sensors
Article
An Experimental Multi-Target Tracking of AM Radio-Based
Passive Bistatic Radar System via Multi-Static Doppler Shifts
Xueqin Zhou 1,2, * , Hong Ma 1, * and Hang Xu 1






Citation: Zhou, X.; Ma, H.; Xu, H.
An Experimental Multi-Target
Tracking of AM Radio-Based Passive
Bistatic Radar System via Multi-Static
Doppler Shifts. Sensors 2021, 21, 6196.
https://doi.org/10.3390/s21186196
Academic Editors: Piotr Samczynski,
Daniel W. O’Hagan,
Jacek Misiurewicz and Lorenzo
Lo Monte
Received: 10 August 2021
Accepted: 10 September 2021
Published: 15 September 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Sensors 2021, 21, 6196. https://doi.org/10.3390/s21186196
1 School of Electronic Information and Communications, Huazhong University of Science and Technology,
Wuhan 430074, China; xuhang90@foxmail.com
2 Department of Electrical and Electronic Engineering, College of Engineering Technology, Hubei University of
Technology, Wuhan 430068, China
* Correspondence: conmie2008@163.com (X.Z.); mahong@hust.edu.cn (H.M.)
Abstract: This paper presents a description of recent research and the multi-target tracking in experi-
mental passive bistatic radar (PBR) system taking advantage of numerous non-cooperative AM radio
signals via multi-static doppler shifts. However, it raises challenges for use by multiple spatially
distributed AM radio illuminators for multi-target tracking in PBR system due to complex data
association hypotheses and no directly used tracking algorithm in the practical scenario. To solve
these problems, after a series of key array signal processing techniques in the self-developed system,
by constructing a nonlinear measurement model, the novel method is proposed to accommodate non-
linear model by using the unscented transformation (UT) in Gaussian mixture (GM) implementation
of iterated-corrector cardinality-balanced multi-target multi-Bernoulli (CBMeMBer). Simulation and
experimental results analysis verify the feasibility of this approach used in a practical PBR system for
moving multi-target tracking.
Keywords: PBR; CBMeMBer; doppler; tracking
1. Introduction
Passive bistatic radar (PBR) is a subset of bistate radars receiving non-cooperative
transmitters of opportunity scattered by potential targets. Research on PBR has attracted
extensive attention because of well-known advantages, such as no additional frequency
channel allocation, lower costs, and lower probability of being detected with respect to
active radars. Although, PBR systems have a long history, there are not enough opera-
tional systems.
Of all the transmitters of opportunity available in the PBR systems, very high fre-
quency/ultra high frequency (VHF/UHF) bands represent some of the most attractive for
surveillance purposes, such as analog television (ATV) [1,2], digital television-terrestrial
(DTV) [3,4], frequency modulation (FM) radio [5], digital audio/video broadcasting
(DAB/DVB) [6–10]. However, relatively little interest has been shown in the high fre-
quency (HF) band (3–30 MHz) due to the propagation complexity and low range resolution.
Especially, the external illuminators in the HF band have excellent range coverage, propa-
gation over the horizon, and stealth target detection. Some preliminary and pioneering
HF-PBR works have been carried out. In the PBR system [11,12], Thomas et al. from
University College London performed an analysis using the HF digital radio Mondiale
(DRM)signal as transmitter of opportunity. The two-dimensional target localization, using
a linear frequency modulated continuous waveform from a non-cooperative OTH radar
located in Longreach, Australia, with a bandwidth of 10 kHz, is presented in [13].
In fact, compared with other opportunity illuminators in the HF band, commercial
amplitude modulation (AM) broadcast signal sources have the advantages of high trans-
mitter power, larger numbers, and wider coverage. Due to the propagation complexity and
bandwidth limited, little attention is paid to AM radio signal for PBR system. The research
https://www.mdpi.com/journal/sensors
Sensors 2021, 21, 6196 2 of 18

development on the practical PBR system named “Sugar Tree” was the first using the AM
broadcast signal as a passive radar illuminator to detect the missiles launched near the
broadcast emitter site. Although this system was shut down decades ago, the performance
and method details have not yet been elaborated [14]. One target was tracked by using
HF-AM radio signals in passive system [15]. We note that to date, there has been little
convincing experimental evidence that AM radio based PBR systems, in particular, can be
used for the purpose of multi-target tracking.
Actually, using the AM broadcast signal as an illuminator for air surveillance raises
new challenges for the multi-target tracking algorithm in PBR system. First, range resolu-
tion is not available or often of extremely poor quality. For example, the regular effective
AM radio broadcast bandwidth is 5 kHz so that the practical radar range resolution is up
to 60km. That would be unacceptable for the PBR systems. Second, the signal-to-noise
ratio (SNR) of echo signal is particularly low and there exists the clutter interference, strong
direct path interference, and the missing alarm. Third, in order to track the target, the
system must observe the target’s Doppler history for an extended time before there is suffi-
cient information. Furthermore, an additional type of data association uncertainty arises
due to the use of multiple AM radio broadcast signals, i.e., not only is it unknown which
measurement belongs to which target, there is also the problem that the data association
hypotheses arise from the number of AM radio broadcast illuminators. In this paper, we
report the newly experimental development of a PBR system using multiple AM radio
broadcast signals as a transmission source.
Since the PBR system can typically collect bistatic range, time-of-arrival (TOA),
direction-of-arrival (DOA), and Doppler shift from the received signals, an alternative
choice is to use multi-static Doppler shift as measurements for multi-target tracking. Target
estimation using only Doppler shift measurements is an old problem studied in different
contexts [16–18]. To achieve the target location accuracy, multiple transmitters can be
utilized simultaneously in the system. Therefore, target tracking in the AM radio based
PBR system is a multi-sensor fusion problem. Some works on localization and tracking
multi-targets in a multi-sensor radar system via multi-static Doppler-shift measurements
only have been studied recently. It is shown in [19] that a multi-sensor Bernoulli filter with
multi-static Doppler-only measurements contaminated by unknown probability of false
alarms and missed detections is able to perform multiple target tracking via numerical
simulation examples. An algorithm for fusing data from a constellation of RF sensors
detecting cellular emanations with the output of a multi-spectral video tracker to localize
and track a target with a specific cell phone is present in [20]; the results show that it is
possible to track multiple targets using Doppler differential measurements. However, few
field experiments have been performed to demonstrate the target tracking performance in
similar scenarios.
In this paper, we went a step further and cast close-in moving multi-target tracking
via multi-static Doppler shifts in the practical PBR system by multiple spatially distributed
AM radio illuminators. We made use of a cardinality-balanced multi-target multi-Bernoulli
(CBMeMBer)-based random finite set (RFS) approach to sequentially estimate targets, with
iterated-corrector (IC) iterating the filter update stage. Compared to data association meth-
ods, a CBMeMBer-based Bayesian algorithm can avoid complex association processes; it
also has significant advantages in less cardinality bias and smaller complexity in solving
multi-target tracking problems. There are two implementations of the CBMeMBer filter,
Gaussian mixtures (GM) and sequential Monte Carlo (SMC). Contrary to SMC implementa-
tion, GM implementation provides more reliable state estimates and a lower computation
complexity in an efficient way [21]. However, the GM-CBMeMBer filter does not directly
accommodate a nonlinear measurement model. To solve this, we extended GM implemen-
tation of the CBMeMBer filter to update the procedure by using unscented transformation
(UT). Simulation performance demonstrates the analysis, and the real data from exper-
imental results indicate feasibility performance of the proposed IC-UT-GM-CBMeMBer
filter. The system described in this paper was constructed on one of the simplest and
 IC-UT-GM-CBMeMBer filter. The system described in this paper was constructed on one
of the simplest and lowest cost architectures; therefore, the main contribution of this paper
Sensors 2021, 21, 6196 is that the proposed multi-target tracking method can provide a reference for similar3 of PBR18
systems.
The rest of this paper is organized as follows. The description of the AM Radio based
PBR system and multi-target tracking formulation are described in Section 2. Section 3
lowest
providescostthe
architectures; therefore, the main contribution
proposed IC-UT-GM-CBMeMBer of this paper
filter. Simulation andisfield
that the proposed
experimental
multi-target tracking method can provide a reference for similar PBR systems.
implementation are given in Section 4. Finally, conclusion and possible future directions
The rest
are drawn inof this paper
Section 5. is organized as follows. The description of the AM Radio based
PBR system and multi-target tracking formulation are described in Section 2. Section 3
provides
2. Problem theFormulation
proposed IC-UT-GM-CBMeMBer filter. Simulation and field experimental
implementation are given in Section 4. Finally, conclusion and possible future directions
2.1. System Description
are drawn in Section 5.
Supposing the ionosphere is homogeneous and spherically symmetric, the bistatic
plane
2. geometry
Problem of an AM-radio-based sky-surface wave PBR system in the scenario can
Formulation
be simplified
2.1. as shown in Figure 1 (two transmitters are shown explicitly simplistically).
System Description
In the system,
Supposing the we ionosphere
exploited one receiving antenna
is homogeneous array approximately
and spherically symmetric, 20 thembistatic
above
ground level located over-the-horizon (farther than 1000 km from the
plane geometry of an AM-radio-based sky-surface wave PBR system in the scenario can be noncooperative
transmitters
simplified as generally) at Hubei
shown in Figure province
1 (two of China,
transmitters which is
are shown equipped
explicitly with a uniform
simplistically). In
circular array (UCA) with 16 antennas.
the system, we exploited one receiving antenna array approximately 20 m above ground
The direct
level located wave (emitter-to-receiver)
over-the-horizon and the
(farther than 1000 illumination
km from wave (emitter-to-target)
the noncooperative transmitters
are reflected
generally) from the
at Hubei ionosphere,
province while
of China, the echo
which wave (target-to-receiver)
is equipped was viaarray
with a uniform circular line-
of-sightwith
(UCA) (LOS) propagation in the surveillance area.
16 antennas.

