Hindawi

International Journal of Antennas and Propagation

Volume 2019, Article ID 1523469, 12 pages
https://doi.org/10.1155/2019/1523469

Research Article
On Calibration and Direction Finding with
Uniform Circular Arrays

Stephan Häfner , Martin Käske, and Reiner Thomä

Electronic Measurements and Signal Processing Group, Technische Universität Ilmenau, 98684 Ilmenau, Germany

Correspondence should be addressed to Stephan Häfner; [email protected]

Received 24 February 2019; Revised 15 May 2019; Accepted 27 May 2019; Published 3 July 2019

Academic Editor: Chien-Jen Wang

Copyright © 2019 Stephan Häfner et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Antenna array calibration methods and narrowband direction finding (DF) techniques will be outlined and compared for a uniform
circular array. DF is stated as an inverse problem, which solution requires a parametric model of the array itself. Because real arrays
suffer from mechanical and electrical imperfections, analytic array models are per se not applicable. Mitigation of such disturbances
by a global calibration matrix will be addressed, and methods to estimate this calibration matrix will be recapped from literature.
Also, a novel method will be presented, which circumvents the problem of a changed noise statistic due to calibration. Furthermore,
local calibration, where array calibration measurements are incorporated in the DF algorithm, is considered as well. Common
DF algorithms will be outlined, their assumptions regarding array properties will be addressed, and required preprocessing steps
such as the beam-space transformation will be presented. Also, two novel DF techniques will be proposed, based on the Capon
beamformer, but with reduced computational effort and higher resolution for bearing estimation. Simulations are used to exemplary
compare calibration and DF methods in conjunction with each other. Furthermore, measurements with a single and two coherent
sources are considered. It turns out that global calibration enables computational efficient DF algorithms but causes biased estimates.
Furthermore, resolution of two coherent sources necessitates array calibration.

1. Introduction as, e.g., mutual coupling between the sensors or the support
structure of the array, unknown sensor gain, and phase or
Direction finding (DF) is a task which occurs in several mechanical imperfections [1]. Consequently, DoA estimation
applications of surveillance, reconnaissance, radar, or sonar. performance degrades, because the assumed array model
Basically, DF can be defined as estimation of the bearing of does not coincide with the real array characteristics. Hence,
one or multiple signal sources with respect to (w.r.t.) a refer- calibration is necessary to mitigate these imperfections.
ence point in space. Typically, an array of spatially distributed All investigations are subject to a uniform circular array
sensors is placed at this reference point and the array output is (UCA). UCAs feature a very attractive geometry, because
exploited for DF. Hence, DF estimation is an inverse problem. their aperture covers the whole azimuth range and hence
Solving the inverse problem requires a parametric model ambiguous free AoA estimates are ensured, on the con-
of the array output in terms of the parameters of interest: trary to, e.g., uniform linear array (ULA). Also, UCAs can
azimuth of arrival (AoA) 𝜑 and elevation of arrival (EoA) 𝜗, be employed to estimate elevation too, but generally not
which together define the direction of arrival (DoA). unambiguous. For simplification, only AoA estimation and
In order to derive a parametric model of the array copolarised sources w.r.t. the array sensors are considered.
output, a model of the sensor array itself is necessary. The For the conducted investigations it is not necessary to con-
array model highly depends on the array geometry and sider elevation and arbitrarily polarised sources. However,
the characteristic of each sensor. Theoretical array models neglecting source polarisation and assuming fix elevation
typically assume omnidirectional sensors and an ideal array may result in biased estimates in real DF applications [2].
geometry, which cannot be assured for real arrays. Apart The goal of this paper is to jointly investigate array
from these assumptions, real arrays suffer from disturbances calibration methods and narrowband DF techniques. Global
2 International Journal of Antennas and Propagation

calibration, where a direction independent calibration matrix z

is used, will be considered. Methods to estimate the global k
calibration matrix from array calibration measurements are
k
reviewed and a new method is proposed, which accounts
for the change of the noise statistic due to the application
of the calibration matrix. Also, local calibration, where the
array calibration data are incorporated in the DF method, 
is considered. Known DF techniques will be outlined and y

two novel DF techniques based on the Capon beamformer x
will be proposed, featuring a reduced computational effort
and better resolution in case of multiple sources. Some of
the considered DF techniques take advantage of special array
structures, which are not provided by UCAs. Hence, beam-
space transformation will be briefly recapped. Simulations
and measurements are employed for the investigations. Mea- Figure 1: Spherical coordinate system and angle definition.
surements with two coherent sources, hence sources with a
fix phase relation, will be considered. Resolution of coherent
sources is crucial in DF [3], because of the rank-degeneration
of the spatial covariance matrix. Hence, the coherent source response is commonly denoted as steering vector. The vectors

푇
푇
case will be used as benchmark to justify calibration necessity s(𝑡) = [𝑠1 (𝑡), . . . , 𝑠
푃 (𝑡)] ∈ C
푃×1 , 𝜑 = [𝜑1 , . . . , 𝜑
푃 ] ∈
and to investigate the DF accuracy. R , and matrix B(𝜑) = [b(𝜑1 ), . . . , b(𝜑
푃 )] ∈ C
푀×
푃

