GHz are formed by magnetic-dipole transitions wavelengths. At one-atmosphere pressure all Nþ and N
between fine-structure energy levels within the 3 S elec- lines, except the isolated 1 line at 118.75 GHz, blend
tronic ground state of the oxygen molecule. Due to together and form a wide band ranging from 50 to 70 GHz.
electronic spin, each rotational level of the oxygen mole- Because of the mixing effect, the resulting intensity differs
cule is split into a spin triplet corresponding to the three from a simple sum of Van Vleck–Weisskopf profiles [1] of
possible space orientations of the total electronic spin isolated lines.
vector S. If N is rotational angular momentum, then the Theoretical and experimental studies of the band have
resultant angular momentum is J ¼ N þ S. Due to symmetry a long history, beginning in the first half of the twentieth
century [2,3]. General theoretical expressions for the
shape of the bands formed by overlapping lines were
 Corresponding author. developed in [4,5]. For the 60-GHz molecular-oxygen
E-mail address: [email protected] (D.S. Makarov). band, a model considering the mixing effect to first-order

0022-4073/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jqsrt.2011.02.018
 D.S. Makarov et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 112 (2011) 1420–1428 1421

in pressure was suggested in [6], and, later, a more measured in the experiment. It is important to decrease
thorough description with extension to second-order resonance-curve width because a wider response curve
was given in [7]. implies greater measurement errors.
Extensive laboratory studies resulted in the develop- With resonator length of 39.4 cm, intermode dist-
ment of an accurate millimeter-wave propagation model ance calculated as c/2L is equal to 380.71 MHz. That allows
(MPM) [8] and its further improvement in [9]. The most about 100 points (eigen-frequencies of the resonator) for
recent update includes line central frequencies and half- absorption measurements in the frequency range between
widths measured in [10] and line amplitudes from the 45 and 85 GHz. While running the measurements, gas
HITRAN database [11]. MPM uses a first-order mixing temperature and humidity, as well as the resonator mirror
model for millimeter-wave oxygen spectra, where mixing temperatures, were recorded to take into account their
coefficients are derived by the method given in [12]. fluctuations in experimental data processing as discussed
The present study continues our series of papers (see in [10].
[13] and references therein) related to the microwave The band absorption profile was recorded at var-
spectrum of molecular oxygen in the atmosphere. In [10] ious temperature values in the range of 245–334 K. Gas
it was shown that the experimental record of the band losses, total resonator losses and baseline signal,
systematically deviates from the first-order MPM predic- recorded at 303 K, are shown in Fig. 1 as a typical
tion. The objective of this study is to develop a second- example. Several absorption profiles obtained at var-
order 60-GHz absorption-band model based on precise ious temperatures are shown in Fig. 2, demonstrating
experimental profiles, to increase modeling accuracy for variation of the band profile with temperature within
atmospheric applications. the studied range.
We decided to use outdoor atmospheric air as a sample
2. Experiment for the experiment. This air contains water vapor which
gives additional absorption in the millimeter- and submilli-
The band absorption profile was recorded by means meter-wave range. Also, at low temperatures moist air
of a resonator spectrometer with fast digital freque- causes hoar-frost on the resonator parts, so the air should
ncy scanning [14]. To allow measurements over a wide be dehydrated before being directed into the climatic
temperature range, the resonator was setup in the climate chamber. For dehydration, the outdoor air, pumped by
chamber equipped with cooling and heating elements, an air compressor, was led through a pipe immersed in
described in [13]. As in [13,14], gas temperature was
measured by a set of temperature sensors with the
measurement accuracy 0.1 1C; gas pressure was measured
by means of a calibrated pressure meter (600–800 mmHg
range), providing measurement accuracy 70.5 mmHg.
The absorption coefficient of the gas sample in the
chamber is obtained at the frequency points correspond-
ing to eigen-frequencies of the resonator, and its value is
calculated from the directly measured Fabry–Perot reso-
nance width:
2p
aðfk Þ ¼ ðDf Df0 Þ ð1Þ
c
where a is the absorption coefficient, fk is the k-th
resonator eigen-frequency, Df and Df0 are Fabry–Perot Fig. 1. Loss profiles at 303 K: total losses (dash line), baseline (dotted
resonance widths measured in air and in pure nitrogen line) and gas losses (solid line).
(having negligibly small absorption in the millimeter-
wave range), respectively.
In comparison to experiments in [13], the resonator
length L was decreased to 39.4 cm to avoid noticeable
increase of diffraction losses at lower frequencies [15]. For
resonator length of 39.4 cm and mirror diameter 14 cm,
one-pass losses (the total relative losses of radiat
ion energy during one traversal of the resonator) due to
diffraction at 45 GHz (the lowest frequency where
absorption was measured) are not greater than 10  5.
Since both baseline losses in the chamber filled by pure
nitrogen and the total losses in the chamber filled with air
are measured, diffraction losses are not considered in
experimental data processing because they are canceled
by subtraction of the baseline from the total resonator
losses [14]. However, lower diffraction losses lead to a
narrower resonance–response curve, which is directly Fig. 2. Absorption profiles in air at several temperatures.
1422 D.S. Makarov et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 112 (2011) 1420–1428

