Anda di halaman 1dari 12

FULL PAPER

Power Balance and Loss Mechanism Analysis in RF


Transmit Coil Arrays
Andre Kuehne,
1,2
* Sigrun Goluch,
1,2
Patrick Waxmann,
3
Frank Seifert,
3
Bernd Ittermann,
3
Ewald Moser,
1,2
and Elmar Laistler
1,2
Purpose: To establish a framework for transmit array power
balance calculations based on power correlation matrices to
accurately quantify the loss contributions from different mech-
anisms such as coupling, lumped components, and radiation.
Theory and Methods: Starting from Poyntings theorem,
power correlation matrices are derived for all terms in the
power balance, which is formulated as a matrix equation.
Finite-difference time-domain simulations of two 7 T eight-
channel head array coils at 297.2 MHz are used to verify the
theoretical considerations and demonstrate their application.
Care is taken to accurately incorporate all loss mechanisms.
The power balance for static B
1
phase shims as well as two-
dimensional spatially selective transmit SENSE pulses is
shown.
Results: The simulated power balance shows an excellent
agreement with theory, with a maximum power imbalance of less
than 0.11%. Power loss contributions from the different loss
mechanisms vary significantly between the investigated setups,
and depending on the excitation mode imposed on the coil.
Conclusion: The presented approach enables a straightfor-
ward loss evaluation for an arbitrary excitation of transmit coil
arrays. Worst-case power imbalance and losses are calculated
in a straightforward manner. This allows for deeper insight into
transmit array loss mechanisms, incorporation of radiated
power components in specific absorption rate calculations and
verification of electromagnetic simulations. Magn Reson Med
000:000000, 2014. VC
2014 Wiley Periodicals, Inc.
Key words: parallel transmission; coil array; power balance;
Poynting theorem; electromagnetic simulation; loss analysis
INTRODUCTION
The power balance, i.e., the breakdown of all coil losses
by loss mechanism, is an important metric for assessing
radiofrequency (RF) coil efficiency. While power deposi-
tion inside a conducting sample cannot be avoided due
to the concomitant electric fields linked to the desired
excitation RF field (B
1
), additional losses occurring in
the coil itself can be minimized by prudent design
choices. To quantify the contributions of coil and sample
losses, the power efficiency, i.e., the fraction of the input
power deposited in the lossy sample, is used as a figure
of merit for single coils (1,2).
Multiple coil elements can be combined to form
phased arrays for reception (3) as well as transmission
(4) in order to increase receive sensitivity and excitation
fidelity, respectively. However, inductive, capacitive,
and resistive interactions between the individual coil
elements incur additional losses, leading to increased
noise, decreased transmission efficiency, and less dis-
tinct coil sensitivity profiles. A variety of methods exists
to reduce coil coupling effects in transmit arrays, ranging
from coil overlap (3), decoupling annexes (5), decoupling
interfaces (6), shared L/C elements (7), transformers (8),
and resonant structures (911) to active techniques
involving modified transmitter hardware (1214).
At ultrahigh field strength (_ 7 T), coils employing
radiative power transfer into the subject have been
shown to yield advantageous B
1
distributions (15), and
traveling-wave magnetic resonance imaging even com-
pletely relies on radio wave propagation at high frequen-
cies (16,17). However, power loss due to unwanted far
field radiation is increasing with frequency as well (18
and references therein). Although electromagnetic energy
leaving the magnet bore is, of course, almost perfectly
contained by the scanners Faraday cage and ultimately
dissipated in the sample and other lossy structures
inside the scanner room, its interference pattern with the
desired coil fields is hardly predictable; and radiative
losses are routinely mitigated through the use of metallic
coil shielding structures (19).
The introduction of shields and additional decoupling
components, however, can introduce further losses, and
it is of high interest to the coil designer to weigh their
benefits against the costs. Apart from low loss require-
ments, component voltage and power limits affect design
choices as well and have to be carefully examined regard-
ing their impact on coil performance. According to manu-
facturer datasheets, high-voltage rated capacitors often
exhibit far lower Q-factors compared to capacitors with
lower working voltages, but conversely feature lower
maximum current limits due to their increased losses.
Ideally, power is mainly dissipated in the subject. This
power deposition, however, critically depends on the
exact superposition of the individual coil elements elec-
tric fields, and the assumption of a lossless coil has been
1
Center for Medical Physics and Biomedical Engineering, Medical Univer-
sity of Vienna, Vienna, Austria.
2
MR Centre of Excellence, Medical University of Vienna, Vienna, Austria.
3
Physikalisch-Technische Bundesanstalt (PTB), Braunschweig und Berlin,
Germany.
Grant sponsor: Austrian BMWFJ FFG; Grant number: 832107 (Vienna
Research Studio for Ultra-High Field Magnetic Resonance Applications);
Grant sponsor: EMRP (jointly funded by the EMRP participating countries
within EURAMET and the European Union); Grant number: HLT06; Grant
sponsor: (E.M.): Siemens Medical Solutions (Erlangen, Germany).
*Correspondence to: Andre Kuehne, M.Sc., Medizinische Universit at Wien,
Zentrum f ur Medizinische Physik und Biomedizinische Technik, Exzellenz-
zentrum Hochfeld-MR, W ahringer Strabe 1820, A-1090 Wien, Austria.
E-mail: andre.kuehne@meduniwien.ac.at
Received 14 April 2014; revised 19 September 2014; accepted 22
September 2014
DOI 10.1002/mrm.25493
Published online 00 Month 2014 in Wiley Online Library (wileyonlinelibrary.
com).
Magnetic Resonance in Medicine 00:0000 (2014)
VC
2014 Wiley Periodicals, Inc. 1
shown to yield significantly different specific absorption
rate (SAR) estimates as compared to more realistic models
(20). The absence of specific loss mechanisms will invari-
ably lead to a different redistribution of the input power
onto the remaining losses, impacting local and global
SAR estimates and thus apparent coil performance.
Electromagnetic simulations are an important tool to
evaluate coil design choices and, additionally, are obliga-
tory for SAR evaluation of multichannel coils used in
parallel transmission (21). However, computational
approaches such as the finite-difference time-domain
(FDTD) method (22) often do not encompass the full
problem space. Instead of including the Faraday cage,
free-space boundary conditions are being used (23,24),
thus introducing a radiated power component that needs
to be accounted for when estimating coil efficiency and
SAR (18,25). This simplification is useful for multiple
reasons. From a purely practical standpoint, the inclu-
sion of the cage greatly raises the memory requirements
for the simulation. Additionally, the simulation time
will increase, as it is proportional to the Q-factor of the
modeled system, and the closed scanner room can be
regarded as a high-Q cavity. Finally, the knowledge
about radiated power leaving the bore is invaluable to
the coil designer, as it allows to quantify and compare
the efficacy of different radiation shielding measures.
Increasingly complex coil setups have spurred the
development of sophisticated simulation approaches
(2628), where data are often exchanged between differ-
ent commercial simulation programs and in-house devel-
oped postprocessing tools. Although this allows a very
flexible and powerful analysis of coil behavior, it is
imperative to ensure the integrity of the results, espe-
cially if patient safety is directly affected. Frequently, B
1
mapping and temperature measurements are used for
simulation validation (29); however, it is desirable to
additionally verify a particular coil simulation in its
entirety based on a single metric. Energy conservation,
expressed for electromagnetic fields in the Poynting the-
orem, is a prime candidate for this purpose. Trivially,
the power balance can be calculated after superimposing
the single-channel fields. However, this approach is tedi-
ous, as it is computationally intensive in practice and
only allows examination of a limited number of driving
modes. As yet, no comprehensive scheme adapted to the
unique loss behavior of transmit arrays has been pub-
lished. It is well known, on the other hand, that the
power correlation matrix formalism can greatly simplify
the estimation of power deposited inside lossy materials,
and it is routinely used for local and global SAR estima-
tion in simulation as well as in situ (3032). Starting out
from our initial investigations (33), the aim of this article
is to extend this formalism by deriving a power correla-
tion matrix for each term contributing to the total power
balance. This will allow to address the aforementioned
coil design and simulation issues by providing a frame-
work for straightforward and consistent calculation of
all loss terms, comparison of different array design
approaches via their respective loss matrices, and integ-
rity validation of electromagnetic field simulations in
order to more accurately predict and evaluate transmit
coil array behavior.
THEORY
Conservation of electromagnetic energy is expressed in
Poyntings theorem (34). For time-harmonic fields, the
real part of the theorem represents the power balance of
the system, P
src
= P
dis
P
rad
, equating the source power
P
src
in a volume V to the sum of the power dissipated
inside, P
dis
, and the power P
rad
radiated through the vol-
umes boundaries:

