Sie sind auf Seite 1von 8

3372 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO.

7, JULY 2009

A Distributed Differentially Encoded OFDM Scheme for


Asynchronous Cooperative Systems with Low Probability of Interception
Zheng Li and Xiang-Gen Xia

Abstract—Recently, Li, Hwu and Ratazzi have proposed a binary coding. In wireless security, one practical objective
physical-layer security design to guarantee low probability of is to design a wireless transmission with low probability of
interception (LPI) for MIMO systems without relying on upper- interception (LPI). Recently, an innovative method has been
layer data encryption. The proposed scheme utilizes antenna
array redundancy to deliberately randomize the transmitted proposed by Li, Hwu and Ratazzi to achieve LPI without
signals to prevent eavesdropping. Motivated by their idea, in this relying on upper-layer data encryption for MIMO systems [6],
paper we design a physical-layer transmission scheme to achieve [7]. The proposed method utilizes antenna array redundancy
LPI in cooperative systems. There are two major differences to deliberately randomize the transmitted signals to prevent
in cooperative systems: 1) each relay node may have only one eavesdropping. In this method, training signals are not trans-
antenna that can not provide antenna array redundancy for
signal randomization; 2) there may exist timing errors due to mitted from the transmitter but from the authorized receiver,
the asynchronous nature of cooperative systems. Considering the thus the transmitter knows the channel while the authorized
two differences, we propose a distributed differentially encoded receiver does not. On the other hand, the eavesdropper may
OFDM transmission scheme with deliberate signal randomization use multiple antennas to obtain better received signal quality
to prevent eavesdropping and exploit the available spatial and than the authorized receiver does and knows all the trans-
frequency diversities in asynchronous cooperative systems. We
use diagonal unitary codes to perform the differential encoding mission protocols. For this reason, the scheme in [6], [7]
in the frequency domain over subcarriers within each OFDM uses a deliberate signal randomization technique to randomize
block, or we use general (not necessarily diagonal) unitary the eavesdropper’s received signals but not the authorized
codes to perform the differential encoding in the frequency receiver’s received signals. Thus, the eavesdropper can neither
domain across several OFDM blocks. By some deliberate signal implement channel estimation [8] for non-blind detection
randomization, the eavesdropper can not detect the transmitted
symbols, while the authorized receiver can perform differential without training signals nor implement blind deconvolution
decoding successfully without the knowledge of the channels or [9] due to the randomized received signals.
the timing errors. In this paper, we are interested in designing a physical-layer
Index Terms—Differential space-time-frequency coding, asyn- transmission scheme to achieve LPI in cooperative systems.
chronous cooperative diversity, unitary codes, OFDM, wireless We propose a deliberate signal randomization technique to
security, low probability of interception.
randomize the eavesdropper’s received signals but not the
authorized receiver’s received signals. However, in cooperative
I. I NTRODUCTION systems, there are two major differences from MIMO systems:

W IRELESS communication security has become more


and more important in recent years due to the wide
wireless communications applications [1], [2]. Unlike wired
1) each relay node may not afford more than one antenna,
thus there may be no antenna array redundancy that can be
employed to implement the deliberate signal randomization;
systems, in wireless communications, any receiver within the 2) there may exist timing errors due to the distributed nature
transmission range can receive the signals and listen/analyze of the relay nodes, i.e., the signals from different relay
the transmission, which makes wireless physical-layer security nodes arrive at the authorized receiver or the eavesdropper
more vulnerable. Although the physical-layer may rely on at different times [10]. These two differences may make the
upper-layer data encryption techniques for security [3], it physical-layer security design in cooperative systems more
is interesting to design wireless physical-layer transmission difficult.
schemes with inherent security which may be helpful and can In this paper, we propose a distributed differentially encoded
be integrated with the traditional data encryption approaches. OFDM transmission scheme with deliberate signal random-
Note that there recently have been a lot of studies on physical- ization to achieve LPI as well as the available spatial and
layer security more from information theoretic perspective, frequency diversities in asynchronous cooperative systems. In
see for example [4], [5]. This paper is, however, more on our proposed scheme, we use OFDM to combat frequency-
signal modulation and processing perspective in stead of selective fading channels and timing errors. The newly pro-
Manuscript received March 16, 2008; revised June 20, 2008 and September posed deliberate signal randomization and the differential
17, 2008; accepted December 7, 2008. The associate editor coordinating the encoding can be performed in the frequency domain over
review of this letter and approving it for publication was H. Shin. subcarriers within each OFDM block using diagonal unitary
The authors are with the Department of Electrical and Computer En-
gineering, University of Delaware, Newark, DE 19716 (e-mail: {zhli, codes (space-frequency coding). Moreover, if channels do not
xxia}@ee.udel.edu). change over several OFDM blocks, i.e., slow fading channels,
This work was supported in part by the Air Force Office of Scientific the differential encoding in the frequency domain can be done
Research (AFOSR) under Grant No. FA9550-08-1-0219 and the National
Science Foundation under Grant CCR-0325180. across several OFDM blocks using general (not necessar-
Digital Object Identifier 10.1109/TWC.2009.080365 ily diagonal) unitary codes (space-time-frequency coding) to
1536-1276/09$25.00 
c 2009 IEEE

