International Journal of Microwave and Wireless Technologies, page 1 of 14.

# Cambridge University Press and the European Microwave Association, 2017

doi:10.1017/S1759078717000903

research paper

Correlation coefficient calculations for

MIMO antenna systems: a comparative study
mohammad s. sharawi1, abdelmoniem t. hassan1 and muhammad u. khan2

The correlation coefficient (r) is an important metric for the evaluation of multiple-input–multiple-output (MIMO) antenna
systems because it describes the relationship between the channels in a particular propagation environment. Highly correlated
channels will degrade the MIMO system performance. Various methods to calculate r are analyzed in this work using three
different types of antennas with different efficiencies and radiating properties. While the field-based method for finding r gives
the most accurate results and should be used all the time. It is independent of antenna efficiency. The the S-parameter-based
method might give reasonable results only when the antenna efficiency is very high and the patterns are separated in space,
but in most cases, it underestimates r values and thus is not a reliable method and should be avoided. Incorporating the
antenna efficiency can improve the r estimates using the S-parameters method under some specific conditions. The equivalent
circuit method provides reasonable results for symmetric antenna structures only, and is the most complex in formulation. As
part of the evaluation, two existing methods incorporating the radiation efficiency of the antennas are generalized to N-ports
for the first time. Although less accurate than the field-based method, these extensions allow the evaluation of r for N-port
antennas using only the radiation efficiency and the S-parameters of the antenna. The effect of the beam tilts on r estimation
is investigated for the first time.

Keywords: Correlation coefficient, MIMO, Isolation, Efficiency, Printed antennas

Received 7 April 2017; Revised 24 July 2017; Accepted 24 July 2017

I. INTRODUCTION uniform (i.e an isotropic environment). Thus, r is mainly a

function of the radiation pattern of each antenna element. A
Higher data rates and reliable communication systems are the simpler method for the computation of r is given in [2].
main pillars for new technologies. Multiple-input–multiple- This method used the S-parameters of the MIMO antenna
output (MIMO) systems are employed in current 4 G and system to find it. This method is derived under the assumption
future 5 G systems. The increase in the data rates and better that the MIMO antenna system is lossless. In [3], an extension
reliability offered by MIMO relies on the number of un- to N-elements is provided.
correlated channels available within the system. Among In practice, all printed antennas are lossy, thus using the
other factors, antenna design is also a contributing factor in S-parameter method to calculate r for such antennas results
inducing the correlation between channels and therefore in wrong values. In [4], the effect of the total radiation effi-
must be characterized. ciency is considered to estimate the worst possible correlation
The correlation coefficient (r) is a performance metric performance. In [1, 5] it is shown that r has no relation with
defined to characterize and compare the performance of the radiation efficiency, hence r should be calculated from
MIMO antenna systems in terms of how much correlation lossless S-parameters, which can be extracted from the mea-
they will add within the channels. A low value for r is sured S-parameters after calculating the losses in the
desired. Several methods were discussed in literature to evalu- antenna system. The challenge in this method is to calculate
ate r for MIMO antenna systems. First described in [1], r the loss resistance, which can be easily obtained if antenna
between two antenna elements is defined as a function of systems can be represented by their equivalent circuits [5].
the radiation pattern of each antenna element and the The equivalent circuit method for serial and parallel circuits
spatial power distribution of the signal in a propagation envir- of antennas to calculate r are investigated and applied for
onment. In most of the subsequent methods for evaluating r, two elements and only the series equivalent circuit was
the spatial power distribution of the signal is considered demonstrated on more than two element antenna systems
that were symmetrically spaced. It should be noted that the
1
Department of Electrical Engineering, King Fahd University of Petroleum and effect of using multiple lossy antennas in close proximity on
Minerals Dhahran, Dhahran, Al-Sharqia 31261, Saudi Arabia the efficiency was investigated in [6, 7]. Large deviation
2
SEECS, National University of Sciences & Technology (NUST), H-12, Islamabad, between field-based and S-parameter-based methods were
Pakistan
Corresponding author:
found in [8, 9] when considering practical printed circuit
M.S. Sharawi boards (PCB) antennas. But no further analysis was conducted
Email: msharawi@kfupm.edu.sa to compare and explain such deviations.

Downloaded from https://www.cambridge.org/core. KFUPM King Fahd University of Petroleum & Minerals, on 18 Sep 2017 at 08:53:50, subject to the Cambridge Core terms of use, available
1 at
https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1759078717000903
 2 mohammad s. sharawi, abdelmoniem t. hassan and muhammad u. khan

In this work, we analyze the performance of various r cal- is adopted), r can be calculated using [1]:
culation methods that appeared in [1–5], when applied to   
high, moderate, and low efficiency (h) antennas. Three   
 4p Fi (u, f)∗ Fj (u, f)dV
MIMO antenna types based on patches (low h), planar rij =     
 

 (4)
inverted F antennas (PIFA) (moderate h), and printed mono- F (u, f)dV 4p  Fj (u, f)dV
4p  i
poles (high h) are used to draw the conclusions on the r cal-
culation accuracy of these methods, their complexity and  
where ∗ is the Hermitian product, Fi (u, f) and Fj (u, f)
validity. This is chosen because most of MIMO antenna are the 3D radiated fields of the antennas when the ith
works in literature are derivatives of these three antenna and jth ports are excited and V is the solid angle. The
types. A new meandered four-element PIFA-like antenna is envelope correlation coefficient (re) can be approximated
designed and analyzed besides the half-circle monopole and by squaring (4).
the four-element patch-based MIMO antenna systems. All This method has one drawback, which is requiring mea-
designs are fabricated and measured. In addition, we extended sured 3D filed patterns of each antenna element. Thus,
the method in [4] to cover N-ports as well as extended the par- it might be relatively expensive (if no measurement
allel equivalent circuit method in [5, 10] to cover more than chamber is available), and requires time in collecting the
two elements. A study on the affect of tilted beams on r calcu- patterns at the frequency points of interest. However, this
lations is also conducted for the first time. Based on this work, is the most accurate and exact method for r calculations,
the usefulness and limitations of each of the four methods and can be applied on any number of antenna elements
along with their validity are highlighted, which will help with any efficiency values. This method should be used
antenna designers when characterizing their MIMO antenna all the time.
designs.

