IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 62, NO.

1, JANUARY 2014 119

On the Analysis and Design of Recongurable
Multimode MIMO Microstrip Antennas
Carlos Redondo and Leandro de Haro, Member, IEEE
AbstractA novel multiport and multimode quasi analyt-
ical model is derived to analyze and to provide comprehensive
guidelines to design recongurable multimode multiple-input
multiple-output (MIMO) microstrip antennas. The model shows
to be an efcient and accurate method to gain physical insight
into the multimode MIMO microstrip behavior. Numerical and
experimental examples illustrate the validity of the model and the
usefulness of the design guidelines.
Index TermsEquivalent circuits, loaded antennas, microstrip
antennas, modeling, multiple-input multiple-output (MIMO)
systems.
I. INTRODUCTION
D
URING the last decade, multiple-input multiple-output
(MIMO) systems have received considerable interest as a
means to enhance the capacity of wireless systems in rich mul-
tipath environments [1]. Among the different main aspects to
be considered for obtaining diversity gain there is the correla-
tion between elements [2]. On the other hand, the main inu-
ence of the propagating scenario on the overall system perfor-
mance justies the considerable interest that this latter subject
has received [3], [4]. Certainly, even with uncorrelated antenna
elements there will not be capacity improvement under some
scenarios [5], [6] and [7]. However, this fact does not diminish
the importance of the correlations role in the design process. In-
deed, in [8] an adaptive antenna system capable of controlling
dynamically its correlation and coupling properties through re-
actance loaded radiators to adapt to the channel distribution is
proposed.
Whereas spatial and polarization diversity has been usually
employed to decrease the correlation between antenna elements,
other solutions based on multimode antennas have been pro-
posed [9], [10]. A structure which shows essentially a multi-
mode behavior is a microstrip antenna and its use for diver-
sity purposes has been already suggested [11]. Recently, there
have been some proposals making use of multimode microstrip
Manuscript received December 03, 2012; revised June 14, 2013; accepted
October 14, 2013. Date of publication November 05, 2013; date of current
version December 31, 2013. This work was supported by Spanish Ministry
of Education and Innovation (MICINN), SICOMORO project with reference
TEC2011-28789-C02-01.
C. Redondo is with the RF Department, Instituto Nacional de Tcnica Aero-
espacial, 28850 Madrid, Spain (e-mail: redondoac@inta.es).
L. de Haro is with the Department of Ingeniera Audiovisual y Comuni-
caciones, Universidad Politcnica de Madrid, E.U.I.T. de Telecomunicacin,
28031 Madrid, Spain (e-mail: leandro.deharo.ariet@upm.es).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TAP.2013.2288975
patch antennas in MIMO systems [12], [13] and [14]. The ap-
proach followed in this paper is of a different nature and it is
mainly based on a combination of an equivalent circuit model
and an extended cavity model (the rst one has currently gained
new attention [15]), instead of resorting to full-wave simula-
tions through commercial tools. Following this modus operandi
would signicantly reduce the overall time effort in the design
process and at the same time would give invaluable physical in-
sight benets of the multimode MIMO microstrip antennas be-
havior. Based on this last advantage, we can provide not only a
specic antenna design, but also general design guide rules, to
conceive this type of MIMO systems, including recongurable
ones. Notwithstanding this fact, it shall be highlighted that these
general design guide rules attempt to improve the systempattern
diversity, not to dynamically maximize the capacity providing
the optimal performance to different scenarios (as in [8]), as they
are provided without assuming any channel distribution.
This paper is organized as follows. In Section II the combi-
nation of the multiport network model and the extended cavity
model is presented. Section III presents the design guidelines
and an example of application, whereas Section IV shows ex-
perimental results. Finally, Section Vprovides some concluding
statements.
II. MULTIMODE MIMO CAVITY MODEL
A. Theoretical Foundations and Overview
As we want to improve the pattern diversity of the MIMO
system by means of the multimode behavior, it is desirable to
nd ways to displace the resonance frequency of each mode
to place them at the same frequency band while keeping them
as independent as possible. This objective can be accomplished
with an appropriate loading of the cavity with reactive loads and
metallic posts, or through the alteration of the geometry of the
radiation patch. These perturbations can also be put in operation
to modify the correlation between the different antenna elements
or the overall geometrical dimensions. In order to analyze the
effects of all these kinds of perturbations and to take advantage
of them in the design process, a quasi analytical model, based
on network and extended cavity models, has been derived to
analyze the elds inside the cavity and the radiated ones.
As the network and the extended cavity model dealt herein
are tightly linked to the cavity model, a few words are neces-
sary regarding the latter. The cavity model [16], [17] supposes
that the microstrip antenna can be modelled as a thin cavity sur-
rounded by magnetic walls in which the only possible modes
are transverse magnetic (TM) ones (only the z-directed elec-
tric eld component along the thin thickness is present due to
the boundary conditions imposed). The z-directed electric eld
0018-926X 2013 IEEE
120 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 62, NO. 1, JANUARY 2014
component is expressed in terms of an innite linear combina-
tion of coefcients and eigenfunctions, with each combination
of coefcient and eigenfunction representing the contribution of
each mode. This is the starting formulation of the cavity
model. However, due to the fact that the summation series con-
verge slowly, common implementations of the cavity model ne-
glect this summation. The approach taken is to model the be-
havior of the microstrip antenna over the narrow band of fre-
quencies around the resonant frequencies of each mode
through a parallel RLC circuit in series with an inductive term
(usually approximated for simplicity, avoiding the double sum-
mation series at the expense of loosing information) which rep-
resents the contribution of the higher order modes. It shall be
noticed that in loaded microstrip antennas this parameter shall
not be approximated as it plays a main role in the resonant fre-
quency of the loaded element, as the impedance is not longer
dominated by the original resonant mode. In the quasi analyt-
ical model, we come back to the original formulation and carry
out the summation series, which is slower than the previously
described one but it is essential for our purposes.
