1

UAV-Aided Jamming for Secure Ground

Communication with Unknown Eavesdropper
Location
Christantus O. Nnamani, Muhammad R. A. Khandaker, Senior Member, IEEE, and Mathini Sellathurai, Senior
Member, IEEE
arXiv:1911.11279v1 [cs.IT] 25 Nov 2019

Abstract—This paper investigates unmanned aerial vehicle protocols and techniques have been developed in the literature
(UAV)-aided jamming technique for enabling physical layer key- for the security in the layers of these models [4]. Public
less security in scenarios where the exact eavesdropper location and private key-based cryptographic security measures are
is unknown. We assume that the unknown eavesdropper location
is within an ellipse characterizing the coverage region of the most widely used in many communication systems. However,
transmitter. By sequentially optimizing the transmit power, the cryptographic security is heavily computation demanding in
flight path of the UAV and its jamming power, we aim at one hand, thus impractical in many IoT applications, and
maximizing the average secrecy rate with arbitrary eavesdrop- subject to sophisticated external attacks with the advent of
per location. Simulation results demonstrate that the optimal modern computing facilities on the other hand. Developing
flight path obtains better secrecy rate performance compared
to that using direct UAV flight path encasing the transmitter novel security measures to combat such attacks is therefore of
and the legitimate receiver. Most importantly, even with the prime interest for many researchers. In this context, the notion
unknown eavesdropper location, we obtained a secrecy rate that of physical layer security has attracted significant attention due
is comparable to a scenario when the eavesdroppers location is to its ability to provide information-theoretic security [4]–[7].
known. However, the average secrecy rate with the unknown The physical layer is similar in most communication models
eavesdropper location varies depending on the proximity of the
eavesdropper to the known location of the transmitter. We also as it deals with processing the encapsulated message for
observe that due to the UAV-aided jamming, the average secrecy transmission via the channel [8]. In wireless communications,
rate stabilizes at some point even though the average received it deals directly with the electromagnetic waves referred to
envelope power of the eavesdropper increases. This essentially as signals. These signals can be compromised via eaves-
demonstrates the effectiveness of the proposed scheme. dropping and jamming of legitimate receivers. Focusing on
Index Terms—Secure communication, jamming, UAV, trajec- the eavesdropping, the security in the physical layer can be
tory optimization, physical layer security. subdivided into key-based and keyless security models. The
primary objective of both models is to reduce the ability of
I. I NTRODUCTION an illegitimate user to gain access to the transmitted mes-
sage. While the key-based models use information obscurity
Protecting sensitive or confidential information is of as its main tool, the keyless models detect the possible
paramount interest to most businesses/organizations pri- information leak in the presence of eavesdropper(s)1 and
vate, public, government, military or intelligence. In the attempts to decrease its intercepted information. The degree
event that such data/information is made public, these busi- of information protection in a keyless physical layer security
nesses/organizations may face legal or financial ramifications. model is measured as the secrecy capacity for delay tolerant
At the very least, they will suffer loss of customer trust (e.g. applications and the outage probability for delay intolerant
companies, etc.); but in the worst case, it could lead to the applications. To maximize the secrecy capacity, [3] proposed
complete annihilation of the organization (e.g. Military, etc.). an on/off algorithm that varies the power transmitted from the
Thus, secure communications are obligatory to most busi- source especially when the eavesdropper have better channel
nesses/organizations and in this sense seen as a primordial re- quality. It relied on the principle that the source knows the
quirement of technological and military exploits. However, as channel state of the eavesdropper based on inherent channel
technologies continue to explode, especially with the develop- monitoring. While this scheme reduces the information content
ment of modern computing technologies, the internet of things received by the legitimate receiver, it also has limited practical
(IoT), 5G and future generation networks, adverse robust ways applications as the channel information of the eavesdropper
of information theft continue to grow [1], [2]. In practice, is usually unknown. Instead of reducing the transmit power,
a total secured communication is unattainable, nevertheless, a more sophisticated approach could be to deliberately jam
theories seem to support a measure that is acceptable [3], [4]. It the eavesdropper’s channel ensuring that it receives little/no
is important that communication be unique in all the layers of information. The major limitation of this technique is that
the communication model - open system interconnection (OSI) the eavesdropper will usually operate at the same band as
and/or the internet model to guarantee its security. Different
1 The kind of eavesdroppers referred to in this paper are considered as
The authors are with the School of Engineering and Physical Sciences, passive Wyner wiretappers [9] which do not attempt to alter the transmitted
Heriot-Watt University, Edinburgh EH14 4AS, UK. message but try to overhear only.
 2

