Progress In Electromagnetics Research, Vol.

132, 369–388, 2012

DESIGN AND VALIDATION OF A TEM CELL USED FOR

RADIOFREQUENCY DOSIMETRIC STUDIES

C. Iftode1, * and S. Miclaus2

1 Department of Measurements and Optical Electronics, Politehnica
University of Timisoara, Timisoara 300223, Romania
2 Department of Technical Sciences, “Nicolae Balcescu” Land Forces
Academy, Sibiu 554011, Romania

Abstract—A Transverse Electromagnetic Mode (TEM) cell is an

interesting option for studying the biological effects of radiofrequency
radiation at reduced scale (in vitro studies). Controlled and
well-characterized exposure conditions are essential for a conclusive
investigation: the biological sample is exposed to a uniform incident
electromagnetic wave and the dose of absorbed radiation is precisely
determined and correlated with the effect. Unfortunately, experimental
dosimetry is often unavailable or inapplicable, so that a pre-
characterized and validated experimental setup is most valuable. As
such, the primary objective of the present work is to experimentally
validate a computational model of a self-built TEM cell designed
for bioelectromagnetic experiments in the 100 MHz–1 GHz frequency
range. Validation is achieved by comparing the computed vs. measured
values for three significant parameters: scattering parameters, incident
electric field distribution, and absorbed power in a set of liquid samples.
Successful validation and characterization is achieved by using CST
Microwave Studio’s finite integration technique (FIT), and respectively
a network analyzer for the experimental setup. The secondary
objective is a dosimetric study of four different liquid samples loaded
in the cell. The absorption coefficient (AC) is used, assimilated to the
specific absorption rate (SAR) of energy deposition in the entire sample
volume. AC is shown to converge in experiment and simulation up to
800 MHz for all samples. AC doesn’t depend directly on the samples’
volume (despite greater volumes frequently showing higher absorption)
but rather upon the internal field distribution, which in turn mostly
depends on the frequency and on the dimensions of the liquid samples.
Received 13 August 2012, Accepted 24 September 2012, Scheduled 2 October 2012
* Corresponding author: Cora Iftode ([email protected]).
370 Iftode and Miclaus

1. INTRODUCTION

The Transverse Electromagnetic Mode (TEM) cell was first introduced

by Crawford in 1974 [1] as a proper model for the simulation of free-
space propagation of plane waves. It is an enclosure consisting of
a rectangular section of transmission line tapered at both ends in
order to adapt to round coaxial connectors. Due to its inner flat
conductor (called septum) it ensures a TEM mode of propagation
inside, up to the frequency at which higher order modes are also
excited. The major advantage of a TEM cell is that it offers a
quasi-uniform electric field (E-field) inside [2, 3]. The cell is used in
electromagnetic compatibility for tests of radiated susceptibility, for
measuring emissions and for exposing (biological) samples in controlled
conditions — allowing further investigation of the biological effects of
electromagnetic fields. The fundamental work regarding TEM mode
propagation in the cell and the analytically calculated field distribution
was done in 1976 by Tippet and Chang [4]. A detailed description of a
TEM cell from the perspective of controlled electromagnetic exposure
and dosimetric approach was previously published by the authors [5].
Interest in using TEM cells increased over the last decades;
both experimental and computer simulation tools were developed, for
various endpoints [6–12]. Since the field homogeneity volume inside
the TEM cell depends on the ratio between the cell’s dimensions
and the frequency band at which it is operated, knowing the internal
field distribution inside, is a requirement for the design of a proper
experiment. Measuring the field strength inside the cell is however
tedious and inaccurate, as the field probe should be isotropic and
electrically isolated, non-perturbing and easy to insert even in rather
inaccessible corners; many practical limitations and inconveniences in
the experimental E-field determination exist [13]. The development
of alternative computerized techniques — simulation software —,
managed to provide repeatable and accurate solutions to the problem
[5, 6, 13]; simulations offer high spatial resolution and allow the
calculation of field data at multiple frequencies, in all directions and
orientations within the volume of interest, simultaneously.
Once a biological object is exposed to the electromagnetic field
inside the TEM cell, it is crucial to have precise knowledge on how
uniform the incident field is on the entire surface of the sample, as
well as on how much of the incident power is absorbed in the sample.
There are also situations where details such as where “hot spots”
of power deposition appear in the sample’s volume are of interest.
In any bioelectromagentic experiment investigating the biological or
biomedical effects of electromagnetic fields, a proper dosimetric tool is
Progress In Electromagnetics Research, Vol. 132, 2012 371