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:38 from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 7, JULY 2009 3373

the destination nodes is that we only consider the case when


R1 the distance between the source and the destination nodes is
far enough so that the destination node cannot receive the
message sent from the source. In this paper, we also assume
that the eavesdropper can be anywhere but far enough from the
source node. When the eavesdropper is close to the source, we
may choose a direct communication between source and the
receiver and use Li, Hwu and Ratazzi’s method [6], [7] directly
RJ to achieve LPI when the source and the receiver are close to
each other, and we may use Li, Hwu and Ratazzi’s method
[6], [7] in the first phase of cooperative communications when
the source and the receiver are far away from each other, and
Fig. 1. Cooperative system architecture.
in the second phase, we use our proposed scheme.
Each relay node employs an OFDM modulator with N
subcarriers. We assume the frequency-selective channels have
obtain better performance. By the deliberate signal random-
L paths. Without timing errors, in the frequency domain, the
ization, the eavesdropper’s received signals change randomly
received signals at the kth (0 ≤ k ≤ N − 1) subcarrier at the
on each subcarrier. Thus, the eavesdropper may not be able
authorized receiver D in one OFDM block can be written as:
to implement a blind deconvolution or differential decoding.
J
Meanwhile, the authorized receiver can implement differential √ 
decoding successfully to obtain full diversity with neither the Y (k) = ρ Xi (k)Hi,D (k) + Z(k) (1)
i=1
knowledge of the channels nor the timing errors. Most of
the literatures on combating timing errors have to assume where ρ is the average received signal to noise ratio, Xi (k)
that the receiver has perfect knowledge of the timing errors, is the transmitted symbol over the kth subcarrier from the ith
the proposed distributed differentially encoded OFDM scheme relay node, Hi,D (k) is the channel frequency response at the
can guarantee transmission security without the knowledge of kth subcarrier between the ith relay node and D, and Z(k) is
timing errors, which is the unique property of the method. the AWGN with zero mean and unit variance.
The remaining of this paper is organized as follows. In The channel frequency response Hi,D (k) can be given by
Section II, the system model is described. In Section III, two L−1
distributed differentially encoded OFDM systems with deliber- 
Hi,D (k) = αi,D (l)e−j2πkΔf τi,D (l)
ate signal randomization are proposed. Distributed differential l=0
diagonal space frequency coding and distributed differential
space-time-frequency nondiagonal coding are discussed sepa- where αi,D (l) is the multipath channel coefficient of the
rately. In Section IV, simulation results are presented. Finally, lth path, τi,D (l) is the lth path delay, Δf = 1/T is the
the conclusions are given in Section V. subcarrier spacing, T is the OFDM symbol period. Each
channel coefficient αi,D (l) is modelled as zero mean, complex
Gaussian random variable with variance E|αi,D (l)|2 = δl2 ,
II. S YSTEM M ODEL
and for convenience,
L−1 the power of the L paths are normalized
Consider a cooperative system with one source node S and such that l=0 δl2 = 1.
J relay nodes Ri (1 ≤ i ≤ J), and we denote the authorized When there exist timing errors, we assume the timing error
receiver as D, and there exists a passive eavesdropper E, as between the ith relay and D is τi,D , which is rounded to the
shown in Fig. 1. Consider the case when every node in the nearest sampling position. If the length of the cyclic prefix
system has only one antenna except the eavesdropper, i.e., (CP) is not less than maxi,l {τi,D (l) + τi,D }, we can still
E may have multiple antennas. To transmit the information maintain the orthogonality between the subcarriers. The delay
from the source node S to the authorized receiver D, there τi,D in the time domain corresponds to a phase change in the
undergo two phases. In the first phase, the source node S frequency domain:
broadcasts the information to the J relay nodes. Meanwhile,
the J relay nodes receive the information. During the second f τi,D = [1, e−j2πτi,D /N , . . . , e−j2πτi,D (N −1)/N ]T
phase, the J relay nodes Ri send the received signals to the
with f = [1, e−j2π/N , . . . , e−j2π(N −1)/N ]T .
authorized receiver D. In this paper, we adopt the decode-and-
Now the received signals with timing errors at the kth (0 ≤
forward (DF) protocol, i.e., the J relay nodes have correctly
k ≤ N − 1) subcarrier at the authorized receiver D can be
detected the information symbols in the first phase. Thus, we
written as:
only consider the transmission in the second phase. Due to
J
the lack of physical boundary, we know that the eavesdropper √ 
E can also receive the signals transmitted from the J relay Y (k) = ρ Xi (k)Hi,D (k)f τi,D (k) + Z(k) (2)
i=1
nodes.
Note that the security for the communication between the where f τi,D (k) = exp(−j2πkτi,D /N ).
source node and relay nodes is equally important. In this Now consider the eavesdropper E, E may use multiple
paper, this problem is omitted because the reason of utilizing antennas for better interception. Therefore, here we assume
relay nodes to help the communication between the source and that E has M antennas. The received signals with timing errors

