PLOS ONE

RESEARCH ARTICLE

Sensing spectrum sharing based massive

MIMO radar for drone tracking and
interception
Milembolo Miantezila Junior ID*, Bin Guo

School of Electronics and Information Engineering, Changchun University of Science and Technology,
Changchun, Jilin, 130022, China

* 2018300111@mails.cust.edu.cn

a1111111111
a1111111111
Abstract
a1111111111 Radar sensors are becoming crucial for environmental perception in a world with the tre-
a1111111111
a1111111111 mendous growth of unmanned aerial vehicles (UAVs) or drones. When public safety is a
concern, the localization of drones are of great significance. However, a drone used for a
wrong motive can cause a serious problem for the environment and public safety, given the
fact that the dynamic movement of a drone’s emission signal and location tracking is differ-
ent from existing positioning. This study proposes a safety zone characterized by the pres-
OPEN ACCESS
ence of N radars sensors with a goal to track and destabilized rogue drones attending to
Citation: Junior MM, Guo B (2022) Sensing
penetrate safety zones (stadium and school). Specifically, a new joint estimation based on a
spectrum sharing based massive MIMO radar for
drone tracking and interception. PLoS ONE 17(5): Gaussian filter has been introduced for spectrum sharing and detection awareness. The
e0268834. https://doi.org/10.1371/journal. profit of this novel sensing method can be clearly seen when the two joint hidden states are
pone.0268834 taken into consideration. Therefore, the drone’s emission state is analyzed by estimating its
Editor: Anandakumar Haldorai, Sri Eshwar College movement jointly. Considering the drone’s unknown states and actual positioning, an algo-
of Engineering, INDIA rithm is developed based on dynamic states space model. Where Bernoulli filter model is
Received: November 10, 2021 designed to estimate recursively the unknown stages of the drone and its changing location
Accepted: May 6, 2022 based on time. Meanwhile a power control acted from the radar to the targeted drones so
that rogue drones are optimally tracked and destabilized over time. Furthermore, an
Published: May 20, 2022
expanding mechanism has been generated to accurately track the drone and enhance
Copyright: © 2022 Junior, Guo. This is an open
detection. A thoughtful result of the experimentation shows clearly that, even when the
access article distributed under the terms of the
Creative Commons Attribution License, which drone is moving, spectral detection can be performed accurately by chasing its positions. Its
permits unrestricted use, distribution, and demonstrates at 90% of credibility that the original signal has a direct effect on the propa-
reproduction in any medium, provided the original gated signal. Therefore, the magnitude of the Doppler shift increases with frequency. And
author and source are credited.
the clue of its positioning can be used for cognitive radio optimization.
Data Availability Statement: Data are available at
https://doi.org/10.6084/m9.figshare.19152635.

Funding: This work was supported by the Natural

Science Foundation of Jilin Province- China under
Grant No.201215133, and the Sci-tech
Development Project of Jilin Province of China 1. Introduction
under Grant No. 20130521015JH. The utilization of drones has increased exponentially with rising technology in recent years.
Competing interests: We have no conflicts of However, unmanned aerial vehicles (UAVs) or drones can become extremely dangerous for
interest to disclose. public safety and people’s privacy, when others applications and tools are added unto it such as

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 1 / 18

PLOS ONE Sensing spectrum sharing based massive MIMO radar for drone tracking and interception

urveillance, agriculture data analysis, movie making, mineral exploration without control and
good monitoring [1]. Therefore, careful studies in UAVs and spectrum access are of greater
importance. A deep analysis of anti-drone systems has been presented in [2,3], where a sce-
nario of drones cooperating in order to track and destabilize rogue drones. The target (rogue
drones) displays a stochastic dynamic movement and their trajectory overtime needs to be
estimated from noisy sensor measurements. It also specified that the mobile agents show a lim-
ited sensing range, and that they can detect the presence of the rogue drones inside their sens-
ing range with a probability of less than one. Consequently, due to the sensing limitation, it is
well-noted that, in addition to the target measurements, the mobile agents receive false-alarm
measurements as well. Recently, there has been a significant research-based on target detection
and the use of radar micro-Doppler [4–11].
In [12–14] studies, an analysis based on TVDs (Time Velocity Diagrams) of small helicop-
ters and multicopters, reveals that both are from simulations and measurements (X-band
radars). The authors presented the properties of a single rotor and multiple rotors, with an
even and odd number of blades, and with short and long integration time. Even though the
system performs well, the on-ground and simulation tests are limited due to the lack of supple-
mentary parameters such as the change of weather or environmental changes. In [15], the
authors presented a Doppler spectrum access without time resolution. The Doppler spectrum
is then used by a boosting classifier. The simulation has been executed at where the radar sig-
nal was generated from a moving helicopter. Its inefficiency lies on the fact that radar signal
and target (drones) are moving at the same time and this will cause some detections problem.
A useful ultra-wideband (UWB) Antenna for UAV applications has been proposed in [16,17],
where antennas with a reflector are used to increase the gain at 2.4 GHz for UAV utilization
and a monopole antenna that operates at 800 MHz was analyzed. In [18], a Pedestrian Dead
Reckon (PDR) structure based on Inertial Navigation System (INS) sensor and UWB system
was analyzed, where a modified zero-velocity detection algorithm and Kalman-type filter was
developed to get the best angle by coupling zero-velocity information and single UWB. In
[19], a mapping antenna array was presented with a circular polarization at the frequency
range of 1.5Ghz to 1.65Ghz.A low-profile antenna structure was proposed in [20], where a
Rogers Duroid 5880LZ material with dimensions of 29mm X 39mm was applied.
Experimental research based on multistatic passive radar with a single antenna for drone
detection has been presented in [21], where the dominant direct-path signal (the strongest
static clutter) in the reference channel was considered as an effective signal [22–26] and a sce-
nario was proposed to utilize a compact single antenna receiver for the UAV detection. In this
paper, we addressed a spectrum access for massive multi-input multi-output (MIMO) radar
covering a safety zone for rogue drones intrusion. Here, MIMO radar and UAV (drone) share
interferences on the same spectral zone. Our main goal here is to detect and track until the
destabilization of the target (drones). The contribution of this work is summed up as follows:
1. The localization tracking of the drone with unknown emission states has been investigated
and a new sensing technique has been proposed to estimate the localization of the drone
and at the same time detect its spectral position. Specifically, in this new technique the
tracking of the drone cannot be interrupted even though its emission states are suspended.
Therefore, this technique came to break the traditional method which doesn’t consider the
dynamic emission states of the drone.
2. Novel algorithm based on massive MIMO radar and drone spectrum sensing which rely on
dynamic Bayesian filter approach. The particularity of this approach is that, the unknown
emission state of the drone is analyzed as an additional hidden state that needs to be ana-
lyzed, instead of its changing locations. Considering the limited information available in

