You are on page 1of 17

- - -

17518792, 0, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/rsn2.12352 by Readcube (Labtiva Inc.), Wiley Online Library on [11/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Received: 5 September 2022 Revised: 10 October 2022 Accepted: 4 November 2022 IET Radar, Sonar & Navigation
DOI: 10.1049/rsn2.12352

ORIGINAL RESEARCH

Phase‐difference measurement‐based angle of arrival estimation


using long‐baseline interferometer

Van‐Sang Doan1 | Thien Huynh‐The2 | Van‐Phuc Hoang3 | Jiri Vesely4

1
Faculty of Communication and Radar, Naval Abstract
Academy, Nha Trang, Vietnam
This article presents a direction‐finding method based on phase difference measurement
2
Faculty of Electrical and Electronics Engineering, for wide‐ranging applications of electronic intelligence. By including three processing
Ho Chi Minh City University of Technology and
Education, Ho Chi Minh City, Vietnam steps of rotating, re‐scaling, and rounding of phase plane, the proposed method is able to
3 obtain the actual phase shift of the longest baseline from the ambiguous phase differ-
Institute of System Integration, Le Quy Don
Technical University, Hanoi, Vietnam ences measured from incoming signals with low computational cost. The method's
4
Faculty of Military Technology, University of
performance is analysed and evaluated in terms of angle of arrival (AOA) estimation
Defence, Brno, Czech Republic accuracy and probability of unambiguity by several interferometers of long baselines
designed by multiple antenna elements. Consequently, an optimised array configuration of
Correspondence four antenna elements is taken into account to obtain high AOA estimation accuracy via
Van‐Sang Doan, Faculty of Communication and several productive simulations and intensively validated with device measurement in the
Radar, Naval Academy, Nha Trang, Vietnam.
Email: doansang.g1@gmail.com
practical environment. Regarding the simulation of angle estimation accuracy, the pro-
posed method remarkably outperforms two other existing methods, including second‐
Funding information order difference array and correlative interferometers. In the experimental measure-
Ministry of Science and Technology, Grant/Award ment, high AOA determination performance is achieved with the transmitter located
Number: NDT/CZ/22/12; National Foundation around 35.5 m far from an antenna array and the small average angle estimation error of
for Science and Technology Development
1.86° at the direction of 20°. For various directions from −60° to 60° with the steps of 5°,
the results satisfy the actual directions of the transmitter with a standard deviation of less
than 0.5°.

KEYWORDS
angle of arrival, antenna array, direction‐finding system, electronic intelligence, interferometer, phase ambiguity

1 | INTRODUCTION minimum, signal‐balance, and mono‐pulse values, however,


their angle accuracy is low [6]. Several TDOA‐based methods,
Electronic Intelligence (ELINT) is an activity using technical comprehensively analysed in refs. [7, 8], were employed to
equipment to collect the useful information of the objects of identify the target location via hyperbolic equations with the
interest [1], such as the presence, the direction, and the position baseline considered over tens of kilometers. Moreover, the
of signal sources as well as the radio frequency spectrum of TDOA measurement can be applied for the AOA estimation
received signals. Angle of arrival (AOA) or direction of arrival with several units to 10‐m baseline [9, 10]. Most of the TDOA‐
(DOA) is the important information to determine the position based AOA estimation methods need only 2 antennas with the
of target [2] which is used for jammer [3] and other weapon baseline distance much greater than the half‐wavelength to
systems [4]. In a general ELINT system, traditional methods determine an unambiguous angle in the 180° field of view
use the amplitude, time‐difference‐of‐arrival (TDOA), and (FOV). However, radio measurement systems are limited by
phase difference of received signals to compute the arrival high error of the tiny TDOA measurement associated with
angle [5]. Some amplitude‐based methods can determine the small distance difference. Additionally, despite capably working
AOA value with a high stability by comparing maximum, with pulse signals, direction finding systems cannot accurately

-
This is an open access article under the terms of the Creative Commons Attribution‐NonCommercial‐NoDerivs License, which permits use and distribution in any medium, provided the
original work is properly cited, the use is non‐commercial and no modifications or adaptations are made.
© 2022 The Authors. IET Radar, Sonar & Navigation published by John Wiley & Sons Ltd on behalf of The Institution of Engineering and Technology.

IET Radar Sonar Navig. 2022;1–17. wileyonlinelibrary.com/journal/rsn2 1


17518792, 0, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/rsn2.12352 by Readcube (Labtiva Inc.), Wiley Online Library on [11/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2
- DOAN ET AL.

measure the TDOA value of continuous wave (CW) signals. On deteriorated along with the worse angular resolution.
the contrary, some systems, which calculate AOA based on the Furthermore, the system will require a higher memory uti-
measurement of phase difference, can achieve high accuracy lisation and a more expensive computational cost to find the
for the pulse and CW signals. For example, a simple phase most appropriate AOA from the higher‐resolution table. In
interferometer system configured with a short baseline d ≤ λ/2 order to accelerate the processing speed, Cui et al. [22] pro-
(where d is the geometric distance between 2 antennas in an posed a two‐dimensional linear interpolation (1DSLUT‐2DLI)
array and λ is the signal wavelength) can effortlessly measure an algorithm with the one‐dimensional sorting mechanism which
unambiguous phase difference and consequently determine the can associate with UCA models. Although this approach can
AOA of a signal source in the 180° FOV. However, besides the yield high AOA estimation accuracy, it performs more slowly
challenges of manufacturing microwave antenna arrays with than the second‐order difference array (SODA) method [23].
d ≤ λ/2, this short‐baseline interferometer suffers from low However, the SODA method provides a low accuracy due to
AOA measurement accuracy caused by small aperture [11]. short second‐order baseline. The phase ambiguity problem is
Because the AOA accuracy is mostly proportional to the also resolved by the Chinese remainder theorem, which is
baseline length, hence a long‐baseline interferometer should be proposed in refs. [24, 25].
studied to improve the accuracy of angle estimation with a In the last decades, the graph (so‐called phase space),
smaller error. Despite the fact that increasing the baseline which is able to competently explain the relationship of phase
length can lead to a phase difference ambiguity in AOA esti- differences, is attracting more attention to solve the phase
mation, this problem was addressed by TDOA based [12] and ambiguity problem. Some useful parameters can be extracted
multi‐long‐baseline interferometer (MLBI) based approaches from the graph, such as baseline ratio and allowable maximum
[13, 14]. It is worth noting that the MLBI is defined as an phase difference error (AMPDE), to optimise the array
antenna array with N elements, where (N − 1)λ/2 ≪ L, where configuration and consequently improve the accuracy of AOA
L is the array aperture. In the literature, MLBI is so‐called by estimation. The meaningful information of AMPDE is useful
sparse interferometer and sparse array [15, 16]. to calculate the maximal detection range of direction finding
Currently, subspace‐based methods, such as multiple signal systems by a one‐way passive radar equation [26]. Meanwhile,
classification and estimation of signal parameters via rotation the baseline ratio provides the most appropriate configuration
invariance (ESPRIT), are other class approaches, which exploit of an antenna array. In ref. [27], Lee et al. developed a phase
the Eigen‐decomposition of the covariance matrix of received space rotation scheme to acquire the information of maximum
signals to determine the arrival angles [17, 18]. These methods phase error in MLBI systems having three and four antenna
have several advantages, including the array gain improvement, elements. However, the rotation method cannot be applied for
multiple AOA estimation, and the ability to reject interference. more than four element arrays, and the lack of an optimal al-
However, these methods also have the disadvantage of gorithm of finding a real phase shift value maintains a chal-
computational cost. Furthermore, high side‐lobes in the array lenging issue. To this end, Younger and Reid [28] studied a
beam pattern for very sparse array configuration may result in novel LUT technique to optimally obtain an unambiguous
incorrect AOA estimation. AOA. However, most of the LUT‐based approaches typically
In the past, Kendall [19] proposed an effective measure- suffer from coarse angular resolution and require an expensive
ment, in which a novel baseline with an additional phase dif- memory for processing.
ference ambiguity is analysed to improve AOA accuracy. Based To handle the aforementioned problems (including angular
on the combination of two ambiguous phase differences, an accuracy, memory and time consumption, and ambiguity), a
unambiguous AOA could be determined more precisely. fast and robust AOA estimation method, which has three
Additionally, the most appropriate antenna configuration is principal steps of rotating, re‐scaling, and rounding the
specified with the spacing ratio M/N = K/(2K + 1) in ambiguous phase differences, is proposed in this paper. With
ambiguous measurements, where M; N; K ∈ N. A deep anal- respect to performance evaluation, different antenna array
ysis of ambiguity‐resistant three‐ and four‐element in- configurations are intensively analysed to validate the robust-
terferometers was reported by Goodwin [20], in which the ness and reliability of the proposed method. Subsequently, the
phase planes (i.e. the relationship of two phase differences) are most appropriate one is considered to benchmark the AOA
taken into account to handle the ambiguity problem success- accuracy by simulation and further validated in practical mea-
fully. The arrays of shared edge elements were accordingly surement with devices. Accordingly, the main contribution of
approved for utilisation in ELINT systems thanks to their this work is listed as follows:
lowest ambiguity probability. Despite giving some relatively
comprehensive investigations, the studies in refs. [19, 20] are � We propose an effective phase difference transformation‐
unable to describe the relationship between phase differences based method to overcome the phase difference ambiguity
to estimate AOA in the 360° FOV. By taking advantage of a problem in MLBI systems. The method can determine the
uniform circular array (UCA) to handle this issue, Pan et al. actual phase shift of the longest baseline thanks to three
[21] significantly enhanced the accuracy of AOA estimation steps of rotating, re‐scaling, and rounding the ambiguous
with a Look Up Table (LUT) mechanism. However, the per- phase differences. As a result, the proper AOA of signal
formance of LUT‐based AOA determination approach is source is obtained through the actual phase shift.
17518792, 0, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/rsn2.12352 by Readcube (Labtiva Inc.), Wiley Online Library on [11/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
DOAN ET AL.
- 3

