ISIT 2009, Seoul, Korea, June 28 - July 3, 2009
Wireless Physical-Layer Security:
The Case of Colluding Eavesdroppers
Pedro C. Pinto, Student Member, IEEE, João Barros, Member, IEEE,
and Moe Z. Win, Fellow, IEEE
Abstract—We consider the fundamental security limits of
stochastic wireless networks in the presence of colluding eaves-
droppers. By establishing a direct connection with the single-
input multiple-output (SIMO) Gaussian wiretap channel, we
are able to provide a complete characterization of the secrecy
capacity for the case in which the eavesdroppers are scattered
according to a spatial Poisson process. Our analysis, which
includes the probabilities of existence and outage of secrecy
capacity, helps clarify how the spatial density of eavesdroppers
can jeopardize the success of wireless physical-layer security
based on information-theoretic principles.
Index Terms—Information-theoretic security, wireless chan-
nels, secrecy capacity, colluding eavesdroppers, stochastic geom-
etry.
I. I NTRODUCTION
Although much has been achieved in terms of securing
the higher layers of the classical protocol stack, protecting
the physical layer of wireless networks from one or mul-
tiple eavesdroppers remains a formidable task. Due to the
properties of the physical medium, any unauthorized receiver
located within the transmission range is capable of observing
the signals sent by the legitimate transmitters. Moreover, the
attacker is free to combine its own observations with those of
other eavesdroppers, thus improving its reception by means of
cooperative inference. On a more positive tone, recent results
in information theory indicate that the physical properties
of wireless channels, such as multipath fading, can be used
effectively to complement the levels of secrecy attained by
means of classical cryptographic primitives.
The theoretical foundation for physical-layer security over
noisy channels, which builds on the notion of perfect se-
crecy [1], was laid in [2] and later in [3]. More recently, space-
time signal processing techniques for secure communication
over wireless links appeared in [4], and the secrecy capacity of
various single-input multiple-output (SIMO) fading channels
was established in [5]. The concept of outage secrecy capacity
of slow fading channels was presented in detail in [6], whereas
the ergodic secrecy capacity of fading channels was derived
in [7], [8]. The presence of colluding eavesdroppers is con-
sidered in [9], but restricting its attention to a fixed number of
eavesdroppers placed at the same spatial location. The secrecy
P. C. Pinto and M. Z. Win are with the Laboratory for Informa-
tion and Decision Systems (LIDS), Massachusetts Institute of Technology,
Room 32-D674, 77 Massachusetts Avenue, Cambridge, MA 02139, USA
(e-mail: ppinto@mit.edu, moewin@mit.edu).
J. Barros is with the Instituto de Telecomunicações, Departamento de
Engenharia Electrotécnica e de Computadores, Faculdade de Engenharia da
Universidade do Porto, Portugal (e-mail: jbarros@fe.up.pt).
978-1-4244-4313-0/09/$25.00 ©2009 IEEE 2442
properties of stochastic wireless networks are discussed in
[10], [11], for the case of non-colluding eavesdroppers.
Intrigued by the fundamental limits of physical-layer secu-
rity in wireless networks with multiple colluding eavesdrop-
pers, we address the issue of how the spatial distribution of
the eavesdroppers and the propagation characteristics of the
channel ultimately determine the achievable secrecy rates. In
large-scale wireless networks, the spatial distribution of the
nodes can be modeled either deterministically or stochasti-
cally. Deterministic models include square, triangular, and
hexagonal lattices in the two-dimensional plane [12], [13],
which are applicable when the location of the nodes in the
network is known exactly or is constrained to a regular struc-
ture. However, in many ad-hoc scenarios, only a statistical
description of the location of the nodes is available, and thus
a stochastic spatial model should be employed. In particular,
the homogeneous Poisson point process [14] is a natural
model when all the points in a region are equally likely
possibilities for the location of a node. The Poisson process
has been successfully used in the context of wireless networks,
to analyze network interference [15], [16], connectivity and
coverage [17], [18], routing [19], and sensor cooperation [20],
among other topics.
The main contributions of this paper are as follows:
• Secrecy capacity in the presence of colluding eavesdrop-
pers: After establishing the equivalence between com-
munication in the presence of colluding eavesdroppers
and the SIMO Gaussian wiretap channel, we obtain an
expression for the corresponding secrecy capacity, when
the eavesdroppers are scattered according to an arbitrary
spatial process.
• Probabilistic characterization of the secrecy capacity:
For the case where the eavesdroppers are scattered ac-
cording to a spatial Poisson process, we provide the
cumulative distribution function (c.d.f.) of the corre-
sponding secrecy capacity.
• Existence and outage of secrecy capacity: We present
expressions for the probabilities of existence and outage
of the secrecy capacity, in the presence of a Poisson field
of colluding eavesdroppers.
The remainder of the paper is organized as follows. Section II
describes the system model. Section III considers the secrecy
capacity for an arbitrary spatial process of eavesdroppers. Sec-
tion IV characterizes the distribution of the secrecy capacity
when the eavesdroppers are spatially distributed according
to a Poisson process. Section V analyzes the corresponding
existence and outage of secrecy capacity. Section VI presents
 2
ISIT 2009, Seoul, Korea, June 28 - July 3, 2009
Main channel
Alice (probe transmitter)
Bob (probe receiver) hM wM
Colluding eavesdroppers