1
2
___
V
Re J
+
E ( )dV =
1
2
___
V
s r ( )[E[
2
dV

1
2
__
_
@V
Re E H
+
( )ds: [1]
Here, E, H, J, and r represent the spatially varying
electric field, magnetic field, current density, and con-
ductivity, respectively, and the asterisk indicates com-
plex conjugation.
As the individual coils of an N-channel transmit array
are driven by N discrete sources, the left-hand integral
can be replaced by the sum
1
2

N
i=1
Re I
+
i
U
i
_ _
over the
source voltages and currents U
i
and I
i
. Treating the sys-
tem as purely linear, all fields, voltages, and currents
can be represented as linear combinations of their
respective single-channel excitation values, e.g.,
E =

N
i=1
a
i
E
i
, where a
i
is a dimensionless complex scal-
ing factor and E
i
represents the electric field generated
by coil element i for a unit excitation. The scaling factors
are commonly regarded as equivalent to driving voltages
that are variable during a magnetic resonance experi-
ment, being updated in discrete intervals long enough to
maintain the time-harmonic field behavior during each
individual step. Equation [1] must necessarily hold for
each of the discrete steps. To separate the static, i.e.,
geometry dependent, from dynamic, i.e., excitation
dependent, contributions, it can be expressed in terms of
quadratic forms v
H
Qv, with a static Hermitian positive-
semidefinite matrix Q specific to each power term, v
constituting the time-dependent coil array driving volt-
age vector common to all quadratic forms, and super-
script H denoting the conjugate transpose. Subsequently,
these matrices will be derived, with a focus on separat-
ing contributions of different loss mechanisms.
Source Power
Most commonly, transmit arrays are driven using a discrete
power amplifier for each channel, with a fixed source
impedance of Z
0
= 50 V, and the individual coil elements
impedance-matched to Z
0
for maximum power transfer.
When describing the power flow within an N-port network
such as a coil array, three distinct notions of power are of
interest. The forward power P
fwd
denotes the power inci-
dent to the system, with a fraction P
src
of this power being
transmitted into the network, and the remainder P
ref
reflected from it due to imperfect impedance match and
coupling, eventually being absorbed in the power amplifier
circulators. The power flow can be analyzed in terms of
normalized forward and reflected voltage wave vectors, a
2 Kuehne et al.
and b, which are related through the systems N N scat-
tering matrix S via b=Sa (35). The source term in Eq. [1]
can be conveniently formulated in terms of S. Power inci-
dent into the system is given by P
fwd
=
1
2
a
H
a. This can be
expressed as a quadratic form of the driving voltage ampli-
tude vector v =

Z
0
_
a:
P
fwd
=
1
2
a
H
I
N
a = v
H
1
2Z
0
I
N
_ _
v
= v
H
Q
fwd
v:
[2]
With I
N
denoting the identity matrix of size N, the result-
ing forward power matrix Q
fwd
is thus purely diagonal.
Power that is reflected from the network due to mis-
match or lost via interelement coupling can be calculated
from the reflected waves b as P
ref
=
1
2
b
H
b, which can be
rewritten in terms of the incident waves and the scattering
matrix to yield the reflected power correlation matrix Q
ref
:
P
ref
=
1
2
a
H
S
H
S
_ _
a = v
H
1
2Z
0
S
H
S
_ _
v
= v
H
Q
ref
v:
[3]
By convention of Eq. [1], source power is regarded as
negative, while all loss terms are positive, thus defining
total source power as P
src
= v
H
Q
fwd
Q
ref
( ) [ [v.
The elements v
i
of vector v are voltages by dimension
and related to the scaling factors a
i
via a
i
= v
i
=

2 Z
0
P
0
_
,
assuming a forward power equal to P
0
was used for the unit
excitation of the single channel fields. To increase readabil-
ity, the factor k = 2 Z
0
P
0
( )
1
is used throughout the fol-
lowing sections.
Dissipative Losses
It is advantageous to categorize power dissipation into
material and lumped element (e.g., capacitor, inductor,
resistor, etc.) losses, as the discrete elements can be
more easily treated via circuit theoretical methods.
Accordingly, power dissipated inside a lumped element
can be calculated using the voltage drop U across the
element and the current I flowing through it via
P =
1
2
Re UI
+
( ). Arranging the voltages and currents of
multiple lumped elements in column vectors u and i,
the dissipated power is given by P
lmp
=
1
2
Re i
H
u
_ _
. For an
N channel coil, matrices U and I can be constructed, in
which the Nth column contains the voltages or currents
of each lumped component for a unit excitation of coil
element N. The total power dissipated inside the discrete
components for an excitation vector v is now
P
lmp
=
k
2
Re v
H
I
H
Uv
_ _
, and taking the real part via addi-
tion of the complex conjugate leads to
P
lmp
=
k
4
v
H
I
H
Uv v
H
I
H
Uv
_ _
H
_ _
= v
H
k
4
I
H
UU
H
I
_ _
_ _
v
= v
H
Q
lmp
v;
[4]
with Q
lmp
denoting the lumped element loss matrix.
Multiple matrices U and I can be utilized to differentiate
element groups, such as matching, tuning and decou-
pling components.
The power dissipated inside a conductive material via
interfering electric fields generated by multiple sources
can be expressed as a quadratic form as previously
described (4,36)
P
mat
= v
H
Q
mat
v: [5]
The entries q
ij
of the power correlation matrix Q
mat
are
defined as
q
ij
=
k
2
___
V
sE
+
i
E
j
dV; [6]
with the indices ij denoting the coil element number. The
diagonal entries of this matrix denote the power dissipation
when driving a single channel only, whereas the off-diagonal
entries provide a measure of electric field orthogonality and
thus coil element interaction. Apart from a scaling factor,
Q
mat
is identical to the noise resistance matrix of a receive
coil array (3,37). Integrating only over the volume of a spe-
cific medium (e.g., coil substrate, load, etc.) will yield a
matrix for losses inside this particular medium.
Radiative Losses
A radiated power correlation matrix Q
rad
similar to the
dissipated power correlation matrix in Eq. [6] can be
constructed from the single coil fields. Starting with
P
rad
=
1
2
Re v
H
~
Q
rad
v
_ _
; [7]
where the elements ~ q
ij
of matrix
~
Q
rad
are defined as
~ q
ij
= k
__
_
@V
E
j
H
+
i
_ _
ds; [8]
and again taking the real part yields
P
rad
= v
H
1
4
~
Q
rad