expression [7]:

X Dni  ð1 þ gi P2 Þ þ Yi  ðf ni dni P 2 Þ

aðf Þ ¼ Ci f 2
i Dn2i þðf ni dni P2 Þ2
!
Dni  ð1 þ gi P2 ÞYi  ðf þ ni þ dni P2 Þ
þ ð2Þ
Dn2i þ ðf þ ni þ dni P 2 Þ2
where the sum is over all lines in the band, i is the individual
line index, f is the frequency of measurement, Ci is the line
amplitude, Dni is the line half-width at half maximum, ni is
the line central frequency, and Yi is the first-order mixing
Fig. 3. Diagram of the cold trap. parameter; Dni and Yi are proportional to pressure P:
Dni ¼ gi  P, Yi ¼ yi  P, where gi is the broadening coefficient
and yi is the first-order mixing coefficient. The gi and dni are
a Dewar with liquid nitrogen, followed by a dense air filter
second-order mixing coefficients accounting for mixing of
to remove frosted water from the air flow (see Fig. 3).
line intensities and line central frequencies, respectively.
Oxygen’s condensation temperature is slightly higher
The non-resonant component of absorption is included in
than liquid-nitrogen temperature, and if cooling of the air
the summation as i¼0, with n0 ¼ 0, Y0 ¼0, g0 ¼0 and
flow is too strong (which happens if the level of liquid
dn0 ¼ 0. Eq. (2) assumes that no first-order line shift is
nitrogen in the Dewar is high), the oxygen concentration
present, which has been confirmed by measurements of
value may decrease due to condensation. As the nitrogen
oxygen at low pressures [10]. (Eq. (2) has been corrected for
evaporates and the level gets low, the air pipe becomes
a sign error in Eq. (17) of [7]; it is evident from Eqs. (10) and
warmer and condensed oxygen also evaporates, which
(A17) of [7] that dni should bear the same sign as ni in Eq.
leads to an increase of oxygen concentration which may
(2).) The coefficients yi, gi and dni are all related, by Eqs. (15),
even exceed the standard value of 21%. To prevent large
(16) and (19) in [7], to combinations of elements of a line-
concentration deviations, the level of liquid nitrogen in
space collisional-relaxation matrix w. Those equations are
the Dewar was kept close to an experimentally deter-
based on the binary-collision approximation, in which w is
mined optimal value during the band-profile recording.
multiplied by the gas pressure; the terms of first- and
Despite the measures taken, oxygen concentration in the
second-order in pressure in Eq. (2) derive from a perturba-
air used for the experiment still varied in the range of
tion expansion of the way in which the lines combine
20–21% from one experimental profile record to another.
through mixing. The temperature dependence of the coeffi-
The reason might be variation of the oxygen concentra-
cients, on the other hand, is not easily obtainable from that
tion in the atmosphere. Weak concentration fluctuations
analysis, so in this work we use an empirical model to fit the
within each record were also possible because of small
experimentally determined temperature dependence of the
variations of the air flow and the liquid-nitrogen level.
mixing coefficients.
Deviation of concentration within the range of 1% gives a
In [10] it was shown that analysis of the band shape
noticeable deviation of the profile magnitude, but negli-
taking only first-order mixing into account (i.e., setting
gible deviation of the profile shape. The average oxygen
gi ¼0 and dni ¼ 0) gives a residual which is regular and
concentration value for each profile record was deter-
looks like a second-order mixing contribution. The
mined by fitting of the MPM profile magnitude to the
noise level of absorption measurements in [10] was
experimental one. Concentration values thus obtained
0.05 dB/km, and this value was insufficient for taking
were taken into account in the mixing-coefficient calcula-
the next order of mixing into account, while the current
tion described below.
noise level of 0.01 dB/km allows it. Fig. 4 shows the
The spectrometer had an absorption-variation sensi-
calculated contribution due to the second-order mixing
tivity of 0.002 dB/km; but for various reasons discussed in
effect, using coefficients from [7] (dotted line) together
[14], including subtraction of a baseline, small tempera-
with two residuals between experimental data at room
ture gradients within the spectrometer, etc., the noise
temperature (300 K) and MPM considering only the
level of the experimental profiles was  0:01 dB=km for
the lowest absorption values (it increased near the band
center, as discussed below). Still, this noise level was
several times smaller than in [10]. This decrease of the
noise level was achieved by using a reference microwave
synthesizer with lower phase noise in comparison to the
one used in [10], and by better stability of gas tempera-
ture and humidity in the chamber during the experiment.