x yM

Γ1
rM
yE

Γ3
hE wE
Γ2 Wiretap channel

Figure 2. SIMO Gaussian wiretap channel, which can be used to analyze

the scenario of colluding eavesdroppers depicted in Fig. 1.

Figure 1. Communication in the presence of colluding eavesdroppers. vectors wM ∈ Cm and wE ∈ Cn , considered mutually
independent and Gaussian distributed with zero mean and
non-singular covariance matrices ΣM and ΣE , respectively.
a case study to illustrate the metrics derived in this paper, as The system of Fig. 2 can then be summarized as
well as their dependence on network parameters such as the !
transmitted power and the spatial density of eavesdroppers. yM = hM x + wM
(1)
Section VII concludes the paper and summarizes important yE = hE x + wE .
findings.
The general SIMO system in Fig. 2 reduces " to the scenario#T
b
in Fig. 1 by setting hM = 1/rM , hE = 1/Γb1 , 1/Γb2 , · · · ,
II. S YSTEM M ODEL ΣM = WM I1 , and ΣE = WE I∞ , where WM and WE are the
We consider the scenario depicted in Fig. 1, where a main noise powers of the legitimate and eavesdropper receivers,
link is composed of two nodes: one transmitter located at respectively, and In is the n × n identity matrix.
the origin (Alice), and one receiver deterministically located
at a distance rM from the origin (Bob). The eavesdroppers III. S ECRECY C APACITY IN THE P RESENCE OF
are scattered in the two-dimensional plane according to an C OLLUDING E AVESDROPPERS
arbitrary spatial process ΠE . The distances of eavesdroppers
to the origin are denoted by {Γi }∞
i=1 , where Γ1 ≤ Γ2 ≤ . . .. In In this section, we determine the secrecy capacity of the
addition, the eavesdroppers are allowed to collude, i.e., they legitimate link, in the presence of colluding eavesdroppers
can exchange and combine the information received by all scattered in the plane according to an arbitrary spatial process.
the eavesdroppers, thus improving their ability to decode the The result is given by following theorem.
secret message. Theorem 3.1: For a given realization of the arbitrary eaves-
To account for the propagation characteristics of the envi- dropper process ΠE , the secrecy capacity of the legitimate link
ronment, we consider that the signal amplitude decays with is given by
the distance r according to k/rb , for some given constant k. ! $ % $ % &
The amplitude loss exponent b is environment-dependent, P PE
Cs = max log2 1 + 2b − log2 1 + ,0 ,
and can approximately range from 0.8 (e.g., hallways inside rM WM WE
buildings) to 4 (e.g., dense urban environments), where b = 1 (2)
corresponds to free space propagation [21].1 in bits per complex dimension, where PE is the aggregate
power received by all the eavesdroppers,
Since the colluding eavesdroppers may gather the received
information and send it to a central processor, the scenario '∞
P
depicted in Fig. 1 is equivalent to the SIMO Gaussian PE = 2b
. (3)
Γ
i=1 i
wiretap channel depicted in Fig. 2. Here, the input is the
signal transmitted by Alice, and the output of the SIMO
Proof: Writing the maximum-a-posteriori rule for the
wiretap channel is the collection of signals received by all the main channel in Fig. 2, it can be shown that a sufficient
eavesdroppers. We consider that Alice sends a symbol x ∈ C
statistic to estimate x from yM is y(M = h†M Σ−1 M yM , and
with power constraint E{|x|2 } ≤ P . The vectors hM ∈ Cm
similarly for the eavesdroppers’ channel [22].3 Since sufficient
and hE ∈ Cn represent, respectively, the gains of the main statistics preserve mutual information, we can equivalently
and eavesdropper channels.2 The noise is represented by the
express (1) in terms of sufficient statistics, for the purpose of
1 Note that the amplitude loss exponent is b, while the corresponding power determining the secrecy capacity. Thus, by left-multiplying
loss exponent is 2b.
2 We use boldface letters to denote vectors and matrices. 3 We use † to denote the conjugate transpose operator.
2443
 3
ISIT 2009, Seoul, Korea, June 28 - July 3, 2009