B) Correlation coefficient calculation using

II. CORRELATION COEFFICIENT S-parameters (method-II)
CALCULATION METHODS In [2], a relation between the radiated field (total power) and
the port parameters (S-parameters) is demonstrated for 100%
A) Correlation coefficient calculation using efficient antennas working in isotropic environments. The
the radiation patterns (method-I) method is valid only for finding r between two element
systems. In [3], it was generalized for N-elements and is
This method calculates r from the three-dimensional (3D)
far-field radiation patterns for each pair of the antenna ele-
ments. For any wireless environment r is found using [1]:
p p
−p 0
(XPRAu,m (u, f)A∗u,n (u, f)pu (u, f) + Af,m (u, f)A∗f,n (u, f)pf (u, f)) sin ud ud f
rmn =   , (1)
d2m d2n

where
 p p
 2
d2n = (XPRAu,n (u, f) pu (u, f) given by:
−p 0
 2
+ Af,n (u, f) pf (u, f)) sin ud ud f, (2)

− Nn=1 S∗ni Snj
rij =

     2 , (5)
 p p 1 − Nn=1 |Sni |2 1 − Nn=1 Snj 
 2
d2m = (XPRAu,m (u, f) pu (u, f)
−p 0
 2
+ Af,m (u, f) pf (u, f))sinud ud f, (3) where rij is the correlation between elements i and j, Sni, and
Snj are the S-parameters between each of the n various ports
Au,m (u, f), Af,m (u, f), Au,n (u, f) and Af,n (u, f) are the and elements i and j of MIMO antenna system.
elevation and azimuthal radiated field components for ele- This method has been used in most of the literature to find
ments n and m, respectively. ()∗ is the conjugate operator. r. The advantages of this method are its fast and ease of use as
XPR is the cross polarization discrimination factor showing it only depends on the port parameters of the antenna system.
the ratio between the vertically and horizontally polarized On the other hand, this method is unreliable when the radi-
power components of the incoming wave, and pu(u, f), ation efficiency of the antenna elements is not 100% and it
pf(u, f) are the angle of arrival power densities for u and f does not account for non-uniform directive patterns that are
angles, respectively. For an environment with a uniform dis- seen in most practical antennas (it should be mentioned
tribution of vertical and horizontal components of the incom- that this method does not account for the beam tilts that are
ing waves (i.e. isotropic environment, which is widely adopted the ones that affect the channel of the incoming signals, and
in rich multipath scenarios where MIMO systems operate, thus has a direct effect on the achieved capacity of the
and most works use this assumption in their derivations, system). Thus, this method usually underestimate the r
thus to compare under the same assumptions, this scenario values and should not be used.

Downloaded from https://www.cambridge.org/core. KFUPM King Fahd University of Petroleum & Minerals, on 18 Sep 2017 at 08:53:50, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1759078717000903
 correlation coefficient calculations for mimo antenna systems 3

C) Correlation coefficient calculation using not accurate. For a four-element (or more) MIMO antenna
S-parameters and radiation efficiencies with identical elements that can be represented as a series
equivalent circuit, the loss resistance is found using [5]:
(method-III)
4
This method gives more reliable calculations than using only h∗rad,1 (1 − hrad,1 ) j=1 k1j Re(Zl )
2
rloss = 4 2 , (7)
the S-parameters because it considers the effects of the radi- (hrad,1 − h∗rad,1 ) j=2 k1j
ation efficiency, but instead of giving the exact value, it gives
an upper bound for r. In [4], the calculation of r was where Zl is the antenna load, which is the impedance of the
derived for two-element MIMO antennas, and here we gener- feeding cable, Re() is the real part operator, and k1j is the
alize it for N-elements as: ratio between the port voltages. For a four-element (or
 more) parallel equivalent circuit-based MIMO antenna with
− Nn=1 S∗ni Snj identical elements, the loss impedance is derived in this
rij,max =

     2
1 − Nn=1 |Sni |2 1 − Nn=1 Snj  hrad,i hrad,j work (Appendix B) and is given as
  

 1  h∗rad,1 (1 − hrad,1 ) 4j=2 k21j
1 rloss = 
+
. (8)
−1 −1 , (hrad,1 − h∗rad,1 ) 4j=1 k1j Re(Zl )
hrad,i hrad,j
(6)

where hrad,1 and hrad,2 are the radiation efficiencies of the two III. LITERATURE REVIEW
antenna elements, and the second term in (6) represents the
degree of uncertainty which is high when the radiation effi- A brief recent literature review (due to limited space) on the
ciencies are low, and thus the value of r can be .1. This methods used for finding r (or re) is conducted to show the
method provides better accuracies compared with method-II, ones often used in evaluating MIMO antenna systems as
but once the efficiency goes below 60% the errors start increas- well as the number and type of antenna elements used in
ing dramatically. The calculation of r using this method
such works. Table 1 shows this brief survey. It shows that
requires one extra measurement, which is the efficiency. This
the majority of works have focused on two-element MIMO
method does not account for the pattern shape or tilt,
similar to method-II, and thus it is also unreliable and
should be avoided. The extension of this method to Table 1. Summary of the literature review.
N-element and the derivation of (6) is given in Appendix A.
Reference No. of Element type Band center(s) r
elements Method