Figure 1. Plane geometry of AM-radio based-sky-surface wave PBR system.

Figure 1. Plane geometry of AM-radio based-sky-surface wave PBR system.
The direct wave (emitter-to-receiver) and the illumination wave (emitter-to-target) are
reflected from the
To obtain ionosphere,Doppler
multi-static while themeasurements
echo wave (target-to-receiver)
in the PBR system, was via
someline-of-sight
key tech-
(LOS) propagation in the surveillance area.
niques in array signal processing are summarized in a block diagram sketched in Figure
To obtain
2. Similar multi-static
to traditional Doppler
passive measurements
radar, surveillance in and
the PBR system,
reference some key
channels aretechniques
needed to
in array signal
receive target processing
echoes andare summarized
reference inrespectively.
signal, a block diagram Thesketched
DOA of in Figure
direct wave2. Similar
can be
to traditional
obtained passive
by the radar,
multiple surveillance
signal and reference
classification (MUSIC) channels
algorithmarefrom
needed thetoreference
receive target
chan-
echoes and reference signal, respectively. The DOA of direct wave can
nel, which is then used to clean the reference signal by using conventional beamformingbe obtained by the
multiple signal classification (MUSIC) algorithm from the reference
(CBF) technology. After scanning the surveillance channel by normalized least meanchannel, which is then
used
squareto (NLMS)
clean thetechnology
reference signal by using
to obtain conventional
the echo signal, webeamforming
calculated the (CBF) technology.
cross-ambiguity
After
function (CAF) of the direct path signal and the scattered signal to estimate range vs. tech-
scanning the surveillance channel by normalized least mean square (NLMS) Dop-
nology to of
pler shift obtain the echo
the targets. signal,the
Finally, wetime
calculated the Doppler
delay and cross-ambiguity function
shift of the targets(CAF) of the
after clutter
direct
removalpath signal
were and the scattered signal to estimate range vs. Doppler shift of the targets.
estimated.
Finally, the time delay and Doppler shift of the targets after clutter removal were estimated.
Sensors 2021,
Sensors 21, 21,
2021, 6196x FOR PEER REVIEW 4 of
4 18
of 19

Figure 2. Signal
Figure processing
2. Signal diagram.
processing diagram.
Furthermore, the greatest limitation on tracking target performance in the
Furthermore, the greatest limitation on tracking target performance in the self-devel-
self-developed experimental PBR system is the interference and clutter in the received
oped experimental PBR system is the interference and clutter in the received signal, for
signal, for example, dense direct path interference and the DOA of echo signal with very
example, dense direct path interference and the DOA of echo signal with very low SNR
low SNR under masking effects. Although the classic DOA estimation and clutter suppres-
under masking effects. Although the classic DOA estimation and clutter suppression
sion methods have been studied over the past decade [22,23], most of them are unsuitable
formethods
processinghave been studied
HF-AM over the
radio signal. We past decade
adopted the [22,23],
methodmost of them[24]
of reference are by
unsuitable
buildingfor
processing HF-AM radio signal. We adopted the method of reference
a single-snapshot virtual array signal. After extending the single-snapshot virtual [24] by building
array a
signal to multi-snapshots and the MUSIC algorithm, the clutter interference could besig-
single-snapshot virtual array signal. After extending the single-snapshot virtual array
nal to multi-snapshots
suppressed andthe
significantly, and thedesired
MUSICecho
algorithm, the enhanced
signal was clutter interference could More
simultaneously. be sup-
pressed significantly, and the desired echo signal was enhanced simultaneously.
details of signal processing and improvement can be found in [24]. Finally, the excellent More
details of signal processing and improvement can be found in [24]. Finally,
Doppler shift information of targets’ corresponding time can be provided on a 2-D time vs. the excellent
Doppler
Doppler shift information of targets’ corresponding time can be provided on a 2-D time
map.
vs. Doppler map.
2.2. CBMeMBer Filter
2.2.The
CBMeMBer
CBMeMBer Filter
filter is first introduced to solve the aforementioned tracking problem
The system
in the PBR CBMeMBerin thisfilter is first
section [21].introducednto solve the aforementioned o tracking prob-
lem in the PBR system in this section [21].
At time k, there are N(k) target states Xk = xk,1 , · · · , xk,N (k) ⊆ F (χ), which denote
space ofAtfinite
timesubsets
k, thereofare N(k) target
χ. Given xk at𝑋time
states
a target = {𝑥
k, it, ,is⋯either
, 𝑥 , ( detected
) } ⊆ ℱ(𝜒),in which denote
the surveil-
space of finite subsets of 𝜒. Given a target 𝑥 at time k, it is either detected
lance area with probability p D,k ( xk ) and generates a Bernoulli RFS Θk ( xk ) with likelihood in the surveil-
lance area
function g (·|with probability
x ) , or it is missed , (𝑥 probability
𝑝 with ) and generates
1 − p a Bernoulli
( x ). Given RFS 𝛩 (𝑥 ) withstate
a multi-target likeli-
k k D,k k
Xkhood
, eachfunction k (⋅ | xk )continues
xk ∈ Xk geither to exist at time k + 1 with 1− pprobability
D,k ( xk ) pSa( xmulti-target
k +1 ) and
, or it is missed with probability . Given
moves to a new state xk+1 with target transition equation f k+1|k ( xk ) or dies with probability
X xk ∈ X k p (x )
1 −state
pS,k ( xkk+, 1each
). Thus, giveneither a targetcontinues
with state to xexist
k ∈ X atk time
at time k +k,1its
with time k +S1 isk +1
probability
behavior
modeled by the Bernoulli RFS Sxk+k +11|k ( xk ), and Γk+1 denotes the multi-Bernoulli f (x ) RFS of new
and moves to a new state with target transition equation k +1|k k or dies with
births at time k + 1. The multi-target state is modeled as [25].
1− pS ,k ( xk +1 ) x ∈ Xk
probability . Thus, given a target with state k at time k, its behavior
Xk+1 = [∪ Sk+1|k ( xk )] ∪ Γk+1 (1)
Sk +1|k ( xk ) Γ
time k + 1 is modeled by the Bernoulli RFS n , and o k +1 denotes the multi-Ber-
noulli RFS ofthere
Similarly, new are
births Nk at time k + 1. The
measurements Zkmulti-target
= zk,1 , . . . ,state
zk,N is, modeled
each taking as [25].
values in an
k
observation space at time k. In addition, the received measurement
X k +1 = [ Sk +1|k ( xk )]  Γk +1 also contains a set of
(1)
missing alarms or clutter that can be modeled as a Poisson RFS Kk . Thus, multi-target
observation at time k + 1 is modeled as finite sets [26].
N Z k = {zk ,1,  , zk , N }
Similarly, there are k measurements k , each taking values in
Z
an observation space at time k. In addition,
k + 1 = [ ∪ Θ k + 1 ( x )] ∪ K
the received measurement also contains(2)
k + 1 a set
x ∈ X k +1
Κ
k Thus, multi-target
of missing
where Θ( xk+1alarms or clutterRFS
) is a Bernoulli thatthat
canisbegenerated
modeled by as atarget
Poisson RFS
state xk+1 ∈. X k +1 .
observation at time k +RFS
A multi-Bernoulli 1 isXmodeled
(i ) on χ as is finite
a unionsetsof[26].
a fixed number M of independent
Bernoulli RFSs with existence probability r (i) and probability density p(i) , X = ∪iM (i )
=1 X .
Z k +1 = [  Θk +1 ( x)]  Κ k +1 (2)
Moreover, the probability density π is [27]:x∈X k +1

Θ( xk +1 ) is a Bernoulli πRFS ) target state xk +1 ∈ X k +1 .