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 2 / 18

PLOS ONE Sensing spectrum sharing based massive MIMO radar for drone tracking and interception

spectrum access, we used received signal strength to estimate recursively the two hidden
states. Meanwhile, this technique can also be extended in other scenario such as spectrum
access between MIMO radar, 5G Communication system, and drone localization.
3. A Soft joint distribution algorithm has been developed, where the emission state of the
drone and other associated state such as its unknown positions, are analyzed like Bernoulli
random finite set (BRFS). We took the advantage of Bayesian assumption algorithm to esti-
mate recursively drone’s existence state and its dynamic positioning,which most of the
times in real live are difficult to analyze. To enhance the tracking scenario of the drone
when its goes off, a horizon analysis was developed, which can be adjusted prior to uncer-
tainty inference process. It is demonstrated by an extensive numerical experimentation
that, the system is not only efficient but also in estimating the drone’s location and detec-
tion, uncertainty of reception can be measured, therefore, the system can constantly be
optimized.
The rest of the paper is organized as follows: the sensing methodology, dynamical states,
dynamical positioning, and statistical detection of the drone is presented in section 2. In Sec-
tion 3, we presented numerical experiment and performance analyses. We discussed the per-
formance of the sytem and Algorithm scenarios in section 4. Section 5 is the conclusion of the
paper.

2. Materials and methods

2.1. System model
By considering simultaneous observation from spectrum sensing of MIMO radar and drone
localization, we addressed a cooperative scenario as presented in Fig 1. Our drone system is
moving as Brownian models continuous motion [27]. For a better analysis and representation,
we denote M, a cooperative MIMO radar in cartesian coordinate, with position of each node
noted by am = [xm, ym]t(m = 1,2, 3,. . .M).
We considered this information to be previously known by the data center. To perform
spectrum sensing and drone localization at the same time, a two-step scenario scheme were
selected. In the first step, the mth MIMO radar antenna will intercept the nearest wireless net-
work at each time discrete t, and receive the information about the local observation ot,m. In
the Following step, all MIMO radar node will send their observation data information to the
data center for analysis. The Information will then be compiled and the observation will be
h it
extracted based on the observation vector ot = ot1 ; ot2 ; . . . ; otM and the emission state of the
0 0�
drone positions will be estimated rt; xt ; yt .
2.1.1. Sensing method. For easy analysis, the dynamic notation can be summarized as,
r t ¼ R ðr t 1 Þ ð1Þ

R(.) is a dynamic function R!R which specifies stochastic progress of drone’s emission
states R. By considering the fact that drone is an agent vulnerable to any movement and exter-
nal influence while in the air, we define two transitional functions R!R,
v t ¼ V ðv t 1 ; h 1 Þ ð2Þ

yt ¼ yðyt 1 ; h2 Þ ð3Þ

These two random stochastics represent the behaviors of drone speed movement vt and
angular orientation θt while in the air, which are moved independently and randomly by

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 3 / 18

PLOS ONE Sensing spectrum sharing based massive MIMO radar for drone tracking and interception

Fig 1. Spectrum sharing for massive MIMO radar and drones detection entering safety zone on brownian motion.
https://doi.org/10.1371/journal.pone.0268834.g001

noises h1 and h2 respectively. Drone still in the air is a dynamic agent, which means we will
have to define its dynamic movement,

Ut ¼ I ðUt 1 ; vt ; yt Þ ð4Þ

Where I(.) is the transition function R2 ! R2, specifies the dynamics drone’s movement with
� 0 0 �t
the vector location Ut ¼ xt ; yt ,and the observation function,

ot ¼ OðUt ; rt ; wt ðnÞÞ ð5Þ

ot is the measurement equation with observation function 0(.): RM! R1,which describes the
relationship between two hidden states rt, ut and the measurement ot,m.
From here, three assumptions have been made to execute the sensing. First, a segment of a
periodic sensing is performed, where the emission state of the drone is assumed to remain
fixed. This means rt will remain unchanged for one sensing period of Tr,after that it will
change.
Secondly, the static Gaussian filter was considered at this stage. The observation ot,m is rela-
tive between the mth radar and the moving drone. The noise random estimation of the nth
portion at discrete time t noted as wt(n) of Eq (5) is assumed to be independent identical distri-
bution with zero mean additive white Gaussian noise, where variance is s2w ,which is also inde-
pendent identical distribution of two hidden states.
� 0 0 �t
Thirdly, we considered the drone as moving with positioning Ut ¼ xt ; yt and its remain
constant during a period of time Tr.