the legitimate receiver, hence the jamming will also affect no-fly regions in [23]. Critical examination reveals that the
the legitimate receiver. A combination of jamming and power methods used in [18]–[23] are similar in principle since they
variation, harnessing their gains is subsequently the bedrock optimized the transmitted power, the UAV jamming power
of modern signal jamming techniques. and its trajectory for the corresponding scenarios. However,
Signal jamming as a physical layer protection strategy is a strong assumption made in [18]–[23] is that the location of
one of the most prominent brute-force methods of limiting the the eavesdropper(s) is known to the source and/or the UAV(s).
information theft in keyless physical layer security exploiting Although this assumption simplifies the respective problem
the fading characteristics of the channel [8]. It entails simul- in each scenario, it is grossly impractical. In most practical
taneous transmission of a signal with similar characteristics communication scenarios, even knowing the presence of an
to the genuine signal but carrying no information content eavesdropper is often very difficult let alone knowing their
to cause interference to the eavesdropper’s received signal. exact locations or channel state information (CSI). This practi-
Although this technique does not guarantee that there will be cal challenge motivates us to investigate secret communication
no information leakage, similar to other security techniques, with unknown eavesdropper location in this paper. We consider
it reduces the probability of successful interception thereby UAV-aided jamming technique for proactively degrading the
increasing the secrecy capacity of the end-to-end communi- eavesdropping channel at unknown ground point for improving
cation. While jamming poses to be an effective technique for the achievable secrecy rate.
improving secrecy, there are some critical issues that affect the An attempt to introduce eavesdropper obscurity has also
effectiveness of signal jamming: been made by Miao Cui, et al. in [24]. The authors in [24]
(a) The degree of transmit power required to increase the considered the UAV as the information source and optimized
secrecy capacity without adversely degrading the infor- its trajectory and transmitting power to a legitimate receiver
mation content of the desired receiver, amidst a group of eavesdroppers located within an independent
(b) The transmitter’s responses to the knowledge of the pos- small uncertainty region. The trajectory of the UAV has been
sible eavesdropper(s), optimized to find the best points in the space to deliver the
(c) The optimal location to deliver the jamming signals from. maximum information to the legitimate receiver while the
Researchers have since investigated these requirements in- eavesdroppers receive minimum information. In contrast, we
dependently as in [10], however, the investigation of the consider the UAV with an opposing role in this paper to
collective effects of (a)-(c) is of practical interest due to degrade the eavesdropper’s channel via cooperative jamming.
their inter-dependency in the context of secrecy performance. Note that our work differs from [24] not just in terms of the
While some recent studies affirm that this technique yields UAV’s role, but also in terms of guaranteed secrecy perfor-
improvement in the secrecy capacity, they are all based on mance. In fact, the achievable secrecy performance in [24]
the impractical assumption that the eavesdropper(s’) location cannot be guaranteed as the uncertainty region expands and
is perfectly known at the transmitter [11]. overlaps with the certainty region of the legitimate receiver.
With respect to the known remote eavesdropper location, Furthermore, a network of legitimate and illegitimate UAVs
mobile means of delivering the jamming signals have recently has been considered in [25] in which a UAV acts as the base
been investigated in the literature. One of the effective methods station to transmit signal to other legitimate UAVs in altitude
proposed is the use of an unmanned aerial vehicle (UAV) in and the eavesdropper UAVs from unknown locations try to
scenarios where the nodes under consideration (the source, overhear the signal. The secrecy outage probability and the
the main receiver and the eavesdropper) are all based on the average secrecy rate performance have been analyzed. Since
ground. This is primarily due to its aerial radio visibility of all the nodes are at the same altitude, the gains of aerial
the ground terminals, its cost efficiency and its availability for visibility of UAV was subdued. In this work, we intend to
low-range applications. The applications of UAVs in commu- explore this opportunity for ground nodes (source, legitimate
nications range from their use as aerial base stations [12]–[14], receiver and eavesdropper) in order to maximize the benefits of
as relay nodes [15], as access/user nodes [11], [12] to channel aerial visibility of the UAV while constrained by the properties
estimation [16], etc. Recently, with the advancement of the of ground propagation.
internet of things (IoT), network of UAVs for UAV-to-UAV We formulate the problem of maximizing the average se-
communications as well as for general data transmission has crecy rate under the unknown eavesdropper location assump-
also been considered [17]. tion by jointly optimizing the source transmit power, the UAV
More recently, UAVs have been deployed for assisting in trajectory and its jamming power. The problem is strictly non-
secure communications between ground terminals [18], [19], convex due to the correlation of the optimization variables
and to act as both relay nodes and security agents between in the problem. Therefore, in this work, we sequentially
ground terminals [20]. In [21], the UAV is deployed with two optimize the flight path of a UAV, its jamming power and
opposing roles namely, to establish favorable and degraded the transmitted power by the source node to ensure secure
channels for the legitimate and the eavesdropping links, re- communication in the considered scenario. One set of variables
spectively. A separate jammer UAV has been considered in are optimized in each step while keeping the others fixed. The
[22] to degrade the eavesdropping channel in addition to the main contributions in this paper can be summarized as:
cooperative UAV for the legitimate channel. Subsequently, (a) Developing the mathematical analysis of UAV-aided jam-
UAVs have also been used to deliver classified messages to ming applications to secure wireless communication when
ground terminals amidst the constraints of eavesdroppers and the location of the eavesdropper is completely unknown.
 3