M
where =∏
(∅)that (1 − r ( j)by
is generated
j =1
(3)
Sensors 2021, 21, 6196 5 of 18

M r (i j ) p (i j ) ( x j )
π ({ x1 , · · · xn }) = π (∅) ∑ ∏ j =1 (4)
1≤i1 6=···6=in ≤ M (1 − r (i j ) )
The multi-target bayes recursion propagates in time [27]:
Z
πk+1|k ( Xk | Z1:k ) = f k+1|k ( Xk+1 | X )πk ( X | Z1:k )δX (5)

gk+1 ( Zk+1 | Xk+1 )πk+1|k ( Xk+1 | Z1:k )

πk+1 ( Xk+1 | Z1:k+1 ) = R (6)
gk+1 ( Zk+1 | X )πk+1|k ( X | Z1:k )δX
where f k+1|k (·|·) is the multi-target transition density and gk+1 (·|·) is the multi-target
likelihood.

2.3. Multi-Target Tracking Model

Tracking model is one of the major problems needing to be considered in the multi-
target system. In this paper, we consider the target tracking scenario performed in a 2D
Cartesian coordinate, with the origin point located at a single receiver antenna array; the
x-axis points East and the y-axis points North. Assume that at time k, the i-th target state is
(i ) (i ) (i ) (i ) T
h i
(i )
represented by the state vector xk = p x,k v x,k py,k vy,k , i = 1, 2, . . . , N (k) Where N (k)
(i ) (i ) T
h i
(i )
is the number of targets, superscript T denotes the matrix transpose, pk = p x,k py,k
(i ) (i ) T
h i
(i )
and vk = v x,k vy,k are the position and velocity of the target, respectively. Each target
dynamic motion is followed by a nearly constant velocity model:

(i ) (i )
xk+1 = Fk xk + uk (7)

where uk ∼ N (u; 0, Qk ) is zero-mean white Gaussian process noise with covariance Qk . We

adopt:
2 ∆ I2 02
   2 
A0 02
Fk = Qk = σv
02 A0 02 ∆2 I2
1 ∆
 
In which ∆ is the sampling interval. A0 = , In , and 0n denote n × n identity
0 1
and zeros matrices, respectively.
In the two-dimensional surveillance area, three spatially distributed non-cooperative
(i )
AM radio illuminators constantly transmit signals with a known carrier frequency f c of
the i-th AM radio station, i = 1 . . . 3, and the receiver places are at the original point, as
illustrated in Figure 3. The direct wave and scattered waves from multiple AM broadcast
stations reflected from the ionosphere (three scattered echo waves and one target are shown
simplistically) reach the target and receiver simultaneously on the condition that the AM
(i ) (i ) T
h i
(i )
broadcast stations located at Rt = xt yt are far away (>1000 km) from the receiver
antenna array. Doppler shift measurements by i-th illumination can be generally modeled
as [28]:
z k (i ) = h (i ) ( x k ) + ε k (i ) (8)
where:
(i ) (i )
" #
pk p − Rt fc
h(i) ( xk ) = −vkT + k (9)
k p k k k p − R (i ) k c
k t
→
q
vk is the constant velocity vector of target at time k, k pk k = p x,k 2 + py,k 2 , c is the
speed of light, ε k (i) is the measurement noise, ε k (i) ∼ N w; 0, σw 2 .


  p p − Rt  f
h ( i ) ( xk ) = −vkT  k + k  c (9)
p
 k pk − Rt( i )  c

vk pk = px,k 2 + py ,k 2
Sensors 2021, 21, 6196 is the constant velocity vector of target at time k, , c is 6the
of 18

speed of light, k
ε (i )
ε
is the measurement noise, k
(i )
 N ( w; 0, σ w )
2

pk


v

Multiple
Figure3.3.Multiple
Figure non-cooperative
non-cooperative illuminators
illuminators ofof
AMAM radiostations
radio stations(red
(redpentacle)
pentacle)reflected
reflectedfrom
from
ionosphere
ionospherereceived
receivedbybysingle
singlereceiver
receiverantenna
antennaarray
array(blue
(bluetriangle)
triangle)ininx-y
x-ycoordinate.
coordinate.

AsAsthe
theAMAMradio
radio stations
stations are
are typically
typically long
long distance
distance from
from thethe receiver
receiver antenna
antenna array,
array,
weapproximately
we approximatelyconsidered
consideredthe thedirection
directionofofthe
thedirect
directwave
wavefromfromthe i-thilluminator
thei-th illuminatorofof
AMradio
AM radiostation
stationtotothe
thereceiver
receiverantenna
antennaarray
arrayasasequal
equaltotothe
thedirection
directionofofthethescattered
scattered
wave from the i-th illumination to the target. According to Equation (9),
wave from the i-th illumination to the target. According to Equation (9), the model can the model canbebe
approximately written
approximately written as: as:

(i )( i )   (i )  → (→

f c f c h  −→ i)
h(i) (hxk )( x=
(i )
k )= vk → vk v⋅ vt·k vt (i)
p k (i)++
i
− v p 
(10)
(10)
cc  k k k k

 (i )  (i ) →
p→ i) vt k vt
Here, kp k (is
Here, the normalized
is the normalizedtarget position
target relative
position to the
relative toreceiver.
the receiver. (i ) is
is kthe nor-the
normalized
malized incident
incident direction
direction vector
vector of the
of the direct
direct wave
wave from
from i-thi-th
thethe illuminator
illuminator ofofthetheAM AM
radiostation
radio stationtotothethereceiver
receiverantenna
antennaarray,
array,which
whichisisindependent
independentononthe thetarget
targetstate
stateand and
couldbebeeasily
could easilyachieved
achievedfromfromDOA
DOAestimation
estimationofofthe
thedirect
directwave,
wave,asasmentioned
mentionedbefore. before.
EachDoppler-shift
Each Doppler-shiftsubsetsubsetincludes
includesatatmost
mostone
onemeasurement
measurementper perilluminator
illuminatorand and
correspondstotothe
corresponds themeasurements
measurementsmade madebybymulti-targets
multi-targetsacross
acrossallallilluminators
illuminatorspractically
practically
contaminatedbybyfalse
contaminated falsealarms
alarmsand
andmisdetections.
misdetections.TheThesubsets
subsetsofofa apartition
partitionare aredisjointed
disjointed

{ }

(i ) ( i ) (i ) (i ) (i )
and comprise measurement space, which is denoted as Zk Z= z(i ) zz(ik,2
k = zkk,1
,1
)
k ,2 · · ·zk(i,)Nz(k,N
i) (i ) ,
k
and comprise measurement space, which is denoted as k, in
(i )
in which i is the i-th illuminator; Nk is the number of the detection values, including false
(i ) (I)
alarms and misdetection; zk,j is the j-th detection value. Therefore, Zk is the measurement
(I)
set involving all Ns illuminators at time k, and Z1:k is the time sequence of measurement
sets encapsulated I = {1, 2, . . . , NS } illuminator characteristics.

3. The Proposed Multi-Target Tracking Method

The GM-CBMeMBer filter has a close-form solution under assumptions of linear
Gaussian models that is difficult to implement on the nonlinear measurement models. To
overcome this limitation, we extended the GM-CBMeMBer filter to a practical nonlinear
measurement model by using unscented transform (UT) techniques [29]. Another straight-
forward extension of the single sensor CBMeMBer filters to the case of multiple illuminators
can be achieved by iterating the filter update stage for each illuminator measurement set.
An IC-UT-GM-CBMeMBer filter can be implemented to accommodate a multi-transmitter
nonlinear Doppler model. However, this IC-CBMeMBer yields final solutions that depend
on the order of the measurement set of illuminators; therefore, the development of efficient
Sensors 2021, 21, 6196 7 of 18

algorithms for the scenario case are left for future investigation. Hence, in this section we
propose the IC-UT-GM-CBMeMBer filter for multi-target tracking in the PBR system.