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 4 / 18

PLOS ONE Sensing spectrum sharing based massive MIMO radar for drone tracking and interception

2.1.2. Drone’s dynamic states. After analysis, we find out that the progress of emission
states of drone over time T, can be represented as a finite states engine and can be described as
two states Markov process [28–32] R = {R0, R1}. If we consider the drone as active and moving
with emission states R1 at time t, then the survival probability of an active drone can be writ-
ten,
� �
Pr ) Pr rtþ1 ¼ 1nrt ¼ 1 ð6Þ

� �
Pr ) Pr rtþ1 ¼ 1nrt ¼ 1 ¼ lt Dt þ oðDt Þ;

Where λt is the survival rate. Dealing with a Markov process, the probability of transition will
depend on only the current state. We can determine the probability of the drone remaining in
its survival movement by adding all the probabilities of its ways of progress:
� � �
Pr rtþ1 ¼ 1 ¼ Pr rtþ1 ¼ 1nrt ¼ 1 Pðrt ¼ 1Þ

� �
Pr rtþ1 ¼ 1nrt ¼ 2 Pðrt ¼ 2Þþ
� � ð7Þ
. . . þ Pr rtþ1 ¼ 1nrt ¼ r Pðrt ¼ rÞ

The computation of Eq (7) can easily lead us to matrix notation. Then the vector of each
survival probability can be written as,
2 3
P ðr t ¼ 1 Þ
6 7
6 P ðr t ¼ 2 Þ 7
6 7
p ðt Þ ¼ 6 .. 7 ð8Þ
6 7
4 . 5
Pððrt ¼ rÞ

And it is transition matrix can be represented as,

2 3
Pð1n1Þ Pð1n2Þ . . . Pð1nrÞ
6 7
6 Pð2n1Þ Pð2n2Þ . . . Pð2nrÞ 7
6 7
A¼6 .. 7 ð9Þ
6 7
4 . 5
Pðrn1Þ Pðrn2Þ . . . PðrnrÞ

The drone will go into sleeping mode on states R0 with a probability 1 − Pr in the following
time t+1. If the drone stays in sleeping mode R0, it will move again into R1 states with a birth
probability,
� �
Pb ) Pr rtþ1 ¼ 1nrt ¼ 0 ð10Þ

� �
Pb ) Pr rtþ1 ¼ 1nrt ¼ 1 ¼ mt Dt þ oðDt Þ;

Where μt is the birth rate and it may remain in states R0 in the next time with a probability
1 − Pb. In the same way, we can determine the probability of the drone remaining in its birth
probability by adding all the probabilities of its ways of maintenance:
� � �
Pr rtþ1 ¼ 1 ¼ Pr rtþ1 ¼ 1nrt ¼ 0 Pðrt ¼ 0Þ

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 5 / 18

PLOS ONE Sensing spectrum sharing based massive MIMO radar for drone tracking and interception

� �
Pr rtþ1 ¼ 1nrt ¼ 1 Pðrt ¼ 1Þþ
� � ð11Þ
. . . þ Pr rtþ1 ¼ 1nrt ¼ r 1 Pðrt ¼ r 1Þ

Then the vector and transition matrix of each birth probability can be written respectively
as,
2 3
P ðr t ¼ 0 Þ
6 7
6 P ðr t ¼ 1 Þ 7
0 6 7
p ðt Þ ¼ 6 .. 7 ð12Þ
6 7
4 . 5
P ðr t ¼ r 1Þ

2 3
Pð1n0Þ Pð1n1Þ . . . Pð1nr 1Þ
6 7
6 Pð2n0Þ Pð2n1Þ . . . Pð2nr 1Þ 7
0 6 7
A ¼6 .. 7 ð13Þ
6 7
4 . 5
Pðrn0Þ Pðrn1Þ . . . Pðrnr 1Þ

It is worthy to note that in the above mentioned dynamic probability, the transitional
matrix is specific with the drone devises. In other wireless devises the dynamic transition
remain invariant for a longer period T [33].
2.1.3. Drone’s dynamic positioning. Firstly, statistical action of the speed and orientation
of the drones were studied, where it was realized that the drone is moving following a random
walking process. As two random variables, the speed and orientation Eqs (2) and (3) at time t
can be written as,
�
vt ¼ vt 1 þ h1 h1 � N 0; s2v ð14Þ

�
yt ¼ yt 1 þ h2 h2 � ε 0; s2y ð15Þ

Where s2v and s2y represent the variances of drone’s acceleration and direction, respectively.
� �
We consider that the two noises N 0; s2v and ε 0; s2y which are Gaussian, are following the
path of random walking. By considering the above equations, based on speed and orientation
of the drone, we can then represent the dynamic cartesian equations of its position by,

xt ¼ xt 1 þ vt cosðyt Þ ð16Þ

yt ¼ y t 1 þ vt sinðyt Þ ð17Þ

Where xt and yt represent the abscissa and ordinate of the cartesian axes position of the drone
respectively.
2.1.4. Statistical detection. In order to derive a decision rule and the detection analysis,
which maximizes Pr[rt+1 = 1\rt = 1]. Based on the observation set ot,m, given this realization,
the conditional probability of correct detection can be written as,
� �
Pr rtþ1 ¼ 1nrt ¼ ot;m

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 6 / 18

PLOS ONE Sensing spectrum sharing based massive MIMO radar for drone tracking and interception

and the observed signal in practice can be represented as,

XN h pffiffiffiffi a=
i2
2
ot;m ¼ n¼1
r t E r at ð n Þd t;m þ wt ðnÞ ð18Þ

Where ot,m is the received signal strength of the mth radar, dt,m is the distance that separate the
mth radar and the stirring drone at time t, α is the path loss fading which supposed to be
greater than 2. ρt represent the received gain of the mth radar which is from radar processing
devises. N = Tƒ is the samples size and ƒ is the sampling frequency. at(n) is the progression of
drone’s message indications, where n = 1,2, ‥‥, N. For easy analysis, binary phase shift keying
(BPSK) has been considered, where at(n) = {+1, -1}, with Er the emission power. For the
absence of drone, the received signal strength is simply,
XN
ot;m ¼ n¼1
wt 2 ðnÞ ð19Þ

With a moving drone, the observation also may continue to be uncertain. Therefor an Euclid-
ean distance between the targeted agent, e.g. drone and the radar are for a greater importance.
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi�ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi�ffiffiffiffi
2 2
dt;m ) kUt am k2 ¼ xt0 xm þ yt0 ym ð20Þ

By considering the distance dt,m and drone’s emission states rt, the component likelihood
density can be written as p(ot,m\dt,m,rt). As the N has to be very huge (such as, N � 100), we
can estimate the likelihood functions by applying Gaussian densities of i.i.d noise. The central
limit theorem (CLT) will give the following approximations,
8 YM �
�< m¼1 p ot;m nUt ; rt ¼ 1 � H1
p ot;m nUt ; rt ð21Þ
: YM �
m¼1
p o t;m nU t ; rt ¼ 0 � H0