sffi�
ffiffiffiffiffiffiffiffiffi�
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi�
ffiffiffiffiffiffiffiffiffiffi�
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi�
ffiffiffiffiffiffiffiffiffiffi�
ffiffiffiffiffiffiffiffiffi
� The most appropriate array configuration is recommended ∂Φ 2 2 ∂Φ 2 2 ∂Φ 2 2
to improve the angular accuracy and ambiguity resolution. σΦ ¼ σd þ σf þ σθ
∂d ∂f ∂θ
The proposed method is evaluated and compared with the
SODA and correlative approaches in terms of AOA accu- sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi� ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi�
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2π 2 2 d sin θ 2 2
racy to prove its efficiency. ¼ ðsin θÞ σ d þ σ f þ ðd cos θÞ2 σ 2θ
� Some practical signal measurements are carried out to λ f
validate the AOA estimation performance of the proposed ð3Þ
method which can be promisingly applied to direction
finding systems. where σd, σf, and σθ are the errors of antenna positioning,
frequency and AOA, respectively, in which σd is the systematic
error (bias), while σf and σΦ are the random errors.
2 | FORMULATION OF PHASE Regarding the phase difference error depending on the
INTERFEROMETER antenna position bias, we consider the formula as follows:

2πsin θ
2.1 | Phase interferometer with two antennas σ Φ;d ¼ σ d ð4Þ
λ
In general, a classical phase interferometer system measures
where σΦ,d is the phase difference error due to the antenna
the phase shift between two antennas to determine the
position error. Obviously, if θ = 0°, there is no error of the
AOA of a signal source. Assume that a far‐field emitter
phase difference even existing antenna position error. For
transmits a narrowband signal over the air to an antenna
θ = 60°, and f = 1090 MHz, the phase difference error σΦ,
array configured by two elements in direction θ as shown in
d will be smaller than 5° if the antenna bias is less than 4.4 mm.
Figure 1.
If we consider d and λ being as constant, then the error of
It is observed that the electromagnetic ray arrives to the
the angle determination process depends on the antenna
antenna A1 earlier than to the antenna A2, which poses the
spacing, phase error, signal wavelength, and arrival angle of the
phase shift (or time delay) Φ determined as follows:
received signal. This dependence is expressed as follows:
2πdsin θ
Φ¼ : ð1Þ λ
λ σθ ¼ σΦ ; ð5Þ
2πdcos θ
Based on Equation (1), the angle θ is calculated as follows: where σΦ is the error of phase shift measurement, which is
� � typically identified via signal‐to‐noise ratio (SNR):
Φλ
θ ¼ arcsin : ð2Þ
2πd pffiffiffiffiffiffiffiffiffiffi
σ Φ ¼ 1= SNR: ð6Þ
Due to the periodical property in Equation (2), the accu-
racy of angle computation reduces significantly if the actual As a result, the error of AOA estimation is accordingly
phase shift is out of the interval ð0; 2πÞ or ð−π; þπÞ for formulated as follows:
d > λ/2. Therefore, the antenna spacing must satisfy the
condition d ≤ λ/2 for an explicit AOA. λ
If the parameters d, f = c/λ and θ in Equation (1) are σ θ ¼ pffiffiffiffiffiffiffiffiffiffi : ð7Þ
SNR 2πd cos θ
considered independent variables, then the error of phase
difference can be defined by using the law of propagation of
uncertainties [29] as follows: Explicitly, widening the baseline d can significantly improve
AOA accuracy via the reduction of error σθ. According to (7),
if d = N � λ/2 with N ∈ N, the accuracy of AOA determi-
nation is N times better than that of d = λ/2; therefore, the
longer the baseline is specified, the higher the AOA estimation
accuracy is. However, a long baseline (d ≫ λ/2) risks an am-
biguity problem of the phase difference measurement because
the measurable values of phase difference are limited in the
range of [−π, +π]. On the contrary, the actual phase shift
might be out of the range of [−π, +π] according to the
following equation:
F I G U R E 1 A direction‐finding system using a classical phase
interferometer with two antennas of an array Φ ¼ ϕ þ 2πn; ð8Þ
17518792, 0, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/rsn2.12352 by Readcube (Labtiva Inc.), Wiley Online Library on [11/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4
- DOAN ET AL.

where ϕ is the measured phase difference and n ∈ N is where n1 ; n2 ∈ N are two coefficients. Notably, the last formula
referred to as a coefficient. To solve the ambiguity problem, it in (10) is a linear function ϕ2 of variable ϕ1 with the slope
is important to identify the most appropriate coefficient n to k = N/M. Considering an angle θ limited in the range of (−90°,
obtain the actual phase shift thanks to additional antenna +90°), n1 and n2 vary in the ranges of ½⌈ − ðM þ 1Þ=2⌉;
elements. ⌊ðM þ 1Þ=2⌋� and ½⌈ − ðN þ 1Þ=2⌉; ⌊ðN þ 1Þ=2⌋� respec-
tively. Graphically, the relationship of phase differences ϕ1 and
ϕ2 is described via a phase plane shown in Figure 3. The phase
2.2 | Multi‐long‐baseline phase lines (in blue) present the ideal phase difference points (i.e.
interferometer without noise). It is noted that the line numbers in the phase
plane correspond to the coefficient n2. The uniform distances
Several MLBI systems have been developed with more than between neighbouring phase lines are calculated as follows:
two antenna elements, where d > λ/2, in an array. For example,
some arrays with the configurations f0; 1; 2; 4gλ=2 and 360°
V ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi: ð11Þ
f0; 3; 13gλ=2 in ref. [30] are introduced to solve the ambiguity M2 þ N2
problems. The first array has the shortest baseline d1 = λ/2
(where d1 is referred to as the distance between the first an- The measured phase difference point (green point) with
tenna and the second one) for preventing AOA ambiguity, random noise might not be positioned in the phase lines,
whereas the second array deploys a phase plane to mitigate the therefore, we need to find the nearest line of this point to
ambiguity problem. However, both the arrays pose several determine n2 (corresponding to the actual phase shift Φ2). For
limitations of low angle accuracy and ambiguity resolution, example, n2 = 1 is determined via the phase plane in Figure 3.
which will be discussed and analysed hereafter. In this study, We propose a novel algorithm for accurately determining n2
we design an array with the configuration f0; M; N gλ=2, where with less computational complexity. In addition, an optimal
M; N ∈ N, M < N, and M, N have the greatest common di- array configuration associated with the algorithm is recom-
visor equal to 1. As illustrated in Figure 2, the array has three mended to obtain a good performance.
antenna elements designated by two baselines d1 = M � λ/2
and d2 = N � λ/2.
The phase differences ϕ1 and ϕ2 of the respective baselines 3 | PHASE DIFFERENCE
d1 and d2 wrapped in [−π, +π] are accordingly determined as TRANSFORMATION‐BASED METHOD
follows: FOR OBTAINING UNAMBIGUOUS PHASE
SHIFT
� �
2πd1 sin θ
ϕ1 ¼ mod − π; 2π þ π; In this section, we propose a novel method for obtaining an
λ
� � ð9Þ unambiguous phase shift, and consequently, a respective ac-
2πd2 sin θ curate AOA is determined. Accordingly, the proposed method
ϕ2 ¼ mod − π; 2π þ π;
λ will be clarified through antenna arrays with three, four and
more than four elements.
where mod(.) denotes the modulo operation. From Equations
(1) and (8), the relationship between Φ1 and Φ2 is presented as
follows:

Φ2 d 2 ϕ2 þ n2 2π N
¼ ⇔ ¼
Φ1 d 1 ϕ1 þ n1 2π M
� � ð10Þ
N Nn1 − Mn2
⇔ ϕ2 ¼ ϕ1 þ 2π ;
M M

F I G U R E 3 Relationship of phase differences corresponding with


FIGURE 2 Phase interferometer of 3‐element array d1 = 5λ/2 and d2 = 8λ/2
17518792, 0, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/rsn2.12352 by Readcube (Labtiva Inc.), Wiley Online Library on [11/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
DOAN ET AL.
- 5

3.1 | Three‐element antenna array α ¼ arctanðM=NÞ: ð13Þ

The distance between neighbouring phase lines is calcu-


With the configuration of recommended antenna array, the
lated by Equation (11).
actual phase shift of the largest baseline d2 is potentially
Rotation step: As shown in Figure 4b, the phase plane and
determined using the phase plane transformation with rota-
measured point are rotated about the angle α by follows:
tion, re‐scaling, and rounding. In particular, the algorithm for
obtaining unambiguous phase shift is expressed by the � � � � � �
following steps: ϕ1n cos α −sin α ϕ1
¼ � : ð14Þ
Initialisation step: According to (10), the phase plane is ϕ2n sin α cos α ϕ2
produced by array configuration parameters, for example,
d1 = 5λ/2 and d2 = 8λ/2. The lines of phase plane are char-
acterised by the coefficient n2. Regarding an �illustration Re‐scaling step: After rotation, the phase lines, which are
� shown
in Figure 4a with the measured point P ϕ1 ; ϕ2 , a one‐
^ ^ being perpendicular to the horizontal axis, are spaced evenly by
distance V. Therefore, the horizontal axis is re‐scaled by
dimensional matrix C containing all labelled numbers in a 2D dividing the phase distance V to the phase lines horizontally
phase plane is given as follows: distributed in the integer values.
2 3 2 3
C ¼ ½2; −1; −4; 1; −2; 3; 0; −3; 2; −1; 4; 1; −2�: ð12Þ ^ 1n
ϕ ϕ1n
6 7 6 V 7
4 5¼4 5: ð15Þ
The angle α, which is defined by phase lines and horizontal ^ 2n
ϕ ϕ2n
axis is defined as follows:

F I G U R E 4 Principle for obtaining unambiguous phase shift using phase difference transformation in 2D phase plane: (a) original phase plane, (b) rotated
phase plane, (c) re‐scaled phase plane, and (d) determined nearest phase line
17518792, 0, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/rsn2.12352 by Readcube (Labtiva Inc.), Wiley Online Library on [11/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6
- DOAN ET AL.

After this step, the phase lines are scattered in a trans- optimal choice of M and N should be N = M + 1 to gain α
formed phase plane for better visualisation and analysis as approaching to 45°, where the distance V will be largest.
illustrated in Figure 4c. Accordingly, several configurations of antenna array derived from
Rounding step: Because of noise, the measured point is not Equation (21) are summarised in Table 1. Despite presenting the
located in any transformed lines. With the last processing step highest value of V, the array f0; 1; 2; 4gλ=2 has a shortest aperture
of rounding the value ϕ ^ 1n , the measured point approaches to (4λ/2), that induces the reduction of AOA accuracy in Equation
the nearest neighbouring phase line (as illustrated in Figure 4d). (7). On the contrary, the array f0; 25; 30; 36gλ=2 provides the
� � largest aperture, which provides a higher AOA accuracy. However,
^ 1n ;
ϕ1nr ¼ nint ϕ ð16Þ a small AMPDE can reduce the detectable range of system.
Therefore, besides array dimension, the phase measurement ac-
curacy of devices in practice should be taken into account when
where nint(.) denotes the nearest integer function (i.e. rounding selecting a most favourable configuration of antenna array [31, 32].
operation). Consequently, d ¼ f0; 9; 12; 16gλ=2 is considered in the MLBI
The value ϕ1nr now becomes an address index of the system to achieve the good trade‐off between AOA accuracy and
corresponding element in the matrix C. Therefore, the nearest workable range.
phase line of measured point is identified as follows: Instead of using phase plane, the algorithm can acquire
! unambiguous phase shift with the four‐element array via three‐
M þN −1
n2;line ¼ C ϕ1nr þ ⌊2 þ1 : ⌋ ð17Þ dimensional (3D) phase space, which effectively presents the
relationship of phase differences ϕ1, ϕ2, and ϕ3 of the base-
lines d1, d2, and d3 respectively. The procedure of 3D phase
Accordingly, an unambiguous phase shift of baseline d2 is space for accurate AOA estimation is described as follows.
written by: Initialisation step: A phase space is established based on
configuration parameters of the interferometer array. Accord-
Φ ^ 2 þ 2πn2;line :
^2 ¼ ϕ ð18Þ ingly, the relationship between actual phase shift and ambiguous
phase difference of the baseline d3 is expressed as follows:
Finally, the correct AOA is estimated as follows:
Φ3 ¼ ϕ3 þ 2πn3 ; ð22Þ
� �
^ where n3 is calculated by the following equation:
^θ ¼ arcsin Φ2 λ : ð19Þ
2πd2 � �
d 3 sin θ
n3 ¼ nint for θ ∈ ½−90o ; 90o �: ð23Þ
3.2 | Four‐element antenna array λ

With regard to a four‐element array, we consider three baselines The values of n3 are labelled on the lines of phase space, as
with a common element as a phase reference, which should be shown in Figure 5a. A two‐dimensional matrix C specialised by
placed at the array edge for leveraging the largest baseline. the number of all annotated phase lines in 3D phase space is
Supposing a four‐element array d ¼ f0; d 1 ; d 2 ; d 3 g can be given as follows:
divided into two three‐element arrays, including d1 ¼ f0; d 1 ; d 2 g
2 3
and d2 ¼ f0; d 2 ; d 3 g, where λ/2 ≪ d1 < d2 < d3. The baseline 0 0 0 −1 3 7 −5
ratios in two arrays d1 and d2 are specified by k1 = d2/d1 and 6 0
6 0 6 −6 −2 2 6 7
7
k2 = d3/d2, by which the phase planes are characterised by the 6 0
6 −3 1 5 −7 −3 1 7
7
line distances V1 and V2 respectively. In order to provide a C¼6
6 4 −8 −4 0 4 8 −4 7
7: ð24Þ
uniform AMPDE, the phase distances V1 and V2 should be 6 −1
6 3 7 −5 −1 3 0 7
7
equal to each other, that means, k1 is equal to k2. By substitution 4 −6 −2 2 6 −6 0 0 5
of k1 = k2 = N/M, where M; N ∈ N, we obtain the baseline ratio 5 −7 −3 1 0 0 0
between two sub‐arrays as follows:

d 3 d 2 N d 3 N 2 N 2 λ=2 TABLE 1 Array configurations with four elements


¼ ¼ ⇔ ¼ ¼ : ð20Þ
d 2 d 1 M d 1 M 2 M 2 λ=2 M, N d = [0, M2, MN, N2]λ/2 V ¼ pffi360°
ffiffiffiffiffiffiffiffiffiffiffi
2 2
M þN
1, 2 f0; 1; 2; 4gλ=2 161°
From Equation (20), the array d is configured as follows:
2, 3 f0; 4; 6; 9gλ=2 100°
� �
d ¼ f0; d1 ; d 2 ; d 3 g ¼ 0; M 2 ; M � N; N 2
λ=2: ð21Þ 3, 4 f0; 9; 12; 16gλ=2 72°