D) Correlation coefficient calculation based [11] 2 PIFA + 0.9 GHz Method I

Monopole
on the antenna equivalent circuit (method-IV) [12] 4 Folded dipoles 5.2 GHz Method I
Basically, when r is calculated using (4) the effect of the radi- [13] 3 dipoles/half 2.45 GHz Method
ation efficiency is canceled out because of normalization. In slot III
[14] 4 Printed 2 GHz Method I
other words, r calculations do not depend on the radiation
monopoles
efficiency [1, 5]. So, the exact value of r using the
[15] 2 PIFA 2.45 GHz Method I
S-parameters can be calculated by excluding the effect of the [16] 2 PIFA GSM, UMTS, Method
loss component in the measured S-parameters. Based on WLAN II
this concept, a method was proposed in [5], which extracts [17] 4 Patch + slot 2.45 GHz Method I
the lossless S-parameters of the antenna from its measured [18] 3 Printed loop WLAN Method
lossy S-parameters and by knowing the lossy equivalent antenna II
circuit of the antenna a more accurate estimation of r can [19] 2 Folded printed WLAN Method
be found. monopoles II
The difficult part in this method is estimating the loss com- [20] 2 Printed WLAN Method
ponents from the equivalent circuits of the antennas. This monopoles II
[21] 2 Printed 760 MHz Method
depends on the representation of the equivalent circuit
monopoles III
whether it is of series or parallel structure based on the [22] 2 Printed WLAN, Method
antenna type (as the loss resistance location will be a function monopoles WiMAX, II
of the location of the radiation resistance). Also, this method UWB
becomes very difficult if different antenna types are used, [23] 2 Printed 2 GHz – 7 GHz Method
because extra steps will be needed to find the different loss monopoles III
values, and thus more computational complexity in setting [24] 2 Loop + slot DVB, LTE Method
up the matrices. The modeling of this method takes time, antenna bands II
and thus it is considered the most complex compared all [25] 4 Slot WLAN, Method
WiMAX II
other methods. This method gives values closer to method-I
[26] 4 Loop LTE, WiMAX, Method
when the antenna elements are uniformly spaced and of the
WLAN, II
same type (two conditions followed in [5, 10]) If the two con- UMTS
ditions are not satisfied, the estimates from this method are

Downloaded from https://www.cambridge.org/core. KFUPM King Fahd University of Petroleum & Minerals, on 18 Sep 2017 at 08:53:50, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1759078717000903
 4 mohammad s. sharawi, abdelmoniem t. hassan and muhammad u. khan

antenna systems due to their need in 4 G wireless terminals. circuits models) with different radiation characteristics are
Most of the designs used similar antennas (i.e. monopoles, first designed in CST Microwave Studio. Later, they are fabri-
PIFAs, etc.) as they will have the worst coupling and correl- cated and their port and far-field parameters are measured.
ation except for [11] where they evaluated two different anten- S-parameters measurements are conducted at KFUPM using
nas on a mobile terminals at 900 MHz in the presence of head an Agilent (N9923A) vector network analyzer (VNA), while
and hand models. Method-II was the most widely used, which radiation patterns and efficiency measurements are conducted
shows that its ease of use was the driving force rather than at MVG-Italy using a Satimo Star-Lab chamber. While each
aiming for more accuracy. port was measured, the other ports were terminated with
50V. The measurement setup was maintained for all port
measurements of a single antenna and the same orientation
as shown in the antenna model geometry and measurement
IV. PERFORMANCE ANALYSIS OF setup picture is followed. The four methods are then applied
VARIOUS METHODS APPLIED TO to compute r from the simulated and measured parameters.
DIFFERENT MIMO ANTENNA
CONFIGURATIONS

To compare the results of different r calculation methods and

A) PIFA-based four-element MIMO antenna
come up with better understanding of their behavior, three Figure 1(a) shows the schematic diagram of a four-element
MIMO antenna systems (different types and equivalent PIFA MIMO antenna, with a new shape that has not appeared

Fig. 1. The four-element PIFA-based MIMO antenna system. (a) Geometry of the antenna (all dimensions are in mm), (b) the measurement setup and the
fabricated antenna, (c) measured S-parameters, and (d) the measured of E-total at 1.95 GHz.

Downloaded from https://www.cambridge.org/core. KFUPM King Fahd University of Petroleum & Minerals, on 18 Sep 2017 at 08:53:50, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1759078717000903
 correlation coefficient calculations for mimo antenna systems 5

Table 2. r for the PIFA-based four-element MIMO antenna at 1.95 GHz. PIFA antennas are widely used in mobile handsets and small
form factor wireless devices.
Method used r12 r13 r14
The r values of the given MIMO antenna system at its res-
Method-I (simulation) 0.1406 0.0851 0.1992 onant frequency are calculated using the four methods.
Measured-I (measured 3D far-field) 0.204 0.207 0.301 Table 2 shows these calculations. r values using method-I
Method-II 0.0847 0.0448 0.0292 are calculated from both the 3D radiation patterns obtained
Method-III 0.0476 0.1959 0.1780 from simulations as well as from measurements. The mea-
Method-IV 0.1476 0.1020 0.1545 sured values showed noticeable increase when compared
with the simulation. This can be attributed to the measure-
ment setup (mast below) and cable location (next to port
being measured). r values calculated using methods II, III,
in literature before. It is designed on an FR-4 substrate and IV are based on the measured S-parameters of the
with dimensions 110 × 60 × 0.8 mm3. A Parallel equivalent fabricated antenna and its measured radiation efficiency.
circuit is used to model this antenna. The measurement Methods II and III give inaccurate values for r. The values
setup and the fabricated design are shown in the inset of obtained using method-IV have close agreement with simu-
Fig. 1(b). Figure 1(c) shows that each antenna element reso- lated method-I results but slight differences start appearing
nates at 1.95 GHz with a bandwidth of 100 MHz and worst between element (1,3) and (1,4) because this method
case isolation of 12 dB. The measured 2D radiation patterns assumes symmetric element positioning. Since the antenna
of each antenna element are shown Fig. 1(d) with a has a moderate radiation efficiency, it is concluded that if an
maximum gain of about 1.65 dBi and efficiency of 60%. S-parameter-based method is to be used, method-IV is the

Fig. 2. The patch-based MIMO antenna system. (a) Geometry of the antenna (all dimensions are in mm), (b) the measurement setup and the fabricated antenna,
(c) measured S-parameters, and (d) the measured of E-total for each element.