(b) Applying the block coordinate descent method and suc- Cs = max Rs (2)
P (hb )
cessive convex approximation (SCA) technique with the
aid of the first-order Taylor series expansion. where Rs , Cs , P (hb ), hb and he = |ge |2 are the average
(c) Unveiling the influence of the unknown eavesdropper’s secrecy rate, secrecy capacity, transmit power from Alice, the
received power on the average secrecy rate between the channel power gain of Bob and Eve, respectively, and [a]+
source and the legitimate receiver. indicates max(0, a) 3 . Note that [·]+ imposes a constraint
(d) Validating the formulations and the solutions by demon- such that Eve cannot receive higher information than Bob at
strating the performance of the proposed algorithm against any time during the communication. Accordingly, the limits
existing UAV-aided secure communication schemes of the integrals in (1) are defined such that when he ≥ hb , the
through extensive numerical simulations. mutual information between Alice and Eve is upper-bounded
The rest of this paper is organized as follows: Section II by log2 (1 + hb P (hb )). The objective of keyless physical layer
describes the UAV-aided communication system model and security is to ensure that (2) is sustained at its optimal value
the problem formulation. The proposed solution is developed over the duration of the communication.
in Section III. Simulation results are presented in Section IV Note that the achievable secrecy rate in (1) describes the
before making the concluding remarks in Section V. secrecy rate as the difference of the average information rates
of Bob and Eve over all fading realizations of Bob and Eve.
The non-negativity assumption on the secrecy rate [·]+ requires
that the location of Eve revolves around that of Bob and not
beyond the coverage region of Alice. However, in practice,
Eve may even be located at positions closer to Alice than Bob
and thereby receive stronger signals than Bob assuming they
both share the same channel model based on the proximity to
the transmitter alone. In such scenarios, the achievable secrecy
rate would be zero.
To ameliorate the aforementioned challenge, we deploy a
UAV that will deliver jamming signals to reduce the informa-
tion acquired by Eve while attempting to sustain that obtained
by Bob. However, in UAV-aided communications, a common
challenge is to optimize the UAV trajectory [12]. In secure
communications, the challenge is further proliferated by the
unknown eavesdropper location. We aim at addressing this
challenge in the following sections. The UAV flight path will
Figure 1. UAV-aided jamming for secure communication. be optimized to ensure that for any location of Eve within
the coverage region of Alice, its information rate will be
continually below that of Bob, thereby, achieving positive
II. S YSTEM M ODEL secrecy rates.
We consider a secure wireless communication scenario We assume that the UAV is not equipped with any tracking
between a base station (BS) acting as a transmitting source devices. Therefore, the UAV will not be able to locate or track
(Alice) located at an a-priori known ground point2 wa = Eve despite having a clear line-of-sight (LoS) to all points
[xa , ya , 0]T and a receiver (Bob) at an a-priori known ground within the coverage region of Alice due to aerial visibility.
point wb = [xb , yb , 0]T as shown in Fig. 1. However, an Furthermore, if the UAV flies horizontally at constant alti-
eavesdropper (Eve) lurks around the area in an unknown tude from an initial point q0 to a final point qf , its ascent and
ground location, w̃e = [x̃e , ỹe , 0]T , but within the area where descent flight path to the initial and final ground points can be
the wireless signal can be received. We denote the complex neglected. The UAV flight duration, T , is sampled at discrete
block-fading channels of Alice with Bob and Eve as gb and time-stamps of N equal time slots with duration of δ = T /N
ge , respectively. Since Eve’s location is unknown, Alice’s [15], [18]. With large number of time slots N , we can assume
transmission power Pa is a function of Bob’s channel power that the UAV maintains constant speed V m/s within a slot δ
gain hb = |gb |2 alone; hence Pa = P (hb ), i.e. Alice varies her and transmits almost continuously. For simplicity, we assume
transmission power depending on the channel state of Bob. that V is constant over the entire flight duration as also as-
The average secrecy rate and secrecy capacity derived from sumed in [18]. If the distance covered in each sample is small
Shannon’s information content are given respectively as [3] enough, we can assume that the UAV is stationary at each
Z Z sample point. Considering a large number of sample points, the
∞ hb
UAV is assumed to send jamming signal continuously. These
Rs = [log2 (1 + hb P (hb ) − log2 (1 + he P (hb )]+  T
0 0 | {z } | {z } sampled points can be denoted as q[n] = x[n], y[n], z[n] ,
information rate of Bob information rate of Eve
n ∈ {1, ..., N }, which satisfies the following constraints:
× f (hb )f (he )dhe dhb , (1)
3 Nota bene: All logarithms used in this work is of base 2 since we refer
2 z-coordinate represents the altitude and the ground point is located at z=0. to digital communications.
 4

A. Problem Formulation
kq[n + 1] − q[n]k2 ≤ (V δ)2 (3a)
kq[1] − q 0 k2 ≤ (V δ)2 (3b) Let Q = {q[n], n ∈ N }, pa = {Pa [n], n ∈ N }, and
q[N ] = q f (3c) pu = {Pu [n], n ∈ N } be the set of UAV sample points
kq[n] − wa k + kq[n] − wb k ≤ 2a (3d) (representing its trajectory when connected by a straight line),
the set of power transmitted by Alice as well as the UAV,
q(xn , yn , zn ) = q(xn , yn , H). (3e) respectively. We aim at solving (2) by alternatingly optimizing
Q, pa and pu . Using Reimann sum and averaging through all
Inequalities (3a) and (3b) ensure that the distance covered
fading realizations of hb , we approximate (1) to obtain
by the UAV within the flight samples does not exceed the
parametric distance. The velocity V m/s is chosen such that
the total distance covered by the UAV through the samples will N Z
1 X hb [n]
be greater than or equal to the Euclidean distance between q0 Rs = [log(1 + hb [n]Pa [n]
N n=1 0
and qf , i.e., (V δ) ≥ kq f − q o k, otherwise the system will
be intractable. This ensures that the UAV travels at least in − log(1 + he [n]Pa [n]]f (he )dhe . (6)
a straight path from its initial to its final points for a given
total flight duration. The equality in (3c) ensures that the final To solve (6), we need to know the possible distribution of the
flight point of the UAV is at the a-priori final destination, fading channel of Eve which can be obtained by historical
while (3d) allows the UAV to remain within the uncertainty measurements collected over the region covered by Alice
region where the eavesdropper can be found. This region is (represented in this model as an ellipsis as in (3d)). Using the
postulated as an ellipse and physically represents a cellular Rayleigh fading realization assumption, we have its squared
coverage region of Alice. a determines the size of the ellipse envelop as
and satisfies {for a > kwb − w a k}, wa and wb are the two
foci of the ellipse, ensuring that Bob is not a cell-edge user. 1 − hey[n]
f (he ) = e e , (7)
Finally, (3e) places the UAV to fly at constant altitude denoted ye
by H meters.
Assuming the ground fading channel between Alice and where ye is the average received envelop power of Eve and can
Bob is Rayleigh distributed, the lower bound of the channel be obtained via measurements and/or estimations. Substituting
power gain (corresponding to the worst channel condition) (7) in (6) and solving the integral, we obtain
with the jamming signal delivered by the UAV is given by
4 as obtained in [21], [19], [22]. We note that 4 is the upper N
bound of the random complex channel gb [21]. 1 X
Rs = log(1 + hb [n]Pa [n])
N n=1 | {z }
information rate of Bob
Z he [n]
ground channel gain hb [n]
Pa [n]e− ye
z }| { − dhe . (8)
βo d−ψ 1 + he [n]Pa [n]
hb [n] = ab
, (4) |0 {z }
Pu [n]βo d−2
qb [n] + 1
information rate of Eve
| {z }
LoS jamming signal attenuation
The secrecy rate in (8) can be further simplified as [26, eq.
3.352.1]
where ψ is the ground path loss component between Alice
and Bob, β0 represents the signal-to-noise ratio (SNR) at N
1 X
a reference distance (d0 = 1)m of the ground channels, Rs = log(1 + hb [n]Pa [n])
dab and dqb are the Euclidean distance between Alice, UAV N n=1
  