mandatory. Experimental dosimetry is rather complicated and affected

by significant uncertainty [2], while software solutions/computational
dosimetry are most valuable after the experimental validation. The
reverberation chamber was recently proven to be an alternative to
the TEM cell use, since a statistically uniform illumination can be
achieved, while the application of computational methods to accurately
describe local specific absorption rate of energy (SAR) deposition may
be available [14].
The main objective of present work is to characterize and validate
a laboratory-manufactured TEM cell that will be further used for
experimental biological samples exposure to radiofrequency (RF) field
in the 100 MHz–1 GHz range. The purpose is to eventually describe
the absorbed power or SAR in any biological sample. By knowing
the usual parameters of a sample — volume, density and dielectric
parameters at the frequency of interest, we aim to provide a reliable
tool for rapid and efficient dosimetric characterization, using the
computational procedure validated in the presented case. The results
achieved by using comparative experimental and computational data
show that a robust TEM cell model was obtained and that the
dosimetric assessment in the computational solution is feasible for any
type of experimental exposure of the samples.

2. TEM CELL CHARACTRISTICS AND VALIDATION

PROCEDURE

We designed and built a TEM cell for biological exposure purposes,

with the goal of using it up to 1 GHz while still to have a significant
volume for the exposure of a biological sample, where the field would be
homogeneous enough; it was previously shown that such an approach
is feasible [15]. The problems regarding characteristic impedance,
resonances, higher order modes of propagation, dimensions and usable
test area were the main concerns. The cell walls and the septum of
the TEM cell were made of 0.8 mm thick copper sheet, as shown in
Figure 1; the geometry and dimensions were chosen such as to obtain
a characteristic impedance of 50 Ω, according to the formula [9]:
377
Zc = £A ¤ (1)
4 B − 2
π ln(sinh( πg
B )) − C
ε0

where C is the fringe capacitance between inner conductor (septum)

and the side walls, and ε0 is the absolute permittivity of vacuum.
The ratio εC0 can be neglected if the ratio A/B is greater than
1 (see Figure 1 for dimension identification). The cell’s dimensions
were chosen as a trade-off between maximum usable frequency
372 Iftode and Miclaus

(a) (b)

(c)

Figure 1. (a) Longitudinal up-view of the TEM cell; (b) transverse

view of the TEM cell; (c) longitudinal front-view of the TEM cell.

Table 1. TEM cell’s dimensions.

Cell’s dimension Length [mm]

G 35
D 0.8
A 260
B 150
W 190
a1 27
a2 32
b1 24

and test volume; they are summarized in Table 1. It is known

that larger TEM cells have a lower useful frequency band. After
construction, electromagnetic characterization followed: the enclosure
was comparatively analyzed by measurements and by simulations. The
numerical model of the cell was experimentally validated by means of
three significant parameters: a) scattering parameters (S-parameters);
b) incident E-field distribution; c) absorbed power in a set of liquid
samples inserted in the cell.
Progress In Electromagnetics Research, Vol. 132, 2012 373

(a) (b)

Figure 2. (a) The model of the TEM cell developed in CST Microwave
Studio; (b) the experimental setup with the TEM cell (with Petri dish
inside) and the VNA connected to its ports.

3. SCATTERING PARAMETERS AND RESONANT

FREQUENCIES OF THE TEM CELL

The designed TEM cell was numerically modeled (Figure 2(a)) using
CST Microwave Studio [16]. The code uses the finite integration
technique, based on discretization of Maxwell’s equations in integral
form [17].
The modeled cell is made of copper, while the feeding source and
the load are modeled by discrete ports having an impedance of 50 Ω.
The cell’s long axis is oriented along the Ox axis, being centered in the
origin of the coordinate system. The height of the cell is on the Oz
axis. All the boundaries of the model are set to electrical.
The experimental setup consisted of the TEM cell connected to
the ports of an Agilent N5230C Vector Network Analyzer (VNA) by
low-loss RF200 coaxial cables (Figure 2(b)). The VNA was calibrated
then used for measuring the scattering parameters (S-parameters)
while the TEM cell was either empty or loaded with a sample; S-
parameters describe the electrical behavior of linear electrical networks
when undergoing various steady state stimuli by electrical signals, the
focus being on the input port voltage reflection coefficient, S11 , and on
the forward voltage gain coefficient, S21 .
The energy is transported in the cell mainly by TEM waves, which
are the dominant mode up to the frequencies where transversal electric
(TE) modes are also excited [4]. The cutoff frequency of a TEmn mode
is given by [18]:
√
c m2 B 2 + n2 A2
fc (TEmn ) = (2)
2AB
As long as the cell’s height is smaller than its width (B < A)
374 Iftode and Miclaus