each side of the equations by the corresponding term h† Σ−1 , where λE is the (constant) spatial density of eavesdroppers,
we obtain ) in nodes per unit area. The following theorem characterizes
y(M = ( hM x + w(M the distribution of the secrecy capacity in this scenario.
(
y(E = hE x + w (E , Theorem 4.1: If ΠE is a Poisson process with density λE ,
the secrecy capacity Cs of the main channel is a r.v. whose
where4 c.d.f. FCs (c) is given by
( 
hM = h†M Σ−1 0,
M hM 

 „ «  c < 0,
( † −1 

hE = hE ΣE hE 1+ 2bP
r WM
2 −1
−c

FCs (c) = 1 − FPeE  (πλMC −1  , 0 ≤ c < CM ,

(M ∼ Nc (0, (
b P
w hM ) 
 E 1/b ) WE



(E ∼ Nc (0, (
w hE ), 1, c ≥ CM ,
6 7 (6)
and w(M is independent of w(E . This (complex) scalar descrip- P
where CM = log2 1 + r2b WM is the capacity of the main
tion corresponds to the Gaussian wiretap channel introduced M

in [23]. If CM and CE denote the capacities of the main channel; Cα is defined as

and eavesdroppers channels, respectively, we know that the 1−α
Cα ! (7)
secrecy capacity Cs of the main channel for some realization Γ(2 − α) cos(πα/2)
of the channels hM and hE is given by with Γ(·) denoting the gamma function; and FPeE (·) is the
Cs = CM − CE c.d.f of a stable r.v. P(E , with parameters6
) * + *+ , $ %
(
h2 (
h2E 1
= max log2 1 + MP − log2 1 + P ,0 (
PE ∼ S α = , β = 1, γ = 1 , (8)
(
hM (
hE b
) * + ,
1 + h†M Σ−1 and b > 1.
M hM P
= max log2 ,0 . (4) Proof: The secrecy capacity Cs of the main channel in
1 + h†E Σ−1E hE P (2) is a function of the r.v. PE , and is therefore also random.
b
" #T In [15], we show that the characteristic function of PE in (3)
Setting hM = 1/rM , hE = 1/Γb1 , 1/Γb2 , · · · , ΣM =
has the form
WM I1 , and ΣE = WE I∞ , (4) reduces to 6 8 6 πα 797
! $ % $ % & φPE (w) = exp −γ|w|α 1 − jβ sign(w) tan ,
P PE 2
Cs = max log2 1 + 2b − log2 1 + ,0 , (9)
rM WM WE
(5) where
-∞ 1 −1 1/b
where PE = i=1 P/Γ2b i is the aggregate power received by α = , β = 1, γ = πλE C1/b P , (10)
b
all the eavesdroppers. This is the result in (2) and the proof
is concluded. and b > 1. R.v.’s with such characteristic function belong
to the class of skewed stable distributions [24]. Stable laws
are a direct generalization of Gaussian distributions, and
IV. P ROBABILISTIC C HARACTERIZATION OF THE include other densities with heavier (algebraic) tails. They
S ECRECY C APACITY share many properties with Gaussian distributions, namely the
Theorem 3.1 is valid for an arbitrary spatial process ΠE , stability property and the generalized central limit theorem.
deterministic or stochastic. In the latter case, the secrecy Equations (9)-(10) can be succinctly expressed as
capacity Cs of the main link in (2) is a random variable $ %
1 −1 1/b
(r.v.), since it is a function the random eavesdropper distances PE ∼ S α = , β = 1, γ = πλE C1/b P . (11)
b
{Γi }∞i=1 . In the rest of the paper, we analyze the case
1
where ΠE is a homogeneous Poisson process in the two- Defining the normalized stable r.v.: P(E ;= PE γ − α =
dimensional plane. Typically, the eavesdropper positions are PE
(πλ C −1 )b P
, we have that P(E ∼ S b , 1, 1 from the scal-
1
E 1/b
unknown a priori, so we may as well treat them as completely
ing property [24]. In general, the c.d.f. FPeE (·) cannot be
random according to a spatial Poisson process.5 Then, the
expressed in closed form except in the case where b = 2,
probability P{n in R} of n eavesdroppers being inside a
which is analyzed in Section VI. However, the character-
region R (not necessarily connected) depends only on the
istic function of P(E has the simple form of φPeE (w) =
total area A of the region, and is given by [14] : " : π ;#;
exp −|w|1/b 1 − j sign(w) tan 2b , and thus FPeE (·) can
(λE A)n −λE A
P{n in R} = e , n ≥ 0, 6 We use S(α, β,γ ) to denote the distribution of a real stable r.v.
n!
with characteristic exponent α ∈ (0, 2], skewness β ∈ [−1, 1], and disper-
4 We use N (0, σ 2 ) to denote a circularly symmetric (CS) complex Gaus-
c
sion γ ∈ [0, ∞). The corresponding characteristic function is
sian distribution, where the real and imaginary parts are i.i.d. N (0, σ2 /2). (
exp −γ|w|α 1 − jβ sign(w) tan πα
` ˆ ` ´˜´
5 The spatial Poisson process is a natural choice in such situation because, 2
, α $= 1,
φ(w) = ` ˆ 2
˜´
given that a node is inside a region R, the probability density function (p.d.f.) exp −γ|w| 1 + j π β sign(w) ln |w| , α = 1.
of its position is conditionally uniform over R.
2444
 4
ISIT 2009, Seoul, Korea, June 28 - July 3, 2009