and Bob respectively and Pu is the UAV jamming signal 1 hb [n] 1
power. To ensure that the power levels of the commu- − e ye Pa [n] Ei − −
ye ye Pa [n]
nication is within acceptable range, Pu and Pa are sub- 
1

jected to average and peak power constraints described as: −Ei − , (9)
ye Pa [n]
0 ≤ Pu [n] ≤ Pumax (5a)
N where
1 X
Pu [n] ≤ P̄ub (5b) Z ∞
e−t
N n=1 Ei (x) = dt
x t
0 ≤ Pa [n] ≤ Pamax (5c)
N
1 X is the exponential integral. We note that (8) is equivalent
Pa [n] ≤ P̄ab . (5d)
N n=1 to (9) and they can be used interchangeably depending on
the parameter been inferred. Thus we substitute the objective
 5

function in (2) with the elaborated form in (9) to obtain the problem (P1) with (8), problem (P1) can be reformulated for
following optimization problem4 any given Q and pu as problem (P2):
N
X
N
X (P 2) : max log(1 + hb [n]Pa [n])
(P 1) : max log(1 + hb [n]Pa [n]) pa
n=1
pa ,pu ,Q
n=1 Z he [n]
hb [n]
1
 
hb [n] 1
 Pa [n]e− ye
−e Ei −
ye Pa [n] − − dhe (11a)
y ye Pa [n] 0 1 + he [n]Pa [n]
  e s.t. (5c) and (5d). (11b)
1
−Ei − (10a)
ye Pa [n] Note that problem (P2) is still non-convex over the entire
s.t. kq[n + 1] − q[n]k2 ≤ (V δ)2 (10b) domain of pa . However, for the region under peak and average
kq[1] − q 0 k2 ≤ (V δ)2 (10c) power constraints, the objective can be shown to be the sum
of a concave and a convex functions. The proof is relegated
q[N ] = q f (10d)
to Appendix A.
kq[n] − wa k + kq[n] − wb k ≤ 2a (10e) Since the objective function of problem (P2) is differentiable
q(xn , yn , zn ) = q(xn , yn , H), (10f) (as demonstrated in Appendix A), it can be solved using
0 ≤ Pu [n] ≤ Pumax (10g) the Karush-Kuhn-Tucker (KKT) conditions for non-convex
N
problems [27]:
1 X
Pu [n] ≤ P̄ub (10h) ∇f0 (x∗ ) + λ∗ ∇fn (x∗ ) = 0, (12a)
N n=1
λ fn (x ) = 0,
∗ ∗
(12b)
0 ≤ Pa [n] ≤ Pamax (10i)
N where f0 is the objective in problem (P2), fn are the con-
1 X
Pa [n] ≤ P̄ab . (10j) straints in (5c) and (5d) and x∗ is the optimal value of Pa .
N n=1 Simultaneously solving (12) using [26, eq. 0.410 and 3.462.17]
respectively, we obtain
Problem (P1) entails that the secrecy capacity of the proposed 
system depends on the optimal transmission power of Alice, hb [n] 1 1
− − e ye Pa [n]
the jamming power delivered by the UAV and the UAV loca- 1 + hb [n]Pa [n] ye (Pa [n])2
    
tion. Unfortunately, (P1) is a non-convex optimization problem 1 hb [n] 1
Γ −1, − Γ −1, +
with respect to the optimization variables (pa , pu , Q) and ye Pa [n] ye ye Pa [n]
cannot be easily solved directly. However, using a sequential N
X 1
and iterative technique under a block coordinate approach, we × Pa [n] − P̄ab = 0, (13)
can obtain suboptimal solutions that satisfy the constraints in n=1
N
(3) and (5). i Pi−1 (−1)k k!
where Γ(−i, z) = (−1)
i! (E1 (z) − e
−z
k=0 z k+1 ) [28, eq.
8.4.15]). Solving (13) with a non-linear solver produces the
suboptimal values of Pa .
III. P ROPOSED S OLUTION
B. Optimizing the UAV Jamming Power (Pu )
We propose solving the non-convex problem (P1) in an
alternating fashion. The proposed solution involves decompos- To optimize the jamming power pu delivered by the UAV,
ing the original problem (P1) into three sub-problems each we consider pu as the optimization variable while fixing the
characterizing a set of optimization variables. In each sub- values of pa and Q. Problem (P1) is then reformulated
problem, we optimize one set of variables while fixing the while substituting for hb [n] as
!
other variables in each iteration. The results obtained from N
βo d−ψ
ab Pa [n]
X
each iteration step are analyzed with the objective value of (P 3) : max log 1 +
(P1) and the iteration stops at the point when the objective
pu
n=1
Pu [n]βo d−2
qb [n] + 1
  −ψ 
value (P1) converges. βo dab
1
−2
  Pu [n]βo dqb [n]+1 1 
−e ye Pa [n] × Ei − − 
ye ye Pa [n]
 