Downloaded from https://www.cambridge.org/core. KFUPM King Fahd University of Petroleum & Minerals, on 18 Sep 2017 at 08:53:50, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1759078717000903
 6 mohammad s. sharawi, abdelmoniem t. hassan and muhammad u. khan

Table 3. r for the patch-based four-element MIMO antenna at 5.87 GHz. Table 4. r for the half-circle printed monopole-based four-element
MIMO antenna.
Method used r12 r13 r14
Method used r12 r13 r14
Method-I (simulation) 0.8291 0.5723 0.8525
Measured-I (measured 3D far-field) 0.903 0.802 0.806 Method-I (simulation) 0.31 0.20 0.26
Method-II 0.0048 0.0047 0.0005 Measured-I (measured 3D far-field) 0.2186 0.2029 0.2685
Method-III 1.3239 1.3461 1.3364 Method-II 0.1869 0.2094 0.2742
Method-IV 0.6728 0.6543 0.6656 Method-III 0.18 0.20 0.27
Method-IV 0.14 0.16 0.15

most suitable one for symmetric structures. Methods II and III

would result in inaccurate values of r for antennas with mod- B) Patch-based four-element MIMO antenna
erate radiation efficiency. None of the methods II–IV accounts A four-element patch-based MIMO antenna system is shown
for the effect of the beam tilts, and this is another factor of the in Fig. 2(a). The antenna is fabricated on an FR4 board of
deviations noticed from their use. Only method-I can account dielectric constant 4 and thickness 0.8 mm. Each patch
for such pattern tilts. element represents a parallel equivalent circuit. Figure 2(b)

Fig. 3. Half-circle printed monopole-based MIMO antenna system. (a) Geometry of the antenna (all dimensions are in mm), (b) the measurement setup and the
fabricated antenna, (c) measured S-parameters, and (d) the measured of E-total for each element.

Downloaded from https://www.cambridge.org/core. KFUPM King Fahd University of Petroleum & Minerals, on 18 Sep 2017 at 08:53:50, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1759078717000903
 correlation coefficient calculations for mimo antenna systems 7

Fig. 4. (a) Two dipoles with different tilting angles (a), (b) S-parameters of the two dipole antennas (a ¼ 0, a ¼ 10, a ¼ 15), and (c) the obtained r curves of the
two dipoles as a function of a using methods I and II.

Downloaded from https://www.cambridge.org/core. KFUPM King Fahd University of Petroleum & Minerals, on 18 Sep 2017 at 08:53:50, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1759078717000903
 8 mohammad s. sharawi, abdelmoniem t. hassan and muhammad u. khan

shows the measurement setup and the fabricated design. C) Half-circle printed monopole-based
Figure 2(c) shows a resonant frequency of 5.8 GHz with a four-element MIMO antenna
measured bandwidth of more than 50 MHz. The worst case
isolation between the antenna elements is 25 dB. The mea- A four-element MIMO antenna system consisting of half-circle
sured radiation patterns are shown in Fig. 2(d). At the reson- printed monopole elements is investigated. The antenna is
ant frequency, the measured maximum gain of each element is designed on a 1.5 mm-thick FR4 substrate. This antenna can
3.77 dBi, while its radiation efficiency is 52%. Patch antennas be represented by a series equivalent circuit. The geometry of
are used on relatively large wireless devices (i.e. behind laptop the antenna is shown in Fig. 3(a). The fabricated antenna and
screens) or vehicular platforms. They are used in some wire- the measurement setup are shown in Fig. 3(b). Figure 3(c)
less access points. shows the measured S-parameters with a resonant frequency
The values obtained for r for the antenna at 5.87 GHz of 2.4 GHz with a bandwidth of 210 MHz and worst case iso-
using different methods are tabulated in Table 3. Since lation of 9.5 dB. Figure 3(d) shows the 2D radiation pattern
method-II is based on 100% efficient antennas, the values of all the antenna elements measured at 2.4 GHz. The mea-
obtained from this method significantly deviate from the sured maximum gain of each element is found to be 3.12
ones obtained using method-I due to the low radiation effi- dBi, while the radiation efficiency is 81.5%. Monopole-based
ciency of the antenna elements and no relations with the antennas are used in wireless handheld devices among other
radiated patterns. The maximum r value is between antenna mobile terminals and wireless access points.
elements 1 and 3 when using method-I (0.45) compared Table 4 shows the r values obtained using different
with 0.0005 using method-II. Method-III gave an unrealistic methods. The maximum value of 0.3 was between elements
value of 1.3. Thus, such method proves itself to be useless to 1 and 2 as obtained from method-I (simulation). Good agree-
estimate the r for antennas with low radiation efficiencies. ment is observed between the values obtained for r using
Method-IV gave the closest values when compared with method-I from simulation and measured 3D radiation pat-
method-I but a deviation of 0.15 is observed. From these terns. Method-II shows an acceptable agreement with
results, it is concluded that method-I should only be used to method-I because the antennas have relatively high radiation
find r for MIMO antenna systems with low radiation efficiency as well as the beams are tilted, although the isolation
efficiencies. value of 9.5 dB was not high. Method-III is in line with

Fig. 5. (a) Two patches with different tilting angle, S-parameters of the two patch antennas, (b)a ¼ 0, (c)a ¼ 90, and (d) the r of the two patches.