~
Q
H
rad
_ _
_ _
v
= v
H
Q
rad
v:
[9]
Power Balance
Poyntings theorem can now be stated in terms of quad-
ratic forms, leading to the matrix equality
Q
fwd
= Q
ref
Q
lmp
Q
mat
Q
rad
: [10]
The individual matrices can be analyzed by estab-
lished mathematical tools, such as Eigendecomposition,
which is routinely applied to investigate the quadratic
forms appearing in (local) SAR calculations (32) or the
evaluation of phased array signal-to-noise ratio (SNR)
(37,38). Eigenvectors form a basis where the power corre-
lation matrices are diagonal; physically, they represent
noninteracting modes (37,39). The corresponding Eigen-
values define the losses occurring when exciting the
Eigenmodes. For a vector v of unit norm, the range of a
Transmit Array Loss Analysis 3
quadratic form v
H
Qv is limited by the largest and small-
est Eigenvalues of the matrix Q. Consequently, these val-
ues can be used to find the maximum and minimum
possible loss contributions pertaining to an arbitrary
power correlation matrix, and thus help in identifying
major loss contributions. A visualization of a possible
practical breakdown of these loss mechanisms into spe-
cific subsets is provided in Figure 1.
Potential violations of the equality can be utilized to
assess the integrity of a given multichannel electromag-
netic coil simulation result. For this purpose, an addi-
tional matrix Q
E
is introduced to the right-hand side,
constituting the error matrix required to balance the
equation. The by magnitude largest Eigenvalue of Q
E
cor-
responds to the maximum power imbalance that might
occur for this simulation. As energy is conserved in elec-
tromagnetic simulations (40), this matrix is expected to
be negligible.
Coupling Effects
Coupling between coil elements manifests as off-diagonal
entries in the Q matrices. As shown in Eq. [10], Poynt-
ings theorem requires that the sum of all loss matrices is
a purely diagonal scaled identity matrix. This implies that
a coil array exhibiting resistive coupling due to shared
lossy current paths, i.e., containing off-diagonal entries in
the material loss matrix Q
mat
, needs to have an associated
loss matrix containing off-diagonal entries with inverted
signs to ensure diagonality of the sum. For a lossless and
nonradiating coil, i.e., Q
lmp
= Q
rad
= 0, this is necessarily
the reflected power matrix Q
ref
, indicating that resistive
coupling cannot be compensated. Consequently, decou-
pling structures eliminating mutual resistances (9) need to
introduce additional losses to fulfill the power balance. In
this case, the lumped element loss matrix Q
lmp
of the
decoupling elements would provide the compensating off-
diagonal entries, thus allowing an (ideally) zero reflected
power matrix. This intrinsic relationship between resis-
tive coupling and losses in transmit arrays is directly
analogous to increased noise and noise correlation in
receive arrays utilizing resistive coupling compensation
(12). Although all decoupling networks will inevitably
introduce some amount of loss in a realistic setup, the
described limitation is fundamental in nature, and defines
a theoretical limit on achievable decoupling with lossless
components. This limitation only pertains to the resistive
part of mutual array element interactions, thus coils
exhibiting only reactive coupling are not affected, and
perfect decoupling is possible without additional losses
required by conservation of power.
METHODS
Coil Setup
To demonstrate the calculation of the loss matrices and
their application to array performance evaluation, two
different eight-channel coil arrays, based on loops and
striplines (41) as their respective building blocks, were
simulated. Both designs, depicted in Figure 2, share their
overall dimensions, consisting of a 1-cm thick cylindrical
former with an outer radius of 13.5 cm, a length of
22.5 cm, and a copper RF shield of equal length with a
radius of 17 cm. These dimensions were chosen to be
similar to a previously investigated setup (18) for com-
parison purposes. Both configurations were loaded with
a 9-cm radius spherical phantom with dielectric proper-
ties similar to brain tissue at 297.2 MHz, and also the
Duke model, which was not truncated to fully capture
potential waveguide mode propagation effects (17). Tun-
ing, matching, and decoupling were adjusted individu-
ally for both arrays and each of the two loading
conditions, and held constant for all further calculations.
The magnet bore was modeled as a 2.45-m long stainless
steel cylindrical sheet with a radius of 44.75 cm, and the
gradient coil and shield assembly approximated as a
122-cm long solid copper annulus with an inner radius
of 34.25 cm and an outer radius of 44.75 cm positioned
at the center of the bore.
The loop array consisted of eight rectangular elements
measuring 9:5 22:5 cm, using 5-mm wide copper strips
as conductive elements. The former material was chosen
as acrylic plastic, with the individual coils arranged con-
formally on the outside. Element decoupling of at least
19 dB for the phantom load and 15 dB for Duke
was achieved with counter-wound air-core solenoid
inductors for nearest-neighbor coils (8).
For the stripline array, the former was composed of
low loss polytetrafluoroethylene (PTFE), with the 1-cm
wide and 22.5-cm long conductors of the stripline ele-
ments mounted on the inside, and the 5-cm wide ground
planes attached to the outside. The required nearest-
neighbor decoupling capacitance value for this configu-
ration was found to be very small ~ 0:3 pF ( ). However,
the addition of copper strips to connect the decoupling
capacitors to the individual elements introduced a paral-
lel parasitic capacitance larger than required for decou-
pling, making a purely capacitive decoupling impossible.
Consequently, a parallel LC circuit tuned to exhibit a
compensating reactance was used to achieve a decou-
pling of at least 15 dB in the phantom and 18 dB
using the Duke model. This decoupling strategy is
FIG. 1. Branching diagram depicting the separation of the funda-
mental loss mechanisms into multiple subsets. Each subset can
be represented by an individual loss matrix and further subdivided
if a more differentiated loss analysis is desired.
4 Kuehne et al.
closely related to the resonant decoupling introduced by
Avdievich et al. (8), while not eliminating mutual resist-
ance. The S-Matrices for all investigated configurations
are shown in Figure 3.
Electromagnetic Simulations
The arrays were simulated using XFdtd 7.3.2 (Remcom,
State College, PA) and ADS (Agilent, Santa Clara) based on
the cosimulation approach introduced by Kozlov and Turner
(26), and described more in depth by Lemdiasov et al. (27).
A detailed practical description of the method is contained
in a paper recently published by Guerin et al. (42).
The grid resolution varied between 1 and 2 mm in the
coil volume and head of the body model up to 25 mm in
the outer regions of the simulation space, which was ter-
minated with a seven cell perfectly matched layer region
(24). Inside the remaining body, the resolution was 1 cm
at most to sufficiently satisfy the 10 cells-per-wavelength
requirement of the FDTD method. The total simulation
extent was 1.4 1.4 3 m, placing the boundary region
more than k=2 away from the coil. Lumped elements
were replaced by 50 V source ports to enable fast tuning,
matching, and decoupling in the ADS cosimulation. Con-
sequently, the loop array was modeled using 48 ports,
whereas only 24 ports were required for the stripline
design. All ports were driven consecutively using a
modulated Gaussian pulse centered at 297.2 MHz with a
bandwidth of 100 MHz; and transient calculations con-
tinued until a convergence of at least 90 dB was
reached. Steady-state fields were extracted at the center
frequency and the sampling rate for the steady-state field
Fourier transform was manually adjusted to 1 GHz to
resolve the complete spectrum of the input pulse and
prevent aliasing errors. This rigorous simulation regime
was chosen to avoid any confounding errors. After tun-
ing, matching, and decoupling, the single channel
lumped element voltages and currents were calculated in
ADS and the excitation fields were calculated in MAT-
LAB (The Mathworks, Natick, MA) as linear combina-
tions of the initial many-port simulations using scaling
factors determined in the cosimulation. Subsequently,
the field and lumped element data were used to deter-
mine all power correlation matrices. The power imbal-
ance for the initial many-port simulations as well as the
fully combined 8-channel data was evaluated separately
to quantify the impact of the cosimulation approach.
Loss Modeling
To allow simulation of lossy metals with realistic conductiv-
ities (copper: s = 5:8 10
7
S=m, steel: s = 1:45 10
6
S=m),
FIG. 2. Single element schematics and 3D models of the simu-
lated 8-channel loop (top) and stripline (bottom) arrays. Both
designs share the overall dimensions, with the loop coil and strip
coil formers consisting of acrylic plastic and PTFE, respectively.
Tuning and matching capacitors are labeled with C
T
and C
M
,
respectively. Nearest-neighbor decoupling for the loop array is
achieved with counterwound solenoidal inductors, which can be
adjusted by varying the coupling between the two solenoids. The
stripline elements are decoupled using an interconnected parallel
LC circuit which can be adjusted using the variable decoupling
capacitor C
D
. The lumped element values used in the simulation
are shown in the schematics. [Color figure can be viewed in the
online issue, which is available at wileyonlinelibrary.com.]
FIG. 3. Simulated scattering matrices for all array/load configura-
tions. For the spherical phantom load, the loop-array exhibits bet-
ter decoupling than the stripline-array, whereas this is reversed for
the full body Duke load. [Color figure can be viewed in the
online issue, which is available at wileyonlinelibrary.com.]
Transmit Array Loss Analysis 5
a good conductor approximation (43) was enabled in XFdtd.
Capacitors were assigned realistic equivalent series resistances
of ATC 100E series capacitors (American Technical Ceramics
Inc., Huntington Station, NY) as provided by the manufacturer
(http://www.atceramics.com/tech-select.aspx). Comparable
component databases are also available from other vendors.
As the exact loss behavior of commonly used variable capaci-
tors depending on their tuning state is not documented by the
manufacturers, a function of the form y = a x
b
c was fitted
to the known equivalent series resistances values of the dis-
crete capacitors used in the simulations to approximate a lossy
capacitor with tuning-dependent equivalent series resistances.
Air-core inductors were assigned a quality factor Q of 570 and
620 for the loop and stripline decoupling components, respec-
tively. These values were approximated for typical air sole-
noid dimensions using wcalc [http://wcalc.sourceforge.net/,
adapted from (35)]. Solder joint resistances were extrapolated
from literature data (44) to 297.2 MHz, assuming a