A. Optimizing the Source Power (Pa ) 1
−Ei − (14a)
ye Pa [n]
We first optimize Alice’s transmit power for arbitrary initial
trajectory and jamming power. Replacing the objective in s.t. (5a) and (5b). (14b)
Under the constraints, the objective of Problem (P3) is a non-
4 We neglected the constant scaling factor 1
in the objective function as convex function with respect to pu due to the non-convexity
N
this does not affect the optimal solution. of the information rate of the Eve. However, the information
 6

rate of Bob is concave with respect to pu . Hence, problem approximation with the first order Taylor expansion. Let
(P3) is solved using successive convex approximation (SCA). Qk [n] = {q k [n], n ∈ N } denote the initial UAV trajectory
Given an initial UAV jamming power in the k-th iteration for the kth iteration. Then the objective function of problem
as pku = {Puk [n], n ∈ N }; we have using first order Taylor (P4) can be rewritten as
expansion that   
   βo d−ψ
ab
−ψ Pu [n]βo
βo dab 1 1 
1   m[n] +1
1   Pu [n]βo d−2
qb [n]+1 1  e ye Pa [n] Ei − − ) − Ei (− 
e ye Pa [n] Ei − −  ye ye Pa [n] ye Pa [n]
ye ye Pa [n]
  ≤ Ok [n] + Wk [n](q[n] − q k [n]) (18)
1
−Ei − ≤ Gk [n] + Tk [n](Pu [n] − Puk [n]),
ye Pa [n]
(15) −kq[n] − wb k2 ≤ S k [n], (19)
where
   where
βo d−ψ
ab   
1   Puk [n]βo d−2
qb [n]+1 1  βo d−ψ
ab
Gk [n] = e ye Pa [n] Ei − −    Pu [n]βo
+1 1 
ye ye Pa [n] 1
Ok [n] = e ye Pa [n] 
mk [n]
Ei − −
 
  ye ye Pa [n] 
1
−Ei −  
ye Pa [n] 1
−Ei −
−(
βo d
−ψ
ab )
ye Pa [n]
−2 k [n]+ye
2 −ψ −2 ye βo d [n]Pu
Pa [n]βo dab dqb [n]e qb
and Tk [n] = . βo d−ψ
(βo d−2 k −ψ −2 k
qb [n]Pu [n]+1)(Pa [n]βo dab +βo dqb [n]Pu [n]+1) βo2 d−ψ
ab Pu [n]e
−  ab
o Pu [n]

Taking only the non-constant terms in (15), problem (P3) can ye 1+ βm [n]
k
Wk [n] =    ,
be reformulated as β d−ψ βo Pu [n]
" ! ye − ye P1a [n] −  o ab
o Pu [n]
 1+ mk [n] m2k [n]
N ye 1+ βm
X βo d−ψ
ab Pa [n]
[n]k
(P 3b) : max log 1 +
pu
n=1
Pu [n]βo d−2
qb [n] + 1 and S [n] = kqk [n]k − 2[qk [n] − wb ]T q[n] − kwb k2 . Under
k 2

−Tk [n]Pu [n]] (16a) similar conditions as of problem (P3), (P4) can be reformu-
s.t. (5a) and (5b). (16b) lated as
 
N
Problem (P3b) is a convex problem within the constrained re- X β o d−ψ
ab P a [n]
gion and can be efficiently solved using interior-point method (P 4b) : max log 1 + P [n]β  − Wk [n]m[n]
u o
Q,M
n=1 m[n] + 1
or a convex solver such as CVX [29], [30].
(20a)
C. Optimizing the UAV Trajectory (Q) s.t. m[n] + S k [n] ≤ 0, (20b)
In this sub-problem, the problem (P1) is recast to ensure and (3). (20c)
that only the UAV trajectory, Q is the optimization parameter.
However, the reformulated problem is non-convex in Q. Problem (P4b) is a convex problem in Q under the specified
Hence, to reduce computational complexity, we introduce a constraints and can be solved using interior-point methods
slack variable M = {m[n] = kq[n] − wb k2 , n ∈ N } such that or with a convex solver. The overall procedure has been
dqb
−2 1
[n] = m[n] . Thus we obtain the following optimization summarized in Algorithm 1.
problem:
  Algorithm 1 Iterative algorithm for solving pa , pu , and Q
N −ψ
X β o dab P a [n] 1 1: Initialize pu and Q such that the constraints in (5a), (5b)
(P 4) : max m log 1 + P [n]β  − e ye Pa [n]
and (3) are satisfied.
m[n] + 1
Q,M u o
n=1
   2: m ← 1.
−ψ
β d o ab
Pu [n]βo
3: repeat
  m[n] +1 1  4: Compute and update pa in (13) with given pu and Q.
× Ei − − 
ye ye Pa [n] 5: Using updated pa and current Q, solve (16) for pu .
  6: With given pa and pu , find Q by solving problem (20).
1 7: Compute Rs as defined in (9).
−Ei − (17a)
ye Pa [n] Rnew −Rold
8: e = s Rnew s .
s.t. m[n] − kq[n] − wb k2 ≤ 0, (17b) 9: m ← m + 1.
s

and (3). (17c) 10: until e < θ OR m ≥ mmax .

11: Output: pa , pu , and Q.
Due to the non-convexity of problem (P4) with respect to the
trajectory, q[n], we reformulate the problem using successive
 7

IV. S IMULATION R ESULTS AND ANALYSIS

0.8
In this section, we evaluate the performance of the pro- ProW 300
ProW 350
posed solution approach through numerical simulations. We 0.7
ProW 400
implement the solution discussed in Section III following 0.6
the procedure described in Algorithm 1. The optimization
parameters are initialized by solving the feasibility problem 0.5
such that the the initial values just satisfy their respective
0.4

e
constraints. The feasibility problem can be formulated by
setting the objective of problem (10) to zero, with all the 0.3
primary constraints unchanged. Then, by iteratively optimizing
each parameter with the knowledge of the others, we obtain 0.2

the suboptimal solution to problem (P1) when the error (e)