Downloaded from https://www.cambridge.org/core. KFUPM King Fahd University of Petroleum & Minerals, on 18 Sep 2017 at 08:53:50, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1759078717000903
 correlation coefficient calculations for mimo antenna systems 9

method-II due to good radiation efficiency, while method-IV waves are coming with equal probability with uniform arrival
shows the largest deviation. This is due to the fact that angles, the antennas and the incident waves are matched in
method-IV assumes symmetric element patterns and posi- the polarization in addition to the fields are independent of
tions. From this study, it is concluded that for antennas the arrival angle [27]. Also the coupling between the elements
with high radiation efficiency, method-II could be used to cal- comes from two sources, the first is due to the current leakage
culate the r since it is computationally very simple only if the between the ports and the second is because of pattern
beams are intentionally tilted, otherwise, it will not work. induced coupling.
Thus, the only reliable method is method-I. Using two different MIMO antenna systems, the effect of
the antenna element beam tilt on r calculations is investigated.
Two wire dipoles and two patches are used.
V. EFFECT OF BEAM TILTS ON r The two dipoles are shown in Fig. 4(a). They resonate at
ESTIMATION 1 GHz with l/8 spacing between them (d), center-to-center.
The simulated port parameters are shown in Fig. 4(b) as a
The methods evaluated (I–IV) for r are based on the assump- function of the tilted angle (a). Note that the curves for Sxx
tion that the radio environment is isotropic, which means the and Sxy are overlapping. r is calculated in the band of

Fig. 6. S-parameters of the two patch antennas: (a) d ¼ l/8, (b) d ¼ l/4, (c) d ¼ l/2, and (d) the r of the two patches.

Downloaded from https://www.cambridge.org/core. KFUPM King Fahd University of Petroleum & Minerals, on 18 Sep 2017 at 08:53:50, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1759078717000903
 10 mohammad s. sharawi, abdelmoniem t. hassan and muhammad u. khan

operation and shown in Fig. 4(c). A notable difference is factor, thus it is the most complex compared with other
observed using methods I and II for the three tilted angles methods.
investigated. It is clear that the isolation is very low
between the dipoles, which only comes from the highly
omnidirectional patterns (as the ports do not share a ACKNOWLEDGEMENTS
common ground). However, r12 values are also low when
considering method-I since the radiation patterns are tilted This work was supported by the deanship of scientific research
due to the tilt angle (a) as well as reflected due to the pres- (DSR) at King Fahd University of Petroleum and Minerals
ence of the adjacent metallic structure. (KFUPM), Dhahran, Saudi Arabia, under project number
For the case of the two patch antennas in Fig. 5(a), the RG1419.
S-parameters when a ¼ 0 and 90 in Figs 5(b) and 5(c) are
almost the same which means the estimation of r using
method-II for both cases will give the same curves as shown
in Fig. 5(d). However, the values of r12 calculated using APPENDIX A
method-I are effected by the beam tilting (a), which is clear
in the same figure. Not only that, also the estimation of r12 To make method-III valid for N-elements, we provide the
using method-IV shows that it does not consider the effect steps to generalize it as part of this work contribution. The
of beam tilting (almost 0.2 difference when compared with total radiated field in the far field can be the sum of each
method-I without tilting). Moreover, when considering differ- element contribution as follows:
ent spacings between the patches, isolation in (Figs 6(b) and
   
6(c)) is high when the spacing is increased; however, the E = E1 + E2 + · · · + EN , (A.1)
value of r is also high because the far-field radiation patterns
are correlated (highly overlapped) as shown in Fig. 6(d). The where the radiated field of element j can be expressed in term
value of r calculated using method-I is affected by the spacing of the incident wave aj, M is the constant, which depends on
between the patch elements, hence its value gets higher as the the antenna dimensions and the directivity as well as the wave
elements get closer together (compare l/8, l/4, and l/2 using number k and Fj (u, f) is the far-field pattern using the
method-I). On the other hand, the evaluation of the r using expression:
method-II for the patch antennas shows that the estimation
does not consider the effect of the element positions as the   e−ikr
values are almost the same. E j = aj M F j u , f . (A.2)
r
It is worth mentioning that for well-isolated antennas, the
estimation of r using method-II will always give low values Then the total power can be written as:
since the numerator of (5) will be very low and this can lead
to wrong values. Increasing the spacing between the elements  2
 
will decrease r since the patterns will be more spatially sepa- Ptot =  E  ds. (A.3)
allspace
rated (in a real non-isotropic environment, the values tend to
oscillate with distance).
The total power can be also expressed in S-parameter form
as:

VI. CONCLUSIONS Ptot = a∗ (I − S∗ S)a, (A.4)

This work evaluates and discusses the accuracy and complex- where a is the incident wave vector of the N elements, S is N ×
ity of four methods for correlation coefficient (r) calculations N scattering matrix and i is the identity vector. The total
in MIMO antenna systems. We generalize two of the existing radiated power from element i with efficiency hrad,i can be
methods (method-III and the parallel equivalent circuit for as follows:
method-IV) to consider N-antenna elements for the first  N 
time. The effect of radiation pattern tilts on the estimation Prad,i = |ai |2 1 − n=1
|S ni | 2
hrad,i . (A.5)
of r is the first to be considered in this work. The analysis con-
ducted is applied on three different MIMO antenna systems
The total input power from element i can be written as:
with various efficiency values. It is found that method-I is
the most accurate and should be used all the time. It requires  N 
3D pattern measurements. Method-II is easy to use but fails Pin,i = |ai |2 1 − n=1
|S ni |2
, (A.6)
most of the times to give accurate r values even for very effi-
cient MIMO antenna systems because it does not consider the Hence, the internal losses power from element i can be
field orientation and tilts. Method-III compensates for some written as:
of the large errors encountered by method-II when the effi-
 N 
ciency is not high, but starts over estimating the r values
Ploss,i = |ai |2 1 − |S ni |2
(1 − hrad,i ). (A.7)
below certain efficiencies. Thus it should not be used due to n=1
its unreliable outcomes. Method-IV provides close estimates
compared with method-I only when the antennas are placed If r can be related to the power, then the correlation matrix
in a symmetric way. It fails to give reliable r values otherwise. (I 2 S∗ S) for different elements i and j can be calculated as in
Method-IV requires extra steps for the calculation of the loss [2]. Then r can be expressed using orthogonality and power

