This is the author's version of an article that has been published in this journal.

Changes were made to this version by the publisher prior to publication.

The final version of record is available at http://dx.doi.org/10.1109/TAP.2016.2633901
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 1

3D Antenna Radiation Pattern Reconstruction in

a Reverberation Chamber Using Spherical Wave
Decomposition
Qian Xu, Yi Huang, Senior Member, IEEE, Lei Xing, Chaoyun Song, Student Member, IEEE, Zhihao
Tian, Student Member, IEEE, Saqer S. Alja’afreh and Manoj Stanley

Abstract—It has been shown that the correlation between the

radiation patterns of two antennas can be measured in a
reverberation chamber (RC). In this paper, it is shown that the
self-correlation coefficient of an antenna, defined as the
correlation between the radiation pattern of an antenna under test
(AUT) and a transformed version of itself, can also be measured in
an RC. It is found that the three-dimensional (3D) radiation
pattern of the AUT can be reconstructed from the measured
self-correlation coefficient by using spherical wave decomposition.
Moreover, the axial ratio of the AUT can also be measured
efficiently in the RC when the pattern is directional. Numerical
simulations and measurements in the RC and in an anechoic
(a)
chamber have been undertaken. Good agreement is obtained
which confirms the validity of the proposed method. Thus this
novel method can become a very useful, cost-effective and efficient
method for 3D antenna radiation pattern measurement.

Index Terms—Antenna measurement, axial ratio, envelop

correlation coefficient, antenna radiation pattern measurement,
reverberation chamber.

I. INTRODUCTION

A REVERBERATION chamber (RC) has been proven to be

an important and useful facility for measuring antenna
properties, such as radiation efficiency [1-5], free-space
impedance/S-parameters [6], and diversity gain [7-10].
(b)
However, it is not easy to measure the radiation pattern of an Fig. 1. Antenna radiation pattern measurement setup in the RC, (a) existing
antenna in an RC because of multipath interference. It has been methods, (b) proposed method.
shown that by using a deconvolution technique [11], the
radiation pattern of the antenna under test (AUT) can be pattern can be extracted by using the K-factor [12-14] or
reconstructed in a non-anechoic environment. It is also possible Doppler shift [15-16]. A typical measurement setup is shown in
to measure the pattern directly in an RC when the line-of-sight Fig. 1(a) where two antennas are connected to the measurement
(LoS) response dominates the unstirred part. In this case, the instrument, a high directivity transmitting (Tx) or receiving
(Rx) antenna is fixed and directed towards the AUT. The
Manuscript received Nov. 29, 2015. Corresponding author: Y. Huang. This non-line-of-sight (NLoS) part can either be averaged out by
work was supported in part by the National Natural Science Foundation of rotating the stirrer [13, 14], or be filtered by using the Doppler
China (61601219) and Natural Science Foundation of Jiangsu Province shift (moving the AUT along a sliding track) [15, 16]. By
(BK20160804).
Q. Xu and L. Xing are with College of Electronic and Information rotating the turntable, the angle dependency of radiation pattern
Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing can be measured. That is, the information hidden behind the
211106, China ([email protected]). LoS/unstirred response is extracted while removing the
Y. Huang, C. Song, Z. Tian and M. Stanley are with the Department of
Electrical Engineering and Electronics, The University of Liverpool,
NLoS/stirred response. It is also possible to extract the
Liverpool, L69 3GJ, UK (e-mail: [email protected]; [email protected]; radiation pattern by using the time-reversal technique in the RC
[email protected]; [email protected]). [17-19], but the time-reversal behavior of the RC need to be
S. Alja’afreh is with the Electrical Engineering Department at Mu’tah
University, Mu’tah, 61710, Jordan (e-mail: [email protected]).
carefully characterized and calibrated.

Copyright (c) 2017 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TAP.2016.2633901
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 2

In this paper, we propose a new method which can extract the

(2𝑙 + 1) (𝑙 − 𝑚)! 𝑙
3D pattern of the AUT from the NLoS/stirred response while 𝑌𝑙𝑙 (𝜃, 𝜑) = � 𝑃 (𝑐𝑜𝑠𝜃)𝑚 𝑗𝑙𝜑 (2)
the LoS/unstirred response is not needed. This is very different 4𝜋 (𝑙 + 𝑚)! 𝑙
from previous methods. The measurement setup is shown in
Fig. 1(b) where the Tx/Rx antenna is not directed to the AUT. 𝑃𝑙𝑙 (𝑐𝑜𝑠𝜃) is the associated Legendre functions of the first kind
Since the RC is inherently a rich multipath environment, using [23]. The vector coefficients 𝒂𝑙𝑙 can be obtained by using the
the stirred part is actually easier than using the unstirred part. orthogonality of the SSHs [22] with each component 𝑚𝑥,𝑙𝑙 ,
As can be seen later, the measurement time using the setup in 𝑚𝑦,𝑙𝑙 and 𝑚𝑧,𝑙𝑙 , ( 𝒂𝑙𝑙 = [𝑚𝑥,𝑙𝑙 𝑚𝑦,𝑙𝑙 𝑚𝑧,𝑙𝑙 ] , 𝐄(𝜃, 𝜑) =
Fig. 1(b) is comparable with the measurement time in an [𝐸𝑥 (𝜃, 𝜑) 𝐸𝑦 (𝜃, 𝜑) 𝐸𝑧 (𝜃, 𝜑)])
anechoic chamber (AC). In Fig. 1(a), the LoS/unstirred part
needs to be extracted by rotating the stirrers or moving the AUT 2𝜋 𝜋
𝑚𝑥,𝑙𝑙 = ∫0 ∫0 𝐸𝑥 (𝜃, 𝜑)𝑌𝑙𝑙∗ (𝜃, 𝜑)𝑠𝑖𝑚𝜃 𝑑𝜃 𝑑𝜑
(along the sliding track) for each AUT angle, this could be very
time consuming.
2𝜋 𝜋
In the proposed method, the radiation pattern of the AUT is 𝑚𝑦,𝑙𝑙 = ∫0 ∫0 𝐸𝑦 (𝜃, 𝜑)𝑌𝑙𝑙∗ (𝜃, 𝜑)𝑠𝑖𝑚𝜃 𝑑𝜃 𝑑𝜑
decomposed into spherical harmonics with unknown
coefficients, and the pattern measurement becomes a 2𝜋 𝜋
𝑚𝑧,𝑙𝑙 = ∫0 ∫0 𝐸𝑧 (𝜃, 𝜑)𝑌𝑙𝑙∗ (𝜃, 𝜑)𝑠𝑖𝑚𝜃 𝑑𝜃 𝑑𝜑 (3)
generalized mode matching problem. By measuring the
self-correlation coefficient of the radiation pattern in the RC,
where * means complex conjugate.
the coefficients of the spherical wave modes/harmonics can be
inverted and thus the radiation pattern can be reconstructed.
The paper is organized in four sections. The theory is given A. Forward Problem
in Section II where the forward problem and inverse problem If the far-field of the radiation pattern is known, using (3), the
are defined, and it is found that the axial ratio (AR) of the AUT far-field can be decomposed into SSHs using (1) with a
can also be measured under a certain approximations. In truncation of level 𝐿 [23]. Suppose we have two antennas,
Section III, simulations and measurements are conducted to where the radiation patterns are 𝐄1 (𝜃, 𝜑) and 𝐄2 (𝜃, 𝜑)
verify the proposed method, results from the RC and AC are respectively, the correlation coefficient 𝜌𝐸 between these two
compared and good agreement is obtained. Finally, discussions antennas is [25, 26]
and conclusions are presented in Section IV.
|∬ 𝐄1 (𝜃, 𝜑) ⋅ 𝐄2∗ (𝜃, 𝜑)𝑑Ω|
II. THEORY 𝜌𝐸 = (4)
�∬|𝐄1 (𝜃, 𝜑)|2 𝑑Ω ∬|𝐄2 (𝜃, 𝜑)|2 𝑑Ω
In this section, we first explain the forward problem: if the
radiation pattern of the antenna is already known, the 2𝜋 𝜋
self-correlation coefficient can be calculated. Then, the inverse where ∬ ∎𝑑Ω = ∫0 ∫0 ∎𝑠𝑖𝑚𝜃 𝑑𝜃 𝑑𝜑 means the integral over
problem is introduced to reconstruct the radiation pattern from a unit spherical surface. Suppose that 𝐄1 (𝜃, 𝜑) and 𝐄2 (𝜃, 𝜑)
the self-correlation coefficient. Finally, the expression for the are expanded by using the SSHs with order 𝐿
axial ratio (AR) is derived under a certain approximation.
𝐿 𝑙
It is well-known that the far-field radiation pattern of an
antenna can be decomposed into a superposition of spherical 𝐄1 (𝜃, 𝜑) = � � 𝒂𝑙𝑙 𝑌𝑙𝑙 (𝜃, 𝜑)
wave harmonics [20-24]. Two types of decomposition have 𝑙=0 𝑙=−𝑙
been used: vector spherical harmonics (VSHs) with scalar
𝐿 𝑙
coefficients and scalar spherical harmonics (SSHs) with vector
coefficients. It has been shown that SSHs with vector 𝐄2 (𝜃, 𝜑) = � � 𝒃𝑙𝑙 𝑌𝑙𝑙 (𝜃, 𝜑) (5)
coefficients can avoid the singularity problem of the antenna 𝑙=0 𝑙=−𝑙