4, 5 f0; 16; 20; 25gλ=2 56°


From Figure 3, it is observed that the smaller M and N must be, 5, 6 f0; 25; 30; 36gλ=2 46°
the larger distance V is obtained. Because M and N are integers, the
17518792, 0, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/rsn2.12352 by Readcube (Labtiva Inc.), Wiley Online Library on [11/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
DOAN ET AL.
- 7

F I G U R E 5 Principle for obtaining unambiguous phase shift using phase difference transformation in 3D phase space: (a) original phase space, (b) rotated
phase space, (c) re‐scaled phase space, and (d) determined nearest phase line

The distance V and angle α are defined� by Equations � (11) and Rounding step: The values of ϕ ^ 1n and ϕ^ 2n might not be in-
(13) respectively. A measured point P ϕ ^1; ϕ
^ 2; ϕ
^ 3 (remarked tegers because of additive noise. Therefore, by rounding values
by the red point in Figure 5a) is also visualised in the 3D phase ^ 1n and ϕ
ϕ ^ 2n , we approximate the measured point to its nearest
space for better illustration. phase line (denoted by green point in Figure 5d), at which the
Rotation step: The phase lines and measured point are annotated number is the need‐to‐find coefficient n3,line for un-
rotated by an angle α as follows: ambiguous phase shift Φ3. The following formula presents the
2 3 2 3 2 3 rounding operation:
ϕ1n cos α −sin α 0 ϕ1 � �
4 ϕ2n 5 ¼ 4 0 cos α −sin α 5 � 4 ϕ2 5: ð25Þ ϕ1nr ¼ nint ϕ^ 1n ;
ϕ3n 0 sin α cos α ϕ3 � � ð27Þ
ϕ2nr ¼ nint ϕ^ 2n :

After rotation, the rotated phase lines are parallel with the
vertical axis, as shown in Figure 4b. The values ϕ1nr and ϕ2nr are now column and row indices
Re‐scaling step: The horizontal axes are re‐scaled by the of the coefficient n3,line in the matrix C. Therefore, to find the
distance V. number that associates to the nearest phase line of measured
point, we perform the following formula:
2 ^ 3 2ϕ 3
ϕ1n 1n 8
MþN −1
6
6

7 6
7 6
^ 2n 7 ¼ 6
6
V 7
ϕ
7
7
7: ð26Þ
>
>
< i ¼ ϕ1nr þ
> ⌊
2 þ 1; ⌋
6
4
7 6
5 4 V 7
2n n3;line ¼ Cði; jÞ; where
5 >
: j ¼ ϕ þ M þ N − 1 þ 1:
⌊ ⌋
>
>
^
ϕ3n ϕ3n 2nr 2
ð28Þ
At this step, all phase lines are arranged for being parallel with
the vertical axis and uniformly located at integer values of two The unambiguous phase shift of baseline d3 is then deter-
horizontal axes, as shown in Figure 5c. mined as follows:
17518792, 0, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/rsn2.12352 by Readcube (Labtiva Inc.), Wiley Online Library on [11/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8
- DOAN ET AL.

Φ ^ 3 þ 2πn3;line :
^3 ¼ ϕ ð29Þ nK−1;line ¼ Cðj1 ; j2 ; …; jK−2 Þ; ð36Þ

Finally, the correct AOA that corresponds to the measured where


point is estimated as follows:


^
� ji ¼ ϕi;nr þ ⌊M þ 2N − 1⌋ þ 1;
^θ ¼ arcsin Φ3 λ : ð30Þ
2πd3 for i = 1, 2, …, K − 2. Here, the elements of matrix C are
calculated as follows:
� �
3.3 | Antenna array with more than four dK−1 sin θ
elements Cðj1 ; j2 ; …; jK−2 Þ ¼ nint ; ð37Þ
λ

In general, we can specify K − 1 baselines with K − 1 where θ ∈ [ − 90°, 90°]. Similar to three‐ or four‐element array,
dimensional phase space (so‐called hyper‐space) from an array the unambiguous phase shift of the baseline dK−1 is deter-
of K antennas. It assumes that K‐element array mined as follows:
dK ¼ f0; d 1 ; d 2 ; …; d K−1 g has baseline ratios, which are
designated by follows: ^ K−1 þ 2πnK−1;line :
^ K−1 ¼ ϕ
Φ ð38Þ

dK−1 d K−2 d 2 N d K−1 N K−2 Subsequently, the accurate AOA of the received signal is
¼ ¼…¼ ¼ ⇔ ¼ K−2 : ð31Þ
dK−2 d K−3 d1 M d1 M estimated:

If d1 = MK−2λ/2, then: � �
^
� ^θ ¼ arcsin ΦK−1 λ : ð39Þ
d 2 ¼ M K−3 � N λ=2; 2πd K−1

d 3 ¼ M K−4 � N 2 λ=2; ð32Þ

dK−1 ¼ N K−2 λ=2: 4 | PERFORMANCE EVALUATION AND
COMPARISON
Based on the array parameters, a phase hyper‐space, which
presents the relationship between phase differences ϕ1, ϕ2, …, In this section, we evaluate the proposed method with
ϕK−1 of the respective baselines d1, d2, …, and dK−1, is different antenna arrays for proving the optimum configura-
accordingly established. The values of coefficient nK−1 that tion. Accordingly, our method competes with other existing
correspond to ϕK−1 are designated on the lines of hyper‐space. approaches in terms of AOA estimation accuracy. It is worth
The unambiguous phase shift of the largest baseline dK−1 is noting that we perform the Monte Carlo simulation with
derived as follows: 10,000 trials for each assessment.
Rotation step:

ϕi;n ¼ ϕi cos α − ϕiþ1 sin α; 4.1 | Optimisation of array configuration


ð33Þ
ϕK−1;n ¼ ϕK−2 sin α þ ϕK−1 cos α;
In general, each DOA estimation method satisfies a particular
for i = 1, 2, …, K − 2. array configuration, therefore choosing a suitable one for the
Re‐scaling step: proposed method is crucial. In ref. [20], four variants of array
baseline were considered for performance analysis, in which
ϕi;n both the three‐element (such as end‐phase left, mid‐phase, and
^ i;n ¼
ϕ ;
V ð34Þ end‐phase right) and four‐element (e.g. independent) arrays
^ K−1;n ¼ ϕK−1;n ; were taken into account. Although they have been compre-
ϕ hensively studied in terms of channel‐pair error parameter in
several previous works, the measured point distribution in
for i = 1, 2, …, K − 2. phase plane and ambiguity probability were evaluated insuffi-
Rounding step: ciently. Regarding the two‐baseline interferometers, three
� � optional configurations are considered: (a) two baselines have a
^ i;n ;
ϕi;nr ¼ nint ϕ ð35Þ shared inner element, (b) two baselines have a shared edge
element, and (c) two baselines have no shared element. As
illustrated in Figure 6, the configurations (b) and (c) reveal the
The values ϕi,nr are now address indices of the coefficient drawback of much‐occupied setup space to obtain the same
nK−1,line in a (K − 2)‐dimensional matrix C. distance d2 as the configurations (a). It should be noticed that
17518792, 0, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/rsn2.12352 by Readcube (Labtiva Inc.), Wiley Online Library on [11/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
DOAN ET AL.
- 9