3.1. IC-UK-GM-CBMeMBer Filter

We supposed that each target follows a linear Gaussian dynamical and observation
mode [21], i.e., 
f k+1|k ( x ζ ) = N ( x; Fk ζ, Qk ) (11)


gk+1 (z| x ) = N (z; Hk+1 x, Rk+1 ) (12)


where f k+1|k (·xk ) is a transition function commonly known as Markov shift [30].N ( · ; m, P)


denotes a Gaussian density with mean m and covariance P, Fk is the state transition matrix,
Qk is the process noise covariance, gk+1 (z| x ) is likelihood function, Hk+1 is the observation
matrix, and Rk+1 is the observation noise covariance.
A multi-Bernoulli RFS is characterized by a posterior distribution with parameters
existence probability r (i) and probability density p(i) of the i-th hypothesized track, i =
n o Mk
(i ) (i )
1, . . . , Mk , i.e.,πk = rk , pk ( xk ) , which is comprised of Gaussian mixtures of the form
i =1
(i )
Jk−1
(i ) (i,j) (i,j) (i,j) (i,j) (i,j) (i,j)
pk ( xk ) = ∑ wk N ( xk ; mk , Pk ), where wk , mk , Pk denote the weights, means,
j =1
and covariances of the j-th Gaussian component by the sample
 time k.
The Bernoulli filter propagates the posterior πk = rk , pk ( xk ) during the whole
time in “prediction” and “update” steps. This effectively means that rk and pk must be
propagated.
Prediction: At time k + 1, spontaneous births are accounted for by appending a
n o MΓ,k+1
(i ) (i )
birth multi-Bernoulli RFS with components rΓ,k+1 , pΓ,k+1 to surviving targets. The
i =1
total number of predicted hypothesized tracks is Mk+1|k = Mk + MΓ,k+1 . The predicted
multi-target density is [21]:
n o MΓ,k+1
(i ) (i ) M (i ) (i )
πk+1|k = {(r P,k+1|k , p P,k+1|k )}i=k1 ∪ (rΓ,k+1 , pΓ,k+1 )
i =1

where:
(i )
JΓ,k+1

∑
(i ) (i,j) (i,j) (i,j)
pΓ,k+1 ( x ) = wΓ,k+1 N ( x; mΓ,k+1 , PΓ,k+1 ) (13)
j =1

(i ) (i )
r P,k+1|k = rk pS,k+1 (14)
(i )
Jk
∑ wk
(i ) (i,j) (i,j) (i,j)
p P,k+1|k ( x ) = N ( x; m P,k+1|k , PP,k+1|k ) (15)
j =1

(i,j) (i,j)
m P,k+1|k = Fk mk
(i,j) (i,j) T
PP,k+1|k = Qk + Fk Pk Fk

Update: In the following, based on nonlinear observations, we propose the unscented

transform implementation of the IC-UK-GM-CBMeMBer filter in the update step.
At time k + 1, the updated multi-Bernoulli density πk+1 is formed by multi-Bernoulli
(i ) (i ) M
RFS of the legacy tracks (r L,k+1 , p L,k+1 )}i=k1+1|k and measurement-corrected tracks (rU,k+1 (W ), pU,k+


as follows:
n o Mk +1 | k 
(i ) (i )
πk+1 = (r L,k+1 , p L,k+1 ) ∪ (rU,k+1 (z), pU,k+1 (·; z)) z∈Z
i =1 k +1
Sensors 2021, 21, 6196 8 of 18

where:
(i ) (i ) 1 − p D,k+1
r L,k+1 = rk+1|k (i )
(16)
1 − rk+1|k p D,k+1
(i ) (i )
p L,k+1 = pk+1|k ( x ) (17)
(i ) (i ) (i )
Mk+1|k rk+1|k (1−rk+1|k )ρU,k+1 (z)
∑ i =1 (i ) 2
(1−rk+1|k p D,k+1 )
rU,k+1 (z) = (i ) (i )
(18)
M r ρ (z)
κk+1 (z) + ∑i=k1+1|k k+1(|ik) U,k+1
1−rk+1|k p D,k+1

(i )
Mk +1 | k J
k +1| k (i,j) (i,j) (i,j)
∑ i =1 ∑ j=1 wU,k+1 (z) N ( x; mU,k+1 , PU,k+1 )
pU,k+1 ( x; z) = (i )
(19)
Mk +1 | k Jk+1|k (i,j)
∑ i =1 ∑ j =1 wU,k+1 (z)

Using unscented transform extends the mean matrix and covariance matrix, respectively,

µ k +1 = [ m k +1| k 00 00 ] 0 (20)

Ck+1 = diag( Pk+1|k , Qk+1 , Rk+1 ) (21)

(`) L
n o n oL
We constructed a set of 2nU + 1 sigma points χk and weights u(`) ,
`=0 `=0
L = 2nU , i.e.,

(`)
 χ k +1 = µ k +1

 u(`) = nUκ+UκU `=0
 (`) p 
χ k +1 = µ k +1 + (nU + κU )Ck+1 u(`) = 2(n 1+κ ) ` = 1, · · · , nU (22)
 p ` U U
 (`) (`) 1

 χ
k +1=µ k +1 − (n + κ )C
U u = U k +1 ` = n + 1, · · · , L 2( nU + κ U ) U
`

where nU is the dimension of µk , and κU is the scaling parameters,nU + κU 6= 0.

(i,j)
q k +1 ( z ) = N ( z k +1 ; z k +1| k , S K +1 ) (23)

(i )
Jk+1|k

∑
(i ) (i,j) (i,j)
ρU,k+1 (z) = p D,k w k +1| k q k +1 ( z ) (24)
j =1

(i,j) (i,j) (i,j)

mU,k+1 (z) = mk+1|k + KU,k+1 (zk+1 − zk+1|k ) (25)
(i,j) (i,j)
PU,k+1 = PU,k+1|k − Gk+1 Sk−+11 G 0 k+1 (26)
L
∑ u(`) zk+1|k
(`)
z k +1| k = (27)
`=0

KU,k+1 = Gk+1 Sk−+11 (28)

L
∑ u(`) (zk+1|k − zk+1|k )(zk+1|k − zk+1|k )0 + Rk
(`) (`)
S k +1 = (29)
`=0
L
∑ u(`) (χk+1|k − mk+1|k )(zk+1|k − zk+1|k )0
(`) (`)
Gk+1 = (30)
`=0
(`) (`)
zk+1|k = hk (χk+1 ), ` = 0, . . . , L (31)
Sensors 2021, 21, 6196 9 of 18

(i )
(i,j)
r k +1| k (i,j) (i,j)
wU,k+1 (z) = (i )
p D,k+1 wk+1|k qk+1 (z) (32)
1 − r k +1| k
(i )
With the i-th illuminator measurement data Zk+1 , the filter is obtained by the afore-
mentioned sequential processing of the measurement set of each illuminator nwith the CB-o
(i ) (i ) (i ) (i )
MeMBer filter corrector. The update operator Ψk+1 (i) from πk+1|k to πk+1 = rk+1 , pk+1
is [21,31]:
(i )
Mk+1|k rk+1|k h pk+1|k (i) ,gk+1 ( Zi | x ) p D,k+1 i
∑ i =1 (i )
h i 1−rk+1|k h pk+1|k (i) ,p D,k+1 i
Ψ k +1 ( i ) r = (i )
(33)
M r h pk+1|k (i) ,gk+1 ( Z(i) | x ) pD,k+1 i
κk+1 ( Z (i) ) + ∑i=k1+1|k k+1|k (i)
1−rk+1|k h pk+1|k (i) ,p D,k+1 i

(i )
Mk+1|k rk+1|k pk+1|k (i) gk+1 ( Z (i) | x ) p D,k+1
∑ i =1 (i )
h i 1−rk+1|k h pk+1|k (i) ,p D,k+1 i
Ψ k +1 ( i ) p = (i )
(34)
Mk+1|k rk+1|k h pk+1|k (i) ,gk+1 ( Z (i) | x ) p D,k+1 i
∑ i =1 (i )
1−rk+1|k h pk+1|k (i) ,p D,k+1 i
R
where h a, bi = χ a( x )b( x )dx denotes the inner product, and the sequential update process-
ing is as shown
(i )
rk+1 = Ψk+1 ( Ns) ◦ · · · ◦ Ψk+1 (2) ◦ Ψk+1 (1) rk+1|k (35)

pk+1 (xk+1 ) = Ψk+1 ( Ns) ◦ · · · ◦ Ψk+1 (2) ◦ Ψk+1 (1) pk+1|k (xk+1 ) (36)
where ◦ denotes a composition.

3.2. State Extraction and Cardinality Biass

Extract multi-target states are the same as that of the GM-MB filter; for more details
see [21]. The number of targets is estimated by:
Mk +1 | k ( i )
ek =
N ∑ i =1 r L,k+1 + ∑ rU,k+1 (z) (37)
z∈ Zk

For completeness, the key steps of the proposed filter are summarized as a block
diagram of the processing algorithm in Table 1.

Table 1. Pseudocode of the proposed filter.

J
0
Input: Initial Gaussian Mixtures {w0 ,m0 ,P0 }i=1 and Doppler Measurement Set Z(I)
1:k
for k = 1:Time duration
predict surviving Gaussian components
for i = 1:Mk
(i ) (i )
r P,k+1|k = rk pS,k+1
for j = 1:Jk
(i,j) (i,j) (i,j) (i,j) T
compute m P,k+1|k = Fk mk PP,k+1|k = Qk + Fk Pk Fk
end
end
construction of birth target Gaussian components using Equation (13)
end
update the legacy tracks
for i = 1:Mk+1|k
(i ) (i ) 1− p D,k+1 (i ) (i ) (i ) (i )
r L,k+1 = rk+1|k (i ) m L,k+1 = m L,k+1|k PL,k+1 = PL,k+1|k
1−rk+1|k p D,k+1
end
Sensors 2021, 21, 6196 10 of 18

Table 1. Cont.