푃×1

The reminder of the paper is organised as follows: a contain the complex envelope of the source signals, the
parametric model of the array output is derived in Section 2. AoAs of all sources, and the steering vectors, respectively.
In Section 3 the DF techniques are presented. The beam- Measurement noise and uncertainties due to, e.g., model
space transformation for UCA is described in Section 4. The errors are accounted for by an additive error term n(𝑡). This
problem of array calibration and its influence on the sensor error term is modelled as a zero-mean and proper complex
characteristic is presented in Section 5. Simulation based normal distributed random process, which is spatially white
comparison of calibration and DF methods is presented in and homogeneous, and uncorrelated with the source signal:
Section 6. In Section 7, the DF methods are compared using n(𝑡) ∼ CN(0, 𝜎2 I
푀). In summary, the model for the observa-
measurements with a single source and two coherent sources. tions y(𝑡) ∈ C
푀×1 is [1]
Section 8 concludes the paper.
Mathematical notation is as follows: scalars are italic y (𝑡) = B (𝜑) ⋅ s (𝑡) + n (𝑡) . (2)
letters. Vectors are in column format and written as boldface,
lower-case, italic letters. Matrices correspond to boldface, In practice 𝑁 snapshots are taken from the 𝑀 sensors.
upper-case letters. The matrix operations (.)
푇 , (.)
퐻, (.)−1 , Accordingly, the model of the array output becomes
and (.)† are defined as the transpose, conjugate transpose,
inverse, and Moore-Penrose pseudo inverse of a matrix, Y = B (𝜑) ⋅ S + N. (3)
respectively. The Frobenius norm of a matrix is stated as ‖.‖F .
The objective of direction finding is to estimate the AoAs
The imaginary unit is defined as 𝚥 = √−1.
from the array observations Y ∈ C
푀×
푁, hence solving the
inverse problem. The source directions can be uniquely and
2. Measurement Data Model ambiguous-freely determined, if this inverse problem is well
posed, e.g., the steering matrix features full column rank. In
DoA estimation requires a parametric model of the measure- the following, the number of sources 𝑃 is assumed as known,
ment data in terms of the DoAs. Consider an array of 𝑀 see, e.g., [5] for a summary of estimation methods.
sensors, having its reference point in the origin of a spherical
coordinate system. Directions of impinging waves are defined
w.r.t. this origin in terms of AoA 𝜑 and EoA 𝜗; see Figure 1. 2.1. UCA Element-Space Model. Consider an uniform circu-
Consider 𝑃 plane waves, emerging from far field sources and lar array with equiangular spaced omnidirectional sensors,
impinging at the array. The waves are assumed to impinge in placed on a circumference of radius 𝑅. The array steering
the azimuth plane, hence 𝜗 = 0∘ holds. Under narrowband vector entry of the 𝑚-th sensor and azimuth only is [6]
assumption [1, 4], the array output y(𝑡) ∈ C
푀×1 in the com-
plex baseband can be approximated as 𝑏
푚 (𝜑) = exp (𝚥𝑘𝑅 cos (𝜑 − 𝑚 ⋅ 𝜙)) , (4)

푃 with wave number 𝑘 = 2𝜋/𝜆
푐 , 𝑚 = 0, . . . , 𝑀 − 1, and the

y (𝑡) ≈ ∑ b (𝜑
푝 ) ⋅ 𝑠
푝 (𝑡) = B (𝜑) ⋅ s (𝑡) , (1) angular spacing of the sensors 𝜙 = 2𝜋/𝑀. In the following,

푝=1 (4) is referred to as the UCA model in element-space. Prac-
tical UCAs do not follow this model due to, e.g., mechanical
with vector b(𝜑
푝 ) denoting the narrowband array response and electrical imperfections. Hence, calibration is necessary
w.r.t. the impingement angle 𝜑
푝 . The narrowband array (see Section 5).
International Journal of Antennas and Propagation 3

Table 1: Considered narrowband DF techniques, the number of identifiable sources, and the considered structure of the steering vectors.

DF Methods Reference No. sources Steering vector

CML [7, 8] ≤𝑀−1 arbitrary
UML [7, 8] ≤𝑀−1 arbitrary
MUSIC-1 [9] ≤𝑀−1 arbitrary
root-MUSIC [10] ≤𝑀−1 Vandermonde
MUSIC-2 [11] ≤𝑀−1 arbitrary
ESPRIT [12] ≤𝑀−1 Vandermonde
MODE [8, 13] ≤𝑀−1 arbitrary
IQML [14] ≤𝑀−1 Vandermonde
Capon-1 [15] 1 arbitrary
Bartlett [15] 1 arbitrary
root-Capon This work 1 Vandermonde
Capon-2 This work 1 arbitrary

3. Narrowband Direction Finding Techniques Subsequently, two estimators are proposed, which employ
polynomial rooting instead of a 1D peak search to estimate
Several DF techniques are known from literature; see [1, 15] the AoAs from the Capon spectrum. The rooting is compu-
for an overview of the most famous ones. The considered tational more efficient, especially if multiple sources shall be
DF methods, the number of sources they are able to resolve, resolved and also has a better resolution compared to spectral
and their assumptions regarding the steering vector struc- methods [10].
ture are summarised in Table 1. The Bartlett and Capon
methods are considered to resolve a single source only,
because their resolution depends on the array aperture [15]. 3.1.1. Root-Capon. Restating cost function (5) as a minimisa-
Hence, the sources have to be well separated in order to tion problem gives
resolve them. According to this restriction, the beamformers
퐻
are not high-resolution estimators and hence not generally ̂−1 ⋅ b (𝜑) .
arg min b (𝜑) ⋅ R (6)
YY

휑
applicable for the resolution of multiple sources. However,
the beamformers will be used to resolve multiple sources If the steering vectors b(𝜑) feature Vandermonde structure,
to show the improved estimation capability by calibration. minimisation is accomplished by estimating the 𝑃 roots
The DF methods constrained maximum-likelihood (CML), closest to the unit circle.
unconstrained maximum-likelihood (UML), and Method of
Direction Estimation (MODE) require the optimisation of
a multidimensional, nonconvex cost function [15]. Opti- 3.1.2. Capon-2. Another DF estimator based on the Capon
misation of nonconvex cost functions requires iterative or beamformer is derived by exploiting array manifold separa-
heuristic methods, which are computational cumbersome tion [11]. In manifold separation the array steering vector is
and may end up in local optima. The Multiple Signal decomposed in the product of a vector d(𝜑) ∈ C2
퐿+1×1 and
Classification (MUSIC)-1, Bartlett, and Capon-1 are spectral an array specific sampling matrix G
methods, which require a 1D peak search for DF. The peak
search is complicated, if multiple peaks have to be detected to b (𝜑) = G ⋅ d (𝜑) . (7)
resolve multiple sources. The iterative quadratic maximum-
푇
likelihood (IQML), root-MUSIC, MUSIC-2, root-Capon, Vector d(𝜑) = [exp(−𝚥𝐿𝜑), . . . , exp(𝚥𝐿𝜑)] features Vander-
Capon-2, and Estimation of Signal Parameters via Rotational monde structure and depends on the direction only.
Invariance Technique (ESPRIT) require polynomial rooting Considering the cost function (6) and plugging in the
or Eigenvalue decomposition, such that these methods are manifold separation (7) gives
computationally more efficient. However, some DF methods
퐻
require a special steering vector structure, which is not ̂−1 ⋅ G ⋅ d (𝜑) .
arg min d (𝜑) ⋅ G
퐻 ⋅ R (8)
YY