3. Data treatment
Fig. 4. Experiment-minus-model residual error using first-order mixing
Considering line mixing to second-order in pressure, models from [10] (gray line) and this work at 300 K (solid black line).
the absorption profile is described by the following Dotted line: theoretical contribution of second-order mixing.
 D.S. Makarov et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 112 (2011) 1420–1428 1423

first-order mixing effect. The first residual (gray) is In each iteration of the estimation algorithm, r is
derived from earlier experimental data [10], the second estimated by using the gi and dni values derived in the
one (black) is derived from the latest data with lower previous iteration to calculate abias
k (on the first iteration,
noise level. Because the introduction of second-order the second-order coefficients are set to zero). The vector y
mixing coefficients makes the estimation problem non- of first-order coefficients is given by y ¼ K  r þb. Then
linear, we used an iterative algorithm to derive both new second-order mixing coefficients are calculated using
first- and second-order mixing coefficients from the the r obtained in the current iteration, considering intra-
new experimental data. branch mixing only. These iterations were repeated until
If we consider the n strongest lines (current model the calculated mixing coefficients converged to consistent
uses n ¼19) in each branch of the oxygen fine-structure values from iteration to iteration.
spectrum, then the dimensions of the line-space matrix w Note that the previously measured [10] broadenings gi ,
are 6n  6n, with two series of lines (N þ and N ) at the as well as the coefficients yi, gi and dni calculated in the
positive frequencies ni , two series at ni , and two series at present work, are pressure-independent by definition and
n0 ¼ 0 [16]. Negative-frequency lines correspond to the Eq. (2) allows one to process the band profile recorded at
second term within the brackets of Eq. (2). The overlap of any particular pressure. The pressure value was measured
the positive-frequency lines with the zero- and negative- and used as a parameter in the model profile (2), which
frequencies is minimal and therefore can be modeled well makes possible precise quantitative characterization of
by a first-order treatment, which results in a small, pressure-dependent effects
though not negligible, bias value bi to be included in yi Rather than using a fixed value of b, we found that
[6,12]. The bi have values of approximately  0.014 bar  1 convergence improved if b=m, where m is the number of
for an effective (i.e., post-mixing) non-resonant broad- measurement points, was set to a relatively high value
ening coefficient g0 ¼ 0:56 GHz=bar (bar¼ 105 Pa). near unity for the first seven iterations and then gradually
Now we define an intra-branch collisional-relaxation lowered to an asymptotic value, e.g.:
matrix, which is an n  n submatrix of w, and assume that (
1:05  m, Z r7
it applies identically to all six sets of lines in w. As b¼ ð7Þ
discussed in [12], to calculate mixing coefficients we need ð0:05 þ 27Z Þ  m, 7 r Z r 20
to specify only half of this submatrix, as the other half is
where Z is the iteration number.
related by the constraint of detailed balancing. The
The Twomey–Tikhonov method provides a solution
diagonal elements are equal to the line-broadening coef-
that fits the measurements within a certain residual-
ficients, which were measured in [10]. Conceptually, we
error level, but it is not a unique solution. With
can arrange the elements of the upper-right off-diagonal
b ¼ 0:05m, the b  I term in the denominator of Eq. (4)
triangle of the intra-branch submatrix as a vector r. This
dominates all but about four or five of the eigenvalues
vector is estimated by the Twomey–Tikhonov method,
of the denominator matrix. Hence there are effectively
which implies minimization of the quantity
X about four to five degrees of freedom in r, although it
e2k þ qb ð3Þ contains n(n-1)/2 ¼171 elements for n ¼19. It is there-
k fore important to distinguish between the uncertainty
where ek is the difference between measured and calcu- of the calculated absorption, and uncertainties in the
lated absorption at frequency fk, q ¼ r>  H  r is a quadratic values of the model’s coefficients. The coefficients are
measure of structure in the collisional-relaxation matrix, adjusted to fit measurements, and underestimation of
and b is a Lagrange multiplier which can be chosen to some coefficients can be counterbalanced by excess
balance between minimal residuals and minimal relaxa- values in other coefficients when absorption is calcu-
tion-matrix structure [12]. Superscript > denotes the lated. However, we consider the set of mixing coeffi-
transpose. Very small b can reduce residuals, but the cients finally obtained here as providing the best
derived mixing coefficients then do not have a smooth fidelity of the model to the whole set of experimen-
dependence on the quantum number N, which conflicts tal profiles as well as corresponding to theoretical
with theoretical assumptions. Methods for selecting the principles.
most suitable value of b are discussed in detail in [12,10]. As in [10], an error-model function was used to pre-
The solution for r is [12] whiten the data, decreasing the weight of points in the
center of the band where measurement errors were
r ¼ H1 K> V> ðVKH1 K> V> þ b  IÞ1 A ð4Þ higher. This is discussed in the following section. We
also accounted for other isotopologues of O2 by multi-
with
plying the calculated absorption by 1.0053 [10]. Unlike
Ak ¼ ðameas
k abias
k Þ=sk ð5Þ the procedure in [10], we do not use any previous
meas
version of y as a baseline here, because gi and dni are
where a k is the absorption measured at frequency fk,
calculated from r, not from an incremental change to it.
abias
k is the absorption calculated by setting Yi ¼ bi  P in
Eq. (2), and the pre-whitening factor sk is discussed
below; matrices H and K are defined in [12] and 4. Discussion
P @aðfk Þ
Vki ¼ ð6Þ In Fig. 5, experiment-minus-model residuals for the
sk @Yi
absorption profile treated by the model including both
1424 D.S. Makarov et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 112 (2011) 1420–1428