J
0
Input: Initial Gaussian Mixtures {w0 ,m0 ,P0 }i=1 and Doppler Measurement Set Z(I)
1:k
update the measurement-corrected tracks
for freq = 1:Ns
f req
for i = 1:Mk
 for j =1: Mk+1|k  
( j) ( j)
m k +1| k Pk+1|k
( j)  ( j)
µ k +1 =   Ck+1 =  Qk+1 
  
0
0 R k +1
each component constructs a set of sigma points and weights using Equation (22) to generate:
n oL
(`)
χk , u(`)
`=0 "
→ → #
f ( f req) (`) → ( f req) (`) → ( f req)
 
(`) (`) (`)
zk+1|k = hk (χk+1 ), ` = 0, . . . , L h χk = c −vk p k + vk · vtk

 end
compute (rU,k+1 , pU,k+1 ) using Equations (18) and (19)
end
prune tracks
end
state extraction and cardinality bias using Equation (37)
end

4. Experimental Results
4.1. Experimental Configuration
We developed the PBR system in Huazhong University of Science and Technology
by tracking a close-in civilian airplane whose working frequency band is 6–30 MHz. The
system is configured to work in multi-transmitter and receiver-only mode. The experiment
was carried out in December 2014, in which three AM radio broadcast stations were
selected as the noncooperative transmitters, namely, Tx1, Tx2, and Tx3, respectively. The
specific parameters can be obtained from the International Telecommunication Union
(ITU) Radiocommunication Sector [32,33] listed in Table 2, including carrier frequency (fc ),
transmitting power, distance with respect to the receiver, and so on.

Table 2. Parameters of the three AM radio broadcast stations used as transmitters.

Station Serial Station Power Latitude Longitude Distance

fc (kHz)
Number Name (kW) (deg) (deg) (km)
Tx1 17770 UDO 250 17.25 102.48 1866
Tx2 15370 SZG 100 38.04 114.28 842
Tx3 15130 BEI 150 39.55 116.25 1027

The ground distance between the AM radio broadcast station and the receiver antenna
array is over 800 km. Thus, the transmitted signals are reflected by the ionosphere to
reach targets over-the-horizon away. Figure 4 shows the geographical distribution of the
illuminators and the receiver station.
Sensors 2021, 21, x FOR PEER REVIEW 12 of 19
Sensors 2021, 21, 6196 11 of 18
Sensors 2021, 21, x FOR PEER REVIEW 12 of 19

Figure 4. The geographical distribution of the illuminators and the receiver.

Figure
Figure 4.
4. The
The geographical distribution of the illuminators and the
the receiver.
Thegeographical distribution
noncooperative targetsofinthethe
illuminators andare
experiment receiver.
two civil aircrafts in the surveil-
lanceThearea with flight numbers CCAXXXX and CSNXXXX, respectively, namely, Target 1
The noncooperative
noncooperative targetstargets in
in the
the experiment
experimentare are two
two civil
civil aircrafts
aircrafts in in the
the surveil-
surveil-
and
lance Target 2. The civil aircrafts parameters were broadcast by the automatic dependent
lance area with flight numbers CCAXXXX and CSNXXXX, respectively, namely, Target 11
area with flight numbers CCAXXXX and CSNXXXX, respectively, namely, Target
surveillance-broadcast
and
and Target
Target 2. The civil (ADS-B)
The civil aircrafts system
aircrafts parameters
parameterswithinwere
werea short interval
broadcast
broadcast by of thetime. The data
automatic sets, in-
dependent
dependent
cluding position, velocity,
surveillance-broadcast
surveillance-broadcast and system
(ADS-B)
(ADS-B) sosystem
on, are the areference
within
within ashort
short to verify
interval
interval of of the
time. tracking
TheThe
time. method,
datadatasets,sets,
in-
which
cluding
includingis recorded
position,
position, by a ground-based
velocity,
velocity,and
and sosoon, AirNav
on,are
arethe Radar Box.
thereference The
referencetotoverify two
verifythe real trajectories
the tracking
tracking method, of the
method,
civil
which
which aircrafts
is during
is recorded
recorded bythe
by experiment are
a ground-based
ground-based plotted
AirNav
AirNav in Figure
Radar
Radar Box. 5a,
Thewhich
Box. The two
two real start
real represented
trajectories
trajectories of by
of the
the
a triangle
civil symbol
civil aircrafts
aircrafts during and
during the end represented
the experiment
experimentare by acircle
are plotted
plotted in symbol.
in Figure As
Figure 5a, the
5a, which recorded
which start data
start representedshow
represented by in
by
Figure 5a,
aa triangle the
triangle symbol two
symbol and targets
and end exist during
end represented
represented bythe whole
by acircle time in
acircle symbol. the
symbol. As surveillance.
As thethe recorded Target
recorded data 1
data showflew
show in at
in
(1) (1)
a constant
Figure 5a,
Figure altitude
5a, the of 8.4 km with
the two targets exist during a nearly
during the constant
the whole
wholetimetimeinspeed
inthe v = 106m/s
thesurveillance.
surveillance.
x
, v
Target = 238m/s
flewatata,
y 1 1flew
Target
1) (1) (1)
aand
constant
constant altitude
altitude
Target ofof
2 flew at 8.4km
a8.4 kmwith
constantwith a anearly
altitudenearlykmconstant
of 5constant speedv(xconstant
with aspeed
nearly vx ==106m/s,
106m/s
speed, vv(yv1x(2)
)
y = = 238m/s
−87m/s,,
238m/s,
(2) Target 2 flew at a constant altitude of 5 km with a nearly constant speed v (2)= −87m/s, (2)
vand
and Target at a constant altitude of 5 km with a nearly constant speed xvx = −and
y = −196m/s in the surveillance area. The height of the transmitting station
87m/s
the,
(2)
(2) = −196m/s in the surveillance area. The height of the transmitting station and the
vyy = −196m/s
v
receiver in the surveillance
are approximately area. The
ignored. Figure height
5b plots theofvariation
the transmitting station
of direction of theand the
targets
receiverthe
during areexperiment.
approximately ignored.
The samplingFigure 5b plots
interval was the
△ variation
= 1 s, ofthe
and direction
total of the targets
experimental
receiver are approximately ignored. Figure 5b plots the variation of direction of the targets
during the
duration wasexperiment.
80 s. The sampling interval was 4 = 1 s, and the total experimental
during the experiment.
duration was 80 s. The sampling interval was △ = 1 s, and the total experimental
duration was 80 s.

(a) (b)
(a) parameters broadcasted by ADS-B in the experiment. (a) The (b)
Figure 5. A part of civil aircraft real civil aircraft traces in x-y
Figure 5. A part of civil aircraft parameters broadcasted by ADS-B in the experiment. (a) The real civil aircraft traces in x-y
coordinate; (b) The direction of the civil aircrafts with respect to the receiver (based on the azimuth with respect to Eastern).
Figure 5. A (b)
coordinate; partThe
of civil aircraft
direction parameters
of the broadcasted
civil aircrafts by ADS-B
with respect to thein the experiment.
receiver (based on(a)
theThe real civil
azimuth withaircraft
respecttraces in x-y
to Eastern).
coordinate; (b) The direction of the civil aircrafts with respect to the receiver (based on the azimuth with respect to Eastern).
 Sensors 2021, 21, x FOR PEER REVIEW 13 of 19