휑
fulfilled by an UCA and therefore beam-space transformation
is necessary; see Section 4. Again, minimisation is accomplished by estimating the 𝑃
roots closest to the unit circle.
3.1. Novel Capon Beamformer Estimators. Generally, the
Capon beamformer attempts to minimise the power con-
tribution from interferer directions, while maintaining the 4. Beam-Space Transformation
gain in the direction of interest. The estimator is given by The ESPRIT, IQML, root-Capon, and root-MUSIC algorithm
maximising a 1D spatial spectrum [15] are naturally applicable for DF with ULAs. Hence, they

퐻 −1 necessitate steering vectors with Vandermonde structure.
̂−1 ⋅ b (𝜑)] .
arg max [b (𝜑) ⋅ R (5)
YY According to (4), the steering vectors of an UCA do not

휑
4 International Journal of Antennas and Propagation

feature a Vandermonde structure in element-space. Utilising 5. Sensor Array Calibration

the concept of phase mode excitation [21, 22], the element-
space steering vectors are transformed to the beam-space, 5.1. Global vs. Local Array Calibration
where the steering vectors feature Vandermonde structure.
Phase mode excitation exploits the Jacobi-Anger expansion, 5.1.1. Global Calibration. In global calibration, the disturbed
which is for the steering vector entry of the 𝑚-th sensor array output is mapped onto a reference array output, whereas
the disturbances are assumed as independent on the direc-
exp (𝚥𝑘𝑅 cos (𝜑 − 𝑚𝜙)) tion of impingement. In the following, a linear relationship
between disturbed and reference array output is assumed.
∞ (9) Then, global array calibration is done by a calibration matrix
= ∑ 𝚥
푛 ⋅ 𝐽
푛 (𝑘𝑅) ⋅ exp (𝚥𝑛𝜑) ⋅ exp (−𝚥𝑚𝜙) ,

푛=−∞
C ∈ C
푀×
푀, mapping the reference array output m(𝜑) onto
the disturbed one m ̃ (𝜑)
with 𝐽
푛 being the Bessel function of first kind and 𝑛-th order.
Approximating the infinite series by a finite one yields C−1 : m (𝜑) 󳨃󳨀→ m
̃ (𝜑) . (15)

exp (𝚥𝑘𝑅 cos (𝜑 − 𝑚𝜙)) Accordingly, calibration of the array is given by

푀󸀠
≈ ∑ 𝚥
푛 ⋅ 𝐽
푛 (𝑘𝑅) ⋅ exp (𝚥𝑛𝜑) ⋅ exp (−𝚥𝑛𝑚𝜙) y
푐 (𝑡) = C ⋅ y (𝑡) , (16)
(10)

푛=−
푀󸀠
with the vector y
푐 (𝑡) of calibrated array outputs. Note that

푇
= d (𝑚𝜙) ⋅ V ⋅ a (𝜑) , applying the calibration matrix changes the second-order
statistics of the noise, which becomes 𝜎2 CC
퐻. In the simula-
with 𝑀耠 = ⌊(𝑀 − 1)/2⌋ [3] and tion and measurement section it will not be accounted for the
changed noise statistics, in order to investigate the influence

푇
d (𝑚𝜙) = [exp (−𝚥𝑀耠 𝑚𝜙) , . . . , exp (𝚥𝑀耠 𝑚𝜙)] (11a) of global calibration on the estimator performance.
The calibration matrix is derived from array calibration
󸀠 󸀠
V = diag {[𝚥−
푀 𝐽−
푀󸀠 (𝑘𝑅) , . . . , 𝚥
푀 𝐽
푀󸀠 (𝑘𝑅)]} (11b) measurements, which are conducted on a test range or in
an anechoic chamber. These measurements are a set of array

푇 outputs for known directions of impingement and comprise
a (𝜑) = [exp (−𝚥𝑀耠 𝜑) , . . . , exp (𝚥𝑀耠 𝜑)] . (11c)
the array characteristics as well as the disturbances. Introduce
Vector a(𝜑) denotes the Vandermonde structured steering the matrices
vector in beam-space. The truncated Jacobi-Anger expansion
for the array steering vector according to (4) is Bref (𝜑) = [bref (𝜑1 ) , . . . , bref (𝜑
푉 )] ∈ C
푀×
푉 (17a)

b (𝜑) = D
푇 ⋅ V ⋅ a (𝜑) , (12) B (𝜑) = [b (𝜑1 ) , . . . , b (𝜑
푉)] ∈ C
푀×
푉 (17b)

󸀠 A (𝜑) = [a (𝜑1 ) , . . . , a (𝜑
푉 )] ∈ C
푀×
푉 , (17c)

with matrix D = [d(0𝜙), . . . , d((𝑀 − 1)𝜙)] ∈ C2
푀 +1×
푀 being
the Discrete Fourier Transform (DFT) matrix. The phase
mode excitation or beam-space transformation is now given whereas Bref comprises the steering vectors from the cali-
by bration measurements, and B and A comprise the steering
vectors from the sampled element- and beam-space array
† model, respectively. The calibration matrix can be calculated
(D
푇 ⋅ V) ⋅ b (𝜑) = a (𝜑) . (13)
in element- and beam-space. Considering the array model
In order to yield the array output in beam-space yBS (𝑡), the in element-space, the calibration matrix is calculated in
transformation is applied to the array output y(𝑡) element-space.