Consequently, all observations from the data center can also be seen as Gaussian distribu-
tion with mean and variance respectively,

X
M
�
φðot nUt ; rt Þ ¼ φ ot;m nUt ; rt
m¼1

X
M
�
Wðot nUt ; rt Þ ¼ W ot;m nUt ; rt
m¼1

2.1.5. Drone’s states prediction. As known, in the Bayesian approach [32,34], we analyse
the unknown quantity, as a random variable. We recursively estimate the conditional posterior
distribution.
Pt-1(rt-1\o1,t-1) at time t-1. In our case the trajectory of the drone‘s emission states at tth dis-
crete time is define by r = {r0, r1, . . ., rt}. Bayesian method is an effective mechanism to analyze
and estimates hidden states. The prediction and updates of the posterior distribution of the
hidden states rt can be computed based on Bayes filter,
Z
Ptnt 1 ðrt 1 no1:t 1 Þ ¼ ptnt 1 ðrt 1 nr1:t 1 Þpt 1nt 1 ðrt 1 no1:t 1 Þdrt 1 ð22Þ

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 7 / 18

PLOS ONE Sensing spectrum sharing based massive MIMO radar for drone tracking and interception

pt ðot nrt Þptnt 1 ðrt no1:t 1 Þ

Ptnt ðrt no1:t Þ ¼ R ð23Þ
pt ðot nrt Þptnt 1 ðrt no1:t 1 Þdrt

Where Eqs (22) and (23) represent the prediction and update respectively, and function.
Pt\t-1(rt-1\o1:t-1) and Pt\t(rt\o1:t) represent the transitional density and likelihood function
respectively. With the above assumption, the joint density can be estimated recursively. The
ordinary estimation process for the sensing may become weak or raise concerns of imperfec-
tion due to drone’s constant changing position. It can be noticed that, the dynamic distance
from Eq (20) may disappear completely by observing from the data center, when a drone goes
off (i.e., H0 or rt = r0). In analysing a Bayesian inference for an unknown position, the related
likelihood involving the drone and radar distances may become unavailable, making the track-
ing of the drone’s dynamic position difficult to analyze. Another important aspect is that, with-
out a clear drone’s positioning, the estimation of the drone states will be inaccurate. This is
because of the imprecise result of the reception, especially for Energy Detection (ED) sensing
method.

2.2. Random finite state

A Random Finite State (RFS) is a random variable that takes values as unordered finites sets
[35]. The Effect of drone’s signal appearing or disappearing can possibly be treated as another
aspect of random states [36–38]. In this present study, for a deeper analyses for dynamic
behaviors of the drone, the two hidden states are studied like one combined random process
called random finite state, represented as F [39].
The cardinality of a RFS F (i.e number of elements) is random and analyzed according to a
discrete distribution ρ(g) = P{|F| = g}, where g 2 N0 and g = |F| is the cardinality of RFS F. A
RFS F is characterized by its cardinality and a group of symmetric joint distribution [35,40] ρ
(F1, . . .Fg), F1, . . .Fg 2 Rg.
According to the current drone sensing, |Ft| 2 {0.1} which means a binary threshold γt
need to be taken into consideration, which stand for γt = 1 (i.e.,H1) when drone emitted a sig-
nal at time t, otherwise γt = 0 (i.e.,H0). Consequently, it was noticed that the random variable
γt and the cardinality distribution ρ(g) are Bernoulli RFS. The Bernoulli RFS can either be
empty (with probability 1 − q) or have one element (with probability q). According to Mahler’s
theorem [39,40], the probability density function (PDF) of the finite set statistics (FISST) for
such Bernoulli RFS can be described as,
(
1 q if � Ft ¼ ; or gt ¼ 0
rðg Þ ¼ ð24Þ
q if � Ft ¼ fUt g or gt ¼ 1

The probability density function (PDF) p(Ft) can farther be developed as a normal random
variable [40] as,
� n o� � �
p Ft ¼ F1 ; . . . ; Fg ¼ g!rðg Þp F1 ; . . . ; Fg

Applying the set integral, we will have,

Z X
1
1
Z � �
pðFt ÞdF ¼ pð;Þ þ p F1 ; . . . ; Fg dF1 ; . . . ; dFg � 1 ð25Þ
t!
tþ1

It is clear to see that p(Ft) integrate to one as it is should be for a PDF.

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 8 / 18

PLOS ONE Sensing spectrum sharing based massive MIMO radar for drone tracking and interception

Thus said, it can further be seen that, the presence of a moving drone during the sensing
can be represented as |Ft| = 1, which correspond to the dynamic position Ut.
Based on cardinality distribution ρ(g) and states distribution PDF p(Ut), the FISST can be
redefined as,
(
1 q if � Ft ¼ ; or gt ¼ 0
rðFt Þ ¼ ð26Þ
q : pðζÞ if � Ft ¼ fUt g or gt ¼ 1

For some cases where the cardinality g is greater than 1, then p(Ft) = 0.

2.3. Dynamic transition agent

According to the actual system, the dynamic transitional model of the Bernoulli RSF Ft shall
also follow Markov process. Thus said, the Eqs (8) and (12) can then be represented as,
(
1 pðt Þ if � Ft ¼ ;
ptnt 1 ðFt nfUt gÞ ¼ ð27Þ
pðtÞptnt 1 ðUt nUt 1 Þ if � Ft ¼ fgt g

And
(
1 p0 ðtÞ if � Ft ¼ ;
ptnt 1 ðFt n;Þ ¼ ð28Þ
p0 ðt Þbtnt 1 Ut if � Ft ¼ fUt g

Where bt\t−1 represent the birth and initial density when the drone is re-detected or re-emit-
ting its signal. And πt\t-1 (Ut\Ut-1) is the dynamic survival transitional density of the drone’s
location, which can be represented by [41]:
( �2 )
jjUt Ut 1 jj2 vt 1
2s2v
1 2 � � 3
ptnt 1 ðUt nUt 1 Þ ¼ pffiffiffiffiffiffiffiffiffi exp 1
yt y t 1 ð29Þ
2psv jtan y j
1 6 x x t 1 7
� 2 exp6
4
t t 1 7
5
y 2s y s2

Where tan-1(.) is the angular vector movement.