FIGURE 6 Three options for two‐baseline interferometer configuration: (a) shared edge element; (b) shared inner element; (c) no shared element

FIGURE 7 Distribution of measured phase points for three cases: (a) shared edge element; (b) shared inner element; (c) no shared element

unambiguity probabilities of coefficient nline (the labelled


number of phase line), which are calculated as follows:
nline;correct
P unambiguity ¼ ; ð40Þ
nline;total

where nline, correct is number of correct coefficients that can be


determined, and nline, total is total number of coefficients that
corresponds to number of measured points in phase plane.
From the numerical results as shown in Figure 8, the shared‐
edge‐element array is better than the shared‐inner‐element
and no‐shared‐element arrays in terms of the rate of correct
coefficient determination at all SNRs. Accordingly, this
assessment offers the shared edge element configuration as the
most appropriate selection for the interferometer baseline.
F I G U R E 8 Probability of correct coefficient determination with
different array variants

4.2 | Optimisation of baseline ratio based on


setting space
the configuration (c) needs four elements for two baselines,
consequently, the system complexity and implementation cost Despite the fact that an antenna array with more elements and
will be more expensive if compared with the others. a broader aperture results in a higher accuracy of AOA
The distribution probability of measured points in phase determination, the system requires a more costly hardware and
plane is shown in Figure 7. We observe that three configuration a more powerful processing unit. Indeed, characterising an-
variants have different measured phase difference distributions. tenna elements for an optimal array configuration remains a
Concretely, the array of configuration (a) (Figure 6a) contrib- challenging task, even though it is only favourable to a specific
utes a southwest‐northeast elliptic distribution (Figure 7a) of algorithm. With the constraint of array aperture and element
measured phase differences due to having a phase‐shared‐ number,
� the proposed� configuration of antenna array
edge‐element, the array of configuration (b) (Figure 6b) pro- d ¼ 0; M 2 ; M � N; N 2 ðλL=2N 2 Þ, where M; N ∈ N and
vides a northwest‐southeast elliptic distribution (Figure 7b) due L ≤ N2, is deployed for performance evaluation. Accordingly,
to phase‐shared‐inner‐element and the array of configuration Table 2 shows the superiority of the proposed configuration in
(c) (Figure 6c) prefers a circle distribution (Figure 7c) due to comparison with the integer array in terms of the phase dis-
having no‐shared‐element. These distributions show different tance between two neighbour phase lines. Our array
17518792, 0, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/rsn2.12352 by Readcube (Labtiva Inc.), Wiley Online Library on [11/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
10
- DOAN ET AL.

T A B L E 2 Comparison of our configuration design with other


possible integer ones in terms of phase distance for four‐element array

L Integer array Vmin Our array Vmin


[λ/2] f0; d 1 ; d 2 ; d 3 g½λ=2� [°] M, N f0; d 1 ; d 2 ; d 3 g½λ=2� [°]
3 f0; 1; 2; 3g 100 1, 2 f0; 0:75; 1:25; 3g 161

4 f0; 1; 2; 4g 161 1, 2 f0; 1; 2; 4g 161

5 f0; 1; 2; 5g 67 2, 3 f0; 2:22; 3:33; 5g 100

6 f0; 1; 5; 6g 46 2, 3 f0; 2:67; 4; 6g 100

7 f0; 1; 3; 7g 47 2, 3 f0; 3:11; 4:67; 7g 100

8 f0; 1; 3; 8g 42 2, 3 f0; 3:56; 5:33; 8g 100

9 f0; 1; 4; 9g 37 2, 3 f0; 4; 6; 9g 100

10 f0; 1; 3; 10g 35 3, 4 f0; 5:625; 7:5; 10g 72

11 f0; 1; 2; 11g 27 3, 4 f0; 6:2; 8:25; 11g 72

12 f0; 1; 5; 12g 28 3, 4 f0; 6:75; 9; 12g 72 F I G U R E 9 Probability of correct coefficient determination with
different baseline ratios
13 f0; 1; 2; 13g 27 3, 4 f0; 7:3; 9:75; 13g 72

14 f0; 1; 3; 14g 25 3, 4 f0; 7:875; 10:5; 14g 72 TABLE 3 Array Configurations with baseline ratio 3:4
15 f0; 1; 2; 15g 24 3, 4 f0; 8:44; 11:25; 15g 72 No. elements dk ¼ f0; d 1 ; d 2 ; …; d K −1 g Aperture V
16 f0; 1; 3; 16g 22 3, 4 f0; 9; 12; 16g 72 3 f0; 3; 4gλ=2 4λ/2 72°
Note: The bold values mean the optimised configurations. 4 f0; 9; 12; 16gλ=2 16λ/2 72°

5 f0; 27; 36; 48; 64gλ=2 64λ/2 72°

configuration method achieves higher phase distance values, 6 f0; 81; 108; 144; 192; 256gλ=2 256λ/2 72°
which implies a higher unambiguity probability. Notably, the
arrays denoted in bold are the best configurations regarding the
comparison is a narrow‐band with 1024 snapshots and has
constraint of number‐of‐element and setting‐space.
arrival angle varying randomly in the range of [−60°, 60°].
Figure 9 presents the probability of unambiguity under
Numerical simulation results at various SNRs in Figure 10
various baseline ratios, such as 3:4, 3:13, and 3:16. Observably,
indicate that the AOA determination accuracy is significantly
the ratio 3:4 remarkably outperforms two other ones thanks to
enhanced when increasing the number of array elements.
a larger phase line distance. For several proved benefits, a four‐
Additionally, thanks to larger array apertures, our proposed
element array with the configuration of d ¼ f0; 9; 12; 6gλ=2 is
configuration remarkably outperforms the uniform linear array
recommended for simulation and measurement evaluation.
with the same element number in terms of AOA accuracy. It is
worth noting that the root mean square error (RMSE) of the
determined AOA is defined as follows:
4.3 | Angle of arrival estimation
performance based on number of elements vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uP
uQ ^ �2
u θ − θi
The number of antenna elements has a substantial impact on ti¼1 i
RMSEAOA ¼ ; ð41Þ
the AOA estimation accuracy [33, 34]. It is evident that more Q
properly antennas result in larger array apertures; consequently,
the AOA estimation accuracy is improved. However, the array where ^θi is the determined AOA, θi is the true AOA, and Q is
with more antennas leads to a higher cost of hardware. In this the number of Monte Carlo simulation.
subsection, we demonstrate the prominence of our array
configuration when increasing the number of array elements in
comparison with a uniform linear array with the same element 4.4 | Angle of arrival estimation
number. As a result of the previous section, the baseline ratio performance with different element spacing
3:4 is selected for the proposed configuration in this simula-
tion. Accordingly, Table 3 presents various arrays with the Array element spacing directly impacts the AOA estimation
same baseline ratio for different element numbers. It is worth accuracy of the interferometer because it makes the change of
noting that the configurations with 3−6 antennas facilitate array aperture. In this simulation, the array is configured with
greatly increasing the apertures from 4λ/2 to 256λ/2 but dr = {0, 9, 12, 16}rλ/2, where r is the element spacing ratio
maintaining the same distance (72°) between phase lines in varying from 0.2 to 1.0 with the step of 0.2. Numerical results
hyper‐space. The signal that is taken into account for this in Figure 11 show that the RMSE of AOA estimation reduces
17518792, 0, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/rsn2.12352 by Readcube (Labtiva Inc.), Wiley Online Library on [11/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
DOAN ET AL.
- 11

F I G U R E 1 0 Angle of arrival (AOA) estimation performance with F I G U R E 1 2 Angle of arrival (AOA) estimation performance with
different element numbers different levels of non‐light of sight (NLOS) signals

power ratios to the LOS signal ρ arranging from 0 to 0.95 with


a step of 0.05. The time delays of NLOS signals are rando-
mised in the range of [1, 100] ns. Results shown in Figure 12
indicate that the relative power of NLOS components to LOS
one has a substantial impact on the error of AOA determi-
nation. Specifically, the AOA estimation accuracy reduces
significantly under multipath environments. With a power ratio
ρ ≤ 0.2, the AOA RMSEs are still good (≤0.3°). However,
when ρ ≥ 0.25, the interferometer cannot define AOA
accurately.