Sensors 2021, 21, 6196 12 of 18

4.2. Field Experimental Results

The targets are observed in the surveillance region with dimensions
4.2.[−Field
40, 40]km × [ −40, 40]km
Experimental . The single-target transition model is a linear Gaussian process
Results
given by Equation (11), in
The targets are observed which ∆ =surveillance
in the 1 s is the sampling period,
region with and σ v = 0.1m/s
dimensions
2
[−40, 40is]km
the ×
[−standard
40, 40]km.deviation of the process
The single-target noise.
transition The birth
model process
is a linear is multiBernoulli
Gaussian process given with
by density
Equation
(i ) ∆ (i ) (i ) 2
(11), in which 2 = 1 s is the sampling
π Γ = {rΓ , pΓ }i =1 , where rΓ = 0.01 , Γ
(i ) (i )
p
period, ( x ) = N
and (
σ x; m
= ,
0.1m/sP ) is
(1) the standard deviation ofT the
Γ = [−20,000,0, 20,000,0]
v Γ Γ , m ,
process noise. The birth process is T multiBernoulli with density
(2)n T PΓ = diag ([1000,10,1000,10] ) 2
m = [20,000,
(i) (i) 0, −20,000,0](i),
o2 . The(1)probability of target sur-
(i ) (i )
πΓ =Γ rΓ , pΓ , where rΓ = 0.01,pΓ (x) = N (x; mΓ , PΓ ),mΓ = [−20, 000, 0, 20, 000, 0]T ,
i=1
pS ,k = 0.95 p = 0.5
(2)
mΓvival = is
[20, 000, 0, −20, . The
000, 0]T , PΓ =ofdiag
probability target detection
([1000, 10, 1000, is 10D]T,k )2 . The . probability of target
survival is pS,k = 0.95. The probability of target detection is pD,k = 0.5. estimation of each
After the aforementioned signal processing, we obtained the DOA
direct
Afterwavethe and the Doppler signal
aforementioned shift measurement
processing, data, we obtained includingthe theDOA
false estimation
alarms andof
misdetections. Figure 6 shows the detected Doppler vs.
each direct wave and the Doppler shift measurement data, including the false alarms andtime obtained from the surveillance
areas using three
misdetections. AM6broadcast
Figure shows thestations
detected with the carrier
Doppler vs. frequency
time obtained of 17.7 MHz,
from the15.37 MHz,
surveillance
and 15.5 MHz, respectively. The Doppler measurement sets have clutter and the missing
areas using three AM broadcast stations with the carrier frequency of 17.7I MHz, 15.37 MHz,
( )
and 15.5 Then,
alarm. MHz, the respectively.
noisy threeThe Doppler
stations measurement
Doppler-shift measurement sets havesets Z1:80
clutter and
, arethe missing
passed to
(I)
alarm. Then, the
the tracking noisy
filter, three stations
as plotted in FigureDoppler-shift
7. The parametersmeasurement
of the tracking Z1:80 are
sets filter , aresetpassed
as fol-to
the tracking filter, as plotted in Figure R = σ ε I1 , where
7. 2
The parameters of the tracking filter are set as
lows: observation noise covariance k σ = 1Hz is the standard deviation of
follows: observation noise covariance Rk = σε2 I1 , whereε σε = 1Hz is the standard deviation
κ k ( z ) =κλc(Vu ( z ) Vu(z),
of the
themeasurement
measurement noise.
noise. Clutter
Clutter parameter
parameter is Poisson
is Poisson with intensity
with intensity k z ) = ,λwhere
c
u( z )u(isz)aisuniform
where a uniform probability density over the surveillance region, V
probability density over the surveillance region, V = 1600 km is the
= 16002 km2 is
(i )
the “volume” of the surveillance region, and the clutter intensity isλ (λi ) c= 0.3
= 0.3; i = 1,2,3 for
“volume” of the surveillance region, and the clutter intensity is c ; i = 1,2,3 for
z ∈ Z = [−30Hz, 30Hz] are all assumed time invariant and independent of the target state.
z ∈ Z = [ −30Hz, 30Hz] are all assumed time invariant and independent of the target state.

(a) (b)

(c)

Figure 6. The results of target detection and Doppler shift measurements vs. time obtained from the surveillance areas
using three AM broadcast stations. (a) Station 1 with a carrier frequency of 17.77 MHz; (b) Station 2 with a carrier frequency
of 15.37 MHz; (c) Station 3 with a carrier frequency of 15.13 MHz.
 Sensors 2021, 21, x FOR PEER REVIEW 14 of 19

Sensors 2021, 21, x FOR PEER REVIEW 14 of 19

Figure 6. The results of target detection and Doppler shift measurements vs. time obtained from the surveillance areas
Sensors using
2021, 21, 6196AM broadcast stations. (a) Station 1 with a carrier frequency of 17.77 MHz ;(b) Station 2 with a carrier frequency
three 13 of 18
Figure 6. The results of target detection and Doppler shift measurements vs. time obtained from the surveillance areas
of 15.37
using MHz;
three AM(c) Station 3stations.
broadcast with a carrier frequency
(a) Station 1 with aofcarrier
15.13 MHz.
frequency of 17.77 MHz ;(b) Station 2 with a carrier frequency
of 15.37 MHz; (c) Station 3 with a carrier frequency of 15.13 MHz.

Figure 7. Doppler shift measurement set in which each color triangle stems from an AM broadcast
transmitter.
Figure 7. Doppler
Doppler shift
shift measurement
measurement set
set in
in which
which each
each color
color triangle
triangle stems
stems from
from an
an AM
AM broadcast
broadcast
Figure 7.
transmitter.
transmitter.
The initial density of the target state 0
p ( x)
is the Gaussian mixture of the form:
The 4
initial density of the target state p
p (
( x
x )
) is
isthe
the Gaussian
Gaussian mixture
mixture of
of the
the form:
)=
0
0
p0 ( xThe initial density of the target state
4ω0 N ( x; mn , Q0 )
form:
4
pp0 ((xx)) == ∑
ω = 1 / 4

= 0
Qn), Q,where is the is weight factorfactor
of each
of initial state vector,
n 1 ω0( xN; (mx;,m
ωN 0 ),where ω0 = 1/4 the weight each initial state
0
Q0 = diag n=10 3
n =1 ([(10 m) , (10m
2n 0 2 ω 3 = )12 /, (10
2/ s ) , (100 m
4 m / s) 2 ])
vector, Q0 = diag ([( 10 3 m ),where
, ( 10m/s ) 2
, ( 10 3 mis 2the weight
) , (10m/s is)2 ])factor
the
is the
of eachcovariance
initial initial state vector,
matrix,
initial covariance matrix,
3 2 2 3 2 2
Qm = diag
= [ p ([(10
,0 m m/ )
s , ,p(10 m
,0 /
m s/) s ,
] T(10
T m ) , (10 m / s ) ])
, n = 1, 2,  , 4 are the initial mean state vectors, and
m 0 = [ p , 0m/s, p , 0m/s] ,n = 1, 2, . . . , 4 are isthethe initial
initial meancovariance
state vectors,matrix,
and
n,0n ,0 xn ,0
xn,0 ,0
ynyn,0
T
(m p xn,0 =
, [
p p ) ,0 m
means/ s , p
four ,0 m /
initials ] positions
= 1, 2,uniformly
 , 4 aredistributed in the surveillance area. In
( np,0 ,0 ,pxnyn,0,0 ) , nparameters
yn,0 yn ,0
GM xnimplementations, some of the the initial mean
used indistributed
the filtering state vectors, and
means four initial positions uniformly inare:
the the target
surveillance existence
area. In
(GMpxn ,0implementations,
threshold ,p ynr,0k )= 0.001; atsome
means foureach time
of
initial the step, Gaussian
parameters
positions used
uniformly components are
in pruned
in the filtering
distributed theand
are:surveillance
the merged
target
−10 merging threshold U = 40, maxi-
area.for
existence
In
each
GMthreshold
r
hypothesized
implementations,
k = 0.001 track somewith weight
oftime threshold
the parameters T = 10
usedcomponents
in the filtering are: the target existence
mum components Jmax ; at = each
200. In step, Gaussian
addition, hypothesized tracks areare pruned
pruned and maximum
with merged for
each r
hypothesized
k = 0.001 track with weight threshold T = 10 −10 merging threshold U = 40, maximum
Tthreshold
max = 5 and weight; threshold at each time 10−5Gaussian
L =step, . Tracking components
results of the are IC-UT-GM-CBMeMBer
pruned and merged for
components
each hypothesized Jmax = 200. In addition, hypothesized tracks
track with weight
filter are shown in Figures 8 and 9, which show the estimated targetthreshold T = 10 are
−10 merging pruned with
threshold maximum
tracesU = and Tmax
40, maximum
four =5
esti-
and
components
mated weight
components threshold
Jmax = 200. L
of the = 10 −5. Tracking results of the IC-UT-GM-CBMeMBer filter are shown
In addition,
state vector: hypothesized
px , py , vx , vtracks arevspruned with maximum Tmaxtrue
=5
y change time compared with the
in Figures
and weight8threshold
trajectories, and 9, which
respectively. L = 10 show
At endingthe estimated
−5. Tracking
time results target traces
of theshort
instants, and four estimated
IC-UT-GM-CBMeMBer
discontinuities occur components
filter tracksof
are shown
in the
inthe
owing state
Figures vector:
to the8 and pxwhich
9,
missing , pyalarm
, vx, show
vy of
change vs timemeasurement.
theDoppler
the estimated compared
target traces with thefour
and
Notice true trajectories,
estimated
that the number respectively.
components
of targetsof
At ending
the statefrom
suffers time
vector: instants,
px, pyproblem
latency short
, vx, vy change discontinuities
at thevsbeginning
time comparedoccur in
of tracking the tracks
with the owing to
true trajectories,
in Figure the
10, because missing alarm
respectively.
the initial
Atof ending
points the are
Dopplertime measurement.
located instants,
arbitrarily.short Notice that the number
discontinuities occur in of thetargets
trackssuffers
owing from
to thelatency
missing problem
alarm
at the beginning of tracking in Figure 10, because the
of the Doppler measurement. Notice that the number of targets suffers from latency problem initial points are located arbitrarily.
at the beginning of tracking in Figure 10, because the initial points are located arbitrarily.