2.3.1. Path loss. Current studies based on radio communications affected by large scale
free space propagation model has proposed several path loss method [21,42]. In this current
work, a single carrier frequency of 3.55Ghz has been adopted. Where more focus has been put
on distance dependency. The Close-in free space reference (CI) path loss models can be
expressed as [42],
PLCI ðd; f Þ ¼ PLFS;ref ðf Þ þ 10nCI log10 ðdÞ þ xs;CI ð30Þ

Where 10nCI log10(d) is the logarithmic distance dependency behavior with nCI path loss expo-
nential (PLE). ξσ,CI represent the shadow fading in decibel and follows Gaussian distribution
with zero mean with standard deviation σ. PLFS,ref (f) represent the carrier frequency and it is
calculated by applying Friis’s law for free space propagation:
4pf
PLFS;ref ðf Þ ¼ 20 log10 ð Þ ð31Þ
c

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 9 / 18

PLOS ONE Sensing spectrum sharing based massive MIMO radar for drone tracking and interception

Where c is the speed of the light. And the doppler angular frequency,
2nr f
fD ¼ ð32Þ
c
Where vr is the radial speed of the target.

2.4. Bernoulli filtering and control

This is very similar with Bayesian prediction and update, where the two posterior densities pt\t
(Ft\o1:t) and ƒt\t (Ft) will be propagated recursively. On the prediction stage, the first predic-
tion densities of the two terms qt\t-1 and ft/t-1(Ut) can be derived as follows,
Z
ptnt 1 ðFt no1:t 1 Þ ¼ Ptnt 1 ðFt nFt 1 Þpt 1nt 1 ðFt 1 no1:t 1 ÞdFt 1
Z
¼ Ptnt 1 ðFt n;Þpt 1nt 1 ð;no1:t 1 Þ þ Ptnt 1 ðFt nUt 1 Þpt 1nt 1 ðUt 1 no1:t 1 ÞdUt 1 ð33Þ

Now we are solving Ft = ; (when the drone is off) with pt\t-1 (; \o1:t-1) = 1-qt\t-1 and Ft =
{Ut} (when the drone went on) with pt\t-1 (Ut\o1:t-1) = qt\t-1 ƒt\t-1 (Ut). And since the predicted
FISST PDF is in the form of Eq (26), we will have:
h � � i
qtnt 1 ¼ 1 ð1 pb Þ 1 qt 1nt 1 þ ð1 ps Þqt 1nt 1

� �
¼ pb 1 qt 1nt 1 þ ps q t 1nt 1 ð34Þ

Similarly, when the drone went on,

� �
pb 1 qt 1nt 1 btnt 1 ðUt Þ
f tnt 1 ðUt Þ ¼ þ
qtnt 1
R ð35Þ
ps qt 1nt 1 pt 1nt 1 ðUt nUt 1 Þft 1nt 1 ðUt 1 ÞdUt 1

qtnt 1

It is worthy to note from the above two equations the predicted density (qt-1\t-1) and spatial
density (ƒt\t-1 (Ut) may involve two important elements, the birth element of a new drone
appearing and a survival element of an already existing drone. The first birth is defined as the
disappearance of the drone (e.g., pb); and the second one which is the survival is define by a
continuing appearance of the drone (ps). The above two equations fully specify the step of Ber-
noulli filter.

3. Numerical results
The results presented in this section are generated from Matlab Simulation and Simulink.
These are more suitable for Dynamic and complex analysis because more parameters can be
added. A dynamic radar detection of a targeted element (Drone), can just penetrate the zone
of detection with its trajectory as shown in 2-D grid Fig 2. The first step was to generate a
radar detection of a moving drone with straight legs of 20km and a turn radius of 2km. The
altitude of the trajectory is 1km, which is defined as –1km by default North-East-Down coor-
dination structure used in this scenario. The radar is mounted on a tower of 5m length, 5m
width, and 30m of height. It is defined as spectrum origin [0,0,0]. A summary of notations pre-
sented in this paper can be found in Table 1.

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 10 / 18

PLOS ONE Sensing spectrum sharing based massive MIMO radar for drone tracking and interception

Fig 2. Target drone racetrack path with straight legs of 20 km and a turn radius of 2 km.
https://doi.org/10.1371/journal.pone.0268834.g002

A monostatic scanning radar sensor has been executed with the step size of update rate
5.5Hz in scene of 0.4 sec. Mounting location [0 0–15], field of view [4 45] and mechanical azi-
muth [–60–60]. The radar coverage zone with its scanning angle can be seen in Fig 3.
Secondly, another radar sensor has been added to the tower to amplify the detection in case
of a huge intrusion of rogue drones in the protected area. It is added with an update rate
5.5Hz. And its performance is very high, as seen from its bleu scanning angle in Fig 4.
In the third approach, there was an intrusion of a second drone in the safety zone, and this
was quickly detected by the two radars as observed in Fig 5. The second drone flew from south-
west to northeast at a height of 1.5km with a time of arrival [0 80]. All the reporting frames of
the radars were sent back to the data center through inertial navigation system (INS). We
noticed that, the second drone was equipped with a sensor which is able to inject anything to
the safety zone. This leads to the next step, which is the destabilization of the drone or making
it to turn back. A Proportional-Integral-Derivative (PID) control was applied for this scenario
with parameters R(.),V(.),θ(.),U(.),O(.). The Euler initial position is defined by (0,0,0), with

Table 1. Massive MIMO radar parameters for test environment.