We shall highlight at this point that the eigenfunctions are
orthogonal ones, which means they behave as independent sub-
channels, so ideally and neglecting at rst the coupling between
modes due to the reactive loading process and the limitation in
feeding independently each of them, increasing the number of
employed modes will increase at the same rate the number of in-
dependent subchannels, and therefore the pattern diversity of the
system. Regarding this last reasoning, it should be commented
that the MIMO capacity of waveguide channels has been exten-
sively dealt with in [18].
In order to account for the effects of a loaded cavity we have
derived a mixture of two analytical models. The rst one is the
equivalent circuit one or network model, in which the overall
system is modelled as an unloaded cavity with N external ports
to which the different loads are attached. The second one is an
extended cavity model based on deriving a new set of mode
functions to represent the eld distribution inside the loaded
cavity, through forcing the boundary conditions imposed by the
reactive loads, and later on computing the coefcients associ-
ated to the electromagnetic coupling between the new mode
functions and the feed ports.
The equivalent circuit option can be developed in different
ways. The one developed herein is based in a mixture between
the generalized cavity model [19] and the multiport network
model [20]. The radiation, conductor and dielectric losses are
considered through the effective loss tangent, like the general-
ized cavity model whereas the external loads are introduced in
the model following a procedure similar to the multiport net-
work model. Perturbation of canonical geometries of the patch
can be modelled through the segmentation and desegmentation
method.
Regarding the second option, it was rst proposed in
[21][23] and more in depth in [24]. As well as in the cavity
model, the rst approach taken is to provide a formulation
with an innite summation series that afterwards is neglected
in order to reduce the computation time and cost. Conversely,
we carry out the summation series, which provides further
exibility.
Concerning both options and in order to calculate the radi-
ated elds, the power radiated and later on the envelope corre-
lation between different feed ports, a combination of the equiv-
alent circuit option and the fast Fourier transform (FFT) will be
employed.
Finally, it shall be pointed out that whereas the network model
is very useful due to its simplicity, its capability to make use of
generic complex impedances, which is essential for example to
compute the envelope correlation, and its versatility provided
through the segmentation/desegmentation methods to analyze
complex composite shapes, it does not provide the same phys-
ical insight as the extended cavity model which deals with a full
modal electromagnetic formulation of the loaded microstrip an-
tenna. Moreover, its use far from the resonant frequency of a
mode could provide inaccurate results, as it is based on
the quality factor of an unloaded cavity. On the other hand, the
extended cavity model can provide accurate results far from the
resonant frequency of a mode and it is more suitable to
properly locate further feeds and reactive loads in an already
loaded cavity or to qualitatively estimate the value of these last
ones, but it is restricted to purely reactive loads. Thus, both op-
tions can be readily combined to avoid its limitations. Notwith-
standing this fact, the quasi analytical model herein presented
inherits the fundamental restrictions of the cavity model: it is
only accurate with electrically thin microstrip antennas, com-
paring them with the wavelength, and it overestimates the feed
reactance when placed in the nearness of the edge of the mi-
crostrip antenna.
B. The Network Model
Let us suppose the more general conguration in which we
have a microstrip antenna with ports, feed ports
and - loads. The input impedance on each of this feed
ports can be calculated as follows.
1) Calculate the self impedance on each of the
ports. The general impedance is computed as follows:
(1)
(2)
(3)
where is the resonant wave number or eigenvalue for
the th mode, is the effective relative dielectric per-
mittivity, is the effective loss tangent, is the elec-
tric eld mode vector or eigenfunction for the th mode,
is the excitation electric current density and is com-
puted as but using instead of and intre-
grating over area.
REDONDO AND DE HARO: ON THE ANALYSIS AND DESIGN OF RECONFIGURABLE MULTIMODE MIMO MICROSTRIP ANTENNAS 121
2) Calculate the mutual impedance between the ports,
taking into account that it is a reciprocal network, and
computing successively the coefcient in each port
and in all the remaining ones, to nally compute in
each iteration .
3) Construct an impedance matrix associated with this
port network taking into account the equivalent circuit
boundary conditions forced by the - loads, i.e., through
matching the voltage on the loads with the voltage due
to the self and mutual impedances, as shown in (4) at the
bottom of the page.
4) Solve the system of equations to obtain the relation be-
tween the voltage and current on each of the feed ports,
which is the sought input impedance, while all the re-
maining feed ports have a matched load connected. In order
to carry out this task, we have to modify the previous
impedance matrix to account for the effects of the matched
loads, each time per each feed port, so in order to com-
pute the input impedance of feed port we have to modify
through all the elements placed in the
diagonal where can be any value included in the closed
interval except . Once the boundary conditions im-
posed by the matched loads are applied, we can compute
the input impedance of each feed port by means of the
following general equation:
(5)
where is the determinant of the modied
impedance matrix and is the cofactor or adjunct
of the element of the impedance matrix which is equal
to the minor , the reduced determinant that is formed
by omitting the th row and column.
As it can be noticed from the previous formulation, a novel
mixed mechanism to account for the losses has been em-
ployed. The radiation, dielectric and conductor losses are
included through the effective loss tangent , like the
generalized cavity model, and through external loads in the
case of generic complex impedances and matched loads, as
from the multiport network model.
C. Basis Functions of the Loaded Cavity
When reactive loads are inserted inside the cavity, it is
necessary to force the new boundary conditions claimed by
these ones. As a consequence, the rst step in this procedure
starts with the formulation of these new boundary conditions
which can be expressed as the determinant of an impedance
matrix, such as the one derived in Section II-B, but this time
only dealing with reactive loads, as shown in (6) at the bottom
of the page, where is the self reactance of the port,
the mutual reactance between ports and and is the load
reactance attached to port .