Figure8.8. The
Figure The estimated
estimated target
targetpath
pathcompared
comparedwith
withthe
thetrue path
true in in
path x-yx-y
coordinate using
coordinate 3 AM
using illu-
3 AM
minators.
illuminators.
Figure 8. The estimated target path compared with the true path in x-y coordinate using 3 AM illu-
minators.
Sensors
Sensors 2021, 21, 6196
2021, 21, x FOR PEER REVIEW 1514of
of 19
18
Sensors 2021, 21, x FOR PEER REVIEW 15 of 19

(a)(a) (b)
(b)

(c) (d)
(c) (d)
Figure
Figure 9. The
9. The estimated
estimated components
components ofofthe
thestate
statevector
vector(dotted
(dotted line)
line) changes
changes as
as functions
functionsofoftime
timecompared
comparedwith
withthethe
true
true
Figure 9.
values The estimated
(solid line). components
(a)The p of the
x component; state
(b) The vector
p (dotted
y component; line)
(c)Thechanges
v as functions
x component; (d) of
The
values (solid line). (a)The px component; (b) The py component; (c)The vx component; (d) The vy component.time
v compared
y component. with the true
values (solid line). (a)The px component; (b) The py component; (c)The vx component; (d) The vy component.

Figure 10. Estimated number of targets for the proposed method.

Figure
Figure 10.
10. Estimated
Estimated number
number of
of targets
targets for
for the
the proposed
proposed method.
method.
Sensors 2021, 21, x FOR PEER REVIEW 16 of 19
Sensors 2021, 21, 6196 15 of 18

Different numbers of illuminators are a problem in multi-target tracking perfor-

Different numbers of illuminators are a problem in multi-target tracking performance
mance when only Doppler measurements are used. To study this, the filter is imple-
when only Doppler measurements are used. To study this, the filter is implemented under
mented under the conditions of AM radio broadcast stations Ns = 2,1, corresponding to
the conditions of AM radio broadcast stations Ns = 2,1, corresponding to stations with
stations with serial numbers of I = {1, 2}, {1}, respectively. The optimal subpattern assign-
serial numbers of I = {1, 2}, {1}, respectively. The optimal subpattern assignment (OSPA)
ment (OSPA) is used to evaluate the tracking miss-distance. The OSPA distances (for c =
is used to evaluate the tracking miss-distance. The OSPA distances (for c = 20 and p = 1)
20 and p = 1) vs. time on conditions of various number of broadcast stations compared
vs. time on conditions of various number of broadcast stations compared with the results
with the results on the condition of Ns = 3, I = {1, 2, 3} is shown in Figure 11. Particularly,
on the condition of Ns = 3, I = {1, 2, 3} is shown in Figure 11. Particularly, the OSPA
the OSPA distances vs. the time between 6s and 32s is plotted. It can be seen that the
distances vs. the time between 6s and 32s is plotted. It can be seen that the estimated
estimated largest OSPA distances are approximately 2880 m, 2922 m, and 5712 m on the
largest OSPA distances are approximately 2880 m, 2922 m, and 5712 m on the condition
condition of Ns = 3, Ns = 2, and Ns = 1, respectively. The obvious error in the period time
of Ns = 3, Ns = 2, and Ns = 1, respectively. The obvious error in the period time from
from k = 27 to 56 is due to the missing detections and clutter of the Doppler measurements
k = 27 to 56 is due to the missing detections and clutter of the Doppler measurements and
and the number
the number of illuminators.
of illuminators. Therefore,
Therefore, we believe
we believe thatmore
that the the more numerous
numerous the AMtheradio
AM
radio
broadcast stations that are exploited, the more accurate the tracking trajectories are. are.
broadcast stations that are exploited, the more accurate the tracking trajectories

Figure
Figure 11.
11. The
The OSPA
OSPA distances vs. time
distances vs. time on
on conditions
conditions of
of different station numbers.
different station numbers.

Simulation Results
4.3. Simulation Results
this subsection,
In this subsection, the performance of the proposed method is verified via simulation
under similar scenarios
scenarios totothose
thoseaforementioned
aforementionedunder
underthe thesituation
situationofofcross
crosstrajectories,
trajectories,in
consideration
in consideration that it usually
that occurs
it usually in real
occurs data
in real processing
data processing on aon2Da Cartesian
2D Cartesian coordinate.
coordi-
ThreeThree
nate. AM broadcast stations
AM broadcast were were
stations chosen, the same
chosen, as Table
the same 2. The2.two
as Table Thetargets’ motion
two targets’ mo-is
assumed to be a nearly constant model adjusted for civil aircrafts, and the flight
tion is assumed to be a nearly constant model adjusted for civil aircrafts, and the flight parameters
are listed inare
parameters Table 3. The
listed false 3.
in Table alarms are uniformly
The false alarms are distributed in the field in
uniformly distributed ofthe
view with
field of
view −30range
rangewith Hz to−3030 Hz,
Hz toand30the
Hz,number
and theof false alarms
number of falseatalarms
each scan follows
at each scanthe Poisson
follows the
distribution
Poisson with a mean
distribution withofa 10.
meanTheofparameters of the tracking
10. The parameters of thefilter are set
tracking the are
filter same setasthe
in
Section 4.2.
same as in Section 4.2.
Table 3. The flight parameters in simulation.
Table 3. The flight parameters in simulation.
Initial
Initial Position InitialInitial
Position Velocity
Velocity Time of Death
Item Time of Birth (s)
Item (km) (m/s) Time of Birth (s) Time of Death
(s) (s)
(km) (m/s)
Target 1 (−5,20) (95, −200) 1
Target 1 (−5,20) (95, −200) 1 8080
Target 2 (5,12) (−150, −120) 1 90
Target 2 (5,12) (−150, −120) 1 90

As shown
As shown inin Figure
Figure 12,
12, despite
despite the
the intersection points, the
intersection points, the two
two targets
targets can
can follow
follow
their trajectories, respectively. In Figure 13, the OSPA metric (p = 1, c = 20) shows the
their trajectories, respectively. In Figure 13, the OSPA metric (p = 1, c = 20) shows the track
 Sensors
Sensors 2021,
2021, 21, 21, x FOR
x FOR PEER
PEER REVIEW
REVIEW 17 17 of 19
of 19
Sensors 2021, 21, 6196 16 of 18

maintenance
maintenance quality
quality of the
of the proposed
proposed method.
method. However,
However, thethe instantaneous
instantaneous peaks
peaks areare
ob-ob-
track
servedmaintenance
from times quality of the proposed method. However, the instantaneous peaks are
served from times k =k59
= 59
to kto=k81
= 81
due due to corresponding
to corresponding intersection
intersection point
point andand track
track termi-
termi-
observed
nation from
latency. times k = 59 to k = 81 due to corresponding intersection point and track
nation latency. TheThe simulation
simulation results
results indicate
indicate that
that thethe proposed
proposed method
method cancan
dealdeal with
with
termination
relatively latency.tracking
complex The simulation
problems. results indicate that the proposed method can deal
relatively complex tracking problems.
with relatively complex tracking problems.

Figure
Figure
Figure 12.
12.12.
The The
The estimated
estimated
estimated target
target path
path
target compared
compared
path with
with
compared the
thethe
with true
true path
path
true inx-y
in in
path x-yx-y coordinate.
coordinate.
coordinate.

Figure
Figure
Figure 13.
13.13.
The The
The OSPA
OSPA
OSPA distances
distances
distances vs.
vs.vs. time
time
time under
under
under the
thethe situation
situation
situation ofcross
of of cross
cross trajectories.
trajectories.
trajectories.