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:38 from IEEE Xplore. Restrictions apply.
3374 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 7, JULY 2009

at the kth (0 ≤ k ≤ N − 1) subcarrier at the mth (1 ≤ m ≤ M) received signals are randomized to prevent eavesdropping.
antenna of the eavesdropper E can be written as: We first describe the distributed differentially space frequency
J coded scheme using diagonal unitary codes, and then we
√  design the random weighting coefficients at the relay nodes.
Ym (k) = ρ Xi (k)Hi,m (k)f τi,m (k) + Zm (k) (3)
i=1
Moreover, we use general (not necessarily diagonal) unitary
codes to construct the distributed differentially space-time-
where Hi,m (k) is the channel frequency response at the kth frequency coded scheme to achieve better performance.
subcarrier between the ith relay node and the mth antenna of
E, and Zm (k) is the AWGN with zero mean and unit variance.
Similarly, Hi,m (k) can be specified as
L−1
 A. A Distributed Differentially Space Frequency Coded
Hi,m (k) = αi,m (l)e−j2πkΔf τi,m (l)
Scheme Using Diagonal Unitary Codes
i=0

where αi,m (l) and τi,m (l) are the multipath channel coefficient Our distributed differentially space frequency coded scheme
of the lth path and the delay of the lth path, respectively. And is based on the space frequency codes in [14], [15] and
τi,m is the timing error between the ith relay and the mth the corresponding differential scheme in [16]. The major
antenna of E. Note that, for equation (3), it has assumed that difference of the scenario here from [16] is that the un-
the length of CP is not less than maxi,l,m {τi,m (l) + τi,m }. known channel at the receiver D is caused by our deliberate
From the above equation, it is obvious that E can try randomized weighting coefficients and the unknown timing
to detect the transmitted symbols based on better received errors, which motivates our following new transceiver design.
signal quality. In the following section, we will design a In the following, we assume the frequency-selective channels
physical-layer secure transmission scheme to prevent E from keep unchanged within each OFDM block, and may change
eavesdropping under the assumption that the relay nodes and independently from one OFDM block to another OFDM block,
D share no additional information such as secret keys [3] i.e., it may be fast fading.
beforehand and know nothing about the eavesdropper E.
Notice that in the frequency domain phase change vector