always be expressed in the integral form and computed which case the c.d.f. of (
√ PE can be expressed in closed form
numerically. as FPeE (x) = erfc(1/ 2x), x ≥ 0. The c.d.f. of Cs in (6)
Using (2), we can now express FCs (c) in terms of FPeE (·), reduces to
for 0 ≤ c < CM , as
FCs (c) = (14)

FCs (c) = P{Cs ≤ c}  0, 
! $ % $ % & 

  c < 0,
P PE 
 >
= P log2 1 + 2b − log2 1 + ≤c P
rM WM WE = erf  √π2 λE C1/2
−1 „ W«  , 0 ≤ c < CM ,
! <$ % =& 
 1+ 4P 2−c −1
P 
 r W
2−c − 1
M
= 1 − P PE ≤ WE 1 + 2b 
1,
r W c ≥ CM .
6 7 M M
P
1 + r2b WM 2−c − 1 In addition, (12) and (13) reduce, respectively, to
= 1 − FPeE  M . $ %
−1 b P
(πλE C1/b ) WE π 2 −1
pexist = erfc √ λE rM C1/2 (15)
2
In addition, FCs (c) = 0 for c < 0 and FCs (c) = 1 for and
c ≥ CM , since the r.v. Cs in (2) satisfies 0 ≤ Cs ≤ CM ,
i.e., the secrecy capacity of the main link in the presence of poutage (Rs ) = (16)
colluding eavesdroppers is a positive quantity which cannot   
 >
be greater that the secrecy capacity of the main link in the 
erf  √π λ C −1 „ P
W«  , 0 < Rs < CM ,
2 E 1/2
absence of eavesdroppers. This is the result in (6) and the = 1+ 4P
2−Rs −1


r W
M
proof is complete. 
1, Rs ≥ CM .

V. E XISTENCE AND O UTAGE OF S ECRECY C APACITY
Based on the results of Section IV, we now obtain the
probabilities of existence and outage of the secrecy capacity,
in the presence of a Poisson field of colluding eavesdroppers.
The following corollary provides such probabilities.
Corollary 5.1: If ΠE is a Poisson process with density λE ,
Figures 3 and 4 quantify, respectively, the c.d.f. and p.d.f.
of the secrecy capacity Cs of the main link, for the considered
case study. We observe that Cs is a positive quantity which
cannot be greater that the secrecy capacity CM of the main
link in the 6absence of eavesdroppers, which in this case is
7
CM = log2 1 + r2bPWM = 3.46 bits per complex dimension.
M