Parameters Values
Radar & Drone Communication RF Band 3550–3650 MHz
Radar Antenna Tx/Rx
Iterations 4/2
1000
Carrier Frequency 3.55 GHZ
Update rate 5.5 Hz
Turn Radius 2 Km
Radial Velocity 1000 m/s
Speed of Light 3x108 m/s
Target point 20 Km
Spectral efficiency(bits/sec/Hz)
Sample Time 1
Gravity 0.02
9.8
Doppler angular frequency 4vrf/c
Path loss PLCI(d, f)
https://doi.org/10.1371/journal.pone.0268834.t001

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 11 / 18

PLOS ONE Sensing spectrum sharing based massive MIMO radar for drone tracking and interception

Fig 3. Plot of radar coverage mounted on 30m tower and signal detection of a moving drone.
https://doi.org/10.1371/journal.pone.0268834.g003

gravity (0,0, -9.8). As shown in Fig 6, the drone can be controlled following the radar position
on a square model, Fig 6A and 6D and by doing so, a significant signal power of approximately
5 MHZ can be straight pointed to the target. However, the drone can progressively start losing
its control as seen in Fig 6C and 6D. Fig 7 demonstrates the signal wideband propagation in a
free space environment. The center frequency is 3 GHz and the frequencies of the three tones
are 750 kHz, 1 GHz, and 1.5 GHz, respectively. The system model applies range-dependent
time delay, gain adjustment, and phase shift to the input signal. Additionally, the model esti-
mates the Doppler shift when the drone is moving. The free-space environment is a boundary-
free medium with a speed of signal propagation independent of position and direction. The
signal is propagated along a straight line from the source to its destination. Therefore, the
model shows the two-way propagation of the signal from the radar to the targets. For this
wideband signal, it was observed that the magnitude of the Doppler shift increased with fre-
quency. In case of narrowband signals, the Doppler shift is assumed to remain constant over

Fig 4. Second radar mounted on the tower. With it scanning detection angle and field of view of (4,45).
https://doi.org/10.1371/journal.pone.0268834.g004

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 12 / 18

PLOS ONE Sensing spectrum sharing based massive MIMO radar for drone tracking and interception

Fig 5. (a) Intrusion of second drone into the safety zone. Flying from southwest to northeast at a height of 1.5 km with a time of arrival [0 80]; (b) The second drone
remain trackable. As its moves, its detection moves as well.
https://doi.org/10.1371/journal.pone.0268834.g005

the band. Lasty, the system model performance has been tested as a series of states spaces
model, to validate the model with real world. Based on time invariant, the simulate states and
observations were estimated. As shown in Fig 8, the true state and simulated states of the target

Fig 6. (a) Drone controlled followed square motion; (b) Drone controlled followed near radar on square motion; (c)
Drone losing control as it received signal power of over 2.5 MHZ; (d) A drone that completely loses control after
receiving a significant jamming power.
https://doi.org/10.1371/journal.pone.0268834.g006

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 13 / 18

PLOS ONE Sensing spectrum sharing based massive MIMO radar for drone tracking and interception

Fig 7. Plot of Wideband propagation signal for the spectrum of original signal and the Doppler-shifted signal
with central frequency, 3Ghz.
https://doi.org/10.1371/journal.pone.0268834.g007

were compared to the observed responses and simulated responses from the radar at a series of
200 observations. It was shown that, the true states values of the target aligned to 90% with the
observed values.

4. Discussion
The development of wireless and control system has accelerated the use of unmanned aerial
vehicles (UAVs) or drones. However, public safety as well as privacy has become a general con-
cern. In this study, we analyzed spectrum access for massive MIMO radar covering a safety
zone for rogue drones’ intrusion, by considering numerous numbers of drones. Here, the
MIMO radar and UAV (drone) shared interferences on the same spectral band zone. We pres-
ent here a novel technique in detecting and tracking the targeted drones until their

Fig 8. Plot of true state values and simulated states alongside with observed responses and simulated responses at
a series of 200 observations.
https://doi.org/10.1371/journal.pone.0268834.g008

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 14 / 18

PLOS ONE Sensing spectrum sharing based massive MIMO radar for drone tracking and interception

destabilization. This was made possible with a deep analyses of drone’s localization tracking,
where the emission states of the drone is estimated. This is followed by dynamic Bayesian filter
approach algorithm, based on massive MIMO radar and drone spectrum sensing. The particu-
larity of this new approach is that, the unknown emission state of the drone is analysed like
another hidden state that needs to be estimated, instead of the changing locations. Considering
the limited information available in spectrum access, we used received signal strength to esti-
mate recursively the two hidden states. Meanwhile, this technique is also very useful in scenar-
ios of spectrum access between MIMO radar, 5G Communication base station system, and
drone localization. A joint distribution algorithm based Bernoulli Random Finite Set (BRFS)
has been developed, where the emission state of the drone and the associate state such as its
unknown locations was analysed. We estimated recursively drone’s existence state and its
dynamic positioning, in which most times it’s difficult to be analysed in reality in real live.
This will enhance the tracking scenario of the drone when it appear and disappear from the
scene. Corresponding to Algorithm summary flow, there were two main parts involved in this
work:
1. By using Observation tracking trajectory O1t, and relying on recursive Bayesian Filter pre-
diction and update, we estimated the posterior density ƒt\t-1(Ut),
2. Tracking of detection uncertainty with length rt-1 = R0 as J, where 0 < j < J and the next
step of state is define as rt-J-1 = R1. Therefore, drone goes collapsing in stage
YM �
m¼1
p ot;m nUt ; rt ¼ 0 � H0 at time t−J, and its prediction estimation is subsequently t–J
+ m,(where m = 1,2,M). If the drone continues its move in stage
YM �
m¼1
p ot;m nUt ; rt ¼ 1 � H1 at time t, a complete likelihood of the birth density is
estimated,