field on the poles [22-24], therefore we use SSHs with vector

Since the SSHs are orthogonal to each other on a unit spherical
coefficients in this paper.
surface, substituting (5) into (4) gives the correlation
The far-field of an antenna can be expressed as [23]
coefficient
𝐿 𝑙

𝐄(𝜃, 𝜑) = lim � � 𝒂𝑙𝑙 𝑌𝑙𝑙 (𝜃, 𝜑) (1) �∑𝐿𝑙=0 ∑𝑙𝑙=−𝑙 𝒂𝑙𝑙 ⋅ 𝒃∗𝒍𝒎 �
𝐿→∞ 𝜌𝐸 = (6)
𝑙=0 𝑙=−𝑙
�∑𝐿𝑙=0 ∑𝑙𝑙=−𝑙 ‖𝒂𝑙𝑙 ‖2 �∑𝐿𝑙=0 ∑𝑙𝑙=−𝑙 ‖𝒃𝑙𝑙 ‖2
where 𝜃 , 𝜑 are the polar angle and azimuthal angle in the
spherical coordinate system respectively, 𝑌𝑙𝑙 (𝜃, 𝜑) is the Since 𝑚𝑥,𝑙𝑙 , 𝑚𝑦,𝑙𝑙 , and 𝑚𝑦,𝑙𝑙 are complex numbers, we use
scalar spherical harmonic with level 𝑙 and mode 𝑚 defined by ‖𝒂𝑙𝑙 ‖2 to represent the square of the magnitude of each
2 2 2
coefficient, which is ‖𝒂𝑙𝑙 ‖2 = �𝑚𝑥,𝑙𝑙 � + �𝑚𝑦,𝑙𝑙 � + �𝑚𝑧,𝑙𝑙 � .
Now we introduce the concept of the self-correlation

Copyright (c) 2017 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TAP.2016.2633901
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 3

coefficient. Suppose 𝐄2 (𝜃, 𝜑) can be transformed from

𝐄1 (𝜃, 𝜑) by using rotations. That is, 𝐄2 (𝜃, 𝜑) is a transformed
version of 𝐄1 (𝜃, 𝜑), they have the same shape but different
reference coordinate systems. It should be noted that, to obtain
a rotated version of 𝐄1 (𝜃, 𝜑), we only need to apply a rotation
matrix to the coefficients at each level 𝑙 [22, 27], which is
Fig. 2. Self-correlation of a directional antenna, the integral result is
[𝑏𝑥𝑙,−𝑙 ⋯ 𝑏𝑥𝑙,𝑙 𝑏𝑦𝑙,−𝑙 ⋯ 𝑏𝑦𝑙,𝑙 𝑏𝑧𝑙,−𝑙 ⋯ 𝑏𝑧𝑙,𝑙 ] dominated by the field in the main beam direction.
= [𝑚𝑥𝑙,−𝑙 ⋯ 𝑚𝑥𝑙,𝑙 𝑚𝑦𝑙,−𝑙 ⋯ 𝑚𝑦𝑙,𝑙 𝑚𝑧𝑙,−𝑙 ⋯ 𝑚𝑧𝑙,𝑙 ]𝐌𝑙 (𝛼, 𝛽, 𝛾) |∑𝑁 ∗
𝑘=1(𝑆31 (𝑘) − ⟨𝑆31 ⟩)(𝑆32 (𝑘) − ⟨𝑆32 ⟩) |
(7) = (9)
�∑𝑁 2 𝑁
𝑘=1|𝑆31 (𝑘) − ⟨𝑆31 ⟩| ∑𝑘=1|𝑆32 (𝑘) − ⟨𝑆32 ⟩|
2

where 𝐌𝑙 (𝛼, 𝛽, 𝛾) is the rotation matrix at level l; (𝛼, 𝛽, 𝛾)