5.
5. Conclusions
Conclusions
5. Conclusions
In this paper, we propose a multi-target tracking filter in ainself-developed PBRPBR
system
In Inthisthis paper,
paper, wewe propose
propose a multi-target
a multi-target tracking
tracking filter
filter a self-developed
in a self-developed PBR sys- sys-
by using spatially distributed multiple AM broadcast stations. Multiple non-cooperative
temtemby by using
using spatially
spatially distributed
distributed multiple
multiple AM AM broadcast
broadcast stations.
stations. Multiple
Multiple non-cooper-
non-cooper-
illuminators
ative with different
illuminators with carriercarrier
different frequencies located
frequencies over-the
located horizon
over-the and one
horizon receiver
ative illuminators with different carrier frequencies located over-the horizon andand
oneone
re-re-
in the surveillance area are involved in the practical system. The direct wave and the
ceiver
ceiver in the
in the surveillance
surveillance area
area areare involved
involved in the
in the practical
practical system.
system. TheThe direct
direct wave
wave and and
illumination wave are reflected from the ionosphere received by a uniform circular array
the illumination wave are reflected from the ionosphere received
the illumination wave are reflected from the ionosphere received by a uniform circular by a uniform circular
located over the horizon, while the echo wave (target-to-receiver) is via LOS propagation
in the surveillance area. After some techniques in array signal processing, the Doppler
Sensors 2021, 21, 6196 17 of 18

measurement sets, including clutter and the missing alarm with corresponding time, can
be collected. To overcome linear Gaussian models, we propose the tracking model and
extend the GM-CBMeMBer filter to a practical nonlinear measurement model by using
unscented transform (UT) techniques by iterating the filter update stage for each illuminator
measurement set in this practical scenario. Three AM broadcast stations were selected
as the non-cooperative illuminators. Two non-cooperative civil aircrafts were chosen
as tracking targets, whose flight parameters were recorded by a ground-based AirNav
Radar Box set. Considering the clutter and missing alarm in the measurement sets, the
OSPA distances are acceptable. Moreover, the performance of simulation has verified the
feasibility of the proposed tracking method. In future work, the unknown clutter rate and
detection probability under unknown background in this practical scenario will be taken
into consideration. Maneuvering target tracking is also worthy of study.

Author Contributions: Conceptualization, X.Z. and H.M.; methodology, X.Z. and H.M.; Software,
X.Z.; Validation, X.Z. and H.X.; Formal analysis, X.Z. and H.X.; Investigation, X.Z.; Resources, X.Z.
and H.X.; Data curation, X.Z. and H.X.; Writing—original draft preparation, X.Z.; Writing—review
and editing, X.Z. and H.M.; Visualization, X.Z.; Supervision, X.Z. and H.M.; Project administration,
H.M.; Funding acquisition, H.M. All authors have read and agreed to the published version of the
manuscript.
Funding: This research was funded by the National Natural Science Foundations of China, grant
numbers 62071188; Equipment Pre-Research Foundation of China, grant numbers 6141A02022364.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Griffiths, H.D.; Long, N.R.W. Television based bistatic radar. IEEE Proc. F Radar Signal Process. 1986, 133, 649–657. [CrossRef]
2. Howland, P.E. Target tracking using television-based bistatic radar. IEE Proc. Radar Sonar Navig. 1999, 146, 166–174. [CrossRef]
3. Saini, R.; Cherniakov, M. DTV signal ambiguity function analysis for radar application. IEE Proc. Radar Sonar Navig. 2005, 152,
133–142. [CrossRef]
4. Saini, R.; Cherniakov, M. Investigation of digital TV-terrestrial signal for radar application. In Proceedings of the 2002 11th
European Signal Processing Conference, Toulouse, France, 3–6 September 2002; pp. 1–4.
5. Howland, P.E.; Maksimiuk, D.; Reitsma, G. FM radio based bistatic radar. IEE Proc. Radar Sonar Navig. 2005, 152, 107–115.
[CrossRef]
6. Harms, H.A.; Davis, L.M.; Palmer, J. Understanding the signal structure in DVB-T signals for passive radar detection. In
Proceedings of the 2010 IEEE Radar Conference, Arlington, VA, USA, 24 June 2010; pp. 532–537.
7. Li, X.; Zhao, C.; Lu, X.; Wei, W. DA-PMHT for Multistatic Passive Radar Multitarget Tracking in Dense Clutter Environment.
IEEE Access 2019, 7, 49316–49326. [CrossRef]
8. Poullin, D.; Palmer, J. Passive detection using digital broadcasters (DAB, DVB) with COFDM modulation. IEE Proc. Radar Sonar
Navig. 2005, 152, 143–152. [CrossRef]
9. Daun, M.; Nickel, U. Tracking in multistatic passive radar systems using DAB/DVB-T illumination. Signal Process. 2012, 92,
1365–1386. [CrossRef]
10. Min, L.; Jianxin, Y. Target tracking in time-division-multifrequency-based passive radar. IEEE Sens. J. 2020, 20, 4382–4394.
11. Thomas, J.M.; Griffiths, H.D.; Baker, C.J. Ambiguity function analysis of digital radio mondiale signals for HF passive bistatic
radar. Electron. Lett. 2006, 42, 1482–1483. [CrossRef]
12. Thomas, J.M.; Baker, C.J.; Griffiths, H.D. DRM signals for HF passive bistatic radar. In Proceedings of the 2007 IET International
Conference on Radar Systems, Edinburgh, UK, 15–18 October 2007; pp. 1–5.
13. Fabrizio, G.; Colone, F.; Lombardo, P.; Farina, A. Adaptive beamforming for high-frequency over-the-horizon passive radar. IET
Radar Sonar Navig. 2009, 3, 384–405. [CrossRef]
14. Willis, N.J.; Griffiths, H. Advances in Bistatic Radar; SciTech Publishing: Raleigh, NC, USA, 2007.
15. Yu, D.; Ma, H.; Cheng, L.; Li, Y.; Qin, Y.; Yin, S.; Chen, W. Method and field experiment of target tracking via multistatic Doppler
shifts in high-frequency passive radar. IET Radar Sonar Navig. 2016, 10, 1201–1212. [CrossRef]
16. Malling, L.R. Radio Doppler effect for aircraft speed measurements. Proc. IRE 1947, 35, 1357–1360. [CrossRef]
Sensors 2021, 21, 6196 18 of 18

17. Fried, W.R. Principles and performance analysis of Doppler navigation systems. IRE Trans. Aeronaut. Navig. Electron. 1957, 4,
176–196. [CrossRef]
18. Salinger, S.N.; Brandstatter, J.J. Application of recursive estimation and Kalman filtering to Doppler tracking. IEEE Trans. Aerosp.
Electron. Syst. 1970, 4, 585–592. [CrossRef]
19. Do, C.-T.; Van Nguyen, H. Tracking Multiple Targets from Multistatic Doppler Radar with Unknown Probability of Detection.
Sensors 2019, 19, 1672. [CrossRef] [PubMed]
20. Demars, C.D.; Roggemann, M.C.; Webb, A.J.; Havens, T.C. Target Localization and Tracking by Fusing Doppler Differentials from
Cellular Emanations with a Multi-Spectral Video Tracker. Sensors 2018, 18, 3687. [CrossRef]
21. Vo, B.-T.; Vo, B.-N.; Cantoni, A. The cardinality balanced multitarget multi-Bernoulli filter and its implementations. IEEE Trans.
Signal Process. 2009, 57, 409–423.
22. Sun, R.; Zhang, W.; Yao, B. Frame Arrival Detection for Low SNR Frequency Selective Fading Channels. In Proceedings of the
2018 IEEE/CIC International Conference on Communications in China (ICCC), Beijing, China, 16–18 August 2018; pp. 374–378.
23. Kim, B.-S.; Jin, Y.; Lee, J.; Kim, S. Low-Complexity MUSIC-Based Direction-of-Arrival Detection Algorithm for Frequency-
Modulated Continuous-Wave. Vital Radar 2020, 20, 4295. [CrossRef]
24. Li, Y.; Ma, H.; Wu, Y.; Cheng, L.; Yu, D. DOA Estimation for Echo Signals and Experimental Results in the AM Radio-Based
Passive Radar. IEEE Access 2018, 6, 73316–73327. [CrossRef]
25. Vo, B.-N.; Singh, S.; Doucet, A. Sequential Monte Carlo methods or multi-target filtering with random finite sets. IEEE Trans.
Aerosp. Electron. Syst. 2005, 41, 1224–1245.
26. Mahler, R. Multi-target Bayes filtering via first-order multi-target moments. IEEE Trans. Aerosp. Electron. Syst. 2003, 39, 1152–1178.
[CrossRef]
27. Mahler, R. Statistical Multisource-Multitarget Information Fusion; Artech House: Norwood, MA, USA, 2007.
28. Guldogan, M.B.; Lindgren, D.; Gustafsson, F.; Habberstad, H.; Orguner, U. Multi-target tracking with PHD filter using Doppler-
only measurements. Digit. Signal Process. 2014, 27, 1–11. [CrossRef]
29. Simon, J.; Uhlmann, J. Unscented filtering and nonlinear estimation. Proc. IEEE 2004, 92, 401–422.
30. Vo, B.-N.; Ma, W.K. The Gaussian mixture probability hypothesis density filter. IEEE Trans. Signal Process. 2006, 54, 11. [CrossRef]
31. Vo, B.T.; See, C.M.; Ma, N. Multi-sensor joint detection and tracking with the Bernoulli filter. IEEE Trans. Aerosp. Electron. Syst.
2012, 48, 1385–1402. [CrossRef]
32. The Comprehensive Shortwave Broadcasting Schedule. Available online: http://www.eibispace.de/ (accessed on 15 October
2014).
33. The Short Wave Schedule. Available online: http://www.shortwaveschedule.com/ (accessed on 15 October 2014).