†
yBS (𝑡) = (D
푇 ⋅ V) ⋅ y (𝑡) . (14) CES : Bref (𝜑) 󳨃󳨀→ B (𝜑) (18)

The truncated Jacobi-Anger expansion introduces a system- Application of matrix CES to the array output yields the
atic error, which depends on the actual AoA and causes an calibrated array output in element-space. Hence, subsequent
estimation bias or increased estimation variance [23]. Fur- transformation to beam-space may be necessary depending
thermore, beam-space transformation changes the second- on the AoA estimator. Considering the array model in beam-
order statistic of the noise, which is no longer homogeneous space, the calibration matrix is calculated in beam-space.
over the array channels.
The beam-space transformation assumes an UCA, which CBS : Bref (𝜑) 󳨃󳨀→ A (𝜑) (19)
follows the element-space model (4). This model does not
apply for real arrays, such that calibration of the array is Application of matrix CBS to the array output yields the
necessary. calibrated array output in beam-space.
International Journal of Antennas and Propagation 5

Table 2: Considered methods to estimate calibration matrix C.

Method Reference Estimation space
Wax [3] element- and beam-space
Sommerkorn [16] element- and beam-space
Kortke [17] element- and beam-space
Friedlander [18] element-space
Ng [19] element-space
See [20] element- and beam-space
Haefner This work element- and beam-space

5.1.2. Local Calibration. In local calibration, the array dis-

turbances are considered as depending on the direction
of impingement. Consideration of direction dependent dis-
turbances is accomplished by using the array calibration
measurements Bref (𝜑) as the reference steering vectors in
the DF algorithms [2, 24]. Consequently, the estimators
ESPRIT, IQML, root-MUSIC, and root-Capon are not appli-
cable under local calibration, because these methods cannot
incorporate arbitrary or measured steering vectors [15].
Because the calibration measurements describe the array
for discrete angles only, whereas DF algorithms require a
continuous description, interpolation is required. Here, the
EADF [25] is used to interpolate the calibration measure-
ments.

5.2. Estimation of Global Calibration Matrix. Table 2 sum- Figure 2: Poynting DF-A0046 UCA (dashed box) featuring 5
vertically polarised dipoles and 5 horizontally polarised monopoles.
marises the considered methods to estimate the global cal-
Only the dipoles will be considered in this paper.
ibration matrix. Depending on the considered array model
and the estimation method, the calibration matrix can be
estimated in element- or beam-space.
First, the influence of global calibration on the array
5.2.1. Haefner-Method. As pointed out, the second-order geometry is investigated. As stated previously, practical arrays
statistic of the noise is changed by applying the calibration suffer from mechanical imperfections, such that the assumed
matrix. This can deteriorate the performance of some DF circular geometry is not assured. The estimated and assumed
methods, because they assume the noise covariance matrix sensor positions are shown in Figure 3. It becomes obvi-
to be diagonal. Introducing the constrain CC
퐻 = I regarding ous that the real array does not feature UCA properties
the estimation of the calibration matrix, the noise statistic as, e.g., equiangular spaced sensors. After calibration the
will not change after calibration. Applying this constraint to sensor positions are slightly corrected. Hence, geometrical
the method of Wax results in novel method to estimate the imperfections can be corrected to some extent by global
calibration matrix. calibration.
󵄩 󵄩2 Furthermore, the effect of calibration on the sensor
arg min 󵄩󵄩󵄩󵄩C ⋅ Bref (𝜑) − M (𝜑)󵄩󵄩󵄩󵄩F , characteristics is investigated. Magnitude and phase of the
C
(20) model and the sensor response before and after calibration
s.t. CC
퐻 = I are shown in Figures 4(a) and 4(b), respectively. From
Figure 4(a) it becomes obvious that the vertical dipole is
Hence, the objective is to minimise the cost function subject shadowed by the array mast, resulting in an attenuation
to the constraint that the resulting matrix is unitary [26]. of up to 10 dB. After calibration, this strong attenuation is
corrected, resulting in a more omnidirectional characteristic
5.3. Impact of Global Calibration on Array Characteristics. of the sensor. Also, the direction cosine of the phase becomes
In order to investigate the effect of calibration on the array much closer to the model after calibration; see Figure 4(b).
characteristic, a real array will be considered. The array
under consideration is the Poynting DF-A0046 UCA (see 6. Simulation Based Studies
Figure 2), which operates at 305 MHz centre frequency. The
array features 5 vertically polarised dipoles and 5 horizontally In order to compare the various calibration matrix estima-
polarised monopoles, whereas the dipoles will be considered tors and the DF techniques, Monte-Carlo simulations with
only. varying signal to noise ratio (SNR) are carried out. Data are
6 International Journal of Antennas and Propagation

0.4 utilised for AoA estimation in beam-space and element-

space, respectively. For the ESPRIT estimator, calibration
according to the array model in beam- and element-space
0.2 with subsequent beam-space transformation is applied. The
resulting RMSEs are shown in Figures 5(b) and 5(a). The
RMSEs for MUSIC-1 based estimation are shown in Figure 6.
Obviously, the RMSE curves for all calibration methods
y / c

0
converge to a certain value, indicating biasedness. Further-
more, the ESPRIT estimates never attain the CRLB, because
the beam-space transformation introduces errors resulting
−0.2 in an increased variance of the estimates. Furthermore, the
RMSEs of the ESPRIT estimates indicate that calibration
w.r.t. the array model in element-space slightly outperforms
−0.4 the calibration w.r.t. the array model in beam-space. An
−0.4 −0.2 0 0.2 0.4 explanation can be given by the global calibration matrix
x/ c itself. Basically, the matrix describes disturbances due to cou-
Assumed pling or electrical and mechanical imperfections. Therefore,
Uncalibrated the calibration matrix has a clear physical meaning in the
Calibrated element-space. Calibration matrix estimation w.r.t. to the
Figure 3: Normalised sensor positions in the x-y-plane (array beam-space model assumes a virtual array, such that the
top view) as assumed by the model, and before and after global calibration matrix has no longer a clear physical meaning.
calibration. The sensor positions are normalised to the centre Hence, the disturbances are not described properly and the
wavelength 𝜆
푐 . calibration becomes less powerful. The proposed estimation
method performs comparably worse, which can be related to
the constraint of a Hermitian calibration matrix causing a less
Table 3: Parameters of the simulations. powerful calibration. Hence, variation of the noise statistic
Parameter Value is not as a crucial for the bearing estimation as remaining
Transmit signal zero-mean, circularly normal distributed calibration errors. Comparison of the RMSEs of the ESPRIT
Centre frequency 305 MHz and MUSIC-1 estimator indicates that the MUSIC-1 estimator
attains the CRLB for SNRs around 0 dB, but the ESPRIT
Receive array Poynting DF-A0046
estimator slightly outperforms the MUSIC-1 estimator in
SNR −10 dB to 40 dB
terms of minimal achievable RMSE. Overall, the estimation
No. snapshots 50 method by Sommerkorn in conjunction with the ESPRIT
AoAs uniformly distributed performs best.