Downloaded from https://www.cambridge.org/core. KFUPM King Fahd University of Petroleum & Minerals, on 18 Sep 2017 at 08:53:50, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1759078717000903
 correlation coefficient calculations for mimo antenna systems 11

conservation laws as [4]:

⎛
 ⎞
     
⎜ r11 1 − Nn=1 |Sn1 |2 hrad,1 ··· r1N 1 − Nn=1 |Sn1 |2 hrad,1 1 − Nn=1 |SnN |2 hrad,N ⎟
⎜ ⎟
⎜ .. .. .. ⎟ = (I − S∗ S).
⎜ . . . ⎟
⎝

 N  N  N ⎠
rN1 1 − n=1 |SnN | hrad,N 1 − n=1 |Sn1 | hrad,1 ···
2 2
rNN 1 − n=1 |SnN | hrad,N
2

(A.8)

However, this equation is with no conduction and dielec-

tric losses. Now, by considering internal losses, the
S-parameters matrix can be divided into S-parameters ⎛ ⎞
r2,loss Z0 2
matrix for the ports and another for the radiation function ⎜ 2 + r2,loss Z0 0 0 ⎟
⎜ 2 + r2,loss Z0 ⎟
and losses [4], hence the reflection power losses and the ⎜ ⎟
⎜ r4,loss Z0 2 ⎟
power affects port i due to coupling from port j can be ⎜ 2 + r4,loss Z0 ⎟
0 0
⎜ 2 + r4,loss Z0 ⎟
expressed as: L24 = ⎜ ⎟,
⎜ 2 r2,loss Z0 ⎟

 ⎜ 0 0 ⎟
N   N  2  ⎜ 2 + r2,loss Z0 2 + r2,loss Z0 ⎟
rloss,ij = 1− |S |2 (1 − hrad,i ) 1 − S  (1 − hrad,j ). ⎜ ⎟
n=1 ni n=1 nj ⎝ 2 r4,loss Z0 ⎠
0 0
2 + r4,loss Z0 2 + r4,loss Z0
This can be applied for all ports. From the power conserva- (B.2)
tion low:

  
N N  N  2 
S∗ S + rloss,ij 1 − |S |2
(1 − h ) 1 − S  (1 − h )
n=1 ni nj n=1 ni rad,i n=1 nj rad,j



   (A.9)
N N  2 
+ rij 1 − |S ni |2
h rad,i 1 −  S nj
 h rad,j = 0
n=1 n=1

by solving and considering the worst case for r, (6) is

obtained. where L13 is the S-parameter matrix of the lossy part in ele-
ments 1 and 3 and L24 is that for elements 2 and 4. It
is worth mentioning that in this configuration, identical
elements and symmetric placement is considered such
that rloss ¼ r1,loss ¼ r2,loss ¼ r3,loss ¼ r4,loss, and then obviously
L13 ¼ L24. The Transmission matrix for both S-parameters
APPENDIX B measured from the VNA and S-parameters for the lossy
network part can be calculated from [28] to extract the lossless
r estimation based on the parallel RLC equivalent circuit will S-parameters as follow:
have a topology as shown in Fig. 7. Vi is the ith element exci-
tation, rloss is the conductance loss, Yrad is the radiation admit-
tance and ZL is the antenna load which is the impedance of the [Slossless ]T = [L1,3 ]T [Slossy ]−1
T [L2,4 ]T , (B.3)
feed cable.
The S-parameters between the ports can be measured using where [Slossy]T is the transmission matrix for the S-parameters
a VNA. The lossy part can be considered as a cascaded measured by the VNA, [Li, j]T is the transmission matrix for
network with the lossless part as shown in Fig. 7. The the lossy part in element i and element j. To calculate the
S-parameters for the lossy part can be calculated value of the loss conductance rloss from Fig. 7, the antenna ele-
by knowing the values of the conductance impedances. For ments can be viewed in two ways; first, it can be defined as
a four-element MIMO antenna system it is given by (for four networks with their inactive port connected to the load
N-elements, the same procedure is followed): impedance, so that only port 1 is active. Thus the total effi-
⎛ ⎞ ciency can be calculated as [3]:
r1,loss Z0 2
⎜ 2 + r1,loss Z0 0 0 ⎟ 
⎜ 2 + r1,loss Z0 ⎟ 
4  
⎜ ⎟ Sj1  hrad1
2 Prad
⎜ r Z
3,loss 0 2 ⎟ htot1 = 1 − , (B.4)
⎜ Prad + Ploss
2 + r3,loss Z0 ⎟
0 0
⎜ 2 + r3,loss Z0 ⎟ j=1  !
L13 = ⎜ ⎟,
⎜ 2 r1,loss Z0 ⎟
⎜ 0 0 ⎟
⎜ 2 + r1,loss Z0 2 + r1,loss Z0 ⎟
⎜ ⎟ where the underbrace term is the radiation efficiency (hrad).
⎝ 2 r3,loss Z0 ⎠
0 0 The relation between V1 and Vj can be calculated from
2 + r3,loss Z0 2 + r3,loss Z0
non-excited elements assuming port 1 is excited as (equations
(B.1) for other ports are removed for space constrains but follow

Downloaded from https://www.cambridge.org/core. KFUPM King Fahd University of Petroleum & Minerals, on 18 Sep 2017 at 08:53:50, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1759078717000903
 12 mohammad s. sharawi, abdelmoniem t. hassan and muhammad u. khan