means the pattern is rotated around x-axis first with angle 𝛼, where 𝑁 represents the total number of samples collected, ⟨∙⟩
then rotated around y-axis with angle 𝛽, and finally rotated means the average value of N samples. In the diversity gain
around z-axis with angle 𝛾 all with the right-hand rule, which measurement, 𝑆31 and 𝑆32 represent the measured
means 𝐌𝑙 (𝛼, 𝛽, 𝛾) = 𝐌𝑙 (𝛼, 0,0)𝐌𝑙 (0, 𝛽, 0)𝐌𝑙 (0,0, 𝛾) . Since S-parameters between the Tx/Rx antenna (in Fig. 1(b)) and two
the total radiated power of 𝐄1 (𝜃, 𝜑) and 𝐄2 (𝜃, 𝜑) are the same, MIMO branches respectively. In this measurement, 𝑆31 and 𝑆32
we have ∑𝐿𝑙=0 ∑𝑙𝑙=−𝑙 |𝒂𝑙𝑙 |2 = ∑𝐿𝑙=0 ∑𝑙𝑙=−𝑙|𝒃𝑙𝑙 |2 , (6) becomes represent the measured S-parameters between the Tx/Rx
antenna and AUT at two different positions respectively. Since
∗ position 2 is the same antenna rotated from position 1, and it
�∑𝐿𝑙=0 ∑𝑙𝑙=−𝑙 𝒂𝑙𝑙 ∙ 𝑇𝛼𝛽𝛾 (𝒂𝑙𝑙 )�
𝜌𝑆𝑆 (𝛼, 𝛽, 𝛾) = (8) has been shown that the correlation of measured S-parameters
∑𝐿𝑙=0 ∑𝑙𝑙=−𝑙‖𝒂𝑙𝑙 ‖2 equals the correlation of the radiation patterns [25, 26], thus we
have
which is the definition of the self-correlation coefficient (𝜌𝑆𝑆 ),
where we use 𝑇𝛼𝛽𝛾 (𝒂𝑙𝑙 ) to represent the transformed version 𝜌𝑆 (𝛼, 𝛽, 𝛾) = 𝜌𝑆𝑆 (𝛼, 𝛽, 𝛾) (10)
of 𝒂𝑙𝑙 by using (7). Since 𝑇𝛼𝛽𝛾 depends on the rotation angles,
𝜌𝑆𝑆 is also a function of rotation angles. To solve 𝒂𝑙𝑙 , two more conditions are needed: the radiated
As expected, the forward problem is well-defined, once the power should be normalized to 1 (otherwise 𝒂𝑙𝑙 can be scaled
radiation pattern of the antenna 𝐄1 (𝜃, 𝜑) is known, the arbitrarily), and the far-field condition should be satisfied. This
self-correlation coefficient can be calculated using (7) and (8). is because we use the SSHs rather than VSHs, while in the
The calculation procedure for the rotation matrix 𝐌𝑙 (𝛼, 𝛽, 𝛾) VSHs, the far-field condition can be satisfied automatically
can be found in [22, 27]. (but the singularity problem needs to be treated separately [22]).
Mathematically, these two conditions are
B. Inverse Problem
∑𝐿𝑙=0 ∑𝑙𝑙=−𝑙‖𝒂𝑙𝑙 ‖2 = 1 (11)
If the angle dependency of 𝜌𝑆𝑆 in (8) is known, by solving (8)
for 𝒂𝑙𝑙 , the 3D radiation pattern of the AUT can be 𝐫�(𝜃, 𝜑) ⋅ ∑𝐿𝑙=0 ∑𝑙𝑙=−𝑙 𝒂𝑙𝑙 𝑌𝑙𝑙 (𝜃, 𝜑) = 0 (12)
reconstructed. As can be seen, the inverse problem is actually a
generalized mode matching problem, that is, the coefficients of where 𝐫�(𝜃, 𝜑) is the unit vector in radial direction.
the spherical wave modes need to be solved to match the angle As can be seen, the inverse problem is a generalized mode
dependency of 𝜌𝑆𝑆 . matching problem defined by the nonlinear equation system
Luckily, this self-correlation coefficient can be measured in (10) - (12), and these can be solved numerically by using the
the RC using the setup as shown in Fig. 1(b). The measurement Levenberg-Marquardt algorithm [29] as shown later in this
procedure is very similar to the diversity gain measurement paper.
[7-10], the only difference is: in the diversity gain measurement,
𝐄1 (𝜃, 𝜑) and 𝐄2 (𝜃, 𝜑) in (4) are the radiation pattern of the two
multiple-input multiple-output (MIMO) antenna branches; in C. Axial Ratio
this measurement, 𝐄2 (𝜃, 𝜑) is a rotated version of 𝐄1 (𝜃, 𝜑). It is interesting to note that, under certain approximations,
We just need to rotate the AUT for a set of angles (𝛼, 𝛽, 𝛾) the problem can be simplified. If the antenna is a directional
instead of switching the receiving/transmitting port in the antenna, the integral in (4) can be approximately evaluated
diversity gain measurement. using the field in the main beam as shown in Fig. 2. Suppose
The correlation coefficient can be obtained by measuring the that the main beam is aligned with z-axis and rotated around it
S-parameters in the RC [25, 26, 28] by an angle 𝛾. The integral (4) can be approximated as

𝜌𝑆 (𝛼, 𝛽, 𝛾) = 𝑐𝑜𝑟𝑟(𝑆31 , 𝑆32 ) |∬ 𝐄1 (𝜃, 𝜑) ⋅ 𝐄2∗ (𝜃, 𝜑)𝑑Ω|

𝜌𝑆𝑆 (𝛾) =
�∬|𝐄1 (𝜃, 𝜑)|2 𝑑Ω ∬|𝐄2 (𝜃, 𝜑)|2 𝑑Ω

Copyright (c) 2017 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TAP.2016.2633901
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 4

Fig. 4. The simulated E-field magnitude pattern (linear scale) at 5 GHz, the
maximum value is normalized to 1 V/m.
Fig. 3. The dimensions of the horn antenna and the definition of the rotation
angles 𝛼, 𝛽, 𝛾 and axes 𝑥, 𝑦, 𝑧. The antenna is excited by using a lumped port
located at the center of the waveguide.