III. T HE D ISTRIBUTED D IFFERENTIALLY E NCODED


Δ
OFDM WITH D ELIBERATE S IGNAL R ANDOMIZATION f τ = [f τ (0), f τ (1), · · · , f τ (N − 1)]T
In order to prevent E from eavesdropping, we can, for = [1, e−j2πτ /N , . . . , e−j2πτ (N −1)/N ]T ,
example, create high BER for E. Thus, as in [6], [7] for
MIMO systems, in the cooperative system, the relay nodes
can not send training signals, otherwise, E can easily utilize there exists a relationship of f τ (k) = (f τ (N − k))∗ , for k =
the training signals for channel estimation and symbol de- 1, ..., N2 − 1. But when k = 0 and k = N2 , this relationship of
tection. However, without training, D also can not have the conjugation does not always hold. We will make use of this
channel state information (CSI) including the channel fading conjugation property in the design to deal with the unknown
coefficients and the timing errors. Thus, we have to design a timing errors. Discarding the 0th and the N2 th subcarriers, we
transmission scheme so that D can detect the symbols without partition the remaining N −2 subcarriers into 2P groups where
the CSI. P = (N − 2)/(2ΓJ), 1 ≤ Γ ≤ L, i.e., the largest integer
In the following proposed schemes, the same as [6], [7] we not greater than (N − 2)/(2ΓJ), where J is the number of
require that the authorized receiver D transmits the training relay nodes involved and Γ is a parameter used in the space
signals to the relay nodes, therefore, the relay nodes know frequency coding to achieve the multipath diversity [14]–[16].
the corresponding channel according to channel reciprocity Thus, each group includes ΓJ subcarriers.
[12], [13], and this procedure gives no useful information In our distributed differentially space frequency coded
to E. Note that, even though the relay nodes can have the scheme, we use half of the useful subcarriers, i.e., the first
knowledge of the corresponding channel, it is hard for each P groups, to send information symbols, and the other half of
relay node to know its relative timing delay, which means that the useful subcarriers, i.e., the second P groups, to send the
in the cooperative system, no one can know the timing errors. corresponding reference symbols. Denote Bp = (p − 1)ΓJ,
Another kind of synchronization errors may be the multiple p = 1, 2, ..., P . Specifically, for relay node i, we send
frequency offsets from the relay nodes. Because the authorized information symbols Xi (k) on subcarrier k, where k =
receiver will send training signals to the relay nodes, each Bp + (i − 1)Γ + γ and Xi (k) = 0 for other k; and we send
relay node can estimate its frequency offset and therefore the corresponding reference symbols Xi (k) on subcarrier k,
may be able to compensate it, which is different from the where k = N − (Bp + (i − 1)Γ + γ) and Xi (k) = 0 for
conventional relay systems. Thus, we omit this problem in other k, in which 1 ≤ p ≤ P , 1 ≤ i ≤ J and 1 ≤ γ ≤ Γ.
this paper. We can see that in this scheme, on each useful subcarrier,
In the following, we propose two transmission schemes with there is only one relay node that sends nonzero symbols. This
random weighting coefficients at the relay nodes. With our scheme can be represented by a matrix form and a distributed
proposed scheme, D can implement the detection successfully differential space frequency code as follows. For group p in the
through differential decoding without the CSI, while E’s first P groups, let Xp be the ΓJ × ΓJ signal matrix including

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:38 from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 7, JULY 2009 3375

information symbols and it has a diagonal form: From the conjugation property, we have fτpD = (fτ D ∗
p ) , thus
⎡ ⎤ we can rewrite (5) as
X1 (Bp + 1)
⎢ .. ⎥ √
⎢ . ⎥ Y∗
p = ρIΓJ W∗ ∗ τD ∗
p ◦ Hp ◦ fp + Zp , p = 1, 2, ..., P (6)
⎢ ⎥
⎢ X (B p + Γ) ⎥
⎢ 1 ⎥ From (4) and (6), if Wp ◦ Hp = W∗ ∗
⎢ .. ⎥ p ◦ Hp , we have
Xp = ⎢ . ⎥,
⎢ ⎥
⎢ XJ (Bp + (J − 1)Γ + 1) ⎥ Yp = Xp Y∗
p + Z̃p (7)
⎢ ⎥
⎢ .. ⎥
⎣ . ⎦
where Z̃p = Zp − Xp Z∗ p is a noise vector where each entry
XJ (Bp + (J − 1)Γ + Γ) is a Gaussian random variable with mean zero and variance