Fig. 7. A lossy network for four-element MIMO antenna as a cascaded parallel equivalent circuit.

same structure):

V2′
Y21 V1′ + (Y22 ZL + 1) + Y23 V3′ + Y24 V4′ = 0. (B.5)
ZL

By defining k1j ¼ |Vj′|/|V1′ | using the matrix;

⎛ 1 ⎞−1
⎛ ⎞ (Y22 ZL + 1) Y23 Y24 ⎛ ⎞
k12 ⎜ ZL ⎟ Y21
⎜ 1 ⎟
⎜ ⎟ ⎜ (Y33 ZL + 1) ⎟ ⎜ ⎟
⎝ k13 ⎠ = −⎜ Y32 Y24 ⎟ ⎝ Y31 ⎠. (B.6)
⎜ ZL ⎟
k14 ⎝ 1 ⎠ Y41
Y42 Y43 (Y44 ZL + 1)
ZL

The radiation efficiencies are calculated from kij using the REFERENCES
following equation:
[1] Vaughan, R.G.; Andersen, J.B.: Antenna diversity in mobile commu-
4 2 nications. IEEE Trans. Veh. Technol., 36 (4) (1987), 149–172.
j=1 k1j Yj,rad
hrad1 = 4 2 4 2 . (B.7)
j=1 k1j Yj,rad + j=1 k1j rj,loss [2] Blanch, S.; Romeu, J.; Corbella, I.: Exact representation of antenna
system diversity performance from input parameter description.
Electron. Lett., 39 (9) (2003), 705–707.
In the second way, the antenna system network is defined
in such a way that one port is active while all others are termi- [3] Stjernman, A.: Relationship between radiation pattern correlation
nated with a loss resistance. The total efficiency in that case is and scattering matrix of lossless and lossy antennas. Electron.
found as: Lett., 41 (12) (2005), 678–680.

4 2 [4] Hallbjorner, P.: The significance of radiation efficiencies when using

j=1 k1j Yj,rad
htot,1 = (1 − |S11 | )4
2
4 4 2 S-parameters to calculate the received signal correlation from two
j=1 k1j Yj,rad + j=1 k1j rj,loss + j=2 k1j Re(Zl )
2 2
antennas. IEEE Antennas Wireless Propag. Lett., 4 (2005), 97–99.
(B.8) [5] Li, H.; Lin, X.; Lau, B.K.; He, S.: Equivalent circuit based calculation
of signal correlation in lossy MIMO antennas. IEEE Trans. Antennas
with some manipulation on (B.8), (8) is obtained. Propag., 61 (10) (2013), 5214–5222.

Downloaded from https://www.cambridge.org/core. KFUPM King Fahd University of Petroleum & Minerals, on 18 Sep 2017 at 08:53:50, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1759078717000903
 correlation coefficient calculations for mimo antenna systems 13