first- and second-order mixing, in comparison to residuals calculated mixing parameters, but should result in better
for a model based on only first-order mixing [10], are general agreement between model profile and measure-
shown for several temperatures by the black and gray ments. Because measurement errors are expected to grow
lines, respectively. In this figure, residuals shown by the along with the gas absorption, we use the band absorp-
solid black lines were calculated with the pre-whitening tion profile as a base for the error model. To define
factor sk set to unity. The dotted lines will be discussed parameters of the model, three pairs of band-profile
later. Systematic discrepancies between the extended records were compared. Each pair consisted of records
model and experiment are still noticeable, especially at observed at close temperatures (252 and 253 K, 300 and
lower temperatures, where the mixing effect should be 303 K, 324 and 327 K). If we take the absolute value of
stronger. These discrepancies are similar for records differences between the residual errors of two records,
repeated at the same experimental conditions, so they the regular part of the residuals will cancel, and the
are not caused by random experimental errors. Never- frequency range where noise increases will be evident
theless, in general, the extended model gives smaller (in this case, the residuals between recorded profiles and
residuals in the temperature range considered. In parti- the absorption model of [10] were taken by reason of the
cular, residuals decreased in the frequency bands from 53 latter’s independence from the current recorded profiles).
to 56 GHz and from 63 to 66 GHz. The lower of these two In Fig. 6 the dots show absolute values of the difference
bands is of special importance because it is widely used between residuals, and the solid line is the noise-model
for atmospheric sensing (see, e.g., [17]). function:
One notices that the residuals have some features near
the center of the band which look like strong noise peaks. Eexp ðf Þ ¼ maxðs,0:015  ðaðf Þ12:0ÞÞ ð8Þ
Two reasons leading to appearance of the features are
with s ¼ 0:01 dB/km; aðf Þ in dB/km is calculated at each
possible. The first is a general increase of the measure-
temperature. At temperatures higher than 310 K, this
ment error due to the growth of the resonator response-
noise model reduces to the constant level of s. Then the
curve width with increase of the radiation losses in the
data are pre-whitened by the factor
gas. The second reason is the aforementioned possible
deviation of the oxygen concentration due to dehydration sk ¼ Eexp ðfk Þ=s ð9Þ
of the air used for the experiment, which should be more
noticeable in the range of maximum absorption. Use of the noise model leads to an increase of the
Since absorption measurement errors are greater in residual in the center of the band but at the same time
the center of the band, it is possible to decrease the decreases the residual in the ranges from 53 to 56 GHz
weight of these points in the data treatment using an and from 63 to 66 GHz, at temperatures below 0 1C, which
error model. It gives additional degree of freedom for will be useful for many applications.