the first mode to propagate is TE01 [19]; however, this propagating

mode is usually not excited unless a large test object is present
inside the cell [20]. The calculated cutoff frequency of TE10 mode
is: fcTE10 = 577 MHz; other higher propagating modes have cutoff
frequencies above 1 GHz and they fall out of the investigated band.
At frequencies for which the effective length of the cell is a
multiple of half the corresponding wavelength the resonance condition
is satisfied [18]. The resonant frequencies are given by:
s µ ¶2
2
pc
fRmnp = fcmn + (3)
2Lmn
where: p is the number of half wavelengths (p = 1, 2, 3. . . ) c =
300.000 km/s, the speed of light; Lmn is the effective length of the cell
Lmn = A + Xmn · d; Xmn is a fraction empirically determined. It can
range from 0.8 for TE01 mode of propagation to 0.5 for TE10 mode [18].
A and d are the dimensions specified in Figure 1 and in Table 1; the
resonant frequencies of TE10 mode are summarized in Table 2.
Figure 3(a) shows a comparison between the measured and
the computed input port voltage reflection coefficient, S11 , while
Figure 3(b) shows a comparison between the forward voltage gain

(a)

(b)

Figure 3. Computed vs. measured S-parameters of the empty cell

((a): S11 ; (b): S21 ).
Progress In Electromagnetics Research, Vol. 132, 2012 375

(a) (b)

Figure 4. Detailed views of the (a) second resonance and (b) fourth
resonance.

coefficients, S21 ; both describe the unloaded cell and the resonances
are evident. Good agreement between the model and the measured
data is observed up to 1 GHz. The computed resonances occurring
at 707 MHz and 978 MHz are very weak, while the experimental ones
are more augmented, and slightly shifted in the experiment. Figure 4
shows in greater detail these two resonances.
The S-parameters were measured (magnitude and phase) by
using the following settings of the VNA: input power Pin = 12 dBm;
frequency span = 100–1200 MHz and the number of frequency samples
(points) = 20.000. A large number of frequency samples were necessary
in order to “underline” the experimental resonances.
Table 2 offers a comparison of the differences between resonant
frequencies obtained in three different ways: by measurement, by
simulation and by using the theoretical formula. The resonant
frequencies obtained by experiment and by simulation are in good
agreement. The analytically resonances also match the measured ones
and those obtained from simulation, except that both, simulation and
experiment, provide two more frequencies which are not predicted
by the theory of waveguides. Garbe and Groh [21] have proven
that resonant frequencies of a standard TEM cell also depend on
taper’s dimensions and slope. The two resonances not predicted
by empty waveguide formulas can be derived from the analytical
approach in [21]. Another observation was that although TE10 mode is
theoretically excited at frequencies above 577 MHz, a 3D representation
of the E-field strength shows that up to 1 GHz the E-field vectors
are perpendicular to the propagation direction, with some (weak)
longitudinal components of the E-field only appearing at frequencies
greater than 1 GHz.
376 Iftode and Miclaus

Table 2. Comparison between resonant frequency values (simulation,

measurement and theory).
Computed Experimental Theoretical
resonance [MHz] resonance [MHz] resonance [MHz]
593.47 602.75
707.54 707.8 698
779.44 797.56
978.8 990.72 976
1000.82 1020.9

The next necessary step in validating the model for dosimetric

purposes was to establish the field uniformity inside the unloaded cell.
It is known that the exposed sample should ideally be placed in a
plane equally distanced between the septum and the floor; moreover
the sample’s height should not exceed one third of the distance between
the septum and the floor [18, 22].

4. INCIDENT FIELD DISTRIBUTION INSIDE THE

TEM CELL

Electric field uniformity in the unloaded TEM cell was computed for
multiple frequencies and planes. Experimental field distribution was
measured by using the passive-scatterer method, proposed by Morioka
in 2007 [23]; this method is based on the perturbation intentionally
introduced by the presence of a passive scatterer into the cell in the
form of a straight wire. The echo fields can be detected by the deviation
of the reflection coefficient S11 from that of the empty cell. This
deviation is related to the incident E-field strength at the location
of the scatterer.
54 holes were drilled in the cell’s floor, each having a diameter
of 2-mm. They were uniformly distributed on the Ox (14-mm apart)
and on the Oy (16-mm apart) axes. The central point of the TEM cell
was in the origin of the coordinate system. Thereby, the grid of holes
ranged from −96 mm to 14 mm on the Ox axis and from −90 mm to
−10 mm on the Oy axis. Afterwards a 25 mm long copper wire, tied
to a thread, was successively inserted through the holes.
Four files, containing magnitude and phase of the reflection (S11 )
and transmission (S21 ) coefficients in two parallel planes were recorded
for each insertion. The two planes were located at 40-mm and 30-mm
respectively below the septum, and parallel to it (for the designed TEM
Progress In Electromagnetics Research, Vol. 132, 2012 377