where Xi (k) is what is transmitted for the ith relay at the kth 2. Since Xp = IΓJ , so Xp = Vp . Without the CSI, the
subcarrier and otherwise there is no tranmission. For group maximum-likelihood (ML) decoding can be performed as
p in the second P groups, the transmission matrix Xp is a
ΓJ × ΓJ signal matrix including the corresponding reference
symbols. Xp also has a diagonal form which can be expressed X̂p = argmin ||Yp − Xp Y∗ 2
p ||F . (8)
Xp ∈Λ
as
⎡  ⎤ We now look at the condition Wp ◦ Hp = W∗ ∗
p ◦ Hp we
X1 (N − (Bp + 1))
⎢ .. ⎥ made for the ML decoding in (8), which is, in fact, easy to
⎢ . ⎥ satisfy since each relay node knows its own channel in the
⎢ ⎥
⎢ X 
(N − (B + Γ)) ⎥ scenario in this paper. Since its own channel is known at a
⎢ 1 p ⎥
 ⎢ . .. ⎥ relay node, the relay node can then design random weighting
Xp = ⎢ ⎥
⎢ ⎥ coefficients so that Wp ◦ Hp = W∗ ∗
⎢ 
XJ (N − (Bp + (J − 1)Γ + 1)) ⎥ p ◦ Hp for the authorized
⎢ ⎥ user D, while the received signals are changing randomly on
⎢ .. ⎥
⎣ . ⎦ each subcarrier in each OFDM block for the eavesdropper E
 as we shall explain below.
XJ (N − (Bp + (J − 1)Γ + Γ))

Now we encode the transmitted signals differentially in the


frequency domain within each OFDM block as follows: Xp = B. A Deliberate Signal Randomization at the Relay Nodes
Vp Xp , and Vp ∈ Λ carries the transmitted information and
is a unitary diagonal matrix. As in [16], the space frequency With the weighting coefficients, our goal is to make
coding can achieve a diversity order of ΓJ. Note that, the wi (k)Hi,D (k) = wi∗ (N − k)Hi,D ∗
(N − k) for D so that D
differential space frequency coding scheme causes rate loss, can perform the differential decoding, while wi (k)Hi,m (k) =
the symbol rate is P ΓJ/N ≈ 12 when N is large. wi∗ (N − k)Hi,m ∗
(N − k) and wi (k)Hi,m (k), wi∗ (N −

Assume the weighting coefficient for the ith relay node on k)Hi,m (N − k) change randomly for E so that E has no way
the useful subcarrier k as wi (k). We can form the received to intercept the information symbols.
signals in a vector-matrix form as follows. We can first choose wi (k) randomly, then wi∗ (N − k) =

For the subcarriers sending information symbols, the re- (wi (k)Hi,D (k))/Hi,D (N − k). Since Hi,D (k) and Hi,m (k)
ceived signal model at D is: are independent fading coefficients, it is almost surely to
have wi (k)Hi,m (k) = wi∗ (N − k)Hi,m∗
(N − k). Therefore,

Yp = ρXp Wp ◦ Hp ◦ fτpD + Zp , p = 1, 2, ..., P (4) in general, our goal can be satisfied by simply choosing
wi (k) to have a specific distribution. However, choosing
where Wp is a ΓJ × 1 weighting vector to be designed later, wi (k) randomly may cause the total transmission power to
whose ((i−1)Γ+γ)th entry is wi (Bp +(i−1)Γ+γ), Hp is the be increased. Thus, we are also interested in preserving
ΓJ × 1 vector of channel coefficients, whose ((i − 1)Γ + γ)th the total transmission power to be unchanged, i.e., the total
entry is Hi,D (Bp + (i − 1)Γ + γ), fτpD is also a ΓJ × 1 vector transmission power is
whose ((i − 1)Γ + γ)th entry is f τi,D (Bp + (i − 1)Γ + γ),
J

Zp is the corresponding noise vector, and ◦ is the Hadamard
product, i.e., the component-wise product. E( |wi (k)|2 ) = Nu (9)
Nu i=1
For the subcarriers sending reference symbols, the received
signal model at D is: where Nu is the number of useful subcarriers. In order to
√ achieve this requirement, we can design the random weighting
Yp = ρXp Wp ◦ Hp ◦ fτ
p
D
+ Zp , p = 1, 2, ..., P (5)
coefficients as follows:
where Xp = IΓJ as the reference symbols. Wp is also 1) To preserve total transmission power, we design the
a weighting vector with its ((i − 1)Γ + γ)th component amplitude of the random coefficients first and the following
wi (N − (Bp + (i − 1)Γ + γ)), the ((i − 1)Γ + γ)th entry two requirements have to be satisfied simultaneously:
of Hp is Hi,D (N − (Bp + (i − 1)Γ + γ)), the ((i − 1)Γ + γ)th

entry of fτ D
is f τi,D (N − (Bp + (i − 1)Γ + γ)), Zp is the |wi (k)||Hi,D (k)| = |wi∗ (N − k)||Hi,D