4.6 | Comparison with second‐order


difference array, second‐order difference array‐
based inference and correlative methods
F I G U R E 1 1 Angle of arrival (AOA) estimation performance with The SODA method is proposed by exploiting the unambiguity
different element spacing of the second‐order phase difference corresponding to the
second‐order baseline. Assume that an interferometer has 2
baselines d21 (distance between antennas 2 and 1) and d32
along with the element array spacing ratio, in which the ratio (distance between antennas 3 and 2), then the second‐order
r = 1 achieves the highest accuracy thanks to the widest baseline is calculated by dΔ = d32 − d21. The second‐order
aperture. phase difference is therefore defined as follows:

2π sin θ 2πdΔ sin θ


4.5 | Angle of arrival estimation ϕΔ ¼ Φ32 − Φ21 ¼ ðd32 − d 21 Þ ¼ ; ð42Þ
performance under multipath propagation λ λ
conditions
where Φ32 and Φ21 are the phase shift of baselines d32 and d21
In practice, the signal travels from emitter to receiver on a light respectively. The following equation determines the AOA of
of sight (LOS) and non‐light of sight (NLOS) paths, which can signal:
impact the AOA estimation accuracy of the interferometer
system. Therefore, in this simulation, we evaluate the influence � �
ϕΔ λ
of NLOS components on the performance of interferometer. θ ¼ arcsin : ð43Þ
2πdΔ
Accordingly, we assume that a LOS signal is incident on the
antenna array in a fixed direction of 10°. Simultaneously, other
unwanted NLOS signals are also incoming to the receiver array It should be noted that the condition for unambiguous
at different random angles. The NLOS signals have relative AOA determination is dΔ ≤ λ/2.
17518792, 0, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/rsn2.12352 by Readcube (Labtiva Inc.), Wiley Online Library on [11/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
12
- DOAN ET AL.

An extended SODA method, so‐called SODA‐Based


Inference (SBI), is developed in ref. [35] to improve the ac-
curacy of AOA estimation based on resolved unambiguous
second‐order phase difference. Accordingly, we consider a set
of baselines {dΔ, d21, d31} corresponding to a set of phase
differences {ϕΔ, ϕ21, ϕ31}. Therefore, the inferred phased shift
^ 21 of baseline d21 can be expressed as follows:
Φ

^ 21 ¼ d 21 ϕΔ :
Φ ð44Þ

Then, the unambiguous coefficient n21 is calculated as


follows:
� �
Φ^ 21 − ϕ21
n21 ¼ nint : ð45Þ
2π F I G U R E 1 3 Comparison of the proposed method with second‐order
difference array (SODA), SODA‐Based Inference (SBI) and correlative
Based on the unambiguous coefficient n21, the unambig- methods at different SNRs
uous phase shift of the d21 baseline, Φ21 is determined:

Φ21 ¼ ϕ21 þ n21 2π: ð46Þ that the proposed method significantly outperforms SODA
and correlative approaches and has the same accuracy as SBI
Next, using the same process, we can determine the un- for SNR ≥ −5 dB. Interestingly, the proposed and CI methods
ambiguous phase shift Φ31 of the d31 baseline. Finally, the have the same performance and both obtain lower errors than
AOA can be estimated as follows: SODA and SBI for SNR ≤ −5 dB. That is because SODA and
SBI are very sensitive with lower SNRs due to errors contained
� � in second‐order phase differences. The correlative method
Φ31 λ
θ ¼ arcsin : ð47Þ using LUT with the resolution of 0.5° results RMSE > 0.1° for
2πd31
all SNRs and seemly do not improve for SNR > 0 dB. Due to
the widest aperture, the proposed and SBI methods obtain the
It is worth noting that any baseline can be used to estimate
same RMSE < 0.1° at SNR > −4 dB, meanwhile SODA re-
the AOA; however, the widest baseline, for instance d31, should
quires a higher SNR > 24 dB for that achievement because of
be chosen to provide the highest accuracy.
using a very short second‐order baseline.
In another work, correlative interferometer (CI) method
Figure 14 presents the RMSE of determined AOA at +5
resolves the ambiguity problem by searching the minimum
dB SNR. Generally, the proposed method, SBI and SODA
least square of measured phase differences with pre‐calculated
estimate more accurately (i.e. small RMSE) at the middle of
values in a table via LUT scheme, in which the reference table
operation range (approximately from −20° to + 20°), being
can be obtained via practical measurement or simulation [36].
worse (i.e. large RMSE) at the border, because the angle res-
A correct AOA is typically computed as follows:
olution of these methods depends only on the resolution of
! frequency and phase difference measurements. Interestingly,
B �
X �2 the proposed method obtains the highest accuracy (RMSE
^θ ¼ arg min ^ b − ϕb ðθÞ
ϕ ; ð48Þ < 0.2°) at all angles from −80° to + 80°, while CI reaches the
i¼1
nearly constant RMSE ≈ 0.15° at different angles. This result
proves that the AOA determination accuracy of CI depends on
where B is the number of interferometer baselines. Interest- the angle resolution of LUT data. With the same wide aperture,
ingly, the CI is further able to estimate the minimum distance SBI achieves the same accuracy compared to the proposed
between the measured point with all points of phase lines in method for angles from −70° to +70°; however, the AOA
phase plane. error rapidly increases when angles are out of this range. With
The interferometer with d ¼ f0; 9; 12; 16gλ=2 is used in respect to SODA, the best accuracy is reported for the angles
the method comparison. The phase differences of synthetic in range ½−40°; 40°� with RMSE < 1°. Similar to the proposed
signals are computed by using the Fast Fourier Transform. algorithm, SODA performs terribly for the angle at the border
With regard to correlative method simulation, a 3D LUT is with RMSE >2°.
configured with the angle range of [−90°, +90°] and the angle For the computational complexity comparison between
resolution of 0.5°. The comparison results of the 4 methods four methods, Big‐O notation is used. Accordingly, the
are reported in Figure 13, in which the accuracy of AOA complexity of SODA, SBI, correlative and proposed methods
determination is measured via RMSE value for randomly are presented in Table 4, where K is the number of antenna
generated AOAs in the range of [−60°, +60°]. It is observed elements, ns is the number of snapshots (samples), nt is the
17518792, 0, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/rsn2.12352 by Readcube (Labtiva Inc.), Wiley Online Library on [11/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
DOAN ET AL.
- 13

(i.e. d ¼ f32 ; 3 � 4; 42 gλ=2). A transmitter, as shown in


Figure 15c, is used to generate the signal of mode‐S with the
frequency of 1090 MHz. The transmitter was spaced at a dis-
tance of 35.5 m from the interferometer system, which satisfies
2
the far‐field condition 2d3 =λ ¼ 35:2 m. The signal generated
from the transmitter is received by four antenna elements, which
connect to four channels of oscilloscope through four 2‐m
coaxial cables. It is worth noting that the phase shifts between
these radio frequency coaxial cables are identified beforehand
for calibration. For example, the phase shifts between cables 2
versus 1, 3 versus 1, and 4 versus 1 are 10.2°, 3.5°, and 12.3°
respectively. The oscilloscope is set at the highest sampling
frequency of 10 GHz. The signal data from four channels are
recorded for lately processing in the Matlab program.