f
_
frequency dependence. The loss tangents of the plastic and
PTFE formers were tan d =0.011 and tan d =0.0002, respec-
tively (45). Dielectric properties of the phantom were modeled
using averaged values for brain tissue at 297.2 MHz
(e = 50:6; s = 0:66 S=m), and the tissues of the Duke
model were assigned the values given in the ITIS tissue data-
base (46).
Parallel Transmission and RF Pulse Calculation
Power balance calculations were carried out for a set of
static B
1
phase shim settings representing the counterclock-
wise rotating circular polarized (CP) Eigenmodes of the
arrays as well as the equal phase (EP) driving mode (37,38).
The CP
1
mode corresponds to the widely used homogene-
ous Birdcage mode. To illustrate the transition between
the discrete modes, nearest neighbor phase increments
were varied between 0 and 180

. Additionally, two-
dimensional spatially selective transmit SENSE (47) pulses
to excite a checkerboard pattern in the spherical phantom
and a homogeneous slice in both human model and phan-
tom were calculated based on the simulated B
1
fields (48).
The pulses were designed for a target flip angle of 20

using
a spiral trajectory on a 32 32 (220 220 mm
2
) field of
excitation with fourfold spatial oversampling, a maximum
slew rate of 200 mT/m/ms, maximum gradient strength of
40 mT/m and an acceleration factor of r =4, resulting in a
pulse length of 2.40 ms (240 segments). Bloch simulations
were performed to visualize the realized spatial excitation
patterns, and the power balance was calculated for every
pulse segment and also averaged over the whole pulse.
RESULTS
The maximum power imbalance as given by the largest
Eigenvalues of the error matrix Q
e
was 0.010% (40 dB)
for the initial 48-port FDTD simulation of the loop-array
and 0.016% (38 dB) for the final 8-loop field data,
respectively. The 24-port simulation of the stripline-
array exhibited a maximum imbalance of 0.003% (45
dB), with the largest error of the combined stripline array
data reaching 0.11% (30 dB). Table 1 shows the com-
plete power balance of the four counterclockwise rotat-
ing excitations and the EP mode for all simulated
configurations, along with the respective theoretical
maximum and minimum losses and power imbalances.
The corresponding B

1
field patterns of these modes are
depicted in Figure 4. Power balance distributions for
nearest-neighbor incremental phase differences between
0

and 180

are shown in Figure 5. Finally, the time


courses of the selective excitation pulses are depicted in
Figure 6, together with Bloch simulations of the excita-
tion pattern and the averaged power balance contribu-
tions of each term.
DISCUSSION
Simulation Power Imbalance
The maximum power imbalance we encountered in our
simulations was 0.11%. This negligible value indicates
excellent agreement with theory. Multiple factors may
contribute to the small residual error, most significantly
the minor reciprocity and power balance violation of the
FDTD method when using graded meshes (49). The
imbalance increase between the initial simulation and
the cosimulated data is small and due to error propaga-
tion from the initial simulations. It is notable that the
simulations of the stripline array exhibit a larger error as
compared to the loop array. This reflects the fact that the
absolute error of the simulations scales with forward
power, and in the cosimulations the 24 virtual ports of
the stripline array need to be driven at higher power in
order to achieve the identical excitation as the loop
array. Nevertheless, this deviation is still very small, and
as care was taken to eliminate confounding simulation
errors, the reported imbalances can be regarded as exem-
plary benchmark lower limit errors in FDTD simulations
of comparable complexity.
Load Power Deposition
As can be seen in both Table 1 and Figure 5, the power
deposited in the spherical phantom and human model
depends sensitively on the excitation mode, but much
less on the coil designs. The agreement is mirrored in
the B