(N − k)| = C
p
corresponding noise vector. |wi (k)|2 + |wi∗ (N − k)|2 = 2.

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:38 from IEEE Xplore. Restrictions apply.
3376 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 7, JULY 2009

Thus, we have C. A Distributed Differentially Space-Time-Frequency Coded


⎧ √ ∗ Scheme Using Nondiagonal Unitary Codes
⎪ 2|Hi,D (k)||Hi,D (N − k)|

⎪ C =  In the previous study, we proposed to use differential

⎨ ∗ (N − k)|2 )
(|Hi,D (k)|2 + |Hi,D space frequency coding across subcarriers when the channels

⎪ |wi (k)| = C/|Hi,D (k)| do not change in one OFDM block, where the differen-


⎩ ∗ ∗ tial space frequency codes are diagonal unitary codes. In
|wi (N − k)| = C/|Hi,D (N − k)|
this subsection, we propose to use differential space-time-
2) For the phases of the random coefficients, we can simply frequency coding across OFDM blocks, where the differential
choose the phase of wi (k) uniformly distributed: wi (k) = space-time-frequency codes may be general (not necessarily
|wi (k)|ejθwi (k) , θwi (k) ∼ U (0, 2π), which is unknown to E. diagonal) unitary codes. In this case, we assume that the
Then, we obtain the phase of wi (N − k) as: θwi (N −k) = channels do not change over J OFDM blocks. For more details
−θHi (N −k) − θwi (k) − θHi (k) . on space frequency and space-time-frequency coding, we refer
Thus, we have designed random coefficients wi (k) and the reader to a tutorial paper [17].
wi (N − k) to make wi (k)Hi (k) = wi∗ (N − k)Hi∗ (N − k) We will use the block repetition coding method to construct
for D as well as preserving the total transmission power. The the nondiagonal unitary codes. Similarly as in Subsection
authorized user D can detect the symbols according to (8). III-A, we partition the useful N −2 subcarriers into 2P groups
On the other hand, if E also tries to detect the symbols by where P = (N − 2)/(2Γ), 1 ≤ Γ ≤ L. We also denote
the differential decoding, then from Bp = (p − 1)Γ, p = 1, 2, ..., P .
√ For group p in the first P group, letting Xp be the ΓJ × ΓJ
Ymp = ρXp Wp ◦ Hm τm m
p ◦ fp + Z p , (10) signal matrix including information symbols, we can express

m∗ √  ∗ 
m∗ τm 
m∗
Yp = ρXp Wp ◦ Hp ◦ fp + Zp (11) Xp as follows:

we can obtain: Xp = IΓ ⊗ X̃p = diag[X̃p , . . . , X̃p ] (14)


 √ m
Ym
p = Xp Yp
m∗
+ ρXp Δp ◦ fτm + Z̃p (12) where ⊗ denotes the Kronecker product and diag[ ] produces a

block diagonal (BD) matrix from the matrices in its argument.
where Δp = Wp ◦ Hm ∗
p − Wp ◦ Hp
m∗
is the channel difference X̃p is a J × J unitary matrix which can be expressed as
which is determined by the difference between wi (k)Hi,m (k) ⎡ p p p ⎤
and wi∗ (N − k)Hi,m

(N − k). The differential decoding for E X1,1 X2,1 . . . XJ,1
p p
⎢ X1,2 X2,2 . . . XJ,2 p ⎥
can be written as: ⎢ ⎥
X̃p = ⎢ . .. .. .. ⎥
⎣ .. . . . ⎦
X̂p = argmin ||Ym

m∗ 2
p − Xp Yp ||F (13) p p p
Xp ∈Λ X1,J X2,J . . . XJ,J
we can see that the decoding performance of E mainly depends where Xm,np
denotes the transmitted signal from the mth relay
on ρE||Δp ||2F . Since wi (k)Hi,m (k) = wi∗ (N − k)Hi,m