F I G U R E 1 4 Comparison of the proposed method with second‐order


difference array (SODA), SODA‐Based Inference (SBI) and correlative
5.2 | Signal processing
methods with different AOAs at signal‐to‐noise ratio (SNR) = 5 dB
The raw signal data are first pre‐processed (with the detailed
TABLE 4 Computational complexity comparison between three workflow as sketched in Figure 16) for computing some regular
methods signal parameters, including average carrier frequency, instan-
taneous phase difference, and the pulse envelops of two signal
Method Complexity
channels In Figure 16, BP1 and BP2 are denoted as the fourth‐
SODA O(Kns log ns + K − 1) order band‐pass filters with a bandwidth of 20 MHz and centre
SBI O(Kns log ns + (K − 1)(M(4) + 2)) frequency of 1090 MHz. Block H(t) is referred to as the Hil-
bert transform and it is used for generating in‐phase and
Correlative O(Kns log ns + (K − 1)(M(1) + 1) + nt))
quadrature signals:
Proposed O(Kns log ns + (K − 2)(M(3) + 1) + 1 + M(1))
Z
1 ∞
sðτÞ
Abbreviations: SODA, second‐order difference array; SBI, SODA‐based inference. hðtÞ ¼ H fsðtÞg ¼ dτ: ð49Þ
π −∞ t − τ

number of LUT cells, and M(.) stands in for the complexity of Then, the signal can be presented in the complex form:
multiplication. O(ns log ns) is the complexity of Fast Fourier
Transform used for calculating the phase differences. It is HðtÞ ¼ sðtÞ þ jhðtÞ: ð50Þ
obvious that the SODA method has the lowest complexity;
however, it provides the worst accuracy. The correlative The output of detector is an envelope signal, which is the
method has the highest complexity due to LUT searching al- amplitude of H(t):
gorithm with nt cells. The proposed and SBI methods have the qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
nearly same computational complexity. Although the proposed AðtÞ ¼ kHðtÞk ¼ s2 ðtÞ þ h2 ðtÞ: ð51Þ
and SBI methods are slightly slower than SODA and faster
than CI, they remarkably outperform those methods in terms Angle(H(t)) defines the signal phase in time, which is
of accuracy. determined as follows:

hðtÞ
φðtÞ ¼ arctan : ð52Þ
5 | EXPERIMENTAL MEASUREMENT sðtÞ
AND RESULTS
To calculate the instantaneous carrier frequency of the
5.1 | Measurement setup signal in time, the diff(φ(t)) block is performed as follows:

An interferometer system configured by four antennas in a linear 1 dφðtÞ


f ðtÞ ¼ : ð53Þ
array is set up for this measurement to prove the AOA esti- 2π dt
mation efficiency in practice. The antenna system is shown in
Figure 15, including (a) an example of representative antenna Furthermore, some remaining blocks, denoted TH, &, Δ,
element with geometrical parameters, (b) characteristic pattern and Average in Figure 16 stand for the threshold, AND logic,
of the antenna element in the azimuth and elevation planes, subtraction, and average operations respectively. The pre‐
which are verified in an anechoic chamber. The interferometer processing module provides the phase difference and car-
system has the baselines d1 = 9λ/2, d2 = 12λ/2, and d3 = 16λ/2 rier frequency in time for AOA estimation. For a better
17518792, 0, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/rsn2.12352 by Readcube (Labtiva Inc.), Wiley Online Library on [11/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
14
- DOAN ET AL.

FIGURE 15 System configuration for measurement: (a) antenna element; (b) antenna patterns; (c) measurement scheme; (d) photography of measurement
place

illustration, Figure 17a,b, and c plot an example of input and correlative methods. Figure 18c presents the estimated
signal, output average carrier frequency, and phase difference AOA results of three methods and therefore proves that the
� respectively. Statistically, the mean ðμϕi Þ and RMSE
in time proposed and CI algorithm can estimate the signal AOA with
σ ϕi of phase differences corresponding to the baselines for relatively high accuracy and stability. The proposed method
one pulse duration are shown in Table 5, where the RMSEs yields the average estimated angle of 21.86° with mean error of
are relatively small ðμϕi < 4°Þ. Regarding the carrier fre- 1.86° that are more accurate than CI method with average
quency, its mean value ðμfc Þ is approximately 1090 MHz with estimated angle of 21.98°, mean error of 1.98°. In contrary, the
� SODA method estimate AOA with very high error and di-
small RMSE σ fc of 0.84 MHz, and maximal error of
versity in this measurement.
3.67 MHz.

5.3.2 | Measurement at various directions


5.3 | Experimental results
This measurement verified the AOA determination efficiency
5.3.1 | Measurement at one direction of system in the case of which the signal is transmitted from
different angles varying from −60° to 60° with the step size of
In this experiment, we measure the phase difference and AOA 5°. The AOA measurement results are shown in Figure 19. It
estimation error for the case of one‐direction signal trans- can be observed that the proposed algorithm achieves the most
mission. Concretely, the interferometer system receives the accurate AOA estimation at angles approaching 0°, for
signal radiated from the transmitter with the direction of 20°. instance, the standard deviations are smaller than 0.2° for
The results of phase difference and their RMSE of the received AOAs from −40° to −45°, meanwhile, the higher AOA de-
signals are shown in Figure 18a and b, respectively. Further- viation occurs when measuring signals are at the border, where
more, the proposed algorithm is compared with the SODA the greatest deviation is 0.49° for AOA = −60°.
17518792, 0, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/rsn2.12352 by Readcube (Labtiva Inc.), Wiley Online Library on [11/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
DOAN ET AL.
- 15

T A B L E 5 Statistical values of phase differences and carrier frequency


in one pulse duration

Baseline d1 Baseline d2 Baseline d3


μϕi −110.22° 52.98° −3.33°

σ ϕi 3.58° 1.89° 1.80°

μfc 1090 MHz

σ fc 0.84 MHz

F I G U R E 1 6 Signal preprocessing scheme for getting instantaneous


phase difference and carrier frequency

F I G U R E 1 8 Experiment measurement for the direction 20°: (a) The


phase differences of received signals; (b) root mean square error (RMSE) of
phase difference measurements; (c) angle of arrival (AOA) determinations

F I G U R E 1 7 Demonstration of (a) raw signals, (b) carrier frequency,


and (c) phase difference frequency varying over time

6 | CONCLUSION
F I G U R E 1 9 The results for the angle of arrival (AOA) measurement
in the direction from −60° to 60° by step of 5°: (a) Measured AOA vs. real
In this paper, a novel AOA estimation approach has been
AOA; (b) root mean square error (RMSE) for each AOA measurement
introduced to address the ambiguity problem of phase differ-
ence. In particular, the proposed method having three primary
transformation steps of rotating, re‐scaling, and rounding has measurement error, which usually depends on the quality of
been evaluated and compared with two existing methods via devices, has to be smaller than the maximum allowable error
numerical simulations as well as practical measurement. Based on calculated by the half‐distance between neighbour lines in the
the numerical results, we conclude that the phase difference phase space. With regard to each interferometer baseline ratio, an
17518792, 0, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/rsn2.12352 by Readcube (Labtiva Inc.), Wiley Online Library on [11/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
16
- DOAN ET AL.