the probability of existence of a non-zero secrecy capacity, Furthermore, as the eavesdropper density λE increases, the
pexist = P{Cs > 0}, is given by probability mass of Cs becomes more concentrated around
* + zero, in the sense that smaller realizations of Cs become more
WE likely. Similarly, the impulses of the p.d.f. at the origin, given
pexist = FPeE 2 C −1 )b W
, (12) by FCs (0)δ(c) (not represented in Fig. 4), also become larger
(πλE rM 1/b M
as λE increases.
and the probability of an outage in secrecy capacity, Figure 5 quantifies the secrecy outage probability poutage
poutage (Rs ) = P{Cs < Rs } for some target secrecy rate Rs > in (16) versus the eavesdropper density λE , for various values
0, is given by of P/W . We observe that as λE increases, an outage in
secrecy capacity of the main link becomes more likely,
poutage (Rs ) = (13)
 „ «  since there are more eavesdroppers that can exchange and

 1+ 2bP
2 −Rs
−1 combine the information, thus improving their ability to
 r WM
1 − FPeE  (πλM C −1  , 0 < Rs < CM ,
= b
E 1/b ) W
P decode the secret message. Furthermore, we observe that as
E

 P/W → ∞, poutage decreases
6 monotonically,7 converging to

1, Rs ≥ CM , the curve poutage = erf √π 2 −1 Rs /2
λ r C 2 .
6 7 2 E M 1/2
where CM = log2 1 + r2bPWM is the capacity of the main
M VII. C ONCLUSION
channel; and FPeE (·) is the c.d.f. of the normalized stable
We established the fundamental security limits when the
r.v. P(E , with parameters given in (8).
eavesdroppers are allowed to collude, by showing that this
Proof: The expressions for pexist and poutage (Rs ) follow
scenario is equivalent to a SIMO Gaussian wiretap channel.
directly from (6).
We derived the secrecy capacity of a legitimate link, consid-
ering that the positions of the illegitimate receivers follow an
VI. C ASE S TUDY arbitrary spatial process ΠE . Then, for the case where ΠE is
We now illustrate the results obtained in the previous a spatial Poisson process, we characterized the distribution of
sections with a simple case study. We consider the case the secrecy capacity, as well as the corresponding probabilities
where WM = WE = W , i.e., the main link and the of existence and outage. Perhaps the most interesting insight
eavesdroppers are subject to the same noise power, introduced to be gained from our results is the exact quantification of
by the electronics of the respective receivers. Furthermore, the impact of the eavesdropper density λE on the achievable
we consider that the amplitude loss exponent is b = 2, in secrecy rates — even a modest number of scattered attackers
2445
 5
ISIT 2009, Seoul, Korea, June 28 - July 3, 2009
1 1

0.9 0.9
P/W = 1.1
0.8 0.8
P/W = 1.5
0.7 0.7
P/W = ∞
0.6 0.6
FCs (c)

poutage
0.5 0.5

0.4 0.4

0.3 0.3

0.2 λE = 0.2 m−2 0.2

λE = 0.1 m−2
0.1 0.1
λE = 0.05 m−2
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.05 0.1 0.15 0.2 0.25 0.3
c (bits) λE (m−2)

Figure 3. C.d.f. FCs (c) of the secrecy capacity Cs of the main link, for Figure 5. Secrecy outage probability poutage versus the eavesdropper
various densities λE of eavesdroppers (b = 2, P/W = 10, rM = 1 m). density λE , for various values of P/W (b = 2, rM = 1 m, Rs = 1 bit).
The vertical line marks the capacity of the main link, which for these system
parameters is CM = 3.46 bits/complex dimension.

[6] M. Bloch, J. Barros, M. R. D. Rodrigues, and S. W. McLaughlin,

1
“Wireless information-theoretic security,” IEEE Trans. Inf. Theory,
λE = 0.2 m−2 vol. 54, no. 6, pp. 2515–2534, 2008.
0.9
λE = 0.1 m−2 [7] Z. Li, R. Yates, and W. Trappe, “Secrecy capacity of independent
0.8 λE = 0.05 m−2 parallel channels,” Proc. Annu. Allerton Conf. Communication, Control
and Computing, pp. 841–848, Sept. 2006.
0.7 [8] P. Gopala, L. Lai, and H. El Gamal, “On the Secrecy Capacity of Fading
Channels,” arxiv preprint cs.IT/0610103, 2006.
0.6 [9] S. Goel and R. Negi, “Secret communication in presence of colluding
fCs (c)