6.2. Comparison of DF Techniques. First, the considered

generated according to (3), whereas a single source is consid- DF techniques in conjunction with global calibration will
ered. The simulation parameters are summarised in Table 3. be compared. The global calibration matrix is estimated in
As receive array the Poynting DF-A0046 UCA is used, and element-space by the method of Sommerkorn. The calculated
calibration data of the array are used as steering vectors for RMSEs are shown in Figure 7(a). All estimators saturate to
the generation of the data. Figure of merit for comparison a certain RMSE for high SNRs, such that the estimators in
is the root mean-square error (RMSE) of AoA estimates. conjunction with the global calibration are not considerable
The estimation error is defined as the orthodromic angular as unbiased. Because of, e.g., the erroneous sensor positions
distance 𝜁 between the given AoA 𝜑 and the estimated AoA due to mechanical imperfections, array disturbances are
𝜑̂ direction dependent. However, global calibration attempts
󵄨 󵄨 to correct the array disturbances according to an average
𝜁 = arc cos (cos (󵄨󵄨󵄨𝜑 − 𝜑̂󵄨󵄨󵄨)) . (21) disturbance over the whole range of impingement directions.
Hence, some model errors remain after calibration, such
The RMSE is calculated by averaging the squared estimation that the estimators cannot be unbiased. The CML, UML,
error over 1000 Monte-Carlo runs, whereas each run features MODE, MUSIC-1, MUSIC-2, and Bartlett beamformer attain
fix SNR and random AoA. The derived RMSEs are compared the CRLB for SNRs from −5 dB to 5 dB. Hence, remaining
to the stochastic Cramér-Rao lower bound (CRLB) [27], model errors are hidden by the noise in that region, such
indicating the lowest achievable RMSE for an unbiased that model errors are only severe for high SNRs. Overall,
estimator. the ESPRIT estimator shows the best performance. Hence,
the errors due to beam-space transformation are not so
6.1. Comparison of Calibration Matrix Estimators. The esti- severe as the remaining errors from the global calibration.
mated global calibration matrices are applied to the gen- The proposed root-Capon and Capon-2 and the literature
erated array outputs and DF is conducted afterwards. The based Capon-1 estimator perform comparably worse. Since
ESPRIT estimator and the MUSIC-1 estimator are exemplary all Capon methods employ the inverse of the covariance of
International Journal of Antennas and Propagation 7

2
100
0
Magnitude [dB]

Phase [degree]
−2 0

−4
−100
−6

−180 −120 −60 0 60 120 180 −180 −120 −60 0 60 120 180
Azimuth [degree] Azimuth [degree]

Model Model
Uncalibrated Uncalibrated
Calibrated Calibrated
(a) (b)

Figure 4: Measured response of the first sensor of the Poynting DF-A0046 UCA, the corresponding response of the element-space model
and the sensor response after calibration with the calibration matrix estimated by the method of Wax. The plots show the (a) magnitude and
(b) phase of the response.

102 102

101 101
RMSE [degree]

RMSE [degree]

100 100

10−1 10−1

10−2 10−2
−10 0 10 20 30 40 −10 0 10 20 30 40
SNR [dB] SNR [dB]

uncalibrated Sommerkorn uncalibrated Haefner

Haefner See Kortke Sommerkorn
Wax Kortke Wax See
Ng Friedlander
(a) (b)

Figure 5: RMSE of AoA estimates with the ESPRIT algorithm and calibration w.r.t. (a) the element-space model and subsequent beam-space
transformation and (b) the beam-space model. The CRLB is plotted as red dashed line.

the measurements and the calibration matrix influences this attained by all DF methods for SNRs above 0 dB. Hence,
covariance, a deteriorated estimation performance seems to the DF methods are considerable as statistically efficient [15].
be caused by calibration. Comparing Figures 7(b) and 7(a) it can be concluded that a
In order to verify the explanation of biased estimates SNR above 10 dB is sufficient to outperform global calibration
due to global calibration, the RMSE for local calibration by local calibration.
will be investigated. Note that the beam-space estimators are
excluded, because they cannot be applied under local cali- 7. Measurement Based Studies
bration. The calculated RMSEs are depicted in Figure 7(b).
The estimators do not saturate to a certain RMSE and hence Calibration and experimental measurements were performed
are considerable as unbiased. Furthermore, the CRLB is on a test range in Paardefontein, South Africa, using the
8 International Journal of Antennas and Propagation

Table 4: Setup for the test measurements.