� �
btnt 1 Xtnt 1 � fv ðvt nJ Þ:fy ðyt nJ Þ ð36Þ

Algorithm Scenario
Iterate
Observation Information Collection,
XN
ot,m = n¼1 t
w 2 ðnÞ, Eq (19)
Estimate Dynamic transitional model RSF Φt
pt\t-1 (Φt\{Ut}) Eq (27) and pt\t-1 (Φt\;) Eq (28)
Confirm Drone’s survival density: πt\t-1 (UtUt-1}
Prepare Bernoulli Filters for prediction
pt\t-1 (Φt\o1:t-1) Eq (33)
For if i = 1: M
Compute recursive Bayesian Filter prediction
qt\t-1 Eq (34)
Update: ft\t-1(Ut) Eq (35)
Forward Eqs (34) and (35) to track
detection uncertainty with length rt-1 = R0,
where 0 < j < J and next step rt-J-1 = R1
End for
Learn Drone’s collapsing stage,
YM
�
p ot;m nUt ; rt ¼ 0 � H0
m¼1
YM �
p ot;m nUt ; rt ¼ 0 � H0
imin ¼ arg max m¼1
1�i�M

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 15 / 18

PLOS ONE Sensing spectrum sharing based massive MIMO radar for drone tracking and interception

If not, estimate the survival stage

QM �
p ot;m nUt ; rt ¼ 1 � H1
m¼1
By completing the birth density likelihood,
� �
btnt 1 Xtnt 1 � fv ðvt nJ Þ:fy ðyt nJ Þ Eq (36)
End
It is demonstrated by an extensive numerical experimentation that the system is not only
efficient but also potentially accurate in estimating the drone’s location and detection. The
uncertainty of reception can be measured, therefore, the system can constantly be optimized.

5. Conclusions
As unmanned aerial vehicles are been used in our daily lives, a greater concent are been placed
on human environment, privacy and public lives.
In this paper, a new approach base on spectrum sensing to realize drone’s tracking, detec-
tion, and destabilization for a safety zone and Cognitive Radio Applications has been pro-
posed. A System model has been established to thoroughly characterize the dynamic
movement of the unknown states of the drone and its moving locations. We get the advantage
of Bernoulli random finite set algorithm to track the drone’s moving positions and detecting
its random emission states. Hence, MIMO radar jointly makes an optimal decision on the
mobility and power control to the targeted drone. Experimentation results demonstrate the
performance and effectiveness of the proposed method, to intercept and destabilize the rogue
drones. Additionally, by taking full consideration of the sensing and dynamic localization, we
can observe and well detect the unknown emission states of the drone even when it is still
moving.
Future investigations are needed for different emission signals of new drones technology, as
5G has brought brand-new cooperative environment of spectrum sensing.

Acknowledgments
We will like to thank the Department of Electronics and Information Engineering of Chang-
chun University of Science and Technology for availing the equipments and materials to con-
tract this experimentation.
Informed consent statement
All subjects involved in the studies have given their consent.

Author Contributions
Conceptualization: Milembolo Miantezila Junior.
Data curation: Bin Guo.
Formal analysis: Bin Guo.
Investigation: Bin Guo.
Methodology: Milembolo Miantezila Junior.
Software: Milembolo Miantezila Junior.
Supervision: Bin Guo.
Writing – original draft: Milembolo Miantezila Junior.

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 16 / 18

PLOS ONE Sensing spectrum sharing based massive MIMO radar for drone tracking and interception

References
1. Nassi B., et al. SoK: Security and privacy in the age of commercial drones. in Proc. IEEE Symp. Security
Privacy (SP). 2021. https://doi.org/10.1109/SP40001.2021.00005.
2. Rizk Y., et al., Decision making in multiagent systems: A survey. 2018. 10(3): p. 514–529 https://doi.
org/10.1109/tcds.2018.2840971.
3. Chung T.H., Hollinger G.A., and Isler V.J.A.r., Search and pursuit-evasion in mobile robotics. 2011. 31
(4): p. 299–316 https://doi.org/10.1007/s10514-011-9241-4.
4. Oh B.-S., et al., Micro-Doppler mini-UAV classification using empirical-mode decomposition features.
2017. 15(2): p. 227–231 https://doi.org/10.1109/lgrs.2017.2781711.
5. Tahmoush D. Detection of small UAV helicopters using micro-Doppler. in Radar Sensor Technol-
ogy XVIII. 2014. International Society for Optics and Photonics https://doi.org/10.1117/12.
2054329.
6. Fioranelli F., et al., Classification of loaded/unloaded micro-drones using multistatic radar. 2015. 51
(22): p. 1813–1815 https://doi.org/10.1049/el.2015.3038.
7. Fuhrmann, L., et al. Micro-Doppler analysis and classification of UAVs at Ka band. in 2017 18th Interna-
tional Radar Symposium (IRS). 2017. IEEE https://doi.org/10.23919/irs.2017.8008142.
8. Rahman, S. and D.A. Robertson. Millimeter-wave micro-Doppler measurements of small UAVs. in
Radar Sensor Technology XXI. 2017. International Society for Optics and Photonics https://doi.org/10.
1117/12.2261942.
9. Jahangir, M. and C. Baker. Characterisation of low observable targets with a multi-beam staring radar.
in International Conference on Radar Systems (Radar 2017). 2017. IET https://doi.org/10.1049/cp.
2017.0430.
10. Regev, N., I. Yoffe, and D. Wulich. Classification of single and multi propelled miniature drones using
multilayer perceptron artificial neural network. in International Conference on Radar Systems (Radar
2017). 2017. IET https://doi.org/10.1049/cp.2017.0378.
11. Björklund, S. Target detection and classification of small drones by boosting on radar micro-Doppler. in
2018 15th European Radar Conference (EuRAD). 2018. IEEE https://doi.org/10.23919/eurad.2018.
8546569.
12. de Wit, J.M., R. Harmanny, and G. Premel-Cabic. Micro-Doppler analysis of small UAVs. in 2012 9th
European Radar Conference. 2012. IEEE.
13. Harmanny, R.I., et al., Radar micro-Doppler mini-UAV classification using spectrograms and cepstro-
grams. 2015. 7(3–4): p. 469–477 https://doi.org/10.1017/s1759078715001002.
14. Molchanov P., et al., Classification of small UAVs and birds by micro-Doppler signatures. 2014. 6(3–4):
p. 435–444 https://doi.org/10.1017/s1759078714000282.
15. Chen, K., Y. Li, and X. Xu. Rotating target classification base on micro-Doppler features using a modi-
fied adaptive boosting algorithm. in 2015 International Conference on Computers, Communications,
and Systems (ICCCS). 2015. IEEE https://doi.org/10.1109/ccoms.2015.7562907.
16. Safaron M., et al., Directional cloverleaf antenna for unmanned aerial vehicle (UAV) application. 2019.
14(2): p. 773–779 https://doi.org/10.11591/ijeecs.v14.i2.pp773-780
17. Liu Z.-Q., et al., A novel broad beamwidth conformal antenna on unmanned aerial vehicle. 2012. 11: p.
196–199 https://doi.org/10.1109/LAWP.2012.2187321
18. Long K., et al., Single UWB anchor aided PDR heading and step length correcting indoor localization
system. 2021. 9: p. 11511–11522 https://doi.org/10.1109/ACCESS.2021.3050405
19. Wei Z., Junfeng Y.J.I.J.o.A., and Propagation, A design of vertical polarized conformal antenna and its
array based on UAV structure. 2017. 2017 https://doi.org/10.1155/2017/9769815.
20. Balderas L.I., et al., Low-profile conformal UWB antenna for UAV applications. 2019. 7: p. 127486–
127494 https://doi.org/10.1109/ACCESS.2019.2939511
21. Fang G., et al., Experimental research of multistatic passive radar with a single antenna for drone detec-
tion. 2018. 6: p. 33542–33551 https://doi.org/10.1109/access.2018.2844556.
22. Searle, S., S. Howard, and J. Palmer. Remodulation of DVB—T signals for use in Passive Bistatic
Radar. in 2010 Conference Record of the Forty Fourth Asilomar Conference on Signals, Systems and
Computers. 2010. IEEE https://doi.org/10.1109/acssc.2010.5757576.
23. Kuschel, H., et al. On the resolution performance of passive radar using DVB-T illuminations. in 11-th
INTERNATIONAL RADAR SYMPOSIUM. 2010. IEEE.
24. Baczyk M.K., Malanowski M.J.I.J.o.E., and Telecommunications, Reconstruction of the reference sig-
nal in DVB-T-based passive radar. 2011. 57: p. 43–48 https://doi.org/10.2478/v10177-011-0006-y.