0.1
between steps is less than θ (where θ = 10−5 ) or the maximum
number of iterations is reached (where mmax = 200). 0
1 1.5 2 2.5 3 3.5 4 4.5 5
Similar to the convergence analysis in [19], Algorithm 1 Number of iteration
is guaranteed to converge for all feasible initial points. This
is shown in Fig. 2 where a fast convergence is observed for Figure 2. Convergence performance of Algorithm 1.
different scenarios of the UAV flight time. In Fig. 2, the ProW
algorithm represents the proposed solution to the unknown
Eve problem while the associated numbers represent the UAV for the ProW algorithm (recall unknown Eve location) and
flight time. In all the simulations, we used the parameters as from Alice to Eve for the JT&P algorithm based on the
described in Table I unless otherwise specified. locations specified in Table I. Nevertheless, from Fig. 4 to
Fig. 8, the numbers attached to the acronyms depict the UAV
Table I flight time in seconds. Results in Fig. 3 illustrate that the direct
S IMULATION PARAMETERS flight path with constant power (Straight) scheme performs the
Simulation parameter Symbol Value worst in terms of the average secrecy rate. Due to the jamming
Alice location wa [0, 0, 0] signals delivered by the UAV, the average secrecy rate of the
Bob location wb [300, 0, 0] JT&P scheme is zero when the Eve is at the same location as
Eve location we [150, 250, 0], [350, 0, 0],
[300, 20, 0], [300, 70, 0] Bob. As Eve moves away from Bob, the average secrecy rate
Initial UAV location qo [−100, 100, H] increases since the UAV locates Eve and stays at an optimum
Final UAV location qf [500, 100, H] location to jam her signal. Nevertheless, since the location of
UAV height(when fixed) H 100m [18] Eve is unknown (as in ProW 300), the average secrecy rate
Velocity per sample(when V 3m/s [18]
fixed) is shown to be close to the JT&P scheme when Eve is closer
Duration per sample(when δ 0.5s [18] to Alice and is supposed to receive more information content
fixed) without the UAV jamming. However, considering that ProW
Signal-to-noise ratio βo 90dB [18]
Average received envelop ye 20dBm
300 is near to a practical scenario, this marginal decrease in
power performance may be considered as the near practical trade-off
Average UAV transmit P̄ub 10dBm [18] to the scheme.
power
Maximum UAV power Pumax 4Pub [18] Also, it is important to note that the information rate of
Average Source power P̄ab 30dBm [18] both Bob and Eve is affected by the jamming signal of the
Maximum source power Pamax 36dBm [18] UAV. However, Eve is affected more, even when it has better
Radius of uncertainty re- a 450m channel condition (measured in terms of its average received
gion (when fixed)
Path loss for ground com- u 3.4 envelope power), as the UAV regularly finds paths such that
munication (urban area it stays further from Bob and estimates as close to Eve as
cellular radio) possible until it flies to its final point. This allows for positive
average secrecy rates shown in Fig. 3.
We then analyze the secrecy rate performance of the pro- The flight trajectory of the UAV with respect to Alice and
posed scheme as compared with the existing schemes. In Bob is shown in Fig. 4. The 2D plot shows that from an aerial
Fig. 3, the performance of the unknown Eve location scenario view, the trajectory of the UAV follows a given pattern bound
using the proposed joint trajectory and power optimization by the uncertainty region of Eve (ellipse) provided it flies at
algorithm (referred to as ProW) is compared the known Eve a constant altitude. However, to demonstrate the effectiveness
location scenario considered in [18] (referred to as JT&P). of the proposed approach in practice, we also plot a 3D view
We also compare the performance with the baseline scheme of the trajectory in Fig. 4. The 3D plot shows that the UAV
without optimizing the UAV trajectory referred to as Straight), trajectory moves towards the opposite of Bob while ensuring
in which the UAV flies straight to the location above the that the jamming signal is still delivered to all points within
eavesdropper. The associated numbers in the legends represent the constrained region of Eve.
the respective Euclidean distances from Alice (source) to Bob Furthermore, Fig. 5 examines the constraints posed by the
 8

0.9 0.4
ProW 300 ProW 250
0.8 JT&P 291.5476 ProW 300

Average secrecy (bps/Hz)

JT&P 350 0.35 ProW 350
Average secrecy (bps/Hz)

0.7
JT&P 300.6659
JT&P 308.0584
0.6
Straight 0.3
0.5

0.4
0.25
0.3

0.2 0.2
0.1

0 0.15
200 220 240 260 280 300 320 340 360 380 400 50 100 150 200
UAV flight duration(s) Height(m)

Figure 3. Average secrecy rate with ‘unknown’ as well as ‘known’ eaves- Figure 6. Influence of UAV altitude (height) on average secrecy rate under
dropper locations, and direct UAV flight path. the proposed scheme.

assumptions on the property of Eve’s channel. We recall that

the only known property of Eve is its average received envelop
power, ye . Hence Fig. 5 presents the effect of varying ye on
the average secrecy rate. It can be observed that increase in ye
decreases the average secrecy rate via a positive exponential
path. Hence, for large values of ye characterizing Eve having
better reception equipment and channel state as compared to
Bob, the decrement in average secrecy rate with respect to
increasing ye becomes negligible. The optimized UAV path
ensures that even when the location of Eve is unknown, the
average secrecy rate of the communication between Alice
and Bob can be guaranteed despite Eve supposedly receiving
signals with high envelope power. While this average secrecy
rate is low, it can be improved by increasing the time of
flight of the UAV or allowing the UAV to fly throughout the
communication duration as shown in Figs. 3 and 5.
Other factors that affect the average secrecy rate in the
Figure 4. UAV flight trajectory in 2D and 3D view while Eve location is
unknown (For clarity, we use δ = 10). considered scenario of unknown eavesdropper location include
the UAV height and speed, and the SNR of the environment.
Figs. 6, 7 and 8 show the average secrecy rate of the proposed
system compared to the UAV altitude, speed and ground node
0.45
ProW 300
SNR, respectively. It can be observed that the impact of flying
0.4 ProW 350 the UAV at higher altitude is minimal in terms of average
ProW 400 secrecy rate, compared to allowing it to fly at longer duration.
ProW 450
Average secrecy (bps/Hz)