Parameter Value
Transmit signal multi–sine
Centre frequency 305 MHz
Bandwidth 10 MHz
Transmit antennas logarithmic periodic dipole antenna
Transmit polarisation vertical
SNR approx. 25 dB
Receive array uniform circular array with 5 dipols
Receive polarisation vertical
AoAs 25∘ ; 25∘ & −30∘ ; 8∘ & −10∘

102 To show the difference between global and local calibra-

tion, the spectrum of the Bartlett, Capon-1, and MUSIC-
1 method is calculated for both calibration schemes. The
101 respective spectra for the measurement with the single source
RMSE [degree]

are shown in Figure 8. Obviously, local calibration results in

more sharpened peaks in the Capon spectrum and reduced
100 side lobes in the Bartlett spectrum.
First, the estimators are applied to the measurements
10−1 without previous calibration. The estimation results for
the single and dual source measurement are shown in
Table 5. The maximum-likelihood estimators (CML, UML,
10−2 and MODE), the MUSIC type estimators (MUSIC-1; MUSIC-
−10 0 10 20 30 40 2), and the Bartlett beamformer are able to resolve the single
SNR [dB] source, whereas the Capon beamformers (Capon-1; Capon-
uncalibrated Sommerkorn 2) and the beam-space estimators (ESPRIT, IQML, root-
Haefner See MUSIC, and root-Capon) show poor results. However, all
Wax Kortke DF methods fail to properly estimate the AoAs in case of
Ng Friedlander two coherent sources. Hence, estimating the directions of two
coherent sources requires array calibration.
Figure 6: RMSE of AoA estimates with the MUSIC-1 algorithm and
calibration w.r.t. the element-space model. The CRLB is plotted as The method by Sommerkorn is utilised to estimate the
red dashed line. global calibration matrix in element-space. Estimated AoAs
for the single and dual source case are shown in Table 6. In
the single source case, all estimators show good estimation
5 vertically polarised dipoles of the Poynting DF-A0046 results for global as well as local calibration. In the dual
UCA as receive array. First, a single source has been placed source case, the maximum-likelihood and MUSIC estimators
at 25∘ azimuth angle and approx. 102 m apart from the show the best accuracy for global calibration. The Capon and
receiving array. Afterwards, a second source has been placed Bartlett methods fail to resolve the sources. Facing the beam-
at −30∘ azimuth angle and approx. 72 m apart from the space estimators, only the ESPRIT shows considerable good
receiving array. The second source is driven by the same results. An explanation is the beam-space transformation and
signal as the first one, such that the phase difference between the few number of sensors, which causes biased estimates
both sources remains fixed (coherent source case). To cope [23]. In case of local calibration, all applied estimators show
with the presence of coherent sources, spatial smoothing in good estimation results for the dual source case. Note that
conjunction with forward-backward averaging is applied as the Capon beamformers properly resolve both sources under
a preprocessing step [28] for the ESPRIT and root-MUSIC local calibration, whereas the Bartlett beamformer fails. An
estimator. Last, two coherent sources were placed at −10∘ and explanation is the reduced leakage of the Capon beamformer
8∘ azimuth angle and 91 m and 61 m apart from the receiver, compared to the Bartlett [15].
respectively. The elevation angles in all measurements were Last, the scenario with the closer located sources is con-
approx. 0∘ . Some details of the measurement setup can be sidered. The estimation results are shown in Table 7. Obvi-
found in [24]. Source antennas were vertically polarised ously, the beam-space estimators fail. Comparing the esti-
logarithmic periodic dipole antenna (LPDA). Transmit sig- mates for global and local calibration indicates an improved
nal was a multi-sine signal with 10 MHz bandwidth at a accuracy by local calibration. Furthermore, the proposed
carrier frequency of 305 MHz. The measurement setup is Capon-2 method is able to slightly resolve both sources,
summarised in Table 4. whereas the Bartlett and Capon-1 method fails.
International Journal of Antennas and Propagation 9

102 102

101 101
RMSE [degree]

RMSE [degree]
100 100

10−1 10−1

10−2 10−2
−10 0 10 20 30 40 −10 0 10 20 30 40
SNR [dB] SNR [dB]
root-Capon ESPRIT CML MUSIC-1
CML MUSIC-1 Capon-1 MUSIC-2
Capon-1 MUSIC-2 MODE
UML
IQML root-MUSIC Bartlett Capon-2
UML MODE
Bartlett Capon-2
(a) (b)

Figure 7: RMSE of AoA estimation with several DF techniques for (a) global calibration using the method of Sommerkorn to estimate the
calibration matrix and (b) local calibration using the EADF as array model. The CRLB is plotted as red dashed line.

0 two coherent sources. Furthermore, beam-space methods are

quite sensitive due to the beam-space transformation and the
few number of sensors, such that these methods are not able
to resolve two closely spaced sources. Overall, DF methods in
−10
Normalised magnitude [dB]

conjunction with local calibration show the best estimation

accuracy.

−20 8. Conclusion
Calibration of and direction finding with uniform circular
arrays has been investigates in this paper. Several assump-
−30
tions have been drawn for the conducted investigations.
First, source signals are assumed to imping in the azimuth
plane. Second, cross-polar sensor characteristics have been
−40 neglected, because the sources are assumed to be copolar.
−150 −100 −50 0 50 100 150 Global and local calibration of the UCA have been
AoA [degree] investigated. As shown by simulations, DF in conjunction
Capon-1 (local)
with global calibration results in biased estimates due to
Bartlett (global)
Bartlett (local) MUSIC-1 (global) remaining model errors, which are especially severe for
Capon-1 (global) MUSIC-1 (local) high SNRs. On the contrary, DF in conjunction with local
calibration results in unbiased estimates. Also, it was shown
Figure 8: Spectra of Bartlett beamformer, Capon-1 beamformer, that local calibration is superior to global calibration in terms
and MUSIC-1 for global and local calibration. The global calibration of the achievable root mean-square error for SNR above
matrix has been calculated by the method of Sommerkorn. Mea-
10 dB. Furthermore, test measurements with a single and
surement data with a single source at 25∘ are used to calculate the
spectra.
two coherent sources have been considered. In case of a
single source, array calibration is not necessary for several
estimation methods. However, calibration was found to be
required in order to resolve coherent sources. Also, local
In summary, calibration is not necessary in the single calibration was found to outperform global calibration in
source case, because some estimators can tackle the array case of closely spaced coherent sources. Comparison of the
disturbances. However, calibration is required in case of DF techniques based on simulations and test measurements
10 International Journal of Antennas and Propagation

Table 5: Estimated AoAs without array calibration. The setup is a single source (25∘ ) and a coherent dual source (−30∘ , 25∘ ) scenario.