Once this equation is formulated, it is necessary to nd the
frequencies to which the boundary conditions equation is satis-
ed. Based on these frequencies (there is at least one frequency
to which the boundary condition is satised per each
mode) newwave numbers are calculated. These newwave num-
bers are employed to formulate the mode functions of the
loaded cavity as a linear combination of the orthonormal base
eigenfunctions of the unloaded one, as stated in the following
equations:
(7)
(8)
where is the electric eld mode vector or eigenfunction
for the th mode of the original unloaded cavity and is
the electric current density on the reactive load. Regarding
the new mode functions , it shall be pointed out that in
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
(4)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
(6)
122 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 62, NO. 1, JANUARY 2014
general they would not be orthornormal. In any case, we should
normalize them through the following operation:
(9)
As the boundary conditions imposed by the reactive loads are
considered in the new mode functions, which account for their
electromagnetic effects, the former ones can be removed from
the problem.
Once done, the following task is to calculate the coupling
coefcient between each function and the feed port:
(10)
Finally it only remains to carry out the linear combination
of each pair of associated coefcient and loaded mode function
to obtain the z-directed electric eld component of the loaded
cavity
(11)
It shall be pointed out that in case we have more than one feed
port, we can straightforwardly apply the previously presented
Section II-B only considering feed ports and making use of the
new mode functions, so mixing up both models. Thus in this
case, the general impedance is computed as follows:
(12)
where and are any of the feed ports, so the associated sym-
metric impedance matrix has a dimension. It only re-
mains to carry out step number 4 of Section II-B to obtain the
input impedance of each feed port.
D. Far Zone Radiated Fields
It shall be highlighted that until now, we have only modelled
the elds inside the loaded cavity. In order to compute the far
zone radiated elds, we may follow a procedure with two main
steps.
1) Compute the near zone elds on the periphery of the
microstrip antenna by means of (12) and (5) and apply
Schelkunoffs Field Equivalence Principle [25] in order to
obtain the equivalent current sources.
2) Calculate the far zone elds from the equivalent current
sources.
Therefore we shall start evaluating the eld on the patch
boundary taking into account the Nyquist theorem between
current equivalent samples. In this case, the maximum allow-
able distance between samples is function among others of the
spatial variation of the loaded modes and the frequency
considered. As an assumption, the equivalent current is sup-
posed to be distributed in a slot of width the thickness of the
microstrip antenna along the border of the patch (hypothesis
usually made to consider the effective extension of the size of
any microstrip antenna to take into account the fringe elds in
the contour).
In order to know the far zone elds from the equivalent cur-
rents, we compute the discrete Fourier transform (DFT) of a
2-D current distribution sequence located on the patch
boundary
(13)
where and represent the number of DFT bins considered
to calculate the DFT per each dimension.
As it can be noticed comparing the DFT equation with the
equation of the radiation vector or vector function, there is a
correspondence between each DFT bin with a spatial angular
position. By means of this correspondence, the far zone elds
of the loaded cavity can be obtained.
Supposing that the equivalent magnetic current is dis-
tributed in the plane, we obtain the following relationship:
(14)
(15)
where and are the spatial distance between equiva-
lent current samples in - and -direction and is the radiation
vector function for the magnetic currents.
It must be highlighted that only the DFT bins which
satisfy the following relationship are representative for our pur-
poses (visible region):
(16)
E. Envelope Correlation Computation
The envelope correlation coefcient provides information re-
garding the diversity performance and in this sense, it is useful
for determining the optimum placements of the feeds and reac-
tive loads. The envelope correlation of a two antennas system
REDONDO AND DE HARO: ON THE ANALYSIS AND DESIGN OF RECONFIGURABLE MULTIMODE MIMO MICROSTRIP ANTENNAS 123
can be computed by an assortment of approaches, with some
differences regarding the application range and the computing
time effort. A method that it is often used, which is based on the
radiated elds, is shown in the following [2]:
(17)
where is the port radiated eld, while all other feed
ports are terminated on their source impedance, and denotes
the Hermitian product.
It shall be noticed, that whereas the former equation is very
time consuming, it can be applied without assuming any distri-
bution of the incoming waves.
Alternatively, the envelope correlation coefcient can be
computed from the antenna system S parameters [26]
(18)
whereas this last equation leads to an important reduction on the
computing process time, it is only applicable when assuming
a uniform distribution of the waves and the use of lossless
antennas.
On the other hand and also based, as the last equation, on
the S parameters and the law of energy conservation, a general
envelope correlation formula for the general case of antenna
elements was derived in [27]
(19)
In order to compute the envelope correlation, we shall force
the boundary conditions imposed by the reactive loads and re-
duce the system of equations to a square matrix with as many
rows and columns as the number of feed ports. Whereas this re-
duction task is already done when employing the modied set of
basis functions of the loaded cavity, it is not the case when using
only the network model. In this last case, this task can be done
in general form through the following set of matrix operations.
Suppose that we have a system of linear equations with a
square impedance matrix where the feed locations corre-
spond to the indexes included in the closed interval , and
the reactive loads to the ones. Forcing the boundary
conditions concerning the reactive loads, the system of linear
equations is transformed into one in which all the impedance
matrix diagonal elements whose index is greater than have
added their corresponding reactive load , where can be
any value included in .
Afterwards we shall split the system of linear equations into
two subsystems through the row, so we can formulate the
reactive loads currents in terms of the currents in the
feed ports , shown in (20) at the bottom of the page.