cell, half the distance between floor and septum is 37.5 mm).
The field distribution in a plane could be determined if the
scatterer were a point. However, in order to provide a detectable echo,
well above the sensitivity threshold of the VNA, a 25 mm scatterer
was chosen. The scatterer wire was oriented along the Oz axis. The
two planes chosen for field uniformity assessment were considered at
half the length of the wire: 52.5 mm below the septum (first plane) and
42.5 mm respectively (second plane). The field strength values in these
planes were obtained by averaging the field values along the length of
the wire.
S-parameters were measured with and without scatterer for each
of the 54 holes. The measurements were repeated 5 times and then
the average value was considered in order to minimize errors caused
by small mechanical oscillations of the thread and wire around the
equilibrium position. The reflection coefficients measured without any
scatterer were then vectorially subtracted from the ones measured with
the scatterer installed. The magnitude of the difference, expressed
in dB, is proportional to the electric field inside the cell, at the
measurement location, when no scatterer is present [23].
For accurate results the VNA had to be properly configured. We
set the source power at +12 dBm (maximum available) and the IF
bandwidth at 70 Hz. S-parameters were recorded for 20000 frequency
samples between 100 MHz and 1200 MHz.
Figures 5 and 6 depict the E-field distributions obtained by
simulation and measurement, in the plane situated at 42.5 mm below
the septum, for 402 MHz (Figure 5) and 930 MHz (Figure 6). Generally,
it was observed that the field was more uniform in the plane situated
closer to the septum as predicted by theory [20].
The advantage of numerical computation is that it allows the

Figure 5. Computed and experimental E-field uniformity at 402 MHz.

378 Iftode and Miclaus

Figure 6. Computed and experimental E-field uniformity at 930 MHz.

visualization of the field distribution in any plane. For the experimental

part we adapted the method described in [23] to the horizontal plane,
which led to a higher uncertainty than in the case of a vertical plane,
where the length of the wire influenced only spatial resolution and
the echo’s intensity. Thus we obtained a rather volumetric field
distribution, due to the wire’s length.

5. DOSIMETRIC VALIDATION OF THE TEM CELL

The validation is considered complete when the power absorbed in the

samples inserted in the cell for exposure (Pa ) converges in experiment
and in simulation.
It is known that, at the same incident E-field strength,
electromagnetic energy absorption in objects is dictated by their
dielectric parameters and by their geometry [2, 24, 25]. Techniques for
determining complex permittivity dispersion of any materials exist [26];
based on this, we chose here a set of four liquids for the dosimetric
approach.
CST Microwave Studio allows dosimetric computations directly,
either in the form of Pa , or in the form of average SAR or max SAR
over the exposed sample. SAR is measured in W/kg, and expressed [2]
as below:
Pa 2
σEint
SAR = = (4)
m ρ
where m is the sample’s mass, σ its conductivity (frequency-
dependent), and Eint the root-mean-square of the E-field strength
inside the sample. Eint itself depends on the ratio between the
dielectric constant of the environment and that of the sample and
Progress In Electromagnetics Research, Vol. 132, 2012 379

on the incident E-field. Average SAR is computed by Microwave

Studio by integrating the absorbed power over the entire volume
of the (dielectric) sample, while max SAR indicates the point SAR
calculated in the form of local SAR without mass or volume averaging.
If measurement provides the average Pa and SAR, by computation,
local SAR values can be determined — being able to visualize the
distribution of Pa or of SAR over the sample’s volume is of significant
interest [26, 27]. Understanding the distribution of absorption could
be essential in microdosimetric investigations of biochemical effects at
the level of cells dimension and below.
The accuracy of the computed SAR depends on: discretization
error, staircase approximation error, truncation error and on the chosen
algorithm used for SAR computation.
In order to get a dosimetric validation of the designed TEM cell
we took three gradual approaches.
In the first approach we chose a number of four liquid samples
to be exposed in the cell: distilled water, ethylene glycol, ethanol and
saline (NaCl solution 0.9%). Their dielectric parameters were extracted
from [28] in case of pure water, while for the rest of the liquids electric
dispersion was extracted from [29]. The samples were contained each
one in a soda-lime glass Petri dish, 90 × 15 mm (Steriplan, Duran
Group). Its dielectric parameter values are given in [30]. The liquid
samples, each of a volume of 20 ml, were inserted one by one in the
TEM cell, with the intention of assessing Pa in the whole liquid.
If the input power Pin is known, than the absorbed power Pa can
be derived from S-parameters according to the formula [31]:
³ ´
Pa = Pin 1 − |S11 |2 − |S21 |2 (5)
During measurements we expressed Pa in the liquid as the
difference between the absorbed power when the liquid was present
in the dish and the absorbed power when TEM cell was loaded with
the empty dish only. Relation (4) was consequently applied two times
and the difference between the two values (loaded dish and empty
dish) provided the final Pa in the liquid itself. Here we chose to use
the absorption coefficient (AC) as an indicator of absorption, which is
the ratio between Pa and Pin .
Figure 7 presents a comparison between the measured and
computed absorption coefficients. The results are in very good
agreement for ethylene glycol and ethanol up to 1.2 GHz; distilled water
and saline only show very good agreement of absorption coefficient
values up to 800 MHz. For frequencies higher than 800 MHz, these
two liquids present significant differences of AC between computed
and measured values; these differences will be considered for further
380 Iftode and Miclaus