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 17 / 18

PLOS ONE Sensing spectrum sharing based massive MIMO radar for drone tracking and interception

25. Xianrong, W., et al. Reconstruction of reference signal for DTMB-based passive radar systems. in Pro-
ceedings of 2011 IEEE CIE International Conference on Radar. 2011. IEEE https://doi.org/10.1109/cie-
radar.2011.6159501.
26. Wan X.-R., et al., Reference signal extraction methods for CMMB-based passive bistatic radar. 2012.
34(2): p. 338–343.
27. Chen, M.-F. and Y.-H. Mao, Introduction to stochastic processes. Vol. 2. 2021: World Scientific.
28. Li, B., Z. Zhou, and A. Nallanathan. Joint estimation based spectrum sensing for cognitive radios in
time-variant fading channels. in 2013 IEEE Global Communications Conference (GLOBECOM). 2013.
IEEE https://doi.org/10.1109/glocom.2013.6831566.
29. Bogale, T.E., L. Vandendorpe, and L.B. Le. Sensing throughput tradeoff for cognitive radio networks
with noise variance uncertainty. in 2014 9th International Conference on Cognitive Radio Oriented Wire-
less Networks and Communications (CROWNCOM). 2014. IEEE https://doi.org/10.4108/icst.
crowncom.2014.255811.
30. Li B., et al., Spectrum sensing for cognitive radios in time-variant flat-fading channels: A joint estimation
approach. 2014. 62(8): p. 2665–2680 https://doi.org/10.1109/tcomm.2014.2325835.
31. Ahmad W.S.H.M.W., et al., 5G technology: Towards dynamic spectrum sharing using cognitive radio
networks. 2020. 8: p. 14460–14488 https://doi.org/10.1109/access.2020.2966271.
32. Miantezila, M., et al. Primary User Channel State Prediction Based on Channel Allocation and DBHMM.
in 2020 International Conference on Cyber-Enabled Distributed Computing and Knowledge Discovery
(CyberC). 2020. IEEE https://doi.org/10.1109/cyberc49757.2020.00063.
33. Moon T.K. and Stirling W.C., Mathematical methods and algorithms for signal processing. 2000.
34. Chen Z.J.S., Bayesian filtering: From Kalman filters to particle filters, and beyond. 2003. 182(1): p. 1–
69.
35. Ristic B., et al., A tutorial on Bernoulli filters: theory, implementation and applications. 2013. 61(13): p.
3406–3430 https://doi.org/10.1109/tsp.2013.2257765.
36. Liang Y.-C., et al., Cognitive radio networking and communications: An overview. 2011. 60(7): p. 3386–
3407 https://doi.org/10.1109/tvt.2011.2158673.
37. Lu L., et al., Ten years of research in spectrum sensing and sharing in cognitive radio. 2012. 2012(1): p.
1–16 https://doi.org/10.1186/1687-1499-2012-28.
38. Jayaweera S.K., Signal processing for cognitive radios. 2014: John Wiley & Sons https://doi.org/10.
1002/9781118824818.index.
39. Mahler R.P., Statistical multisource-multitarget information fusion. 2007: Artech House, Inc. https://dl.
acm.org/doi/abs/10.5555/1512927.
40. Vo, B.-T. and B.-N. Vo. A random finite set conjugate prior and application to multi-target tracking. in
2011 Seventh International Conference on Intelligent Sensors, Sensor Networks and Information Pro-
cessing. 2011. IEEE https://doi.org/10.1109/issnip.2011.6146549.
41. Vo B.-T., Vo B.-N., and Cantoni A.J.I.T.o.s.p., Bayesian filtering with random finite set observations.
2008. 56(4): p. 1313–1326 https://doi.org/10.1109/tsp.2007.908968.
42. Jeong W.H., et al., Empirical Path-Loss Modeling and a RF Detection Scheme for Various Drones.
2018. 2018 https://doi.org/10.1155/2018/6795931.

PLOS ONE | https://doi.org/10.1371/journal.pone.0268834 May 20, 2022 18 / 18