The former matrix equation can be reformulated as follows:
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
(21)
Or more compactly
(22)
Thus, we have obtained the current in the reactive loads in
terms of the current in the feed ports; it just needs to solve the
other subsystem in terms of to reduce the overall system of
equations to an square matrix
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
(23)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
(20)
124 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 62, NO. 1, JANUARY 2014
Or more succinctly
(24)
Thus, we have condensed the information of the original
impedance matrix into an matrix , making
it suitable to be directly employed to compute the envelope
correlation.
Finally, the associated scattering parameters matrix can be
obtained from the impedance one through the well known
relationship
(25)
where is the identity matrix and the normalized
impedance one.
III. DESIGN GUIDELINES OF MULTIMODE MIMO
MICROSTRIP ANTENNAS
Along the present paper we have provided the foundations
required to analyze multimode MIMO microstrip antennas in
quasi analytical form and to outline a general synthesis proce-
dure to design them. Throughout this section, our intention is
to provide an easy to follow guideline regarding the design pro-
cedure and an example applied to a rectangular multimode mi-
crostrip antenna operating in the 2.42.5 GHz industrial, scien-
tic and medical (ISM) band.
It shall be highlighted that thanks to the quasi analytical
model presented, all the procedures described from here on out
can be carried out such as the overall design time effort will be
signicantly lower than making use of full-wave simulations
through commercial tools, with which the time effort would not
be acceptable. Moreover the knowledge acquired of the mode
distribution and the loading effects behavior allows nding
better solutions to given objectives.
A. On the Selection of the Involved Cavity Modes
The rst step is to select the patch dimensions and the sub-
strate permittivity and thickness, due to their main role deter-
mining the resonant frequency and the bandwidth of the mi-
crostrip antenna. This subject has been thoroughly studied (e.g.,
[28], [29]), so it will not be repeated here. In any case, it should
be taken into account which modes are desired to work simulta-
neously at the operating frequency. As previously commented,
the modes eigenfunctions are orthogonal ones, so at rst
an ideally any choice will be an optimal one as they behave as
independent subchannels. Notwithstanding this fact, special at-
tention is required regarding the frequency separation between
their resonant frequency and the operational one, because later
on we will have to displace them to this last one. Concerning
this feature and as a general rule of thumb, a 15% of frequency
displacement can easily be achieved, so it can be considered as
a reasonable starting point.
In the matter of the rectangular microstrip antennas resonant
frequencies, a glance to the predicted ones (see Fig. 1) for the
Fig. 1. Square patch on foam (17.5 17.5 0.3 cm): modes resonant
frequencies.
modes based on the formula provided in [28] shows that
using high modes provides many advantages regarding
the number of orthogonal modes that can be made working in
the same frequency band, the relative proximity between their
resonant frequency and thus, concerning the potential pattern
diversity of the overall system.
Concerning Fig. 1, it has been obtained using [28]
(26)
where is the speed of light, and are integer values corre-
sponding to the mode, and are the dimensions of
the rectangular patch, and are respectively the length
and width extension due to the fringe elds calculated through
[30] and is the effective relative dielectric permittivity com-
puted as described in [28], [31], and [32].
Notwithstanding this conclusion, if we want to employ high
modes for a xed frequency band, we will need to use
bigger microstrip antennas at rst. Whereas physical constraints
could not be a problem in particular applications (e.g., a vehicle
in a dense urban area operating in the ISM band), the size incre-
ment would not be desirable beforehand for many applications,
so it will be necessary to reach a compromise between the high
modes employed and the physical size restrictions. In
any case, it should be noticed that there are many options that
can be used to reduce the microstrip antenna dimensions ranging
from increasing the substrate permittivity (at the expense of re-
ducing the radiation efciency and the bandwidth) through the
use of reactive loads, such as shorting pin diodes, up to me-
andering the microstrip patch or the ground plane. An option
closely linked with this paper is the usage of reactive loads or
physical perturbations (which actually have an equivalent re-
active load behavior associated to them) to displace downwards
the resonance frequency of a specic set of cavity modes. All the
guidelines respecting the location of the reactive loads or phys-
ical perturbations that are going to be provided in Section III-C
are as well readily applicable to this task of size reduction.
On the subject of the example of application, we have selected
a square microstrip antenna, so we have pairs of modes
REDONDO AND DE HARO: ON THE ANALYSIS AND DESIGN OF RECONFIGURABLE MULTIMODE MIMO MICROSTRIP ANTENNAS 125
with the same resonance frequency. We have selected for illus-
trative purposes and simplicity the , and
modes, due to their resonance frequency and radiation pattern.
Thus, the designed patch on a foamsubstrate has the dimensions
17.5 17.5 0.3 cm (1.43 1.43 0.02 ) and accordingly,
the resonance frequencies as predicted by (26) for the
, and modes are respectively 2.302 and 2.440 GHz
(this last one for both and modes). Consequently,
later on we will have to perturb the original cavity with a re-
active load in order to displace the resonance frequency of the
mode to the 2.42.5 GHz ISM band.
Regarding the substrate and taking into account the accuracy
limitations of the cavity model in terms of thickness, it has been
chosen foam 3 mm. Thus, whereas the microstrip antenna will
not be made thick in terms of wavelength, due to the degradation
of the model accuracy, the radiated power and the impedance
bandwidth can be improved using a low permittivity dielectric
substrate.
B. On the Location of the Feed Ports
As the computation time effort associated to (3) is extremely
lowand based on the resonant frequency estimation provided by
(26), the rst task to carry out is to mesh the microstrip patch an-
tenna and to calculate the reection coefcient concerning every
feed position in the frequencies around the estimated resonant
ones.
On the other hand, as we look for low values of envelope
correlation, it is essential that the location of each of the feed
ports (one per involved mode) is selected so one
mode impedance is good matched while all the remaining ones
not, so as less as possible energy is coupled to those last ones.