Figure 7. Absorption coefficient in four different liquid samples.

analysis below. One should however note that, if the exposed object is
small compared to the wavelength of the incident field, little scattering
occurs, whereas when the object’s size is comparable to or larger
than a wavelength, significant scattering generally occurs [2]. This is
clearly noticed in Figure 7 for all liquids, above 800 MHz. Also, at low
frequencies the absorption in the samples is very weak (10−3 or less)
and thus cannot be accurately determined from S-parameters, for both
measurements and simulation. The sample produces only very small
changes in S-parameters and this makes the results more sensitive to
measurement errors and truncation errors in the computation.
Concerning the numerical model, in order to get reliable results,
S-parameters have to be computed accurately. Due to its geometry,
the cell behaves like a resonant cavity at certain frequencies, which
leads to long simulation times. The simulation is stopped when the
energy in the analyzed structure decay beyond an acceptable limit,
but this decay is retarded by resonances. Microwave Studio provides
an auto-regressive (AR) filter, for use with resonant structures, which
is basically a signal-processing method of predicting time signals
to save simulation time. These predicted signals are then used
to calculate the scattering parameters’ matrices. The usefulness
of AR filter is demonstrated in Figure 8, where the reflection and
transmission coefficients are plotted with the AR filter enabled and
Progress In Electromagnetics Research, Vol. 132, 2012 381

disabled, respectively. Using the AR filter removes the ripple from

the waveforms, thus decreasing the computational instability; this
approach was used.
Based on the results obtained up to this stage, it became of
interest to further analyze the dependence of AC on frequency and
on the volume of the sample. Since we used a very low quantity
of liquid in the previous dosimetric experiments (20 ml), and since
AC was generally very low, at the sensitivity limit of experimental
and computational capabilities, we focused (Figure 7), as a second
approach, on analyzing the influence of sample’s volume on AC values
throughout a broader frequency band (100 MHz–3 GHz), using only
the experimental approach. The results are shown in Figure 9;
three volumes of distilled water were used, filling either 1/3, 2/3 or
completely the recipient exposed in the TEM cell. As seen in Figure
9, AC cannot be correlated with the quantity of liquid, even though,
frequently, greater volumes do show higher absorption.
The third approach aimed at verifying the hypothesis that AC
doesn’t depend upon sample’s volume but rather upon the internal
field distribution in the sample, distribution that mostly depends on
the frequency. If and when the inner field is uniform, it is however
expected that AC would be proportional to the volume’s sample;
the hypothesis was verified in tandem, by measurements and by

Figure 8. The advantage of using the auto-regressive filter in

Microwave Studio.

Figure 9. Absorption coefficient frequency dependence for 3 different

volumes of distilled water filling a recipient.
382 Iftode and Miclaus

computations. Experimentally, three relative volumes of distilled water

(quoted as 1/3, 2/3 and 3/3 = full volume) were exposed in the dish
inside the TEM cell. AC values were calculated at three different fixed
frequencies: 707 MHz, 804 MHz and 930 MHz chosen based on previous
results in Figure 7. Figure 10 shows the comparative results obtained
for AC values in this scenario. “VNA” indicate measurements made
by the vector network analyzer, while “CST” indicates simulated data.
As seen, generally all computed values exceeded the measured ones,
thus the possibility of a reliable dosimetric prediction by computation
is demonstrated. The differences in AC values obtained by the two
methods are due to the fact that in the simulation absorptions in the
TEM cell (dish walls, dish support, glass absorption, losses along the
cell’s door, etc.) were completely neglected and they become higher
with increasing frequency. The second significant result at this stage
is that no correlation of AC with volume could be observed.
We then verified that AC depends on the E-field (and power)
distribution inside the volume of the sample — which further indirectly
depends on the frequency; it is at this stage that computational
dosimetry proves its versatility compared to the experimental
dosimetry. For this approach we studied the volumetric cases of
water exposed at 930 MHz by simulation only. The six snapshots in
Figure 11 illustrate the E-field distribution (vector-orientation) for top
and lateral views of the dish filled by the three different water volumes.
As it was predicted by computed AC in Figure 10 at 930 MHz, big
differences were not observed between the 1/3 and 2/3 filling levels,
but the field strength increased considerably when the dish was filled