Single Source Dual Source

Estimator
𝐴𝑜𝐴 [∘ ] 𝐴𝑜𝐴 [∘ ] 𝐴𝑜𝐴 [∘ ]
ESPRIT 29.82 −61.17 52.91
root-MUSIC 22.22 −50.01 44.05
IQML 26.38 −24.03 79.29
root-Capon 29.41 −39.36 38.82
Capon-1 28.42 −39.99 35.64
Bartlett 25.04 −135.28 16.06
MUSIC-1 25.04 −33.98 28.82
CML 25.04 −33.25 22.4
UML 25.04 −178.18 102.63
MODE 25.04 −34.24 22.08
Capon-2 28.42 −40.69 35.82
MUSIC-2 25 −35.12 29.8

Table 6: Estimated AoAs under global and local array calibration. The setup is a single source (25∘ ) and a coherent dual source (−30∘ , 25∘ )
scenario.
Single Source Dual Source
Estimator Calibration
𝐴𝑜𝐴 [∘ ] ∘
𝐴𝑜𝐴 [ ] 𝐴𝑜𝐴 [∘ ]
ESPRIT global 25.17 −31.18 25.39
root-MUSIC global 25.66 −27.5 23.26
IQML global 25.19 −27.7 26.04
root-Capon global 25.82 −26.3 24.42
Capon-1 global 24.4 −23.81 20.17
Bartlett global 25.15 −138.18 15.93
MUSIC-1 global 25.15 −30.42 25.46
CML global 25.15 −30.2 25.8
UML global 25.15 −30.2 25.8
MODE global 25.15 −30.2 25.8
Capon-2 global 24.39 −23.55 20.09
MUSIC-2 global 25.14 −30.45 25.48
Capon-1 local 25.31 −30.1 25.13
Bartlett local 25.37 −134.48 15.4
MUSIC-1 local 25.37 −30.09 25.56
CML local 25.37 −30.17 25.89
UML local 25.37 −30.17 25.89
MODE local 25.37 −30.17 25.89
Capon-2 local 25.31 −30.12 25.18
MUSIC-2 local 25.37 −30.11 25.59

indicates that maximum-likelihood estimators (CML, UML, Data Availability

and MODE) and MUSIC type estimators (MUSIC-1 and
MUSIC-2) show better estimation accuracies than beam- The used data have not been made available due to confiden-
space estimators (ESPRIT, IQML, and root-MUSIC). tiality agreements with research collaborators.
In summary, choosing the appropriate DF technique
and calibration method is an application specific trade- Conflicts of Interest
off between required estimation accuracy, computational
complexity, and also calibration measurement effort. The authors declare that they have no conflicts of interest.
International Journal of Antennas and Propagation 11

Table 7: Estimated AoAs under global and local array calibration. [5] J. da Costa, A. Thakre, F. Roemer, and M. Haardt, “Compar-
The setup is a coherent dual source (−10∘ , 8∘ ) scenario. ison of model order selection techniques for high-resolution
parameter estimation algorithms,” in Proceedings of the 54th
Dual Source Internationales Wissenschaftliches Kolloquium, September 2009.
Estimator Calibration
𝐴𝑜𝐴 [∘ ] 𝐴𝑜𝐴 [∘ ] [6] C. A. Balanis, Antenna Theory: Analysis and Design, John Wiley
ESPRIT global −22.27 19.86 Sons, 2nd edition, 1997.
root-MUSIC global −17.78 13.62 [7] P. Stoica and A. Nehorai, “Performance study of conditional and
IQML global −22.73 19.72 unconditional direction-of-arrival estimation,” IEEE Transac-
root-Capon global −161.45 −7.61 tions on Acoustics, Speech, and Signal Processing, vol. 38, no. 10,
pp. 1783–1795, 1990.
Capon-1 global −16.88 15.62
[8] P. Stoica and K. C. Sharman, “Maximum likelihood methods
Bartlett global 2.5 177.67 for direction-of-arrival estimation,” IEEE Transactions on Signal
MUSIC-1 global −10.69 9.78 Processing, vol. 38, no. 7, pp. 1132–1143, 1990.
CML global −10.69 9.72 [9] R. Schmidt, “Multiple emitter location and signal parameter
UML global −10.69 9.72 estimation,” IEEE Transactions on Antennas and Propagation,
MODE global −10.69 9.72 vol. 34, no. 3, pp. 276–280, 1986.
Capon-2 global −16.91 15.66 [10] B. D. Rao and K. V. S. Hari, “Performance analysis of root-
MUSIC-2 global −11.23 10.18 music,” IEEE Transactions on Signal Processing, vol. 37, no. 12,
pp. 1939–1949, 1989.
Capon-1 local −10.03 6.87
[11] F. Belloni, A. Richter, and V. Koivunen, “DoA estimation
Bartlett local 2.61 177.86
via manifold separation for arbitrary array structures,” IEEE
MUSIC-1 local −9.79 8.44 Transactions on Signal Processing, vol. 55, no. 10, pp. 4800–4810,
CML local −9.81 8.83 2007.
UML local −9.81 8.83 [12] R. Roy and T. Kailath, “ESPRIT-estimation of signal parameters
MODE local −9.81 8.83 via rotational invariance techniques,” IEEE Transactions on
Capon-2 local −10.61 9.7 Signal Processing, vol. 37, no. 7, pp. 984–995, 1989.
MUSIC-2 local −10.34 9.46 [13] P. Stoica and K. Sharman, “Novel eigenanalysis method for
direction estimation,” IEEE Proceedings - Radar and Signal
Processing, vol. 137, no. 1, pp. 19–26, 1990.
[14] Y. Bresler and A. Macovski, “Exact maximum likelihood param-
Acknowledgments eter estimation of superimposed exponential signals in noise,”
This work was partially funded by the Bundesministerium für IEEE Transactions on Signal Processing, vol. 34, no. 5, pp. 1081–
1089, 1986.
Bildung und Forschung [grant number 16BN1203]. Stephan
Häfner is funded by the Deutsche Forschungsgemeinschaft [15] H. Krim and M. Viberg, “Two decades of array signal processing
research: the parametric approach,” IEEE Signal Processing
[grant number 317632307]. The authors acknowledge support
Magazine, vol. 13, no. 4, pp. 67–94, 1996.
for the Article Processing Charge by the German Research
[16] G. Sommerkorn, D. Hampicke, R. Klukas et al., “Uniform
Foundation (DFG) and the Open Access Publication Fund of
rectangular antenna array design and calibration issues for 2-
the Technische Universität Ilmenau. The authors like to thank D ESPRIT application,” in Proceedings of the in 4th European
Dipl.-Ing Uwe Trautwein for supporting the calibration and Personal and Mobile Communications Conference, EPMCC, vol.
test measurements. 1, pp. 169–173, February 2001.
[17] A. Kortke, “A new calibration algorithm for smart antenna
References arrays,” in Proceedings of the 57th IEEE Semiannual Vehicular
Technology Conference. VTC 2003, vol. 2, pp. 1030–1034, Jeju,
[1] T. E. Tuncer and B. Friedlander, Classical and Modern Direction- South Korea, 2003.
of-Arrival Estimation, Academic Press, Amsterdam, Nether- [18] B. Friedlander and A. Weiss, “Direction finding in the presence
lands, 2009. of mutual coupling,” IEEE Transactions on Antennas and Prop-
[2] M. Landmann, M. Kaske, and R. S. Thoma, “Impact of agation, vol. 39, no. 3, pp. 273–284.
incomplete and inaccurate data models on high resolution [19] B. C. Ng and C. M. Samson, “Sensor-array calibration using a
parameter estimation in multidimensional channel sounding,” maximum-likelihood approach,” IEEE Transactions on Anten-
IEEE Transactions on Antennas and Propagation, vol. 60, no. 2, nas and Propagation, vol. 44, no. 6, pp. 827–835, 1996.
pp. 557–573, 2012. [20] C. M. S. See, “Sensor array calibration in the presence of
[3] M. Wax and J. Sheinvald, “Direction finding of coherent mutual coupling and unknown sensor gains and phases,” IEEE
signals via spatial smoothing for uniform circular arrays,” IEEE Electronics Letters, vol. 30, no. 5, pp. 373-374, 1994.
Transactions on Antennas and Propagation, vol. 42, no. 5, pp. [21] A. Tewfik and W. Hong, “On the application of uniform linear
613–620. array bearing estimation techniques to uniform circular arrays,”
[4] S. Häfner and R. Thomä, “Application of tensor decomposition IEEE Transactions on Signal Processing, vol. 40, no. 4, pp. 1008–
methods to antenna array calibration measurements for de- 1011, 1992.
noising and narrowband modelling,” in Proceedings of the 13th [22] C. Mathews and M. Zoltowski, “Direction finding with circular
European Conference on Antennas and Propagation (EuCAP), arrays via phase mode excitation and Root-MUSIC,” in Proceed-
pp. 1–5, 2019. ings of the IEEE Antennas and Propagation Society International
12 International Journal of Antennas and Propagation