𝑐𝑜𝑠𝛾 −𝑠𝑖𝑚𝛾 𝐸𝑥 𝐸𝑥 ∗
�� � �𝑗𝐸 � ⋅ �𝑗𝐸 � �
𝑠𝑖𝑚𝛾 𝑐𝑜𝑠𝛾 𝑦 𝑦
≈
𝐸𝑥2 + 𝐸𝑦2
��𝐸𝑥2 + 𝐸𝑦2 �𝑐𝑜𝑠𝛾 − 2𝑗𝐸𝑥 𝐸𝑦 𝑠𝑖𝑚𝛾�
= (13)
𝐸𝑥2 + 𝐸𝑦2

where 𝐄1 (𝜃, 𝜑) is the original pattern, 𝐄2 (𝜃, 𝜑) is the rotated

pattern with angle 𝛾. It can be found from (13) that, when
𝛾 = 90° the 𝜌𝑆𝑆 (𝛾) has the minimum value

2𝐸𝑥 𝐸𝑦
𝜌𝑆𝑆 (𝛾)𝑙𝑚𝑚 ≈ (14) Fig. 5. The decomposed magnitude (dB scale, 20log|∎|) and phase (degree)
𝐸𝑥2 + 𝐸𝑦2
of the SSHs of the radiation pattern in Fig. 4, 𝒂𝑙𝑙 is normalized to make sure
∑𝐿𝑙=0 ∑𝑙𝑚=−𝑙‖𝒂𝑙𝑙 ‖2 = 1.
Note that the AR is defined as 𝐸𝑥 /𝐸𝑦 , thus the minimum of
𝜌𝑆𝑆 (𝛾) can be related to the AR of the antenna compare the reconstructed pattern (obtained from the
self-correlation coefficients) with the pattern obtained by using
2𝐸𝑥 𝐸𝑦 2�𝐸𝑥 /𝐸𝑦 � 2�𝐸𝑦 /𝐸𝑥 � CST to verify the proposed method. Then measurements in the
𝜌𝑆𝑆 (𝛾)𝑙𝑚𝑚 ≈ 2 2
= 2 = 2 RC and AC are conducted and the results are compared. We use
𝐸𝑥 + 𝐸𝑦 1 + �𝐸 /𝐸 � 1 + �𝐸𝑦 /𝐸𝑥 �
𝑥 𝑦 the measured results obtained in AC as the reference to confirm
the effectiveness of the proposed method.
2𝐴𝑅
= (15) A. Simulations
1 + 𝐴𝑅2
A typical rectangular horn antenna shown in Fig. 3 is used as
and the AR can be obtained from the AUT in the numerical simulation. The definitions of the
rotation angles are also given (right-hand rule). The simulated
1 + �1 − 𝜌𝑆𝑆2 (𝛾)
𝑙𝑚𝑚
radiation pattern at 5 GHz is obtained by using CST Microwave
𝐴𝑅 ≈ (16) Studio and given in Fig. 4 with maximum E-field magnitude
𝜌𝑆𝑆 (𝛾)𝑙𝑚𝑚
normalized to 1V/m.
This offers an opportunity to measure the AR in the RC for The radiation pattern can be decomposed into SSHs with
directional antennas. coefficients 𝒂𝑙𝑙 (the forward problem) by using (3). The
magnitude and phase of each component of 𝒂𝑙𝑙 are shown in
Fig. 5. We use 𝐿 up to 15 in Fig. 5, as can be seen, when 𝑙 > 8,
III. SIMULATIONS AND MEASUREMENTS the magnitude of the SSHs is already very small (<-27 dB).
Then, (8) is used to obtain the self-correlation coefficients of
In this section, simulations and measurements are conducted
the radiation pattern, since the AUT is rotated only around the
to verify the proposed method. Before conducting the
x-, y-, and z-axes, we use 𝜌𝑆𝑆 (𝛼), 𝜌𝑆𝑆 (𝛽) and 𝜌𝑆𝑆 (𝛾) instead of
measurements, numerical simulations are carried out. Since in
𝜌𝑆𝑆 (𝛼, 0,0), 𝜌𝑆𝑆 (0, 𝛽, 0) and 𝜌𝑆𝑆 (0,0, 𝛾) respectively.
the numerical simulation, once the geometrical structure of the
After the self-correlation coefficients of the radiation pattern
AUT is defined, the radiation pattern of the AUT can be
are obtained, the radiation pattern of the AUT can be
obtained by using the full wave simulation software (CST
reconstructed (the inverse problem). Mathematically, by
Microwave Studio is used in this paper). Thus the simulated
combining (10), (11) and (12), the inverse problem can be
radiation pattern can be used as the reference, and we can
expressed as

Copyright (c) 2017 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TAP.2016.2633901
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 5

(a)

Fig. 6. Reconstructed 𝒂𝑙𝑙 from (17).

�∑𝐿 ∑𝑙 𝒂 ∙ 𝑇 ∗ (𝒂 )�
⎧ 𝑙=0 𝑙=−𝑙 𝑙𝑙 𝛼 𝑙𝑙 = 𝜌𝑆𝑆 (𝛼), 𝛼 = 1, 2, … , 360°
⎪ ∑𝐿𝑙=0 ∑𝑙𝑙=−𝑙‖𝒂𝑙𝑙 ‖2
(b)
⎪ �∑𝑙=0 ∑𝑙𝑙=−𝑙 𝒂𝑙𝑙 ∙ 𝑇𝛽∗ (𝒂𝑙𝑙 )�
𝐿
⎪ = 𝜌𝑆𝑆 (𝛽), 𝛽 = 1, 2, … , 360°
⎪ ∑𝐿𝑙=0 ∑𝑙𝑙=−𝑙‖𝒂𝑙𝑙 ‖2
⎪ �∑𝐿𝑙=0 ∑𝑙𝑙=−𝑙 𝒂𝑙𝑙 ∙ 𝑇𝛾∗ (𝒂𝑙𝑙 )�
= 𝜌𝑆𝑆 (𝛾), 𝛾 = 1, 2, … , 360°
∑𝐿𝑙=0 ∑𝑙𝑙=−𝑙‖𝒂𝑙𝑙 ‖2
⎨ 𝐿 𝑙
⎪
⎪ � � ‖𝒂𝑙𝑙 ‖2 = 1 ,
⎪ 𝑙=0 𝑙=−𝑙
𝐿 𝑙
⎪
⎪ 𝐫�(𝜃, 𝜑) ⋅ � � 𝒂𝑙𝑙 𝑌𝑙𝑙 (𝜃, 𝜑) = 0, (𝜃, 𝜑) ∈ 𝕊
⎩ 𝑙=0 𝑙=−𝑙
(c)
(17) Fig. 7. Reconstructed 𝜌𝑆𝑆 and the original 𝜌𝑆𝑆 . (a), (b) and (c) are the
self-correlation rotated around x-, y- and z-axis respectively.