Figure 10. Measured and computed absorption coefficient (AC) in

three water samples at three different frequencies.
Progress In Electromagnetics Research, Vol. 132, 2012 383

(a) Level 1 --- top view (c) Level 2 --- top view (e) Level 3 --- top view

(b) Level 1 --- side view (d) Level 2 --- side view (f) Level 3 --- side view

Figure 11. E-field vector distribution in two views at 930 MHz for
three water filling levels.

entirely. Also it is worth noting that the longitudinal components of

the E-field are also observable, in all snapshots — implying that TE
modes are also propagating in the TEM cell; they were very weak or
even absent in an empty cell, up to 1 GHz, but they are observable when
an obstacle is encountered. The abundant presence of longitudinal E-
components in the full-water dish in the last snapshots set of Figure 11,
converges to the conclusion that SAR is higher when the E-field is
more parallel to the longer axes of the target than perpendicular, as
described by the literature [2] — a point proven using the power of
vector-field representation in computations. Furthermore, since AC is
proportional to SAR (at a fixed frequency), AC is demonstrated to
be the highest in this third case of Figure 11. Also, the convergence
between expectations and results is obvious.
Figure 12 shows the SAR distribution in two different volumes of
distilled water for a frequency in the mid-range utility of the TEM
cell (402 MHz). The SAR values were averaged over 10 grams of
liquid, in two cases: Petri dish filled by 66% water and fully filled
dish respectively. The SAR is higher in the higher volume sample
as predicted by the previously computed AC. Therefore AC is an
adequate descriptor of the whole volume average SAR. Computations
in the case of two neighboring Petri dishes inserted in the TEM cell,
tangent to each other and situated on the same horizontal plane, as
well as the case of only one dish being present, were performed; the
results are presented in Figure 13. SAR distribution is very similar
for the case of double dishes and single dish respectively at frequency
of 402 MHz. However, at higher frequencies (930 MHz) the position
in the cell significantly modifies the SAR distribution. The field is
384 Iftode and Miclaus

(a) (b)
Figure 12. SAR distribution at 402 MHz (a) in 66% water filled Petri
and (b) in fully filled Petri.

(a) (b)
Figure 13. SAR distribution in two neighboring distilled water filled
dishes (a) at 402 MHz and (b) at 930 MHz.

not uniform in large areas when frequencies exceed 800 MHz, so in

this case it is not recommended to use more than one Petri dish at
a time. Theoretically, for frequencies higher than 800 MHz one dish
above the septum and another one below, in the same uniform-field
region (due to symmetry) could be used; but, for the cell in question,
this is not possible since the septum could not mechanically-support it.
This approach proved once again the efficiency of our exposure system
up to 800 MHz, mainly when larger volumes are to be exposed at once.
The dosimetric results converge with recent ones in [32]: SAR
distributions in Petri-dish samples are affected by the presence of the
meniscus (different liquids of the same volume but different meniscus
showed different AC), by the amount of liquid in the dish and we also
checked the influence of the shape of the dish and this modified the
SAR distribution.
Dosimetric applications are valuable if the precision is known Ex-
Progress In Electromagnetics Research, Vol. 132, 2012 385

perimental validation of the computational dosimetry was performed

during present work. Each of these two steps is affected by uncer-
tainties. The uncertainty of S-parameters, as measured by a vector
network analyzer was expressed, following procedures in [33, 34]. The
following errors are connected to the measurement system and the
experimental setup: analyzer calibration; linearity; system detection
limits; readout electronics; response time; integration time; RF ambi-
ent noise; liquid dielectric parameters. Combined standard uncertainty
was then calculated as +/ − 25%. However, computational analysis of
the SAR features many modeling and approximation phases that may
introduce error and uncertainty. Uncertainty factors of computed SAR
values are [35]: discretization error (depends on the ratio of cell size and
wavelength); sufficiently fine resolution must be used; staircasing error
related to the cell size and target accuracy; absorbing boundary condi-
tion/reflection error tested and tuned accordingly; truncation error —
long enough simulation time was provided; SAR computing algorithm
validated by the software’s IEEE-1528.1 compliance”; voxel size was
chosen according to the height of the liquid; material properties were
carefully chosen; SAR spatial averaging was used as defined by CST
software; dish positioning and contact with conductors were carefully
chosen to minimize errors. From these perspectives, all known errors
were minimized and a good agreement was obtained between experi-
ment results and numerical solutions.