eavesdroppers,” in Proc. Military Commun. Conf., Oct. 2005, pp. 1501–

0.5
0.4
0.3
0.2
0.1
0 [13] R. Mathar and J. Mattfeldt, “On the distribution of cumulated interfer-
0 0.5 1 1.5 2 2.5 3 3.5
c (bits)
Figure 4. P.d.f. fCs (c) of the secrecy capacity Cs of the main link, for var-
ious eavesdropper densities λE (b = 2, P/W = 10, rM = 1 m). Note that
the Dirac impulses at the origin, given by FCs (0)δ(c), are not plotted in the
figure. For the chosen values of λE , we have FCs (0)|λE ={0.05,0.1,0.2} =
{0.1561, 0.3062, 0.5690}.
can dramatically reduce the levels of security that can be
provided at the physical layer of a wireless communication
network.
R EFERENCES
[1] C. E. Shannon, “Communication theory of secrecy systems,” Bell
System Technical Journal, vol. 29, pp. 656–715, 1949.
[2] A. D. Wyner, “The Wire-Tap Channel,” Bell System Technical Journal,
vol. 54, no. 8, pp. 1355–1367, October 1975.
[3] R. Ahlswede and I. Csiszar, “Common randomness in information
theory and cryptography - Part I: Secret sharing,” IEEE Trans. Inf.
Theory, vol. 39, no. 4, pp. 1121–1132, July 1993.
[4] A. Hero, “Secure space-time communication,” IEEE Transactions on
Information, vol. 49, no. 12, pp. 3235–3249, Dec. 2003.
[5] P. Parada and R. Blahut, “Secrecy capacity of SIMO and slow fading
channels,” in Proc. IEEE Int. Symp. on Inf. Theory, Adelaide, Australia,
Sept. 2005, pp. 2152–2155.
1506.
[10] P. C. Pinto, J. Barros, and M. Z. Win, “Physical-layer security in
stochastic wireless networks,” in Proc. IEEE Int. Conf. on Commun.
Systems, Guangzhou, CHINA, Nov. 2008, pp. 974–979.
[11] M. Haenggi, “The secrecy graph and some of its properties,” in Proc.
IEEE Int. Symp. on Inf. Theory, Toronto, Canada, July 2008.
[12] J. Silvester and L. Kleinrock, “On the capacity of multihop slotted
ALOHA networks with regular structure,” IEEE Trans. Commun.,
vol. 31, no. 8, pp. 974–982, Aug. 1983.
4
ence power in Rayleigh fading channels,” Wireless Networks, vol. 1,
pp. 31–36, Feb. 1995.
[14] J. Kingman, Poisson Processes. Oxford University Press, 1993.
[15] P. C. Pinto and M. Z. Win, “Communication in a Poisson field of
interferers - Part I: Interference distribution and error probability,” IEEE
Trans. Wireless Commun., 2009, accepted pending revision.
[16] ——, “Communication in a Poisson field of interferers - Part II: Channel
capacity and interference spectrum,” IEEE Trans. Wireless Commun.,
2009, accepted pending revision.
[17] C. Bettstetter and C. Hartmann, “Connectivity of wireless multihop
networks in a shadow fading environment,” Wireless Networks, vol. 11,
no. 5, pp. 571–579, Sept. 2005.
[18] D. Miorandi and E. Altman, “Coverage and connectivity of ad hoc
networks in presence of channel randomness,” in Proc. IEEE Conf. on
Computer Commun., vol. 1, Mar. 2005, pp. 491–502.
[19] M. Haenggi and D. Puccinelli, “Routing in ad hoc networks: A case
for long hops,” IEEE Commun. Mag., vol. 43, no. 10, pp. 93–101, Oct.
2005.
[20] T. Q. S. Quek, D. Dardari, and M. Z. Win, “Energy efficiency of dense
wireless sensor networks: To cooperate or not to cooperate,” IEEE J.
Sel. Areas Commun., vol. 25, no. 2, pp. 459–470, Feb. 2007.
[21] A. Goldsmith, Wireless Communications. Cambridge University Press,
2005.
[22] D. Tse and P. Viswanath, Fundamentals of Wireless Communication.
Cambridge University Press, 2005.
[23] S. Leung-Yan-Cheong and M. Hellman, “The Gaussian wire-tap chan-
nel,” IEEE Trans. Inf. Theory, vol. 24, no. 4, pp. 451–456, July 1978.
[24] G. Samoradnitsky and M. Taqqu, Stable Non-Gaussian Random Pro-
cesses. Chapman and Hall, 1994.
2446