appropriate phase space, a maximum allowable error for phase 8. Spencer, S.J.: Closed‐form analytical solutions of the time difference of
difference measurement, and a coefficient matrix are selected to arrival source location problem for minimal element monitoring arrays. J.
gain the most accurate AOA estimation. Finally, a practical Acoust. Soc. Am. 127(5), 2943–2954 (2010). https://doi.org/10.1121/1.
3365313
measurement was performed to prove the efficiency of the 9. Cui, X., Yu, K., Lu, S.: Evolutionary TDOA‐based direction finding
proposed method, where the realistic system with the most methods with 3‐D acoustic array. IEEE Trans. Instrum. Meas. 64(9),
appropriate antenna array configuration has achieved high‐ 2347–2359 (2015). https://doi.org/10.1109/tim.2015.2415051
accurate AOA estimation with the standard deviation lower 10. Farmani, M., et al.: Informed TDoA‐based direction of arrival estimation
than 0.5°. for hearing aid applications. In: Proc. 2015 IEEE Global Conf. Signal
Inf. Process, pp. 953–957. (GlobalSIP), Orlando (2015)
11. Guerin, D., Jackson, S., Kelly, J.: Passive Direction Finding A Phase
AUT HO R CO N TR I BU T I ON Interferometry Direction Finding System for an Airborne Platform.
Van‐Sang Doan: Conceptualisation; Data curation; Formal Worchester Polytechnic Institute (2012)
analysis; Investigation; Methodology; Validation; Visualisation; 12. Report ITU‐R SM.2211‐2: Comparison of Time‐Difference‐Of‐Arrival
and Angle‐Of‐Arrival Methods of Signal Geolocation. ITU‐R (2018)
Writing – original draft; Writing – review & editing. Thien
13. Jacobs, E., Ralston, E.W.: Ambiguity resolution in interferometry. IEEE
Huynh‐The: Data curation; Resources; Visualisation; Writing Trans. Aero. Electron. Syst. AES‐17(6), 766–780 (1981). https://doi.org/
– review & editing. Van‐Phuc Hoang: Funding acquisition; 10.1109/taes.1981.309127
Methodology; Project administration; Resources; Writing – 14. Liu, Z., Guo, F.: Azimuth and elevation estimation with rotating long‐
review & editing. Jiri Vesely: Conceptualisation; Data curation; baseline interferometers. IEEE Trans. Signal Process. 63(9), 2405–2419
Investigation. (2015). https://doi.org/10.1109/tsp.2015.2405506
15. Holder: Angle‐of‐Arrival Estimation Using Radar Interferometry:
Methods and Applications. Institution of Engineering and Technology
ACKNO W L E DG EMEN T S (2013)
The research is supported by the Naval Academy and the 16. Nessel, J.A., Acosta, R.J.: Predicting sparse array performance from two‐
Institute of System Integration, Le Quy Don Technical Uni- element interferometer data. IEEE Trans. Antenn. Propag. 60(2),
886–894 (2012). https://doi.org/10.1109/tap.2011.2173110
versity, Vietnam in collaboration with the Department of
17. Van Trees, H.L.: In: Van Trees, H.L. (ed.) Optimal Array Processing, ser.
Communication Technologies, Electronic Warfare and Radio- Detection, Estimation and Modulation Theory IV. John Wiley & Sons,
location, Faculty of Military Technology, the University of Inc. (2002)
Defence in Brno, Czech Republic. This work is funded by 18. Doan, V.‐S., Kim, D.‐S.: DOA estimation of multiple non‐coherent
Vietnam ‐ Czech bilateral project “NEO classification of sig- and coherent signals using element transposition of covariance ma-
nals (NEOCLASSIG) for radio surveillance systems” under trix. ICT Express 6(2), 67–75 (2020). https://doi.org/10.1016/j.icte.
2020.03.008
grant number NDT/CZ/22/12. 19. Kendall, W.B.: Unambiguous accuracy of an interferometer angle‐
measuring system. IEEE Trans. Space Electron. Telemetry SET‐11(2),
CONF L IC T S OF I N T E R ES T 62–70 (1965). https://doi.org/10.1109/tset.1965.5009646
All authors declare that they have no conflicts of interest. 20. Goodwin, R.L.: Ambiguity‐resistant Three‐ and Four‐Channel Inter-
fereometers. Naval Res. Lab., Washington, DC, USA, NRL‐8005 (1976)
21. Pan, Y.J., et al.: An ultra‐fast DOA estimator with circular array inter-
DATA AVAI L A BI LI T Y S TAT E M E NT ferometer using lookup table method. Radioengineering 24(3), 850–856
Data available on request from the authors. (2015). https://doi.org/10.13164/re.2015.0850
22. Cui, K.B., et al.: 2D DOA estimation of UCA correlative interferometer
based on one dimensional sorting lookup table‐two dimensional linear
OR CID
interpolation algorithm. Radioengineering 26(2), 562–572 (2017).
Van‐Sang Doan https://orcid.org/0000-0001-9048-4341 https://doi.org/10.13164/re.2017.0562
Thien Huynh‐The https://orcid.org/0000-0002-9172-2935 23. Searle, S.: An examination of bias in SODA interferometry. IEEE Signal
Van‐Phuc Hoang https://orcid.org/0000-0003-0944-8701 Process. Lett. 24(4), 470–474 (2017). https://doi.org/10.1109/lsp.2017.
2672742
24. Towers, C.E., Towers, D.P., Jones, J.D.C.: Time efficient Chinese
RE FE R ENC ES remainder theorem algorithm for full‐field fringe phase analysis in multi‐
1. Wiley, R.G.: ELINT: The Interception and Analysis of Radar Signals. wavelength interferometry. Opt Express 12(6), 1136–1143 (2004).
Artech House, Boston (2006) https://doi.org/10.1364/opex.12.001136
2. Lipsky, S.E.: Microwave Passive Direction Finding. SciTech Pub., Raleigh 25. Jiang, Z., et al.: Multibaseline phase unwrapping through robust Chinese
(2004) remainder theorem. In: Proc. 7th IEEE Int. Symp. Microw., Antenna,
3. Adamy, D.L.: EW 101: A First Course in Electronic Warfare. Artech Propag., and EMC Technol, pp. 462–466. (MAPE), Xi’an (2017)
House, Boston (2008) 26. O’Neill, S.: Electronic Warfare and Radar Systems Engineering Hand-
4. Martino, A.D.: Introduction to Modern EW Systems, Second Edition. book. Military Studies Press (2013)
Artech House, Boston (2018) 27. Lee, J., Lee, J., Woo, J.: Method for obtaining three‐ and four‐element
5. Vaccaro, D.D.: Electronic Warfare Receiving Systems. Artech House, array spacing for interferometer direction‐finding system. IEEE
Boston (1993) Antenn. Wireless Propag. Lett. 15, 897–900 (2016). https://doi.org/10.
6. An, D.‐J., Lee, J.‐H.: Performance analysis of amplitude comparison 1109/lawp.2015.2479224
monopulse direction‐of‐arrival estimation. Appl. Sci. 10(4), 1246 (2020). 28. Younger, J.P., Reid, I.M.: Interferometer angle‐of‐arrival determination
https://doi.org/10.3390/app10041246 using precalculated phases. Radio Sci. 52(9), 1058–1066 (2017). https://
7. Spencer, S.J.: The two‐dimensional source location problem for time doi.org/10.1002/2017rs006284
differences of arrival at minimal element monitoring arrays. J. Acoust. 29. Taylor, J.R.: Introduction to Error Analysis: The Study of Uncertainties in
Soc. Am. 121(6), 3579 (2007). https://doi.org/10.1121/1.2734404 Physical Measurements. UNIV SCIENCE BOOKS (1997)
17518792, 0, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/rsn2.12352 by Readcube (Labtiva Inc.), Wiley Online Library on [11/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
DOAN ET AL.
- 17

30. Rouabhia, M.N.E.I., et al.: On the phase interferometry direction 35. Ly, P.Q.C., et al.: Computationally fast AOA estimation using sparse large
finding: performance comparison and FPGA implementations. In: aperture arrays for electronic surveillance. In: 2013 Int. Conf. Radar, pp.
Proc. 2017 Seminar Detect. Syst. Archit. Technol, pp. 1–5. (DAT), 526–531. (2013)
Algiers (2017) 36. Therese, A., Babu, N., Krishna, M.: Passive localisation based on
31. Zhang, M., et al.: Phase difference measurement method based on correlative interferometry. Procedia Comput. Sci. 143, 2–9 (2018).
progressive phase shift. Electronics 7(86), 1–11 (2018). https://doi.org/ https://doi.org/10.1016/j.procs.2018.10.345
10.3390/electronics7060086
32. Covles, J., Gilbert, B.: Accurate gain/phase measurement at radio fre-
quencies up to 2.5 GHz. Analog. Dialogue 35(5), 1–4 (2001)
33. Zhang, M., et al.: Angle‐domain channel and DOA estimation in airborne How to cite this article: Doan, V.‐S., et al.: Phase‐
massive MIMO system with eavesdropper. In: Proc. 2019 28th Wirel. difference measurement‐based angle of arrival
Opt. Commun. Conf. (WOCC), pp. 1–6. Beijing (2019) estimation using long‐baseline interferometer. IET
34. Zhang, M., et al.: Multi‐L‐shape array division based 2‐D localization for
incoherently distributed sources in airborne massive MIMO systems. In:
Radar Sonar Navig. 1–17 (2022). https://doi.org/10.
Proc. IEEE 19th Int. Conf. Commun. Technol, pp. 430–435. Xi’an, 1049/rsn2.12352
China (2019)

You might also like