As it is necessary to compute the eld distribution of isolated
modes, we carry out this activity through (1), (2) and (3)
which are evaluated on the locations of the previous patch mesh,
once per each selected mode of interest and only con-
sidering each mode alone (thus without performing the double
summation series). Thus, we obtain a correspondence between
feed placements and electric eld distribution and impedance
of a set of specic modes, which with a threshold con-
cerning its value we termed as masking map.
Afterwards, we focus on each mode of interest at its res-
onant frequency and we set a minimum admissible threshold
to it concerning the reection coefcient and another one re-
garding the maximum allowable level of electric eld distribu-
tion and reection coefcient for the remaining modes
of interest. Applying these thresholds to the feed positions and
cross mapping the placements that satisfy both requirements per
each mode, it nally provides us the best regions for each
feed port. It shall be pointed out that many times, mostly while
dealing with several high order modes, it is necessary to
reach a compromise between both thresholds, as not always the
feed can be placed at the best position concerning the reection
coefcient and at the same in all the electric eld nulls regarding
the remaining modes. The nal feed placements are determined
based on (19).
In order to set the thresholds values, which play a main role in
the design process, it is suggested to apply an iterative method
Fig. 2. Reection coefcient versus feed location. (a) modefre-
quency: 2.32 GHz (b) and modesfrequency: 2.46 GHz.
Fig. 3. Masking mapfeed location. (a) . (b) . (c) .
starting from very stringent requirements concerning the reec-
tion coefcient of the mode under consideration (e.g.,
20 dB) and a null value for the remaining
modes of interest. If none feed placements satisfy both require-
ments, they are iteratively and progressively relaxed until there
are feed placements which satisfy them. It shall be noticed that
the time required to carry out this task is negligible as all the
necessary information is already computed and it only remains
to perform cross mapping between feed locations applying the
specied thresholds.
Applying the procedure described on the suggested example,
we start meshing the microstrip patch and getting a correspon-
dence between feed positions along the microstrip patch and
a reection coefcient level associated to them. In Fig. 2 it is
shown this correspondence per each resonant frequency where
the -coordinates represent the feed position whereas the
-coordinate illustrate the associated reection coefcient (trun-
cated for values lower than 35 dB).
On the other hand, in Fig. 3 it is shown the masking map
achieved per each mode of interest after reaching a com-
promise regarding the thresholds. For example, for the
mode it is represented those positions where simultaneously the
associated and modes satisfy the electric eld and
impedance limit, whereas for the mode the map is built
based on and modes and so on for the mode.
Applying cross mapping between Figs. 2 and 3 (once per each
mode) and considering each feed isolated, we nd feed locations
whose reection coefcients are better than 15 dB. Among
126 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 62, NO. 1, JANUARY 2014
the regions of feed positions which full both thresholds, (19)
is used in order to know which combination provides lower en-
velope correlation level. Finally, the best feed locations
in cm are respectively for the and modes
(9.44, 8.06), (8.06, 12.19), and (12.19, 8.06).
C. On the Location of the Reactive Loads and Their Associated
Reactive Values
At this step in the procedure, it is time to get on with the fre-
quency displacement of the modes of interest whose as-
sociated resonant frequency is different than the desired one.
It shall be noticed that this task can be performed without sig-
nicantly altering the radiation patterns of the unloaded cavity
(e.g., in [33] the resonant frequency is shifted over a 1.5 to 1
range without seriously changing the radiation pattern).
In order to carry out our purpose, we compute the eld distri-
bution concerning each isolated mode on the positions
of the microstrip patch mesh making use of (1) and (2) in order
to nd locations where the electric eld distribution regarding
each mode is highest (a threshold value is used to mark off these
regions), so the inuence of the perturbation applied to it will be
more important. On the other hand, we make use of the masking
map described in Section III-B, but this time the limit value can
be a different one. Once again, it is necessary to reach a com-
promise between both thresholds (an iterative procedure similar
to the one described in Section III-B is suggested), or what it is
the same, within obtaining a high inuence in a specic
mode while altering as slightly as possible the remaining modes
of interest. Applying these thresholds and cross mapping the
remaining placements that satisfy both requirements per each
mode, it nally provides us the best regions for each re-
active load.
Concerning the value of the reactive load to cause the desired
frequency shift, a thorough analysis on this theme can be done
and valuable conclusions can be obtained using (5) and (12).
Summing it up, the greatest displacement will be caused when
the overall contribution of the remaining modes and the reac-
tive load shows a very low inductive or capacitive value and
conversely, as it can be deduced from the qualitative analysis
provided in [21] and [24]. Moreover, it should be pointed out
that a net inductive contribution will cause a frequency shift to-
wards upper frequencies, whereas a net capacitive one to lower
ones.
Applying this procedure on the proposed example and in
order to displace the resonance frequency of the mode to
the 2.42.5 GHz ISM band, it is found out that the best location
for the reactive load is the centre of the square patch, in which
the mode shows its -directed electric eld intensity
maximum value, while the and modes have a null
eld.
Regarding the reactive value, a net inductive contribution
with a relatively low reactance value will be needed. Based on
this primary seed, a simple brute-force search procedure is used,
as the associated elapsed time is negligible, to seek the reactive
value, using as objective function parameters the desired nal
frequency and the reection coefcient level. In this case, the
optimum inductive reactance is j14 and this causes that at the
frequency of 2.406 GHz the reection coefcient is 24.98 dB.
Fig. 4. Reection coefcient. Isolated feed+load pairs.
On the other hand, whereas it is not necessary to our spe-
cic application to displace the and modes, as they
are already placed at the 2.42.5 GHz ISM band, in case we
wanted to make them recongurable, crossing the placements
where their eld is higher than 98% of its peak value with
a masking map under the compromise of reection coefcients
between 0 and 1.75 dB and eld value between 0%4%
of the maximum value for the remaining modes of in-
terest, it provides the reactive loads locations in cm(3.10, 11.78)
and (11.78, 3.10) for the and the modes respec-
tively. Regarding the reactive load values, any high reactance
value will be appropriate. For example, a value of pro-
duces respectively a negligible displacement of the and
the modes to 2.444 GHz, while keeping a reection co-
efcient below 15 dB.