Symposium 1992 Digest, vol. 2, pp. 1019–1022, Chicago, IL, USA,

June 1992.
[23] F. Belloni and V. Koivunen, “Beamspace transform for UCA:
error analysis and bias reduction,” IEEE Transactions on Signal
Processing, vol. 54, no. 8, pp. 3078–3089, 2006.
[24] S. Häfner, M. Käske, R. S. Thomä et al., “Selection of antenna
array configuration for polarimetric direction finding in corre-
lated signal environments,” in Proceedings of the 19th Interna-
tional ITG Workshop on Smart Antennas, pp. 1–8, March 2015.
[25] M. Landmann and G. Del Galdo, “Efficient antenna description
for MIMO channel modelling and estimation,” in Proceed-
ings of the 7th European Conference on Wireless Technology,
ECWT2004, pp. 217–220, October 2004.
[26] T. Abrudan, J. Eriksson, and V. Koivunen, “Conjugate gradient
algorithm for optimization under unitary matrix constraint,”
Signal Processing, vol. 89, no. 9, pp. 1704–1714, 2009.
[27] P. Stoica, E. G. Larsson, and A. B. Gershman, “The stochastic
CRB for array processing: a textbook derivation,” IEEE Signal
Processing Letters, vol. 8, no. 5, pp. 148–150, 2001.
[28] R. T. Williams, S. Prasad, A. K. Mahalanabis, and L. H. Sibul, “An
improved spatial smoothing technique for bearing estimation
in a multipath environment,” IEEE Transactions on Signal
Processing, vol. 36, no. 4, pp. 425–432, 1988.
 International Journal of

Rotating Advances in
Machinery Multimedia

The Scientific
Engineering
Journal of
Journal of

Hindawi
World Journal
Hindawi Publishing Corporation Hindawi
Sensors
Hindawi Hindawi
www.hindawi.com Volume 2018 http://www.hindawi.com
www.hindawi.com Volume 2018
2013 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018

Journal of

Control Science
and Engineering

Advances in
Civil Engineering
Hindawi Hindawi
www.hindawi.com Volume 2018 www.hindawi.com Volume 2018

Submit your manuscripts at

www.hindawi.com

Journal of
Journal of Electrical and Computer
Robotics
Hindawi
Engineering
Hindawi
www.hindawi.com Volume 2018 www.hindawi.com Volume 2018

VLSI Design
Advances in
OptoElectronics
International Journal of

International Journal of
Modelling &
Simulation
Aerospace
Hindawi Volume 2018
Navigation and
Observation
Hindawi
www.hindawi.com Volume 2018
in Engineering
Hindawi
www.hindawi.com Volume 2018
Engineering
Hindawi
www.hindawi.com Volume 2018
Hindawi
www.hindawi.com www.hindawi.com Volume 2018

International Journal of
International Journal of Antennas and Active and Passive Advances in
Chemical Engineering Propagation Electronic Components Shock and Vibration Acoustics and Vibration
Hindawi Hindawi Hindawi Hindawi Hindawi
www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018