where 𝐿 = 8 is chosen, 𝜌𝑆𝑆 (𝛼) , 𝜌𝑆𝑆 (𝛽) and 𝜌𝑆𝑆 (𝛾) are
sampled at 1 degree/step, 𝕊 is the point set chosen on the sphere,
5 degree/step is used for both 𝜃 and 𝜑 angles.
The nonlinear system of equations (17) can be
solved/optimized using the Levenberg-Marquardt algorithm
[29] in Matlab. After optimization, the reconstructed 𝒂𝑙𝑙 are
obtained and shown in Fig. 6. As can be seen, compared with
Fig. 5, they have a very similar magnitude but different phase,
which is not an issue since the phase is a relative value. The Fig. 8. Reconstructed E-field magnitude pattern (linear scale) at 5 GHz, the
reconstructed 𝜌𝑆𝑆 calculated from the reconstructed 𝒂𝑙𝑙 are is maximum value is normalized to 1 V/m.
shown in Fig. 7, comparisons between the reconstructed 𝜌𝑆𝑆
and 𝜌𝑆𝑆 of the original pattern (Fig. 4) are shown. As can be
seen, a very good agreement has been obtained except at some
angles. The 3D radiation pattern can be obtained quickly using
𝐄(𝜃, 𝜑) = ∑8𝑙=0 ∑𝑙𝑙=−𝑙 𝒂𝑙𝑙 𝑌𝑙𝑙 (𝜃, 𝜑) and is shown in Fig. 8.
Compared with Fig. 4, a very similar pattern is reconstructed.
Comparisons of co-polarization and cross-polarization
components in the YOZ plane and the relative error in all angles
are given in Fig. 9. As can be seen, a good agreement is
obtained in the main lobe of the pattern. However, the back lobe, (a) (b)
Fig. 9. (a) Co-polarization (CP) and cross-polarization (XP) of the original
side lobe and the cross-polarization component are not exactly
pattern (Fig. 4) and the reconstructed pattern (Fig. 8) in YOZ plane,
the same, which will be discussed in the next section. normalized to the peak value in dB; (b) 20log�|𝐄| − |𝐄′|�, relative error in all
The AR measurement result in the RC can also be verified angles in dB, where 𝐄 and 𝐄′ are the original pattern (Fig. 4) and the
through simulation. We use a conical horn antenna with two reconstructed pattern (Fig. 8) respectively (𝐄 and 𝐄′ are normalized to the
peak value of 1V/m).

Copyright (c) 2017 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TAP.2016.2633901
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 6

Fig. 13. Radiation pattern measurement in the AC and the definition of x-, y-,
and z-axis.

Fig. 10. The conical antenna and a typical radiation pattern at 1.5 GHz, two
perpendicular lumped ports are used to synthesize waves with different AR
values.

Fig. 14. AR measurement in the AC.

Fig. 11. 𝜌𝑆𝑆 (𝛾) with different AR values, the AR values in the legend are read
from CST directly.

Fig. 15. Measurement setup in the RC, AUT rotated around z-, y-, and x-axis
are shown in (a), (b) and (c) respectively, AR measurement is shown in (d).

Fig. 12. AR obtained from 𝜌𝑆𝑆 (𝛾)𝑚𝑚𝑚 using (16) and AR obtained from CST.

perpendicular excited ports to generate a directional pattern

with different AR values. To improve the radiation pattern, a
layer of absorbing material is used to suppress the side lobe and
back lobe. The model is shown in Fig. 10, by tuning the phase
difference between port 1 and port 2, waves with different AR
can be simulated.
Patterns with different AR values are generated and
decomposed into SSHs with 𝒂𝑙𝑙 , thus 𝜌𝑆𝑆 (𝛾) can be obtained
by using (8) with different rotation angles. Results are given in
Fig. 11 with minimum value markers. Finally, (16) is used to Fig. 16. Self-correlation measurement flowchart in the RC.
extract the AR values from 𝜌𝑆𝑆 (𝛾)𝑙𝑚𝑚 and compared with the
AR values obtained from CST (shown in Fig. 12). As can be horn antenna was used as the AUT. In the AR measurement, we
seen, a very good agreement is obtained. used a homemade wide band log-periodic cross dipole as the
AUT (Fig. 14). Measurement setup in the RC is also shown in
Fig. 15, foams were used to hold the AUT for three different
B. Measurements rotation axes. The measurement scenarios and procedures were
Measurements were conducted in the AC and RC at the the same for different AUTs and different rotation axes.
University of Liverpool, the radiation pattern measurement In the pattern reconstruction, the procedure was the same as
setup in the AC is shown in Fig. 13, the SATIMO® SH 2000 in the simulation, the only difference was that, 𝜌𝑆𝑆 (𝛼), 𝜌𝑆𝑆 (𝛽)

Copyright (c) 2017 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TAP.2016.2633901
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 7

(a)
(a)

(b)
Fig. 19. Measured and reconstructed radiation pattern in XOY and XOZ
plane, peak value is normalized to 0 dB, CP means co-polarization
(b) component, XP means cross-polarization component, (a) radiation pattern in
XOY plane, (b) radiation pattern in XOZ plane.

rotated for a full revolution, the stirrer position was fixed at one
position to make sure the environment was the same (shown in
Fig. 16), and we assumed that the size of the antenna was not so
large that rotating the antenna would perturb the field in the RC
greatly, otherwise the results would be decorrelated and always
gave small 𝜌𝑆 . After all the S-parameters were collected, (9)
was used to obtain the self-correlation coefficient for different
angles of 𝛼, 𝛽, and 𝛾.
(c) The measured 𝜌𝑆𝑆 are shown in Fig. 17, and the
Fig. 17. Measured and reconstructed 𝜌𝑆𝑆 of the AUT (SATIMO® SH 2000),
reconstructed 𝜌𝑆𝑆 from 𝒂𝑙𝑙 are also given. As expected they
(a), (b) and (c) are the self-correlation rotated around x-, y- and z-axis
respectively, the definition of three axes are shown in Fig. 13. agree well with the measured results. The reconstructed 3D
pattern from 𝒂𝑙𝑙 by using the summation of SSHs is also
shown in Fig. 18. To validate the results, the pattern in the XOY
plane and the XOZ plane were also measured in the AC, results
are compared in Fig. 19. As can be seen, very good agreement
is obtained for the main beam; however, the error becomes
large when the magnitude of the pattern becomes small (side
lobes, back lobes and cross-polarization). The maximum error
for the co-polarization component in the XOY plane occurs at
𝜃 = 90°, 𝜑 = 252° where the measured value is -23.8 dB and
the reconstructed value is -15.0 dB. In the XOZ plane, the
maximum error for the co-polarization component occurs at
𝜃 = 113°, 𝜑 = 180° where the measured value is -34.4 dB and
Fig. 18. Reconstructed E-field magnitude pattern (linear scale) of the AUT
(SATIMO® SH 2000), the maximum value is normalized to 1 V/m. the reconstructed value is -26.1 dB. This phenomenon is very
similar to that in the numerical simulation and will be discussed
and 𝜌𝑆𝑆 (𝛾) were measured in the RC rather than simulated. The in the next section.
self-correlation coefficients were measured at 4 GHz with 50 In the AR measurement in Fig. 14, the AUT (the measured
points of frequency stir (in 10 MHz bandwidth), the turntable S11 is given in Fig. 20) was rotated on the turntable with 1
was rotated with 1 degree/step for 360 degrees and the stirrers degree/step in the frequency range from 200 MHz to 5 GHz,
were rotated with 3 stirrer positions. Therefore, we had S-parameters were collected for each degree between the AUT
frequency stir [30], source stir [31-33] and mechanical stir for and the Tx antenna, the maximum and minimum values in a
each rotation angle, and 𝑁 = 50 × 360 × 3 = 54000 sample revolution are shown in Fig. 21. The AR values can be obtained
points in (9) for each angle of self-correlation coefficient using [34]
calculation. It should be noted that, the turntable and the stirrer
were not rotated simultaneously. When the turntable was