1
maps of the rotating modes (Fig. 4), indicating
that load power deposition is indeed a suitable proxy for
transmit field strength. As opposed to results obtained
for an 8-channel loop array using a quasistatic approach
in a cylindrical phantom (37), we found that the homo-
geneous CP
1
excitation does not exhibit the strongest
power deposition in the load, but rather the next highest
CP
2
, which is very close to the theoretical maximum
load power in the phantom. This difference can be par-
tially attributed to wave effects that are not captured by
the quasistatic calculations, but are mainly due to the
relatively strong radiative losses occurring in the CP
1
mode. In the EP mode, the loop array interacts with the
load almost exclusively via the loop segments resembling
the endrings, resulting in a very low power deposition.
As the stripline array does not have any endring seg-
ments, the power deposition is even lower.
The human model simulations show a stronger power
deposition in the head compared to the spherical phan-
tom, however, the overall qualitative progression depicted
in Figure 5 shows a strong resemblance between the two
load configurations. The CP
2
mode, while still exhibiting
6 Kuehne et al.
the strongest losses of the investigated CP modes, does
not represent the theoretical maximum losses. This is due
to the ellipsoid shape of the head that would require cor-
respondingly varying amplitudes fed into the respective
coil elements to achieve maximum power deposition.
The stripline array displays a strong power deposition
in the body of the Duke model when driven in the EP
mode which is not seen in the loop array, indicating that
the whole body represents a significant load to the stri-
pline elements. This power is not just absorbed in the
shoulders, but distributed throughout the whole body,
underlining the need to model the full extent of the
human body in electromagnetic simulations. A more
thorough discussion of this phenomenon will be given
in the upcoming section on radiated power.
Coupling/Decoupling Losses
For the phantom load configurations, the loop array dis-
plays a superior decoupling behavior compared to the
stripline configuration. However, the losses introduced
by the decoupling structures strongly differ, and it is evi-
dent that the decoupling structure losses for the stripline
array depend strongly on the imposed excitation mode,
whereas the decoupling losses of the loop array show a
much less varying behavior. This is due to the fact that
the decoupling inductances of the loop array simply rep-
resent a series resistance in each coil element, with the
associated losses behaving similar to the other coil series
resistance terms. Conversely, the current in the decou-
pling LC circuits of the striplines is proportional to the
voltage differential across its terminals, which in turn
strongly depends on the neighboring coil voltages and
thus the excitation mode used to drive the array. This
implies corresponding losses, which are in fact partly
responsible for the apparent decoupling performance.
Increased losses in the LC circuit, as could be introduced
by choosing a lower inductance solenoid, would further
increase the channel isolation in terms of smaller off-
Table 1
Relative Contributions of the Investigated Loss Mechanisms for All Array/Load Configurations and Selected Excitation Modes.
Excitation mode Extrema
Losses ([%)] CP
1
CP
2
CP
3
CP
4
EP Max. Min.
Loop Phantom Phantom 53.92 66.37 41.69 24.01 19.36 66.80 19.36
Plastic 7.33 7.62 13.83 17.71 13.73 17.78 7.31
Metal 1.91 2.09 3.81 4.89 3.28 4.95 1.89
Coupling 0.98 2.06 2.99 5.94 25.23 25.30 0.96
Capacitors 14.48 15.45 27.50 35.05 25.63 35.05 14.47
Dec. Inductors 6.84 6.40 10.18 12.40 12.76 12.80 6.38
Radiated 14.53 0.00 0.01 0.00 0.00 14.55 0.00
Error 0.015 0.002 0.001 0.002 0.007 0.018 0.001
Stripline Phantom Phantom 49.41 63.62 36.81 22.43 1.43 63.77 1.43
PTFE 1.47 1.17 1.65 1.86 0.01 1.86 0.01
Metal 10.49 8.26 11.47 12.81 0.11 12.96 0.11
Coupling 2.01 0.71 2.88 5.53 79.84 79.85 0.67
Capacitors 18.13 14.20 19.45 21.65 0.27 21.65 0.26
Dec. LCs 4.47 12.04 27.75 35.73 0.00 35.75 0.00
Radiated 14.03 0.01 0.00 0.00 18.32 18.34 0.00
Error 0.005 0.010 0.013 0.011 0.014 0.088 0.008
Loop Duke Head 59.15 64.93 42.28 27.50 26.42 75.99 16.99
Body 4.30 1.94 0.42 0.17 2.29 4.99 0.03
Plastic 5.84 6.28 12.86 15.55 12.90 17.69 4.65
Metal 1.55 1.75 3.61 4.39 3.18 5.10 1.14
Coupling 4.14 6.65 5.37 10.44 18.35 20.89 1.19
Capacitors 11.75 12.90 26.03 31.31 24.70 35.86 9.11
Dec. Inductors 5.41 5.15 9.24 10.63 12.03 12.56 3.67
Radiated 7.84 0.39 0.18 0.02 0.14 9.38 0.00
Error 0.005 0.000 0.001 0.005 0.001 0.021 0.001
Stripline Duke Head 60.30 65.51 39.04 26.19 13.64 76.78 13.14
Body 7.73 0.76 0.40 0.12 56.07 56.29 0.02
PTFE 1.09 1.04 1.60 1.80 0.78 2.11 0.60
Metal 7.80 7.36 11.14 12.49 5.56 14.90 4.06
Coupling 1.35 1.68 2.04 3.76 5.03 5.48 0.10
Coil Caps 13.52 12.58 18.85 21.08 9.69 24.57 7.39
Dec. LCs 3.35 11.02 26.86 34.54 0.03 38.79 0.00
Radiated 4.88 0.07 0.10 0.03 9.21 10.81 0.00
Error 0.011 0.016 0.029 0.029 0.003 0.111 0.002
Additionally, the power imbalance is shown in the Error row. In the last two columns, the theoretical limits for any arbitrary driv-
ing vector v as given by the minimum and maximum Eigenvalues of the corresponding loss matrices are given. Power deposition
in the simulations involving the human model have been split into head and body power, with the neck being included in the head
volume.
Transmit Array Loss Analysis 7
diagonal elements of the scattering matrix, however, at
the cost of higher losses in the decoupling structure and
reduced coil performance. This observation challenges
the common assumption of low coupling being equiva-
lent to high coil efficiency. It rather highlights the need
to carefully evaluate and compare losses introduced into
the array by the various decoupling structures available.
Both arrays show strong coupling losses in the EP mode,
which is explained by the previously mentioned weak
interaction with the coil load in this mode. For the stri-
plines, this leads to the majority of the input power
being reflected and thus lost due to coupling.
Using the Duke model, the loop array shows a small
increase in coil element coupling, however, no significant
qualitative changes for coupling and decoupling loss dis-
tributions can be observed. This behavior is contrasted by
the stripline array showing marked differences close to the
EP mode. The strong reflection and accompanying reduc-
tion of all dissipative loss terms (metal, substrate, capaci-
tors) is replaced by the dominating power deposition in
the body and intrinsic resistive coil losses comparable to
other driving modes. This suggests that only the body of
the Duke model represents a significant load to the array
in this mode, which allows electromagnetic energy to
propagate along the coil elements and into the body,
which is not possible for the phantom load and thus
explains the lack of power deposition.
Radiative Losses
The radiated power component of the CP
1
mode is sim-
ilar to previously published results for both arrays when
loaded with the spherical phantom. A similarly sized
and loaded free-standing transverse electromagnetic
coil (18) radiates 18% of the input power as compared to
approximately 14% in our investigation, which included
the bore and gradient shield, and thus partially mitigated
radiation losses in comparison. Similarly, we found that
simulations including the full human body exhibited
lower radiative losses. A part of the power being radiated
in phantom experiments is absorbed in the body outside
the head when the coil is loaded with a human as can
be seen from Table 1, where strong radiative losses in
phantom simulations correspond to a relatively higher
power deposition in the body. The radiative losses can
be understood in terms of the coil arrays coupling to
waveguide modes supported by the bore and gradient
shield (17). Separating the CP
1
mode into its two
orthogonal linear constituents, it can be seen that each of
the linear modes excites a TE11 mode in the gradient
shield and bore, which explains the relatively strong
radiative loss of the homogeneous mode.
When driven in the EP mode, the phantom-loaded stri-
pline array radiates 18% of the forward power, which is
lowered to 9% with a human load. Simultaneously, this
comes with the already mentioned 56% power deposi-
tion in the body. The radiation and power deposition in
the body again indicate a traveling wave. Using the EP
excitation, the stripline-array couples to the TM01 mode,
which can be supported by the bore but not by the
empty gradient shield. When loaded with the phantom,
coil fields extending out of the gradient shield excite a
dampened TM01 mode leading to the observed 18%
radiation. With a human load, the TM01 mode can prop-
agate inside the gradient shield and body as well, lead-
ing to the strong power deposition throughout the whole
model. Conventional coils coupling to traveling wave
modes have been previously observed (17,50), and our
findings further underline the necessity to exactly model
not only the coil and patient geometries, but the sur-
rounding structures as well. Furthermore, these results
suggest that the cylindrical RF shields commonly used
to minimize radiative losses are insufficient at suppress-
ing power coupling into waveguide modes, and addi-
tional shielding structures should be considered.
To aid an intuitive understanding of the radiative and
traveling wave phenomena, a visualization of the arrays
behavior under different driving conditions and loads is
FIG. 4. Simulated B