Fig. 5. Experiment-minus-model residual errors at several temperatures, for second-order mixing models fitted to each temperature (black lines), the
model of [10] (gray lines), and the model using coefficients from Table 1 (dotted lines).
 D.S. Makarov et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 112 (2011) 1420–1428 1425

Fig. 7. Reduced first-order mixing coefficients vs. inverse temperature

(solid lines), calculated coefficients’ values (circles) and reduced first-
order mixing coefficients vs. inverse temperature from [10] (dashed
lines). The factor (300/T)0.8 has been removed from the coefficients.

graphical form in Fig. 8. Other coefficients necessary for

computation of absorption can be found in Table 5 of [10].
The value of x ¼0.8 was measured experimentally for
Fig. 6. Absolute value of the close temperature records’ residuals differ- broadening of the 9 þ line [8] and is still used in MPM and
ence (points) and error model (solid line).
in the current model as a uniform temperature coefficient
for all lines, i.e. gi ¼ g0i  ð300=TÞ0:8 . In [13] the value of
Measurements were carried out over a wide tempera- x ¼0.785(35) (which coincides with x ¼0.8 within experi-
ture range, so one should pay attention to temperature mental uncertainty) was measured for the 1 line. In our
dependence of the mixing coefficients. Ref. [9], says that earlier work [10] it was shown that broadening coeffi-
‘‘if the temperature variation in the Y’s contains a term cients (g) of the discussed fine-structure lines are very
that varies with (300/T)x, then for at least some of close to ones of the corresponding lines of the A-band (see
the coefficients there will also be a term varying with Fig. 5 in [10]) at room temperature. This fact suggests that
(300/T)x þ 1. This argument does not prove that two terms temperature coefficients x for corresponding lines in the
in powers of (300/T) will always be sufficient to represent fine-structure band and A-band might also have close
the temperature dependence of the Yk, but at least two are values. Experimentally measured x-coefficients for
necessary to maintain detailed balance’’. According to this A-band [18] range from 0.6 to 0.77 and they appear to
reasoning, an expression accounting for temperature depend regularly on the quantum number N, so it would
dependence of mixing coefficients was written as be natural to suppose that the 60-GHz absorption band
     may behave in the same way. We tried to use the
300 300 x
Zi ðTÞ ¼ Zi0 þZi1  1  ð10Þ temperature coefficients’ dependence on N from [18] for
T T
line-widths and first-order mixing coefficients in our
where Zi is a first- or second-order coefficient yi, gi or dni , calculations. It resulted in insignificant differences in
Zi0 is the mixing coefficient value at 300 K, and the Zi1 term residuals and slight differences (about 5%) in derived
accounts for the ð300=TÞx þ 1 dependence; x¼0.8 for yi (the mixing coefficients, in comparison to using the uniform
origin of this value for the temperature exponent is x value for all lines. Therefore, we decided to keep the
discussed below) and x¼ 1.6 for gi and dni . The reason uniform value for all lines until new experimental data on
for the difference in x values is the fact that second-order the dependence for individual lines is obtained.
mixing coefficients are proportional to sums of terms The mixing coefficients in Table 1 are the result of
involving products of two elements of the collisional- fitting the calculated coefficients for each temperature
relaxation matrix [7]; while first-order mixing coefficients point by Eq. (10), so using coefficients from the table for
are proportional to sums of terms involving individual absorption-band modeling gives residuals which differ
matrix elements. from those which were derived by treatment of each
Our algorithm code for estimation of mixing coeffi- isolated record and without pre-whitening. In Fig. 5, the
cients produces values reduced to the exponential depen- dashed black line plots residuals with the new second-
dence (300/T)x, thus they should depend on T as order model using parameters from Table 1. Use of pre-
Zi0 þZi1  ð300=T1Þ. As one can see in Fig. 7, the depen- whitening tends to reduce residuals in the band wings,
dence of these values on (300/T) can be considered but increase them in the band center (because the
as linear within the accuracy of estimation. With the weight of measured points in the center of the band
second-order mixing effect taken into account, first-order was decreased). One notices that residuals in the cen-
mixing coefficients changed only slightly. Temperature ter of the band, obtained with pre-whitening, look
dependences of y-coefficients from MPM [10] are shown systematic, but their magnitude is not greater than that
by the dashed lines in Fig. 7 for comparison. of systematic residuals given by MPM in the same
Table 1 lists the coefficients thus obtained, with pre- figure. With decreased residuals in the band wings, this
whitening applied. Data from the table are presented in means that systematic errors in reproducing the
1426 D.S. Makarov et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 112 (2011) 1420–1428