In Fig. 4 it is shown the reection coefcient per each iso-
lated pair of feed port and reactive load in the frequency band
of 1.53.5 GHz. It shall be pointed out that the reection co-
efcients from both and feed-load pairs match as
expected due to the symmetry.
D. On the Verication and Final Adjustments
Once the appropriate locations of the feed ports and reac-
tive loads have been established (alongside the corresponding
suitable values of these last ones), it is time to test the antenna
system as a whole in terms of the reection coefcient level as-
sociated to each feed port and the envelope correlation between
them, making use of (5), (12), and (19).
Usually, even after having followed the previously presented
guidelines, it is advisable to slightly readjust the location of the
feed ports, as there are several undesired effects related to the
loss of orthogonality in the feeding process. Among them, the
most harmful ones regarding our objectives are the increment
of the envelope correlation level and the deterioration of the re-
ection coefcient value associated to a specic feed
mode.
Notwithstanding this fact, there are manifold procedures
to compensate the aforementioned undesirable effects. Con-
REDONDO AND DE HARO: ON THE ANALYSIS AND DESIGN OF RECONFIGURABLE MULTIMODE MIMO MICROSTRIP ANTENNAS 127
Fig. 5. Multimode MIMO microstrip system. (a) Reection coefcient. (b) En-
velope correlation.
cerning the growth of the envelope correlation coefcient,
some perturbations can be applied to the cavity (e.g., reactive
loads or geometrical modications) in order to decouple the
feeds. On the other hand and regarding the reection coef-
cient level associated to each specic feed mode, the
presence of the remaining ports and loads has slightly altered
it; indeed the level of perturbation is highly linked with the
fullment or not of the presented requirements (essentially
based on the orthogonality preservation). Usually, the level of
correlation between the feeds shall be very low, if negligible
effect in the reection coefcient associated to each port is
wanted when assembling the whole system. The procedure
to improve the reection coefcient consists in applying an
optimization process based on the gradient descent algorithm
(it is a good option in terms of the number of iterations and
the computation time due to the presumable proximity to the
local maxima) which takes into account the feed position, the
reection coefcient and the envelope correlation level. Our
objective function will be to improve the reection coefcient
while maintaining low values of envelope correlation. This last
requirement must be imposed because we could be improving
the reection coefcient increasing the coupling of energy from
a feed to more than the one desired mode.
Concerning our example, the reection coefcient and en-
velope correlation shown in Fig. 5 are obtained when all the
ports and reactive loads are attached. As it can be seen at the
2.42.5 GHz ISM band, the reection coefcient levels asso-
ciated with the and feed ports have been signif-
icantly degraded with respect to the case of isolated pairs of
feed ports and reactive loads presented in Fig. 4. The reasons
surrounding this undesired effect must be mainly sought in the
mutual coupling between these feed ports, due to the inability
of exciting only and in complete isolation each of the desired
modes, and in a lesser extent in this example, because of
the loss of orthogonality of the modes due to the reactive loading
process. Indeed, as it can be noticed in Fig. 5, the higher values
of envelope correlation between the and feed ports
Fig. 6. MIMO system: (a) variable length stub used (b) connectors view.
reveal this loss of independence between these feed ports, fact
that is not present with the feed port.
As the reection coefcient and envelope correlation levels
obtained with this initial design are good enough, in this case it
is not necessary to readjust the feeds and reactive loads place-
ments, so the previously presented optimization procedure is not
applied in this example. However, this one should be useful in
a more complex design.
On the other hand, it shall be pointed out that the envelope
correlation showed in Fig. 5(b) and the ones presented in Fig. 8
are computed using (19), assuming that the hypothetical deploy-
ment environment is a dense urban area, where the radiation is
coming with uniform angular distribution. Whereas, using (19)
in our example is not completely rigorous and accurate, as we
are not dealing with a lossless antenna, due to the fact that it is a
highly efcient one (the radiation efciency depends primarily
on the substrate thickness and permittivity [34] and we are using
a thick foam substrate and the losses associated to
the experimental implementation of the reactive load are negli-
gible) the loss in accuracy is not signicant [35].
IV. EXPERIMENTAL EXAMPLE
In order to verify the accuracy of the quasi analytical model
presented, the designed MIMO system has been built. It shall
be noticed that only the reactive load is used as it is not
necessary to our specic application to displace the and
modes. Concerning the desired reactive load value
of j14 at 2.4 GHz, a variable length stub nished in a short load
has been used. The precise length and thus the sought reactance
value have been obtained through a coaxial line stretcher with
an overall adjustable length of 10 mm, as presented in Fig. 6(a).
This option has been chosen as a nearly constant phase value
in the frequency band of 1.53.5 GHz cannot be easily ob-
tained, due to the connectors length and the thickness of the
ground plane. Moreover, for illustration purposes, in Fig. 6(b),
it is shown as well the overall MIMO system port connections
during the vector network analyzer (VNA) measuring process.
In order to compare the measured results with the appropriate
simulated ones, once xed the stub length that provides the de-
sired reactance value at 2.4 GHz for the load, the stub
associated parameter measured in the frequency band of
1.53.5 GHz is used instead of the j14 xed one in the imple-
mented quasi analytical model simulations carried out.
The comparison between the simulated and measured reec-
tion coefcient is presented in Fig. 7. It must be highlighted,
that the results provided are exactly the simulated and measured
ones without any correction or adjustment on them. As it can
128 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 62, NO. 1, JANUARY 2014
Fig. 7. Reection coefcient. Simulated versus measured results. (a)
feed. (b) feed. (c) feed.