1
maps of
the counterclockwise rotating
(CP) and equal phase (EP)
modes for all investigated coil
array and load configurations. All
images are scaled to 1 kW for-
ward power. The enhancement
occurring in the occipital part of
the head for the stripline array
can be attributed to the close
proximity of the conductor seg-
ments being positioned on the
inside of the former annulus.
[Color figure can be viewed in
the online issue, which is avail-
able at wileyonlinelibrary.com.]
8 Kuehne et al.
provided as a supporting video animation (Supporting
Information Video V1).
Metal Losses
The metal conductor losses of approximately 2% for the
loop array and 8% for the stripline array when driven in
the CP
1
mode are larger than those reported by Liu et al.
(18) of less than 0.1% for a high-pass birdcage, where the
difficulty of metallic loss modeling in FDTD was already
acknowledged. Despite the use of the good conductor
approximation, the losses reported in our work may still
have a large associated uncertainty. Especially, the cor-
rect approximation of lateral skin effect contributions
(51) in copper strip conductors is questionable, as the
copper loss models utilized only consider the classical
skin effect and neglect the discontinuities appearing at
the conductor boundaries. Initial investigations have
shown that the deviation from theoretical loss expecta-
tions strongly varies with the conductor geometry, com-
putational mesh, and even software package used for the
simulation (52). For flat conductors with a rectangular
cross section, the losses were found to be generally
underestimated by a factor of approximately 23 at 128
MHz for conductor widths and mesh resolutions used in
our loop array simulations. As the effect increases with
frequency, these values represent a lower limit for the
error. As the relative contribution of the lateral skin effect
decreases with increasing conductor width due to a
smaller current distribution curvature at the conductor
boundaries, the stripline array simulation is expected to
be less affected. However, further research is required to
exactly quantify these errors. Instead of relying on the
results from three-dimensional (3D)-electromagnetic sim-
ulations, it may be advantageous to model coil conduc-
tors as perfectly conducting in the 3D simulation, and
introduce losses as lumped resistors derived from analyti-
cal considerations or measurements for the given conduc-
tor geometry and coil size, as the latter approach has
been shown to provide accurate estimates for intrinsic
coil losses (44) in receive arrays. Power deposition occur-
ring in the shield and bore structures is not affected by
these computational difficulties, however, they only play
a very minor role and the results are strongly dominated
by the coil conductor losses.
The higher metal losses in the stripline array seem coun-
terintuitive at first glance, as the conductor widths of the
loop array are smaller and the resistivity is consequently
higher. This discrepancy can be partially explained by the
aforementioned loss underestimation occurring in narrow
flat conductors, which impacts metal losses in the loop
array copper traces more strongly than the wider stripline
conductors. Additionally, metal losses constitute a higher
fraction of a striplines intrinsic coil losses as compared to
a loop. The loop losses are dominated by the relatively
FIG. 5. Power balance distribution
for all simulated configurations
with varying nearest neighbor
phase increments and equal
amplitudes for all channels. The
positions of the Birdcage (CP)
modes are marked, a 0

phase
difference corresponds to an
equal phase drive (EP). The
parameter space between 180

and 360

(not shown), containing


the clockwise rotating modes, dis-
plays an exactly mirrored behavior
for the phantom simulations, and
only minor deviations from the mir-
rored progression for the Duke
simulations. This is due to the
(approximate) cylindrical symmetry
of the setup and the resultingly
degenerate Eigenmodes.
Transmit Array Loss Analysis 9
high number of capacitors and accompanying solder joints,
followed by inductor and former substrate losses. The use
of lower loss substrate, fewer solder joints, and parallel
capacitors (44) lead to a relatively higher impact of metal
conductor losses in the stripline array.
Substrate Losses
Losses in the coil former are significantly higher for the
loop array. This is due to the loop arrays plastic former
having a dissipation factor almost two orders of magnitude
above the PTFE former of the stripline array. The strong
FIG. 6. Simulated transmit SENSE pulse power balances for selective excitation patterns with a 20