(N − node in the nth OFDM block in group p, in which 1 ≤ m, n ≤
2
k), ρE||Δp ||F can be large. Thus E can hardly detect the J. We can notice that X̃p is a unitary space time code and we
symbols by the differential decoding according to (13). repeat it across the subcarriers to construct the BD code Xp ,
However, E may try to implement blind deconvolution to thus Xp is a space-time-frequency code, which is different
detect Xp and Xp directly from the received signals on each from the space frequency code in Subsection III-A. In order
subcarrier, and then using Xp = Vp Xp to intercept the infor- to get BD codes with better performance than the diagonal
mation. Since wi (k)Hi,m (k), wi∗ (N − k)Hi,m ∗
(N − k) change unitary codes, we will choose X̃p as the unitary codes with
randomly on each subcarrier in each OFDM block and the larger coding gain (diversity product), such as the codes in
distribution of wi (k) is unknown to E and there exist unknown [18], [19].
phase change caused by timing errors, as proved in [6], [7], we Assume the weighting coefficient for the ith relay node on
conclude that E can not implement blind deconvolution while subcarrier k as wi (k). Now we can give the received signal
it is well known that successful blind deconvolution requires model at D. For the subcarriers sending information symbols,
the receiver knows some special and sufficient statistics or the received signal model at D is:
structure of the signals [9]. Thus, we can guarantee LPI by √
the proposed deliberate signal randomization. Yp = ρ(Xp Wp Hp ) ◦ fτpD + Zp , p = 1, 2, ..., P (15)
Note that, when designing the random coefficients, we where Wp is a ΓJ ×ΓJ diagonal matrix which can be designed
introduce random phases according to the uniform distribution, similarly as in Subsection III-B, whose (i + J(γ − 1))th
the random phases can help to keep the peak-to-average power diagonal entry is wi (Bp + γ), Hp is the ΓJ × 1 vector
ratio (PAPR) of the scheme in an acceptable level even we of channel coefficients, whose (i + J(γ − 1))th entry is
simply send 1 as the reference symbol on half of the useful Hi,D (Bp +γ), fτpD is also a ΓJ ×1 vector whose (i+J(γ−1))th
subcarriers. However, the proposed design of the random entry is f τi,D (Bp + γ).
coefficients focuses primarily on preserving total transmission For the subcarriers sending reference symbols, we still use
power, if considering other aspects, such as obtaining the Xp = IΓJ as the reference symbols, then the received signal
best performance of preventing eavesdropping or reducing the model at D can be written as:
PAPR of the scheme, there might exist better design than the

proposed one here. Y∗
p = ρ(IΓJ W∗ ∗ τD ∗
p Hp ) ◦ fp + Zp , p = 1, 2, ..., P (16)

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:38 from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 7, JULY 2009 3377

where the parameters are similarly defined as in (15). 10

Similar to Subsection III-A, if Wp Hp = W∗ ∗


p Hp , from (15)
and (16), we can perform the ML differential decoding as

X̂p = argmin ||Yp − Xp Y∗ 2


−1

p ||F . (17)
10

Xp ∈Λ

We can see that (17) has the same form as (8) except
we utilize the nondiagonal unitary matrix Xp to carry the −2

BER
10
transmitted information. The received signal model and the
differential decoding for E can be similarly obtained as in
(10)-(13).
−3
10
E, Γ=1, 0.48 bits/s/Hz
IV. S IMULATION R ESULTS E,
D,
Γ=2, 0.47 bits/s/Hz
Γ=1, 0.48 bits/s/Hz
D, Γ=2, 0.47 bits/s/Hz
In this section, we present some simulation results to show
the effectiveness of the proposed transmission schemes by 10
−4