0.35 The trend in Fig. 6 suggests that the average secrecy rate
increases with increase in the UAV altitude/height. However,
0.3 we observed from our simulations that for large values of
0.25
UAV flight altitude, the trajectory optimization problem (P4)
becomes infeasible. Increasing the UAV speed and ground
0.2 nodes SNR tend to increase the average secrecy rate with a
logarithmic path. Nevertheless, the rate of increase is higher
0.15 with the UAV speed than the SNR. As the UAV speed
0.1
increases, its sample points increase allowing it to deliver more
0 0.5 1 1.5 2 2.5 3 jamming signal to Eve within its flight time. Similar to results
y (W) observed in [25], increasing the ground SNR improves the
e
secrecy, however, this parameter is subject to characteristics
Figure 5. Effect of average received envelop power of Eve on average secrecy of the outdoor environment which cannot be easily controlled.
rate.
 9

The optimized transmitting power of Alice (Pa ) and the

optimized UAV jamming power (Pu ) are plotted in Fig. 9.
0.55
ProW 300 While Alice transmits at its maximum power when the UAV
ProW 350 is close to Bob, the UAV transmits minimum jamming signal.
0.5
Average secrecy (bps/Hz)

ProW 400
This ensures that the UAV interference to Bob is minimal and
Bob continues to receive the information sent by Alice.
0.45

0.4 V. C ONCLUSION
0.35 In this paper, we have exploited UAV-aided jamming tech-
nique in reducing the information rate received by an eaves-
0.3 dropper in an unknown location. We solved the achievable
secrecy rate maximization problem using sequential block
0.25 coordinate optimization method. While we were constrained
3 4 5 6 7
Speed(m/s)
by the elusive nature of the eavesdropper location, we obtained
a secrecy rate that is comparable to a scenario when the
eavesdropper’s location is known. We also showed that the
Figure 7. Influence of UAV flying speed on average secrecy rate with obscure
Eve. UAV speed and flight duration are amongst the main parame-
ters to consider while using UAV to increase physical layer
security. Most importantly, we have demonstrated that the
average received envelope power of the eavesdropper cannot
0.4 guarantee better information content as the secrecy rate tends
ProW 300
ProW 350 to stabilize with large envelope power. We propose that future
0.38
works investigate predicting the eavesdropper location with
Average secrecy (bps/Hz)

ProW 400

0.36 the aid of deep learning techniques in order to update the

UAV flight path in real-time. This could reduce the latency
0.34 in continuously solving the optimization problem for each
communication block.
0.32

0.3
A PPENDIX A
0.28
In this section, we show that the non-convexity of (8) is
0.26 the sum of a concave and a convex functions in terms of Pa .
82 84 86 88 90 From (8), we obtain
SNR(dB)
N
X
Figure 8. Average secrecy rate versus signal-noise-ratio (SNR) with obscure Rs = log(1 + hb [n]Pa [n])
Eve. | {z }
n=1
f1 (Pa )
Z he [n]
hb [n]
Pa [n]e− ye
− dhe . (21)
0.2 -16 1 + he [n]Pa [n]
Pa |0 {z }
Pu f2 (Pa )
0.15 -17

We consider (21) in two parts, showing their convexity with

0.1 -18
the second derivative method. In general, the convexity of a
function is defined as [27]
Power (dBW)

Power(dBW)

0.05 -19

 Convex :> 0
0 -20 ′′
f (x) = Concave :< 0

-0.05 -21
Af f ine := 0.

Thus we have from (21) that

-0.1 -22
 2
′′ hb
-0.15 -23 f1 (Pa ) = − .
0 5 10 15 20 25 30 35 40 1 + hb Pa
N

We then show the convexity of the f2 (Pa ) using the princi-

Figure 9. Comparing transmitted power between Alice and UAV.
ple that the nonnegative weighted-sum of a convex (concave)
 10