flip angle, expressed in percent of


forward power. For each transmit array, three different use cases were investigated: a checkerboard excitation in the spherical phantom
(top), a homogeneous slice excitation of the same phantom (middle) and a homogeneous slice excitation in the Duke model (bottom).
The averaged power balance contributions are given in parentheses in the legends, and the simulated excitation patterns are shown in
the top right corners. Comparing both arrays reveals that while the individual power loss contributions significantly vary, the absolute
power deposited in the phantom/head is almost identical for the same excitation pattern. This indicates a highly similar array behavior
in parallel excitation performance with respect to global SAR. [Color figure can be viewed in the online issue, which is available at
wileyonlinelibrary.com.]
10 Kuehne et al.
electric fields inside the stripline substrate partially offset
this advantage, and prevent the use of plastic or other rela-
tively high loss materials as former materials, as the coil
performance would be severely degraded by the substrate
losses. However, it becomes evident that the loop array
would benefit from a lower loss substrate as well.
Transmit SENSE
Comparing the power balance for the checkerboard and
homogeneous excitation patterns in the phantom using
the same coil (Fig. 6), it becomes evident that up to about
2.1 ms, i.e., for more than 85% of the pulse duration, the
individual loss terms vary within relatively narrow bands
around their average value. In between the two pulses,
these average values differ, albeit not dramatically. Only
for the last 0.3 ms, corresponding to central k-space exci-
tation, this behavior changes: in both pulses the high-
frequency fluctuations of the loss distribution disappear
and all terms approach stable values, possibly well out-
side their previous variation bands. In the phantom, these
final values are almost identical for both pulses; they are
characterized by a relative increase in phantom power as
well as radiated power, with a simultaneous decrease in
the remaining loss terms. The similarity in the late seg-
ments of the transmit SENSE pulses can be attributed to
the fact that central k-space excitation requires a relatively
even B
1
distribution, which is reached by driving vectors
similar to the CP
1
mode. Although the individual aver-
ages of the loss contributions vary between the two arrays,
this qualitative behavior is evident in both configurations.
The simulations involving the human model show a
stronger fluctuation, owing to the reduced symmetry and
more complex field distributions in the human body.
Further Applications
Besides the applications shown, power correlation matri-
ces can be utilized to adjust SAR values to account for
the radiated power component occurring in simulations,
which is important for scenarios relying on 3D electro-
magnetic simulations for safety assessment (32). For
whole-body SAR calculations, the sum of the tissue loss
and radiated power matrices would provide a conserva-
tive estimate of power deposited in the body. Similarly,
local SAR for any arbitrary excitation could be scaled by
a factor (P
rad
P
mat
)=P
mat
, which can be easily calculated
using the power correlation matrices.
Furthermore, power dissipation in sensitive compo-
nents can be tracked for arbitrary pulses, thus helping to
prevent equipment damage and remain within the coils
operating specifications. Similar to power and SAR con-
straints, these hardware limits could be directly enforced
inside parallel transmit pulse calculation algorithms (53).
Finally, owing to the strong parallels between power
deposition during transmission and noise during recep-
tion via the fluctuation-dissipation theorem (54), the pre-
sented formalism could be as well applied to receive coil
arrays to investigate array noise contributions.
CONCLUSION
Electromagnetic losses in multichannel transmit coil arrays
depend on the amplitudes and phases used to drive each
individual channel. We have derived a framework of
power correlation matrices that provides straightforward
access to the power balance and all loss terms for arbitrary
array excitations using simple matrix calculations. Worst
case power imbalance estimates enable a straight-forward
plausibility check for electromagnetic simulations of arbi-
trary complexity; and an upper limit on the imbalance
could be used as one criterion for safety-relevant simula-
tions. Loss contributions have been shown to vary signifi-
cantly between different excitations, allowing a deeper
insight into loss mechanisms than considering a single
mode only. Components and structures exhibiting domi-
nating losses are readily identified, which can assist in
developing informed decisions on coil array component
and material choice. The complexity of loss mechanism
interplay as evidenced here highlights the need for meticu-
lous loss modeling to capture all power deposition phe-
nomena. Inclusion of all losses leads to more accurate
predictions of transmit array coil performance and SAR
behavior, and the methods presented will assist in design-
ing more efficient and safe transmit coil arrays.
ACKNOWLEDGMENT
The funding sources had no influence on the scientific
scope nor the outcome of this study.
REFERENCES
1. Edelstein WA, Glover GH, Hardy CJ, Redington RW. The intrinsic
signal-to-noise ratio in NMR imaging. Magn Reson Med 1986;3:604618.
2. Vernickel P, R oschmann P, Findeklee C, L udeke KM, Leussler C,
Overweg J, Katscher U, Grasslin I, Sch unemann K. Eight-channel trans-
mit/receive body MRI coil at 3T. Magn Reson Med 2007;58:381389.
3. Roemer PB, Edelstein WA, Hayes CE, Souza SP, Mueller OM. The
NMR phased array. Magn Reson Med 1990;16:192225.
4. Zhu Y. Parallel excitation with an array of transmit coils. Magn Reson
Med 2004;51:775784.
5. Kriegl R, Ginefri J-C, Poirier-Quinot M, Darrasse L, Goluch S, Kuehne
A, Moser E, Laistler E. Novel inductive decoupling technique for flex-
ible transceiver arrays of monolithic transmission line resonators.
Magn Reson Med 2014. doi: 10.1002/mrm.25260.
6. Lee RF, Giaquinto RO, Hardy CJ. Coupling and decoupling theory
and its application to the MRI phased array. Magn Reson Med 2002;
48:203213.
7. Wu B, Qu P, Wang C, Yuan J, Shen GX. Interconnecting L/C compo-
nents for decoupling and its application to low-field open MRI array.
Concepts Magn Reson Part B Magn Reson Eng 2007;31B:116126.
8. Avdievich NI. Transceiver-phased arrays for human brain studies at 7
T. Appl Magn Reson 2011;41:483506.
9. Avdievich NI, Pan JW, Hetherington HP. Resonant inductive decou-
pling (RID) for transceiver arrays to compensate for both reactive and
resistive components of the mutual impedance. NMR Biomed 2013;
26:15471554.
10. Constantinides C, Angeli S. Elimination of mutual inductance in
NMR phased arrays: the paddle design revisited. J Magn Reson 2012;
222:5967.
11. Li Y, Xie Z, Pang Y, Vigneron D, Zhang X. ICE decoupling technique
for RF coil array designs. Med Phys 2011;38:4086.
12. Hoult DI, Kolansky G, Kripiakevich D, King SB. The NMR multi-
transmit phased array: a Cartesian feedback approach. J Magn Reson
2004;171:6470.
13. Kurpad KN, Wright SM, Boskamp EB. RF current element design for