Copyright (c) 2017 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TAP.2016.2633901
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 8

Fig. 20. Measured 𝑆11 of the AUT (wide band log-periodic cross dipole).
Fig. 24. Average and maximum pattern error with different 𝐿, the average
error is defined as 20log�𝑚𝑚𝑚𝑚�|𝐄| − |𝐄′|�� and the maximum error is
defined as 20log�𝑚𝑚𝑚�|𝐄| − |𝐄′|��, where 𝐄 and 𝐄′ are the original pattern
and the pattern approximated by using SSHs respectively ( 𝐄 and 𝐄′ are
normalized to the peak value of 1V/m), mean and max means obtaining the
average and maximum value over all angles respectively; the error patterns
(20log�|𝐄| − |𝐄′|�) are also given when 𝐿 = 2, 4, 6, 8.

IV. DISCUSSION
It should be noted that, in both simulation and measurements,
the reconstructed patterns are similar to the original pattern but
Fig. 21. Measured maximum and minimum |𝑆21 | in a revolution in the AC.
not exactly the same (Fig. 8 and Fig. 18), possible error sources
and error analysis are discussed in this section.
1) The SSHs are truncated at level 𝐿.To analyze the error
caused by 𝐿, different 𝐿 are used to decompose the far-field in
Fig. 4, by comparing the far-field calculated from 𝒂𝑙𝑙 in (1)
with the original pattern, the truncation effect is shown in
Fig. 24, both average error and maximum error are given. As
can be seen, for the antenna pattern in Fig. 4, when 𝐿 = 8, the
average pattern error is quite small.
2) The inverse problem is a complex nonlinear problem and
could have multiple solutions. 𝜌𝑆𝑆 for all 𝛼, 𝛽, 𝛾 angles in Fig.
Fig. 22. Measured 𝜌𝑆𝑆 (𝛾) at all frequencies in the RC (linear scale). 5 and Fig. 6 can be calculated using (8) and is shown in Fig. 25.
In the measurements, because of the limitation of the facility,
only three cut planes were measured which correspond to the
values on the three axes in Fig. 25 (𝛼 = 1° ~ 360°, 𝛽 = 0,
𝛾 = 0 ; 𝛼 = 0 , 𝛽 = 1° ~ 360° , 𝛾 = 0 ; 𝛼 = 0 , 𝛽 = 0 , 𝛾 =
1° ~ 360°). As can be seen, at some regions these two 𝜌𝑆𝑆 are
different, which means that it could be possible to have two sets
of 𝜌𝑆𝑆 with the same value on three axes but have differences in
some regions. To quantify the relation between the pattern error
and the error in the self-correlation coefficients, 𝜌𝑆𝑆 was
perturbed with random values 2000 times. The pattern error
was calculated in each case. Results are given in Fig. 26 which
Fig. 23. Measured 𝜌𝑆𝑆 (𝛾) at all frequencies in the RC (linear scale). The shows a direct statistical understanding between the error in 𝜌
directivity in 1 GHz ~ 3.5 GHz is about 5 dBi ~ 6.5 dBi.
and the error in the pattern. It can be seen that, when the
|𝑆21 |𝑙𝑎𝑥 reconstructed 𝜌 is accurate, statistically it is more likely to get a
𝐴𝑅 = (18)
|𝑆21 |𝑙𝑚𝑚 more accurate pattern. It is also possible to sample 𝜌𝑆𝑆 (𝛼, 𝛽, 𝛾)
in 3D to reconstruct the radiation pattern. To simulate this
Similarly, the AR measurement was conducted in the RC using procedure, the original 𝜌𝑆𝑆 (𝛼, 𝛽, 𝛾) in Fig. 25(a) was sampled
(16), measured 𝜌𝑆𝑆 (𝛾) in the RC are shown in Fig. 22 at each in 3D with different degrees per step for all 𝛼, 𝛽, and 𝛾. Using
frequency and the minimum values were extracted to calculate the resampled 𝜌𝑆𝑆 to reconstruct the pattern, the pattern errors
the AR. Finally, the obtained AR in the AC and RC are shown with different step size can be obtained and are shown in
in Fig. 23, as can be seen, a very good agreement is obtained Fig. 27. As expected, more 3D samples can improve the
when the AUT is close to the circular polarization (1 GHz ~ 3.5 accuracy of the reconstructed pattern, but more time is needed
GHz). in the optimization.

Copyright (c) 2017 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TAP.2016.2633901
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 9

(a) (b)
Fig. 25. Calculated 𝜌𝑆𝑆 (𝛼, 𝛽, 𝛾) from (a) original 𝒂𝑙𝑙 in Fig. 5 and (b) Fig. 28. The average 𝜌 error (averaged over all selected sample angles) and
reconstructed 𝒂𝑙𝑙 in Fig. 6, a different region is marked with dotted circle (the the sample number, 𝜌 values with 54000 samples are used as the reference.
color represents the value of 10log𝜌𝑆𝑆 (𝛼, 𝛽, 𝛾), dB scale).