Table 1
First- and second-order mixing parameters.

N y0, bar  1 y1, bar  1 g0, bar  2 g1, bar  2 dn0 , GHz bar2 dn1 , GHz bar2

1  0.041a 0.000 0.000 0.000 0.000 0.000

1þ 0.265  0.0132  0.0835 0.00843 0.00545 0.0017
3  0.354 0.0582  0.0947 0.0234  0.0175  0.0013
3þ 0.53  0.0981  0.218 0.0596 0.0286 0.00107
5  0.547 0.114  0.16 0.0326  0.043  0.00444
5þ 0.589  0.153  0.162 0.041 0.0491 0.00493
7  0.359  0.0494 0.0197 0.0174  0.0229  0.0529
7þ 0.276 0.0476 0.132 0.039 0.0162 0.061
9  0.0942 0.186 0.128 0.262  0.000866  0.0193
9þ  0.0124  0.235 0.166 0.305  0.00609 0.0168
11  0.0647 0.611 0.0883 0.126  0.00115 0.0393
11þ  0.155  0.674 0.0713 0.0799  0.00256  0.0448
13  0.224 0.628 0.0735  0.156 0.00221 0.0247
13þ  0.299  0.662 0.0582  0.216  0.00447  0.0265
15  0.365 0.583  0.00369  0.253 0.00605 0.0102
15þ  0.426  0.598 0.00439  0.296  0.00746  0.0106
17  0.466 0.596  0.0213  0.287 0.00675 0.0104
17þ  0.515  0.6  0.0634  0.324  0.00762  0.0105
19  0.546 0.617  0.0868  0.374 0.00732 0.0163
19þ  0.587  0.615  0.105  0.394  0.00784  0.0163
21  0.608 0.515  0.114  0.393 0.00802 0.00835
21þ  0.642  0.509  0.134  0.4  0.00833  0.00815
23  0.64 0.406  0.163  0.329 0.00696 0.000922
23þ  0.668  0.399  0.174  0.328  0.00714  0.00076
25  0.655 0.348  0.186  0.261 0.00517  0.00174
25þ  0.679  0.34  0.195  0.258  0.00526 0.00184
27  0.669 0.335  0.179  0.237 0.00368  0.0015
27þ  0.689  0.327  0.186  0.232  0.00376 0.00157
29  0.689 0.342  0.184  0.244 0.00277  0.000561
29þ  0.706  0.332  0.189  0.239  0.00283 0.000641
31  0.715 0.346  0.197  0.262 0.0023  0.0000939
31þ  0.729  0.336  0.2  0.254  0.00235 0.000173
33  0.743 0.344  0.212  0.278 0.00209  0.0000366
33þ  0.754  0.334  0.214  0.268  0.00211 0.000099
35  0.771 0.338  0.227  0.295 0.00194  0.000156
35þ  0.78  0.327  0.228  0.283  0.00197 0.000224
37  0.802 0.32  0.215  0.354 0.00199  0.000708
37þ  0.807  0.309  0.216  0.336  0.00199 0.000743

a
Value from [13].