independent control of current amplitude and phase in transmit
phased arrays. Concepts Magn Reson Part B Magn Reson Eng 2006;
29B:7583.
14. Kirilina E, K uhne A, Lindel T, Hoffmann W, Rhein KH, Riemer T,
Seifert F. Current CONtrolled Transmit And Receive Coil Elements
(CONTAR) for parallel acquisition and parallel excitation techniques
at high-field MRI. Appl Magn Reson 2011;41:507523.
Transmit Array Loss Analysis 11
15. Raaijmakers AJE, Ipek O, Klomp DWJ, Possanzini C, Harvey PR,
Lagendijk JJW, van den Berg CAT. Design of a radiative surface coil
array element at 7 T: the single-side adapted dipole antenna. Magn
Reson Med 2011;66:14881497.
16. Brunner DO, De Zanche N, Fr ohlich J, Paska J, Pruessmann KP. Trav-
elling-wave nuclear magnetic resonance. Nature 2009;457:994998.
17. Zhang B, Sodickson DK, Lattanzi R, Duan Q, Stoeckel B, Wiggins GC.
Whole body traveling wave magnetic resonance imaging at high field
strength: homogeneity, efficiency, and energy deposition as compared
with traditional excitation mechanisms. Magn Reson Med 2012;67:
11831193.
18. Liu W, Kao CP, Collins CM, Smith MB, Yang QX. On consideration
of radiated power in RF field simulations for MRI. Magn Reson Med
2013;69:290294.
19. Haase A, Odoj F, Von Kienlin M, et al. NMR probeheads for in vivo
applications. Concepts Magn Reson 2000;12:361388.
20. Seifert F, Ittermann B. Uncertainties of local SAR determination in
parallel transmission MRI. In Proceedings of the 21st Annual Meeting
of ISMRM, Salt Lake City, Utah, USA, 2013. p. 2827.
21. Collins CM, Wang Z. Calculation of radiofrequency electromagnetic
fields and their effects in MRI of human subjects. Magn Reson Med
2011;65:14701482.
22. Yee KS. Numerical solution of initial boundary value problems
involving Maxwells equations in isotropic media. IEEE Trans Anten-
nas Propag 1966;14:302307.
23. Liao ZF, Kong-Liang H, Bai-Po Y, Yuan YF. A transmitting boundary
for transient wave analyses. Sci Sin A 1984;27:10631076.
24. Berenger JP. A perfectly matched layer for the absorption of electro-
magnetic waves. J Comput Phys 1994;114:185200.
25. Kozlov M, Turner R. Optimization of geometry for a dual-row MRI
array at 400 MHz. In 2013 IEEE MTT-S International Microwave
Workshop Series on RF and Wireless Technologies for Biomedical
Healthcare Applications, Singapore, 2013. pp. 13.
26. Kozlov M, Turner R. Fast MRI coil analysis based on 3-D electromag-
netic and RF circuit co-simulation. J Magn Reson 2009;200:147152.
27. Lemdiasov RA, Obi AA, Ludwig R. A numerical postprocessing pro-
cedure for analyzing radio frequency MRI coils. Concepts Magn
Reson Part A 2011;38A:133147.
28. Seifert F, Lindel T, Ullmann P. Efficient and comprehensive EM field
simulation procedure for pTX experiments. In Proceedings of the
20th Annual Meeting of ISMRM, Melbourne, Victoria, Australia,
2012. p. 2777.
29. Seifert F, W ubbeler G, Junge S, Ittermann B, Rinneberg H. Patient
safety concept for multichannel transmit coils. J Magn Reson Imaging
2007;26:13151321.
30. Zhu Y, Alon L, Deniz CM, Brown R, Sodickson DK. System and SAR
characterization in parallel RF transmission. Magn Reson Med 2012;
67:13671378.
31. Deniz CM, Alon L, Brown R, Sodickson DK, Zhu Y. Specific absorp-
tion rate benefits of including measured electric field interactions in
parallel excitation pulse design. Magn Reson Med 2012;67:164174.
32. Eichfelder G, Gebhardt M. Local specific absorption rate control for
parallel transmission by virtual observation points. Magn Reson Med
2011;66:14681476.
33. Kuehne A, Goluch S, Moser E, Laistler E. Power balance considera-
tions for RF transmit coil arrays. In Proceedings of the ISMRM,
Milan, Italy, 2014. p. 0551.
34. Balanis CA. Advanced engineering electromagnetics, 2nd ed. Hobo-
ken, New Jersey: Wiley; 2012.
35. Davis WA. Radio frequency circuit design, 2nd ed. Hoboken, New
Jersey: Wiley; 2011. pp 2528.
36. Bardati F, Borrani A, Gerardino A, Lovisolo GA. SAR optimization in
a phased array radiofrequency hyperthermia system. IEEE Trans
Biomed Eng 1995;42:12011207.
37. King SB, Varosi SM, Duensing GR. Eigenmode analysis for under-
standing phased array coils and their limits. Concepts Magn Reson
Part B Magn Reson Eng 2006;29B:4249.
38. King SB, Varosi SM, Duensing GR. Optimum SNR data compression
in hardware using an Eigencoil array. Magn Reson Med 2010;63:
13461356.
39. Alagappan V, Nistler J, Adalsteinsson E, et al. Degenerate mode
band-pass birdcage coil for accelerated parallel excitation. Magn
Reson Med 2007;57:11481158.
40. De Moerloose J, De Zutter D. Poyntings theorem for the finite-
difference-time-domain method. Microwave Opt Technol Lett 1995;8:
257260.
41. Adriany G, Van de Moortele PF, Wiesinger F, et al. Transmit and
receive transmission line arrays for 7 Tesla parallel imaging. Magn
Reson Med 2005;53:434445.
42. Guerin B, Gebhardt M, Serano P, Adalsteinsson E, Hamm M, Pfeuffer
J, Nistler J, Wald LL. Comparison of simulated parallel transmit body
arrays at 3 T using excitation uniformity, global SAR, local SAR, and
power efficiency metrics. Magn Reson Med, in press. Doi: 10.1002/
mrm.25243.
43. Chamberlin K, Gordon L. Modeling good conductors using the finite-
difference, time-domain technique. IEEE Trans Electromagn Compat
1995;37:210216.
44. Kumar A, Edelstein WA, Bottomley PA. Noise figure limits for circu-
lar loop MR coils. Magn Reson Med 2009;61:12011209.
45. Bolz RE. CRC handbook of tables for applied engineering science,
2nd ed. Boca Raton, Florida: CRC Press; 1973.
46. Hasgall P, Neufeld E, Gosselin M, Klingenb ock A, Kuster N. ITIS
database for thermal and electromagnetic parameters of biological tis-
sues, Version 2.4. Available at www.itis.ethz.ch/database. Accessed
on July 24, 2014.
47. Katscher U, B ornert P, Leussler C, van den Brink JS. Transmit
SENSE. Magn Reson Med 2003;49:144150.
48. Grissom W, Yip CY, Zhang Z, Stenger VA, Fessler JA, Noll DC. Spa-
tial domain method for the design of RF pulses in multicoil parallel
excitation. Magn Reson Med 2006;56:620629.
49. Celuch-Marcysiak M. Extended study of Poynting theorem and reci-
procity on nonuniform FDTD meshes. IEE Proc Sci Meas Technol
2004;151:452455.
50. Zhang B, Sodickson DK, Duan Q, Wiggins G. Contribution of the
inherent traveling wave in 7T to large FOV imaging. In Proceedings
of the 18th Annual Meeting of ISMRM, Stockholm, Sweden, 2010. p.
1480.
51. Giovannetti G, Hartwig V, Landini L, Santarelli MF. Classical and lat-
eral skin effect contributions estimation in strip MR coils. Concepts
Magn Reson Part B Magn Reson Eng 2012;41B:5761.
52. Kuehne A, Waiczies H, Moser E, Laistler E. Skin effect estimation
accuracy in FDTD coil simulations. In Proceedings of the ISMRM,
Milan, Italy, 2014. p. 1363.
53. Brunner DO, Pruessmann KP. Optimal design of multiple-channel RF
pulses under strict power and SAR constraints. Magn Reson Med
2010;63:12801291.
54. Brown R, Wang Y, Spincemaille P, Lee RF. On the noise correlation
matrix for multiple radio frequency coils. Magn Reson Med 2007;58:
218224.
SUPPORTING INFORMATION
Additional Supporting Information may be found in the online
version of this article.
Supporting Video S1: Animation of the propagating transverse
electromagnetic elds due to different excitation modes for both
coil arrays and loads.
12 Kuehne et al.