Fig. 29. Degradation curves with different directivity, AR curves deviate (16)
(AR from 𝜌𝑚𝑚𝑚 ) when directivity reduces in +z direction.
Fig. 26. The average 𝜌 error (averaged over all selected sample angles) and
the pattern error in dB: the average error is defined as 20log�𝑚𝑚𝑚𝑚�|𝐄| − cost-effective than an AC. There are also other advantages. For
|𝐄′|�� and the maximum error is defined as 20log�𝑚𝑚𝑚�|𝐄| − |𝐄′|��, where 𝐄 example, the proposed method can be conducted in 2D while
and 𝐄′ are the original pattern and the reconstructed pattern respectively (𝐄 the reconstructed radiation pattern is always in 3D. The
and 𝐄′ are normalized to the peak value of 1V/m): mean and max means the proposed method is based on the NLoS/stirred part of the
average and maximum value over all angles respectively.
measured S-parameters. The AUT and the Tx/Rx antenna do
not need to be carefully aligned as they would need to be in the
AC, which makes the measurement setup more robust and
insensitive to antenna positions. The measurement time is
shorter than some of the methods using the LoS/unstirred part
[13-16] (at each angle, the stirred part needs to be cancelled out
by averaging S-parameters at many stirrer positions or using the
Doppler shift). The measurement time could be even shorter
than the direct measurement of the 3D radiation pattern in the
(a) (b)
Fig. 27. (a) Typical reconstructed pattern error and the step size of 3D AC with an acceptable loss of accuracy, since 3D sample points
(𝛼, 𝛽, 𝛾), e.g. 10 degrees/step means all 𝛼, 𝛽 and 𝛾 are sampled every 10 could be much larger than 2D sample points. Also, more time is
degrees in the range of 0° to 360° , thus has 37 samples in each angle needed in the post processing of the measurement data.
dimension and 50653 samples in total. Definitions of average and maximum
are the same as in Fig. 26; (b) optimization time in hours for different
The AR of a directional antenna has also been measured
degrees/step, the results are based on the same initial values and convergence approximately in the RC. This is under the assumption that the
tolerance in the Levenberg-Marquardt algorithm. pattern integral is dominated by the main beam in (13). To
investigate how the accuracy degrades over the directivity, we
3) The inverse problem is a multi-goal optimization problem, use a pyramid to block the wave in +z direction. By tuning the
the reconstruction accuracy is limited by the size of the pyramid we can tune the directivity but not change
Levenberg-Marquardt algorithm; the optimized 𝒂𝑙𝑙 could be at the AR in +z direction. Results with different directivity are
a local minimum rather than the global minimum. Normally, shown in Fig. 29. As can be seen, even when the directivity is 0
the main beam with co-polarization has the highest magnitude dBi at +z direction, the maximum error of AR in (16) is 1.3.
thus has higher weight than the other components (side lobes, Therefore (16) is a very good approximation and different
back lobes, cross-polarizations, etc.). antenna may have different degradation curves.
4) To investigate the convergence of 𝜌, different sample It is also interesting to note that, by combing the existing
numbers are used to repeat the calculation. The average 𝜌 error measurement methods in the RC, with the same measurement
with different sample numbers is shown in Fig. 28. As can be setups but different data post-processing techniques, nearly all
seen, when the sample number is large the measured 𝜌 antenna parameters (such as radiation efficiency [1], 3D pattern
converges with small uncertainties. in this paper, 𝑆11 [6] and gain) can be obtained in one
Although the reconstructed pattern in the RC is not as measurement.
accurate as that measured in the AC, an RC is more There are also potential issues: when the directivity of the

Copyright (c) 2017 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TAP.2016.2633901
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 10