[6] Yun, J.X., Vaughan, R.G.: Space efficiency of multiple element anten- wideband microstrip slot antenna all for portable wireless DTV
nas. IEEE Trans. Antennas Propag., 60 (6) (2012), 3066–3071. media player. IEEE Trans. Antennas Propag., 61 (2) (2013), 964–968.
[7] Yun, J.X.; Vaughan, R.G.: MEA efficiency and impact on diversity [25] Karimian, R.; Oraizi, H.; Fakhte, S.; Farahani, M.: Novel F-shaped
and capacity. IEEE Trans. Antennas Propag., 60 (2) (2012), 529–539. quad-band printed slot antenna for WLAN and WiMAX MIMO
systems. IEEE Antennas Wireless Propag. Lett., 12 (2013), 405–408.
[8] Razmhosseini, M.; Vaughan, R.G.: Accuracy of numerical modelling
for evaluation of an integrated diversity wireless system on a small [26] Fernandez, S.C.; Sharma, S.K.: Multiband printed meandered loop
PCB, in 2015 IEEE Int. Symp. on Antennas and Propagation & antennas with MIMO implementations for wireless routers. IEEE
USNC/URSI National Radio Science Meeting, Vancouver, BC, Antennas Wireless Propag. Lett., 12 (2013), 96–99.
2015, 1230–1231.
[27] Wallace, J.W.; Jensen, M.A.: Termination-dependent diversity per-
[9] Bhattacharya, A.; Vaughan, R.G.: Daughter board antenna for formance of coupled antennas: network theory analysis. IEEE
compact polarization diversity on MIMO devices, in 2015 IEEE Trans. Antennas Propag., 52 (1) (2004), 98–105.
Int. Symp. on Antennas and Propagation & USNC/URSI National
[28] Frei, J.; Cai, X.; Muller, S.: Multiport S-parameter and T-parameter
Radio Science Meeting, Vancouver, BC, 2015, 1218–1219.
conversion With symmetry extension. IEEE Trans. Microw.
[10] Li, H.; Lin, X.; Lau, B.K.; He, S.: Calculating signal correlation in lossy Theory Tech., 56 (11) (2008), 2493–2504.
dipole arrays using scattering parameters and efficiencies, in 2013
7th European Conf. on Antennas and Propagation (EuCAP), Mohammad S. Sharawi received the
Gothenburg, 2013, 519–523. Ph.D. degree in RF systems engineering
[11] Ogawa, K.; Matsuyoshi, T.; Monma, K.: An analysis of the perform- from Oakland University, Oakland, MI,
ance of a handset diversity antenna influenced by head, hand, and USA, in 2006. He was a Hardware
shoulder effects at 900 MHz: Part II—correlation characteristics. Design Engineer with Silicon Graphics
IEEE Trans. Veh. Technol., 50 (3) (2001), 845–853. Inc., Mountain View, CA, USA, from
[12] Yeap, S.B.; Chen, X.; Dupuy, J.A.; Chiau, C.C.; Parini, C.G.:
2002 to 2003. From 2006 to 2008, he
Integrated Diversity Antenna for Laptop and PDA Terminal in a became a Faculty Member with the
MIMO System. IEE Proc. – Microw. Antennas Propag., 152 (6) Computer Engineering Department,
(2005), 495–504. German Jordanian and Philadelphia Universities in Amman,
Jordan. He was a Research Scientist with the Applied Electro-
[13] Chiu, C.Y.; Yan, J.B.; Murch, R.D.: Compact three-port orthogonally
polarized MIMO antennas. IEEE Antennas Wireless Propag. Lett., 6
magnetics and Wireless Laboratory, Oakland University, from
(2007), 619–622. 2008 to 2009 and in 2013. In 2014, he joined the iRadio La-
boratory, University of Calgary, Calgary, AB, Canada, as a vis-
[14] Ding, Y.; Du, Z.; Gong, K.; Feng, Z.: A four-element antenna system iting research Professor until 2015. He is currently a Professor
for mobile phones. IEEE Antennas Wireless Propag. Lett., 6 (2007),
of Electrical Engineering with King Fahd University of Petrol-
655–658.
eum and Minerals (KFUPM), Dhahran, Saudi Arabia. Dr.
[15] Gao, Y.; Chen, X.; Ying, Z.; Parini, C.: Design and performance Sharawi is the Founder and Director of the Antennas and
investigation of a dual-element PIFA array at 2.5 GHz for MIMO Microwave Structure Design Laboratory (AMSDL). He is
terminal. IEEE Trans. Antennas Propag., 55 (12) (2007), 3433–3441. the single author of the book, Printed MIMO Antenna Engin-
[16] Głogowski, R.; Peixeiro, C.: Multiple printed antennas for integration eering, Artech House, 2014. He has authored/co-authored
into small multistandard handsets. IEEE Antennas Wireless Propag. seven book chapters in antenna design, MIMO antenna sys-
Lett., 7 (2008), 632–635. tems and RF systems, along with over 200 refereed inter-
[17] Li, H.; Xiong, J.; He, S.: A compact planar MIMO antenna system of national journal and conference paper publications mostly
four elements with similar radiation characteristics and isolation with the IEEE. He holds 12 issued and 14 pending patents
structure. IEEE Antennas Wireless Propag. Lett., 8 (2009), 1107–1110. from the US-Patent Office. His current research interests in-
clude printed multiple-input–multiple-output antenna sys-
[18] Su, S.W.: High-gain dual-loop antennas for MIMO access points in
the 2.4/5.2/5.8 GHz bands. IEEE Trans. Antennas Propag., 58 (7)
tems, miniaturized printed antennas and antenna arrays,
(2010), 2412–2419. active integrated antennas, reconfigurable antennas, micro-
wave circuits and electronics, millimeter-wave antennas and
[19] Cui, S.; Liu, Y.; Jiang, W.; Gong, S.X.: Compact dual-band monopole antenna arrays, and applied electromagnetics. Dr. Sharawi is
antennas with high port isolation. Electron. Lett., 47 (10) (2011),
an IET Fellow and Senior member of IEEE. He was a recipient
579–580.
of the prestigious Excellence in Scientific Research and Best
[20] Ling, X.M.; Li, R.L.: Novel dual-band MIMO antenna array with low Research Project Awards from KFUPM in 2015 and 2017.
mutual coupling for portable wireless devices. IEEE Antennas He received conference paper awards in IEEE LAPC 2014,
Wireless Propag. Lett., 10 (2011), 1039–1042. and IEEE MeCAP 2016. He has served/serving on the tech-
[21] Sharawi, M.S.; Iqbal, S.S.; Faouri, Y.S.: An 800 MHz 2×1 compact nical and organizational committees of several international
MIMO antenna system for LTE handsets. IEEE Trans. Antennas IEEE conferences, especially EuCAP, APS, APWC, APCAP,
Propag., 59 (8) (2011), 3128–3131. and ICCE.
[22] Lee, J.M.; Kim, K.B.; Ryu, H.K.; Woo, J.M.: A compact ultra wideband
MIMO antenna with WLAN band-rejected operation for mobile
devices. IEEE Antennas Wireless Propag. Lett., 11 (2012), 990–993.
Abdelmonim T. Hassan is currently pursuing his Ph.D.
[23] Li, J.F.; Chu, Q.X.; Huang, T.G.: A compact wideband MIMO degree in Electrical Engineering at Concordia University,
antenna with two novel bent slits. IEEE Trans. Antennas Propag., Canada. He did his M.Sc. degree under the supervision of
60 (2) (2012), 482–489. Professor M. S. Sharawi at KFUPM, Saudi Arabia. His re-
[24] Kulkarni, A.N.; Sharma, S.K.: Frequency reconfigurable microstrip search interests are printed MIMO antenna systems and
loop antenna covering LTE bands with MIMO implementation and mobile communications.

Downloaded from https://www.cambridge.org/core. KFUPM King Fahd University of Petroleum & Minerals, on 18 Sep 2017 at 08:53:50, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1759078717000903
 14 mohammad s. sharawi, abdelmoniem t. hassan and muhammad u. khan

Muhammad U. Khan received the B.S. (KFUPM), Dhahran, Saudi Arabia, in 2014. He is currently
degree in Electrical Engineering from an Assistant Professor at the National University of Science
the National University of Sciences and Technology (NUST), Islamabad, Pakistan. His current
and Technology (NUST), Pakistan, in research interests include miniaturization of printed
2006 and the M.S. degree in Electrical antennas, MIMO antennas, and metamaterial antennas.
Engineering from the GIK Institute of
Engineering Sciences and Technology,
Pakistan, in 2008. He received his
Ph.D. in Electrical Engineering from
King Fahd University of Petroleum and Minerals

Downloaded from https://www.cambridge.org/core. KFUPM King Fahd University of Petroleum & Minerals, on 18 Sep 2017 at 08:53:50, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1759078717000903