Fig. 8. Envelope correlation. Simulated versus computed from measured S pa-
rameters. (a) feed- feed. (b) feed- feed. (c)
feed- feed.
be seen, the maximum percentage relative error is below 1.4%
concerning the maximum adaptation frequency location in the
frequency band of interest.
On the other hand and regarding this time the envelope corre-
lation, the associated results are shown in Fig. 8. Anew, there is
a good agreement between the results pointing up the great level
of accuracy that can be obtained through the presented quasi an-
alytical model.
Besides, the low levels of envelope correlation between the
different feed ports obtained and its corresponding reection co-
efcient values validate the correctness of the designed MIMO
system to enhance the system diversity.
V. CONCLUSION
As far as we know, this is the rst time that a combination of
quasi analytical models (network and extended cavity ones) are
adapted and applied to the analysis and design of recongurable
multimode MIMO microstrip antennas with any number of feed
ports and complex impedance loads. It shall be pointed out, that
the way of combining and customizing the presented quasi an-
alytical models is itself a novel approach. Moreover, the usage
of the FFT implementation of the DFT in the extended cavity
model is a reasonable compromise in terms of precision and
computation time and it constitutes itself a new way to tackle
the problem of the radiated elds by a loaded cavity instead of
the choice suggested in [24].
On the other hand, an original comprehensive general
synthesis procedure to benet from the pattern diversity im-
provement achievable through the multimode antenna MIMO
system has been provided. This procedure has been validated
and showed to be very useful and accurate both through simula-
tions and experimental measurements. Taking into account the
accuracy of the quasi analytical model proposed and the overall
computation time effort associated to our implementation, it
makes this quasi analytical model an efcient alternative in
the design process over full-wave simulations by means of
commercial tools.
REFERENCES
[1] J. G. Foschini and M. J. Gans, On limits of wireless communications
in a fading environment when using multiple antennas, Wireless Per-
sonal Commun., vol. 6, no. 3, pp. 311335, Mar. 1998.
[2] R. G. Vaughan and J. B. Andersen, Antenna diversity in mobile
communications, IEEE Trans. Veh. Technol., vol. VT-36, no. 4, pp.
149172, Nov. 1987.
[3] D.-S. Shiu, G. J. Foschini, M. J. Gans, and J. M. Kahn, Fading corre-
lation and its effect on the capacity of multielement antenna systems,
IEEE Trans. Commun., vol. 48, no. 3, pp. 502513, Mar. 2000.
[4] P. Almers et al., Survey of channel and radio propagation models for
wireless MIMO systems, Eurasip J. Wireless Commun. Networking,
2007.
[5] D. Chizhik, G. J. Foschini, and R. A. Valenzuela, Capacities of mul-
tielement transmit and receive antennas: Correlations and keyholes,
Electron. Lett., vol. 36, no. 13, pp. 10991100, Jun. 22, 2000.
[6] D. Gesbert, H. Bolcskei, D. Gore, and A. Paulraj, MIMO wireless
channels: Capacity and performance prediction, in Proc. IEEE Global
Telecommun. Conf. (GLOBECOM), 2000, vol. 2, pp. 10831088.
[7] D. Chizhik, G. J. Foschini, M. J. Gans, and R. A. Valenzuela, Key-
holes, correlations, and capacities of multielement transmit and receive
antennas, IEEE Trans. Wireless Commun., vol. 1, no. 2, pp. 361368,
Apr. 2002.
[8] E. P. Tsakalaki, O. N. Alrabadi, C. B. Papadias, and R. Prasad, Adap-
tive reactance-controlled antenna systems for multi-input multi-output
applications, IET Microw. Antennas Propag., vol. 5, no. 8, pp.
975984, Jun. 2011.
[9] T. Svantesson, An antenna solution for MIMO channels: The multi-
mode antenna, in Conf. Rec. Asilomar Conf. Signals, Syst., Comput.,
Oct. 29Nov. 1 2000, vol. 2, pp. 16171621.
[10] C. Waldschmidt and W. Wiesbeck, Compact wide-band multimode
antennas for MIMOand diversity, IEEETrans. Antennas Propag., vol.
52, no. 8, pp. 19631969, Aug. 2004.
[11] R. G. Vaughan, Two port higher mode circular microstrip antennas,
IEEE Trans. Antennas Propag., vol. 36, no. 3, pp. 309321, Mar. 1988.
[12] J. Sarrazin, Y. Mahe, S. Avrillon, and S. Toutain, A new multimode
antenna for MIMO systems using a mode frequency convergence con-
cept, IEEE Trans. Antennas Propag., vol. 59, no. 12, pp. 44814489,
Dec. 2011.
[13] N. Herscovici, C. Christodoulou, E. Rajo-Iglesias, O. Quevedo-Teruel,
and M. Sanchez-Fernandez, Compact multimode patch antennas for
MIMO applications [wireless corner], IEEE Antennas Propag. Mag.,
vol. 50, no. 2, pp. 197205, Apr. 2008.
[14] O. Quevedo-Teruel, M. Sanchez-Fernandez, M. L. Pablo-Gonzalez,
and E. Rajo-Iglesias, Alternating radiation patterns to overcome
angle-of-arrival uncertainty [wireless corner], IEEE Antennas
Propag. Mag., vol. 52, no. 1, pp. 236242, Feb. 2010.
REDONDO AND DE HARO: ON THE ANALYSIS AND DESIGN OF RECONFIGURABLE MULTIMODE MIMO MICROSTRIP ANTENNAS 129
[15] J. W. Wallace and R. Mehmood, On the accuracy of equivalent circuit
models for multi-antenna systems, IEEE Trans. Antennas Propag.,
vol. 60, no. 2, pp. 540547, Feb. 2012.