function is a convex (concave) [27, Section 3.2.1]. The second [12] S. Zhang, Y. Zeng, and R. Zhang, “Cellular-enabled UAV communi-
part can be rewritten as cation: A connectivity-constrained trajectory optimization Perspective,”
IEEE Trans. Commun., vol. 67, no. 4, pp. 2580–2604, Mar. 2019.
Z hb [n] [13] Y. Zeng and R. Zhang, “Energy-efficient UAV communication with
he [n] Pa [n]
e− ye dhe trajectory optimization,” IEEE Trans. Wireless Commun., vol. 16, no. 6,
0 1 + h e [n]Pa [n] pp. 3747–3759, Jun. 2017.
Z hb [n] [14] X. Zhou, Q. Wu, S. Yan, F. Shu, and J. Li, “UAV-enabled secure
≡ w(he )f (Pa , he )dhe . communications: Joint trajectory and transmit power optimization,”
0 IEEE Trans. Veh. Technol., vol. 68, no. 4, pp. 4069–4073, Apr. 2019.
It has been shown in [27] that if f (Pa , he ) is convex (concave), [15] Y. Zeng, R. Zhang, and T. J. Lim, “Throughput maximization for
UAV-enabled mobile relaying systems,” IEEE Trans. Commun., vol. 64,
then f2 (Pa ) is convex (concave). Thus, we have that the no. 12, pp. 4983–4996, Dec. 2016.
second derivative of f2 (Pa ) as [16] A. Ahmed, S. Zhang, and Y. D. Zhang, “Multi-target motion parameter
estimation exploiting collaborative UAV network,” in IEEE Int. Conf.
′′ 2he Acoustics, Speech and Signal Processing (ICASSP ’19). Brighton,
f2 (Pa ) = − . United Kingdom: IEEE, May 2019, pp. 4459–4463.
(1 + he Pa )
[17] H. Wang, J. Wang, J. Chen, Y. Gong, and G. Ding, “Network-connected
Thus both parts of (21) are concave functions independently UAV communications: Potentials and Challenges,” China Commun.,
under the constraint of hb ≥ 0 and Pa ≥ 0. These are the vol. 15, no. 12, pp. 111–121, Dec. 2018.
positive semi-definite constraints that guarantees communica- [18] A. Li, Q. Wu, and R. Zhang, “UAV-enabled cooperative jamming
tion between the source and the destination. If hb < 0 and/or for improving Secrecy of Ground Wiretap Channel,” IEEE Wireless
Commun. Lett., vol. 8, no. 1, pp. 181–184, Feb. 2019.
Pa < 0 then no information could be transmitted successfully.
[19] Z. Li, M. Chen, C. Pan, N. Huang, Z. Yang, and A. Nallanathan, “Joint
Therefore, we have that (21) is the sum of a concave and trajectory and communication design for secure UAV networks,” IEEE
a convex function (−f (x) = convex if f (x) = concave) in Commun. Lett., vol. 23, no. 4, pp. 636–639, Apr. 2019.
terms of Pa . [20] L. Shen, N. Wang, and X. Mu, “Iterative UAV trajectory optimiza-
tion for physical layer secure mobile relaying,” in Int. Conf. Cyber-
Enabled Distributed Computing and Knowledge Discovery (CyberC
R EFERENCES ’18). Zhengzhou, China: IEEE, Oct. 2018, pp. 19–23.
[1] F. Tariq, M. R. A. Khandaker, K.-K. Wong, M. Imran, M. Bennis, , [21] G. Zhang, Q. Wu, M. Cui, and R. Zhang, “Securing UAV communi-
and M. Debbah, “A speculative study on 6G,” IEEE Wireless Commun. cations via joint trajectory and power control,” IEEE Trans. Wireless
Mag., revised, 2019. Commun., vol. 18, no. 2, pp. 1376–1389, Feb. 2019.
[2] A. Mukherjee, “Physical-layer security in the internet of things: Sensing [22] A. Li and W. Zhang, “Mobile jammer-aided secure UAV communica-
and communication confidentiality under resource constraints,” Proc. tions via trajectory design and power control,” China Commun., vol. 15,
IEEE, vol. 103, no. 10, pp. 1747–1761, Oct. 2015. no. 8, pp. 141–151, Aug. 2018.
[3] P. K. Gopala, L. Lai, and H. E. Gamal, “On the secrecy capacity of
fading channels,” IEEE Trans. Info. Theory, vol. 54, no. 10, pp. 4687– [23] Y. Gao, H. Tang, B. Li, and X. Yuan, “Joint trajectory and power design
4698, Oct. 2008. for UAV-enabled secure communications with no-fly zone Constraints,”
[4] Y. Zou, J. Zhu, X. Wang, and L. Hanzo, “A survey on wireless security: IEEE Access, vol. 7, pp. 44 459–44 470, Apr. 2019.
Technical challenges, recent advances, and future trends,” Proc. IEEE, [24] M. Cui, G. Zhang, Q. Wu, and D. W. K. Ng, “Robust trajectory and
vol. 104, no. 9, pp. 1727–1765, Sep. 2016. transmit power design for secure UAV communications,” IEEE Trans.
[5] M. R. A. Khandaker and K.-K. Wong, “Masked beamforming in the Veh. Technol. , vol. 67, no. 9, pp. 9042–9046, Sep. 2018.
presence of energy-harvesting eavesdroppers,” IEEE Trans. Inf. Foren- [25] J. Ye, C. Zhang, H. Lei, G. Pan, and Z. Ding, “Secure UAV-to-UAV
sics Security, vol. 10, pp. 40–54, Jan. 2015. systems with spatially random UAVs,” IEEE WIreless Commun. Lett.,
[6] ——, “Robust secrecy beamforming with energy-harvesting eavesdrop- vol. 8, no. 2, pp. 564–567, Apr. 2019.
pers,” IEEE Wireless Commun. Letters, vol. 4, pp. 10–13, Feb. 2015.
[7] A. A. Okandeji, M. R. A. Khandaker, and K.-K. Wong, “Secure full- [26] I. S. Gradshteyn and I. M. Ryzhik, Table of integrals, series and
duplex two-way relaying for SWIPT,” IEEE Wireless Commun. Letters, products, 7th ed., A. Jeffrey and D. Zwillinger, Eds. USA: Academic
vol. 7, pp. 336–339, June 2018. Press, 2007.
[8] L. Sun and Q. Du, “Physical layer security with its applications in 5G [27] S. Boyd and L. Vandenberghe, Convex optimization. Cambridge, U. K.:
networks: A review,” China Commun., vol. 14, no. 12, pp. 1–14, Dec. Cambridge University Press, 2004.
2017. [28] F. W. J. Olver, W. Lozier, Daniel, F. Boisvert, Ronald, and W. Clark,
[9] A. D. Wyner, “The wire-tap channel,” The Bell System Tech. J., vol. 54, Charles, Eds., NIST handbook of mathematical functions. Cambridge
no. 8, pp. 1355–1387, Oct. 1975. University Press, Jul. 2010. [Online]. Available: https://dlmf.nist.gov/8.4
[10] G. Pan, C. Tang, X. Zhang, T. Li, Y. Weng, and Y. Chen, “Physical-
layer security over non-small-scale fading channels,” IEEE Trans. Veh. [29] M. Grant and S. Boyd, Recent advances in learning and control, ser.
Technol., vol. 65, no. 3, pp. 1326–1339, Mar. 2016. Lecture Notes in Control and Information Sciences. Springer-Verlag
[11] F. Cheng and S. Zhang and Z. Li and Y. Chen and N. Zhao and F. R. Yu Limited, Mar. 2008, ch. Graph implementations for nonsmooth convex
and V. C. M. Leung, “UAV trajectory optimization for data offloading programs, pp. 95–110.
at the edge of multiple cells,” IEEE Trans. Veh. Technol., vol. 67, no. 7, [30] ——, “CVX: Matlab software for disciplined convex programming,
pp. 6732–6736, Jul. 2018. version 2.1,” http://cvxr.com/cvx, Mar. 2014.
This figure "bullet.png" is available in "png" format from:

http://arxiv.org/ps/1911.11279v1