6. CONCLUSION
Bioelectromagnetic experiments in the radiofrequency region of the
spectrum require proper instruments for accurate dosimetry. Present
work made a series of steps in developing such a set of instruments,
beginning with building a TEM cell for future controlled biological
exposure and ending with its in — depth characterization. The
modeled cell was successfully validated by measurements based on
a vector network analyzer, while the simulations were made by
using CST Microwave Studio. Simplifications in the modeled TEM
cell, without taking into consideration supplementary absorption or
losses, conducted to some differences between simulation — predicted
absorbed power and the measured one. However, very good agreement
was obtained at all stages of present work, regarding scattering
parameters, resonances, incident E-field level distribution in the
unloaded cell and dosimetric characterization of samples up to 800 MHz
frequency. The higher frequencies region was shown to be sensitive to
differences between simulation and experiment, but the causes were
analyzed and explained. Therefore, the results encourage us to use
the developed set of instruments for proper radiofrequency exposure
386 Iftode and Miclaus

experiments of small biological samples with great confidence in the

computational approach alone — (for frequencies lower than 800 MHz),
and in tandem computational — experimental approach, for higher
frequencies.

REFERENCES
1. Crawford, M. L., “Generation of standard EM fields using
TEM transmission cells,” IEEE Transactions on Electromagnetic
Compatibility, Vol. 16, No. 4, 189–195, Nov. 1974.
2. Durney, C. H., H. Massoudi, and M. F. Iskander, Radiofrequency
Radiation Dosimetry Handbook, 4th Edition, USAFSAM-TR-85-
73, Brooks AFB, Texas, USA, 1986, http://www.radhaz.com/doc
s/RF%20Dosimetry%20Handbook%20ver%204.pdf.
3. Guy, A. W., C. K. Chou, and J. A. McDougall, “A quarter
century of in vitro research: A new look at exposure methods,”
Bioelectromagnetics, Vol. 20, No. S4, 21–39, 1999.
4. Tippet, J. C. and D. C. Chang, “Radiation characteristics of elec-
trically small devices in a TEM transmission cell,” IEEE Transac-
tions on Electromagnetic Compatibility, Vol. 18, No. 4, 134–140,
1976, http://www.jpier.org/PIERB/pierb29/13.11022506.pdf.
5. Iftode, C., S. Miclaus, P. Bechet, and E. Surducan, “A TEM
cell model analysis for radiofrequency dosimetry improvement
by computational means,” Proceedings of the 7th International
Symposium on Advanced Topics in Electrical Engineering,
Bucharest, May 12–15, 2011.
6. Burkhardt, M., K. Poković, M. Gnos, T. Schmid, and N. Kuster,
“Numerical and experimental dosimetry of Petri dish exposure
setups,” Bioelectromagnetics, Vol. 17, No. 6, 483–493, 1996.
7. Green, H. E., “The TEM-mode bandwidth of two-conductor
open transmission lines,” Progress In Electromagnetics Research,
Vol. 40, 1–28, 2003.
8. Nikoloski, N., J. Fröhlich, T. Samaras, J. Schuderer, and
N. Kuster, “Reevaluation and improved design of the TEM cell in
vitro exposure unit for replication studies,” Bioelectromagnetics,
Vol. 26, No. 3, 215–224, 2005.
9. Boriraksantikul, N., P. Kirawanich, and N. E. Islam, “Near-field
radiation from commercial cellular phones using a TEM cell,”
Progress In Electromagnetics Research B, Vol. 11, 15–28, 2009.
10. Lee, S. U., H. J. Eom, and J. H. Kwon, “TEM mode in the GTEM
cell,” Journal of Electromagnetic Waves and Applications, Vol. 25,
No. 4, 519–526, 2011.
Progress In Electromagnetics Research, Vol. 132, 2012 387