experimental data by means of the new second-order and 65 GHz, the residuals using coefficients from
mixing model are in general better than those of MPM. Table 1 are on the order of 2% or less of the measured
Modeling the temperature dependence of the coeffi- absorption at each frequency.
cients with an analytic function Eq. (10) leads to We would like to point out some possible experimental
increased residuals generally, since the function does improvements which could further improve the accuracy of
not reproduce each temperature exactly. But since the band absorption-profile modeling. (We do not discuss
measurement errors in the center of the band are most here improvement of the spectrometer sensitivity, which
likely related to the fluctuation of the oxygen concen- obviously leads to that goal but it is not easily attainable.)
tration, they should be considered as statistical, not The first would be to use artificial air having a known
systematic. Fitting an analytical temperature depen- concentration of oxygen and nitrogen instead of outdoor air.
dence to the mixing coefficients is a form of averaging, This would avoid the trouble with oxygen concentration
which is known to reduce the statistical errors. Hence, mentioned in the experimental section. The second
even if the accuracy of reproducing the experimental improvement could be direct measurement of temperature
profiles seems to be reduced, there is a reason to dependence of individual line-broadening coefficients. That
believe that it actually is beneficial. One also notices can be done at low pressures using a spectrometer with
that the modeled second-order mixing still leaves some radio-acoustic detection [10] additionally equipped with a
systematic residuals at the band wings. This might be temperature-regulated thermostat for the absorption cell.
due to limitations of the model, such as neglecting of
mixing between the N þ and N  branches, or to under- 5. Comparison with other work
estimated temperature dependence of the lines’
collisional parameters, as discussed in the previous Line mixing has the proportionately greatest effect on the
paragraph. However, as shown in Fig. 9, between 54 wings of a band. Therefore, measurements at frequencies
 D.S. Makarov et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 112 (2011) 1420–1428 1427

higher than those from which the new model was derived absorption has been scaled upward by a factor of 4.774,
are a stringent test of its accuracy. At frequencies far from the ignoring possible differences in broadening and mixing
main resonances, absorption in dry air is dominated by the coefficients between air and pure O2. The calculation
collision-induced absorption by nitrogen, so measurements included the contribution from a weak 16O18O line at
of pure oxygen provide a better test of an oxygen-band 234 GHz. We used a half-width value of 2.8 GHz/bar for this
model. Fig. 10 shows a comparison to the pure-oxygen line, taken from [19], although it is 68% larger than another
measurements from Fig. 11 of [19]. The air model’s predicted recent measurement of that line [20]. The ‘‘total’’ curve also
includes O2–O2 collision-induced absorption, based on the
work in [21]. Absorption due to the submillimeter rotational
lines of O2 is barely significant in the plotted frequency
0
range; however, it is included in the total also.
Although we would not vouch for the accuracy of the
model at three atmospheres, it is in fair agreement with
the data in Fig. 10. It is worth mentioning that the
contribution of the non-resonant or Debye component
of the oxygen spectrum in Fig. 10 is 0.065 km  1; the total
absorption is less than this value as a consequence of the
mixing-coefficient biases bi, which account for mixing
between the resonant and non-resonant branches. This
effect is confirmed by the measurements of [19], which
also are below the Debye value over most of this
frequency range.

6. Conclusions

Improvements of the resonator spectrometer and the

method of measurement presented in [13,14] as well as
algorithm extensions in the present work permitted a
refined analysis of the 60-GHz atmospheric oxygen

Fig. 10. Power absorption coefficient in pure oxygen at 3 atm. Measure-

ments from [19] (triangles), calculation based on the discussed second-
order mixing model (solid line), collision-induced absorption (dashed
Fig. 8. Mixing coefficients from Table 1. line).

Fig. 9. Residuals between recorded profiles and second-order model using coefficients from Table 1, relative to the measured absorption value, at three
temperatures: 245 K (solid line), 288 K (dashed line) and 334 K (dotted line). The highest, the lowest and the central temperature of the measured
temperature range are chosen for the figure.
1428 D.S. Makarov et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 112 (2011) 1420–1428