reverberation chamber,” IEEE Antennas Wireless Propag. Lett., vol. 15,

pp. 336-339, 2016.
[6] P. -S. Kildal, C. Carlsson and J. Yang, “Measurement of free-space
impedances of small antennas in reverberation chambers,” Microw. Opt.
Technol. Lett., vol. 32, pp. 112-115. 2002.
[7] P.-S. Kildal and K. Rosengren, “Correlation and capacity of MIMO
systems and mutual coupling, radiation efficiency, and diversity gain of
their antennas: Simulations and measurements in a reverberation
chamber,” IEEE Commun. Mag., vol. 42, no. 12, pp. 104–112, Dec. 2004.
[8] K. Rosengren and P.-S. Kildal, “Radiation efficiency, correlation,
diversity gain and capacity of a six-monopole antenna array for a MIMO
system: Theory, simulation and measurement in reverberation chamber,”
Proc. Inst. Elect. Eng., Microw. Antennas Propag., vol. 152, no. 1, pp. 7–
Fig. 30. An overview of the framework in this paper. 16, Feb. 2005.
[9] P.-S. Kildal, K. Rosengren, J. Byun, and J. Lee, “Definition of effective
AUT is high (e.g. a parabolic antenna with a large reflector), gain and how to measure it in a reverberation chamber,” Microw. Opt.
𝜌𝑆𝑆 becomes uncorrelated very quickly, and a very fine rotation Technol. Lett., vol. 34, no. 1, pp. 56–59, Jul. 2002.
[10] Q. Xu, Y. Huang, X. Zhu, S. S. Alja’afreh and L. Xing, “A new antenna
step for the turntable is needed which could be time consuming diversity gain measurement method using a reverberation chamber,”
for a whole revolution. Also, to approximate a high directivity IEEE Antennas Wireless Propag. Lett., vol. 14, pp. 935-938, 2015.
pattern, more coefficients are required [35], which further [11] J. Koh, A. De, T. K. Sarkar, H. Moon, W. Zhao and M. S.-Palma, “Free
space radiation pattern reconstruction from non-anechoic measurements
increases the optimization time when solving (17), and the using an impulse response of the environment,” IEEE Trans. Antennas
unknown numbers increase quickly at the order of 𝑂(𝐿2 ). For Propag., vol. 60, no. 2, pp. 821–831, Feb. 2012.
𝐿 = 8 in this paper, the optimization time is around 2 hours on [12] V. Fiumara, A. Fusco, V. Matta, and I. M. Pinto, “Free-space antenna
field-pattern retrieval in reverberation environments,” IEEE Antennas
a personal computer. How to quickly reconstruct 𝒂𝑙𝑙 of high Wireless Propag. Lett., vol. 4, pp. 329–332, 2011.
directivity antenna could be challenging. [13] P. Besnier, C. Lemonie, J. Sol and J. –M. Floc’h, “Radiation pattern
measurements in reverberation chamber based on estimation of coherent
and diffuse electromagnetic fields,” in IEEE Conference on Antenna
Measurements & Applications (CAMA), pp.1-4, 16-19 Nov. 2014.
V. CONCLUSIONS [14] C. Lemoine, E. Amador, P. Besnier, J. –M. Floc’h and A. Laisne,
“Antenna directivity measurement in reverberation chamber from Rician
The concept of the self-correlation coefficient has been K-factor estimation,” IEEE Trans. Antennas Propagat., vol. 61, no. 10,
introduced in this paper. It has been shown for the first time that pp. 5307-5310, Oct. 2013.
the 3D radiation pattern of the AUT can be reconstructed using [15] M. Á. García-Fernández, D. Carsenat, and C. Decroze, “Antenna
radiation pattern measurements in reverberation chamber using plane
the self-correlation coefficient of the antenna pattern which can wave decomposition,” IEEE Trans. Antennas Propag., vol. 61, no. 10, pp.
be measured in an RC. The proposed method only used the 5000–5007, Oct. 2013.
NLoS components, it can be considered as a generalized mode [16] M. Á. García-Fernández, D. Carsenat and C. Decroze, “Antenna gain and
radiation pattern measurements in reverberation chamber using Doppler
matching method which optimizes the SSH coefficients to effect,” IEEE Trans. Antennas Propagat., vol. 62, no. 10, pp. 5389-5394,
match the measured self-correlation coefficients. This results in Oct. 2014.
a system of nonlinear equations which can be solved/optimized [17] H. Moussa, A. Cozza and M. Cauterman, “A novel way of using
using the well-developed Levenberg-Marquardt algorithm [29]. reverberation chambers through time reversal,” in ESA Workshop on
Aerospace EMC (ESA’09), pp. 10-2, Mar. 2009.
It has also been shown that the AR of an antenna can be [18] A. Cozza and A. Abou el-Aileh, “Accurate radiation-pattern
measured in an efficient manner (rotated around one axis) in an measurements in a time-reversal electromagnetic chamber,” IEEE
RC. An overview of the framework of this novel method is Antennas and Propagation Magazine, vol. 52, no. 2, pp. 186-193, Apr.
2010.
shown in Fig. 30. Simulations and measurements have been [19] F. Monsef, A. Cozza, P. Meteon and M. Djedidi, “Preliminary results on
conducted to verify the proposed theory. Error sources have antenna testing in reverberating environments,” in IEEE Conference on
been analyzed and quantified, and have been shown to be small Antenna Measurements & Applications (CAMA), pp. 1-4, 16-19 Nov.
2014.
for all of the cases considered in this paper. [20] J. Jackson, Classical Electrodynamics. New York: Wiley, 1983.
[21] Y. Chen and T. Simpson, “Radiation pattern analysis of arbitrary wire
antennas using spherical mode expansions with vector coefficients,”
IEEE Trans. Antennas Propag., vol. 39, no. 12, pp. 1716–1721, Dec.
REFERENCES 1991.
[1] C. L. Holloway, H. A. Shah, R. J. Pirkl, W. F. Young, D. A. Hill and J. [22] G. D. Galdo, J. Lotze, M. Landmann and M. Haardt, “Modelling and
Ladbury, “Reverberation chamber techniques for determining the manipulation of polarimetric antenna beam patterns via spherical
radiation and total efficiency of antennas,” IEEE Trans. Antennas harmonics,” in 14th European Signal Processing Conference, pp.1-5, 4-8
Propagat., vol. 60, no. 4, pp. 1758-1770, Apr. 2012. Sep. 2006.
[2] S. J. Boyes, P. J. Soh, Y. Huang, G. A. E. Vandenbosch and N. Khiabani, [23] J. Rahola, F. Belloni and A. Richter, “Modelling of radiation patterns
“Measurement and performance of textile antenna efficiency on a human using scalar spherical harmonics with vector coefficients,” in 3rd
body in a reverberation chamber,” IEEE Trans. Antennas Propagat., vol. European Conference on Antennas and Propagation (EuCAP),
61, no. 2, pp. 871-881, Feb. 2013. pp.3361-3365, 23-27 Mar. 2009.
[3] X. Chen, “On statistics of the measured antenna efficiency in a [24] M. Mhedhbi, S. Avrillon and B. Uguen, “Comparison of vector and scalar
reverberation chamber,” IEEE Trans. Antennas Propagat., vol. 61, no. spherical harmonics expansions of UWB antenna patterns,” in
11, pp. 5417-5424, Nov. 2013. International Conference on Electromagnetics in Advanced Applications
[4] X. Chen, “Generalized statistics of antenna efficiency measurement in a (ICEAA), pp.1040-1043, 9-13 Sep. 2013.
reverberation chamber,” IEEE Trans. Antennas Propagat., vol. 62, no. 3, [25] J. Yang, S. Pivnenko, T. Laitinen, J. Carlsson and X. Chen,
pp. 1504-1507, Mar. 2014. “Measurements of diversity gain and radiation efficiency of the eleven
[5] Q. Xu, Y. Huang, X. Zhu, L. Xing, Z. Tian and C. Song, “A modified antenna by using different measurement techniques,” in 4th European
two-antenna method to measure the radiation efficiency of antennas in a

Copyright (c) 2017 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TAP.2016.2633901
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 11

Conference on Antennas and Propagation (EuCAP), pp.1-5, 12-16 Apr.

2010.
[26] X. Chen, P. –S. Kildal and J. Carlsson, “Comparisons of different
methods to determine correlation applied to multi-port UWB eleven
antenna,” in 5th European Conference on Antennas and Propagation
(EuCAP), pp.1776-1780, 11-15 Apr. 2011.
[27] Z. Gimbutas, L. Greengard, “A fast and stable method for rotating
spherical harmonic expansions,” Journal of Computational Physics, vol.
228, no. 16, pp. 5621-5627, 2009.
[28] P. Hallbjorner, “Accuracy in reverberation chamber antenna correlation
measurements," in International Workshop on Antenna Technology:
Small and Smart Antennas Metamaterials and Applications (IWAT '07),
pp.170-173, 21-23 Mar. 2007 .
[29] D. W. Marquardt, “An algorithm for least-squares estimation of nonlinear
parameters,” Journal of the Society for Industrial and Applied
Mathematics, vol. 11, no. 2, pp.431–441, 1963.
[30] D. A. Hill, “Electronic mode stirring for reverberation chambers,” IEEE
Trans. EMC, vol. 36, no. 4, pp. 294-299, Nov. 1994.
[31] P. –S. Kildal, C. Carlsson and J. Yang, “Measurement of free-space
impedances of small antennas in reverberation chambers,” Microwave
and Optical Technology Letters, vol. 32, no. 2, pp. 112-115, Dec. 2001.
[32] Y. Huang and D. J. Edwards, “A novel reverberating chamber:
source-stirred chamber,” in Proceeding of IEE 8th International
Conference on EMC, pp. 120-124, Sep. 1992.
[33] K. Rosengren, P. –S. Kildal, C. Carlsson and J. Carlsson,
“Characterization of antennas for mobile and wireless terminals in
reverberation chambers: improved accuracy by platform stirring,”
Microwave and Optical Technology Letters, vol. 39, no. 6, pp. 391-397,
Sep. 2001.
[34] IEEE Standard Test Procedures for Antennas, IEEE Std, 149-1979.
[35] F. Jensen and A. Frandsen, “On the number of modes in spherical wave
expansions,” TICRA, Copenhagen, Denmark, Jan. 2005 [Online].
Available: http://www.ticra.com.

Copyright (c) 2017 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].