11. Calo, G. and V. Petruzzelli, “Electromagnetic and thermal analy-

ses of improved GTEM cells for bioelectromagnetic experiments,”
Progress In Electromagnetics Research, Vol. 125, 503–526, 2012.
12. Zhao, J. X., H. M. Lu, and J. Deng, “Dosimetry and temperature
evaluations of a 1800 MHz TEM cell for in vitro exposure with
standing waves,” Progress In Electromagnetics Research, Vol. 124,
487–510, 2012.
13. Rostamzadeh, C., B. Archambeault, and S. Connor, “FDTD
analysis of symmetric TEM cell,” IEEE International Symposium
on Electromagnetic Compatibility, Aug. 8–12, 2005.
14. Moglie, F., V. M. Primiani, and A. P. Pastore, “Modeling of the
human exposure inside a random plane wave field,” Progress In
Electromagnetic Research B, Vol. 29, 251–267, 2011.
15. Malaric, K. and J. Bartolic, “Design of a TEM-cell with increased
usable test area,” Turkish Journal Electrical Engineering, Vol. 11,
No. 2, 143–154, 2003.
16. http://www.cst.com/Content/Products/MWS/Overview.aspx.
17. Clemens, M. and T. Weiland, “Discrete electromagnetism with
the finite integration technique,” Progress In Electromagnetics
Research, Vol. 32, 65–87, 2001.
18. Sevat, P., “Design of a TEM cell EMP simulator,” Report
No. 1084, Defence Research Establishment, Ottawa, Jun. 1991.
19. Zhong, C., “Examinations of higher order mode cutoff frequencies
in symmetrical TEM cells,” Proc. of International Symposium on
Electromagnetic Compatibility, EMC 2009, 6–11, 2009.
20. Wilson, P. F. and M. T. Ma, “Simple approximate expressions for
higher order mode cut-off and resonant frequencies in TEM cells,”
IEEE Transactions EMC, Vol. 28, No. 3, 125–130, Aug. 1986.
21. Garbe, H. and C. Groh, “Calculating the usable frequency range
of TEM-waveguides,” Proceedings of the 27th General Assembly
of the International Union of Radio Science, Maastricht, NL,
Aug. 2002.
22. Morega, M., S. Miclaus, and A. Machedon, “Analysis of the
electromagnetic field in a controlled enclosure for biological
dosimetry,” Rev. Roum. Sci. Techn. — Électrotechn. et Énerg.,
Vol. 52, No. 2, 225–235, Bucarest, 2008.
23. Morioka, T., “A field uniformity study of a TEM cell by
using a short wire scatterer,” IEEE International Symposium on
Electromagnetic Compatibility, Jul. 9–13, 2007.
24. Islam, M. T., H. Z. Abidin, M. R. I. Faruque, and N. Misran,
“Analysis of materials effects on radio frequency electromagnetic
388 Iftode and Miclaus

fields in human head,” Progress In Electromagnetics Research,

Vol. 128, 121–136, 2012.
25. Wang, Y. and M. N. Afsar, “Measurement of complex
permittivity of liquids using waveguide techniques,” Progress In
Electromagnetics Research, Vol. 42, 131–142, 2003.
26. Angulo, L. D., S. G. Garcia, M. F. Pantoja, C. C. Sanchez, and
R. G. Martin, “Improving the SAR distribution in petri-dish cell
cultures,” Journal of Electromagnetic Waves and Applications,
Vol. 24, Nos. 5–6, 815–826, 2010.
27. Zhang, M. and A. Alden, “Calculation of whole-body SAR from a
100 MHz dipole antenna,” Progress In Electromagnetics Research,
Vol. 119, 133–153, 2011.
28. Complex Dielectric Constant of Water Calculator, http://www.r
andom-science-tools.com/electronics/water dielectric.htm.
29. Hill, R., “Performance tests of indexsar ‘Di-line’ TEM sensors for
dielectric property measurement of tissue-simulant liquids for SAR
testing,” Surrey RH5 5DR, Indexsar, UK, Jul. 2002
30. Aziz M. S., A. G. Mostafa, A. M. Youssef, and S. M. S. Yous-
sif, “Electrical conductivity and dielectric properties of bulk
glass V2O5 (ZnO, PbO) SrO FeO,” Physics Research In-
ternational, Vol. 2011, 10 pages, 2011, Article ID 583420,
DOI:10.1155/2011/583420.
31. Li M.-H., H.-L. Yang, and X.-W. Hou, “Perfect metamaterial
absorber with dual bands,” Progress In Electromagnetics
Research, Vol. 108, 37–49, 2010.
32. Gangwar, R. K., S. P. Singh, and D. Kumar, “SAR distribution
in a bio-medium in close proximity with rectangular dielectric
resonator antenna,” Progress In Electromagnetics Research B,
Vol. 31, 157–173, 2011.
33. ISO/IEC, “Guide of the expression of uncertainty in measure-
ment,” Geneva, 1995.
34. EA Guideline 4/16, “Expression of uncertainty in quantitative
testing,” 2003.
35. Laakso, I., “Uncertainty in computational RF dosimetry,”
Doctoral Dissertation, Aalto University, 2011.