absorption band. Precise experimental data on the band [7] Smith EW. Absorption and dispersion in the O2 microwave spec-
profile were obtained at pressures near one atmosphere in trum at atmospheric pressures. J Chem Phys 1981;74(12):6658–73.
[8] Liebe HJ. MPM—an atmospheric millimeter-wave propagation
the temperature range of 245–334 K, which covers a model. Int J Infrared Mill Waves 1989;10:631–50.
substantial part of the Earth’s atmospheric temperature [9] Liebe HJ, Rosenkranz PW, Hufford GA. Atmospheric 60 GHz oxygen
range. The second-order mixing effect was taken into spectrum: new laboratory measurement and line parameters.
J Quant Spectrosc Radiat Transfer 1992;48(5–6):629–43.
account for experimental data processing, and first- and
[10] Tretyakov MY, Koshelev MA, Dorovskikh VV, Makarov DS, Rosenk-
second-order mixing coefficients were derived. The mix- ranz PW. 60 GHz oxygen band: precise broadening and central
ing coefficients’ dependences on temperature and on frequencies of fine structure lines, absolute absorption profile at
quantum number are consistent with earlier data and atmospheric pressure, revision of mixing coefficients. J Molec
Spectrosc 2005;231:1–14.
theoretical predictions. Between 54 and 65 GHz, residual [11] Rothman LS, Jacquemart D, Barbe A, Benner DC, Birk M, Brown LR,
errors between the extended model and the observed et al. The HITRAN 2004 molecular spectroscopic database. J Quant
absorption profiles are typically 2% or less, which is an Spectrosc Radiat Transfer 2005;96:139–204.
[12] Rosenkranz PW. Interference coefficients for overlapping oxygen
improvement on MPM.
lines in air. J Quant Spectrosc Radiat Transfer 1988;39:287–97.
The information obtained in the course of this work [13] Makarov DS, Koval IA, Koshelev MA, Parshin VV, Tretyakov MY.
can be applied to Earth-atmosphere remote sensing, Collisional parameters of the 118 GHz oxygen line: temperature
microwave wireless communication, and industrial appli- dependence. Journal of Molecular Spectroscopy 2008;252:242–3.
[14] Tretyakov MY, Krupnov AF, Koshelev MA, Makarov DS, Serov EA,
cations, as well as for fundamental spectroscopic studies. Parshin VV. Resonator spectrometer for precise broadband inves-
tigations of atmospheric absorption in discrete lines and water
vapor related continuum in millimeter wave range. Rev Sci Instrum
Acknowledgment 2009;80:093106-1–0.
[15] Kogelnik H, Li T. Laser beams and resonators. Appl Opt 1966;5(10):
1550–67.
The work presented in this paper was supported in [16] Lam KS. Application of pressure broadening theory to the calcula-
part by the Russian Foundation for Basic Research. tion of atmospheric oxygen and water vapor microwave absorp-
tion. J Quant Spectrosc Radiat Transfer 1977;17:351–83.
[17] Rosenkranz PW. Retrieval of temperature and moisture profiles
References from AMSU-A and AMSU-B measurements. IEEE Trans Geosci
Remote Sensing 2001;39(11):2429–35.
[1] van Vleck JH, Weisskopf VF. On the shape of collision-broadened [18] Brown LR, Plymate C. Experimental line parameters of the oxygen
lines. Rev Mod Phys 1945;17:227–36. A-band at 760 nm. J Mol Spectrosc 2000;199:166–79.
[2] van Vleck JH. Magnetic dipole radiation and atmospheric absorp- [19] Meshkov AI, Lucia FCD. Laboratory measurements of dry air atmo-
tion bands of oxygen. Astrophys J 1934;80:161–70. spheric absorption with a millimeter wave cavity ringdown spec-
[3] vanVleck JH. The absorption of microwaves by oxygen. Phys Rev trometer. J Quant Spectrosc Radiat Transfer 2007;108(2):256–76.
1947;71:413–24. [20] Drouin BJ. Temperature dependent pressure induced linewidths of
[4] Baranger M. Problem of overlapping lines in the theory of pressure and 18O16O transitions in nitrogen, oxygen and air. J Quant
broadening. Phys Rev 1958;111(2):494–504. Spectrosc Radiat Transfer 2007;105(3):450–8.
[5] Fano U. Pressure broadening as a prototype of relaxation. Phys Rev [21] Boissoles J, Boulet C, Tipping RH, Brown A, Ma Q. Theoretical
1963;131(1):259–68. calculation of the translation-rotation collision-induced absorption
[6] Rosenkranz PW. Shape of the 5 mm oxygen band in the atmo- in N2–N2, O2–O2, and N2–O2 pairs. J Quant Spectrosc Radiat
sphere. IEEE Trans Antennas Propagat 1975;23(4):498–506. Transfer 2003;82(1–4):505–16.