[16] Y. Lo, D. Solomon, and W. Richards, Theory and experiment on mi-
crostrip antennas, IEEE Trans. Antennas Propag., vol. AP-27, no. 2,
pp. 137145, Mar. 1979.
[17] W. Richards, Y. Lo, and D. Harrison, An improved theory for mi-
crostrip antennas and applications, IEEE Trans. Antennas Propag.,
vol. 29, no. 1, pp. 3846, Jan. 1981.
[18] S. Loyka, Multiantenna capacities of waveguide and cavity channels,
IEEE Trans. Veh. Technol., vol. 54, no. 3, pp. 863872, May 2005.
[19] V. Palanisamy and R. Garg, Analysis of arbitrarily shaped microstrip
patch antennas using segmentation technique and cavity model, IEEE
Trans. Antennas Propag., vol. 34, no. 10, pp. 12081213, Oct. 1986.
[20] A. Benalla and K. C. Gupta, Multiport network model and transmis-
sion characteristics of two-port rectangular microstrip patch antennas,
IEEE Trans. Antennas Propag., vol. 36, no. 10, pp. 13371342, Oct.
1988.
[21] W. Richards, S. Davidson, and S. Long, Dual-band reactively loaded
microstrip antenna, IEEE Trans. Antennas Propag., vol. 33, no. 5, pp.
556561, May 1985.
[22] W. Richards and S. Long, Pattern adaptation using loaded, dual-mode
microstrip antennas, in Proc. Antennas Propag. Soc. Int. Symp., Jun.
1985, vol. 23, pp. 9396.
[23] W. Richards and S. Long, Matching microstrip antennas using reac-
tive loads, in Antennas Propag. Soc. Int. Symp., Jun. 1986, vol. 24,
pp. 431434.
[24] W. F. Richards and Y. T. Lo, Theoretical and experimental investiga-
tion of a microstrip radiator with multiple lumped linear loads, Elec-
tromagnetics, vol. 3, no. 34, pp. 371385, Jul.Dec. 1983.
[25] S. A. Schelkunoff, Some equivalence theorems of electromagnetics
and their application to radiation problems, Bell Syst. Technol. J., vol.
15, no. 1, pp. 92112, Jan. 1936.
[26] S. Blanch, J. Romeu, and I. Corbella, Exact representation of antenna
system diversity performance from input parameter description, IEE
Electron. Lett., vol. 39, no. 9, pp. 705707, May 1, 2003.
[27] J. Thaysen and K. B. Jakobsen, Envelope correlation in
MIMO antenna array from scattering parameters, Microw. Opt.
Technol. Lett., vol. 48, no. 5, pp. 832834, May 2006.
[28] R. W. Dearnley and A. R. F. Barel, A comparison of models to de-
termine the resonant frequencies of a rectangular microstrip antenna,
IEEE Trans. Antennas Propag., vol. 37, no. 1, pp. 114118, Jan. 1989.
[29] P. Mythili and A. Das, Simple approach to determine resonant
frequencies of microstrip antennas, IEE Proc. Microw. Antennas
Propag., vol. 145, no. 2, pp. 159162, Apr. 1998.
[30] M. Kirschning, R. H. Jansen, and N. H. L. Koster, Accurate model for
open end effect of microstrip lines, Electron. Lett., vol. 17, no. 3, pp.
123125, Feb. 5, 1981.
[31] E. Hammerstad and O. Jensen, Accurate models for microstrip com-
puter-aided design, in 1980 IEEE MTT-S Int. Microwave Symp. Dig.,
May 2830, 1980, pp. 407409.
[32] M. Kirschning and R. H. Jansen, Accurate model for effective dielec-
tric constant of microstrip with validity up to millimetre-wave frequen-
cies, Electron. Lett., vol. 18, no. 6, pp. 272273, Mar. 18, 1982.
[33] D. Schaubert, F. Farrar, A. Sindoris, and S. Hayes, Microstrip an-
tennas with frequency agility and polarization diversity, IEEE Trans.
Antennas Propag., vol. 29, no. 1, pp. 118123, Jan. 1981.
[34] D. M. Pozar, Rigorous closed-form expressions for the surface wave
loss of printed antennas, Electron. Lett., vol. 26, no. 13, pp. 954956,
Jun. 21, 1990.
[35] P. Hallbjrner, The signicance of radiation efciencies when using
S-parameters to calculate the received signal correlation from two an-
tennas, IEEEAntennas Wireless Propag. Lett., vol. 4, pp. 9799, 2005.
Carlos Redondo was born in Oviedo, Spain, in
1978. He received the M.Sc. degree in electrical
engineering from Universidad Alfonso X el Sabio,
Villanueva de la Caada, Madrid, Spain, in 2001.
He is currently working towards the Ph.D. degree in
electrical engineering from Universidad Politcnica
de Madrid.
Currently, he is working with the RF Department,
Instituto Nacional de Tcnica Aeroespacial (INTA),
Torrejn de Ardoz, Madrid, Spain. His main research
interests are focused on tunable microstrip antenna
design, characterization and modeling.
Leandro de Haro (S90M90) was born in
Barcelona, Spain, in 1962. He received the M.Sc.
and Ph.D. degrees from the Universidad Politc-
nica de Madrid, Madrid, Spain, in 1986 and 1990,
respectively, both in electrical engineering.
He joined the Department of Seales, Sistemas
y Radiocomunicaciones, Universidad Politcnica
de Madrid, where he was an Assistant Professor
from 1990 to 2012, where he is currently a Professor
with the Department of Ingeniera Audiovisual y
Comunicaciones. His research areas are focused on
advanced antennas design and measurement, RF receivers modeling and radio
channel modeling and measurement.