0 2 4 6 8 10 12 14 16 18 20
comparing the BERs of D and E. In the simulations, we ρ(dB)

assume that E directly uses D’s decoding method, i.e., (13), so


that the BER of E mainly depends on ρE||Δp ||2F , where ρ is Fig. 2. BER comparison between D and E with different Γ using diagonal
the averaged signal to noise ratio at the receiver. For simplicity, code.
we assume that the number of the relay nodes is 2, and E has
2 antennas. Each OFDM modulator has N = 128 subcarriers
with the total bandwidth of 1 MHz. The corresponding OFDM group. Thus the similarity of adjacent subcarriers will reduce
block interval is T = 1/Δf = 128μs. The length of CP is 32 ρE||Δp ||2F . Moreover, although the decoding performance of
μs for combating intersymbol interferences caused by channel E mainly depends on ρE||Δp ||2F , the specific code used in
multipath and the timing errors. We consider a simple two-ray the scheme also has an influence on the performance. When
equal power delay profile (L = 2) with a delay of τ = 20μs ρE||Δp ||2F is large, the decoding performance of E is poor,
between the two rays. The timing errors τ are chosen randomly however, the G4,16 code with diversity order 4 can still obtain
from 0 to 10 μs with the uniform distribution. a slightly better performance than the G2,4 code with only
We first show the performance of using diagonal unitary diversity order 2. Thus, when using G4,16 code, E’s BER
codes. The diagonal unitary codes we use in the simulations is slightly lower than 0.5. However, it still guarantees a low
are the cyclic group codes designed in [20], [21]. Specif- probability of interception.
ically, for an M × M unitary matrix and a transmission Then we increase the information bit rate to compare the
rate R such that the group size L = 2RM , the set of performance between D and E. We only consider the case
parameters used to fully specify Vp is denoted as GM,L = of Γ = 1 for simplicity. In the simulation, we use cyclic
(M, L, [u1 , u2 , ..., uM ]) [20]. In the simulations, the fading group code G2,8 = (2, 8, [1, 3]) in which each signal matrix
channels are assumed to be constant within each OFDM block represents 3 information bits and G2,16 = (2, 16, [1, 7]) in
and may change independently from one OFDM block to which each signal matrix represents 4 information bits. Thus,
another. the information bit rate is 0.73 bits/s/Hz and 0.97 bits/s/Hz,
We investigate the effect of varying Γ to the performance of respectively. As shown in Fig. 3, we can see that in both cases,
D and E first. When Γ = 1, we choose the cyclic group code D can implement the differential decoding successfully while
G2,4 = (2, 4, [1, 1]), in which each signal matrix represents 2 E’s BER is 0.5.
information bits, i.e., the information bit rate is 0.48 bits/s/Hz Now we give the performance of using nondiagonal unitary
(omitting the CP). When Γ = 2, we choose the cyclic group codes and compare the performance with the diagonal codes.
code G4,16 = (4, 16, [1, 3, 5, 7]), in which each signal matrix The nondiagonal unitary codes we use in the simulations are
represents 4 information bits, i.e., the information bit rate is the parametric codes proposed in [19]. The fading channels
0.47 bits/s/Hz. In either case, the P groups which are sending are assumed to be constant within two OFDM blocks in the
information symbols transmit P information bits. When Γ = 1, simulations.
we can see from Fig. 2 that D can detect the symbols We first compare the performance when Γ = 1. For the
successfully. However, E’s BER is 0.5 even with 2 antennas, diagonal code, we choose the cyclic group code G2,32 =
which means E can not do better than guessing. When Γ = 2, (2, 32, [1, 7]) in which each signal matrix represents 5 infor-
it is obvious that D’s diversity order increases especially when mation bits and this code is the optimal cyclic group code
ρ is large. However, E’s BER is slightly lower than 0.5, it at the rate 2.5. For the nondiagonal code, we use the 2 × 2
can be explained as follows: if we do not exploit frequency parametric code V(30, 6, 0) which has a signal constellation
diversity (Γ = 1), then each relay will use only one subcarrier of size 32, i.e., the rate is also 2.5, denoted as P2,32 in the
to perform encodng/decoding in the same group p, in this simulation. Thus, the information bit rate is 1.23 bits/s/Hz for
case, the channel coefficients on different subcarriers within both cases. From Fig. 4, we can see the nondiagonal code
the same group are independent. However, when making outperforms the diagonal code about 2dB when ρ is large at
use of frequency diversity (Γ > 1), each relay will use the same rate. And for E, they all guarantee LPI as E’s BER
adjacent subcarriers to perform encodng/decoding in the same is 0.5 in both cases.

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:38 from IEEE Xplore. Restrictions apply.
3378 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 7, JULY 2009

0
10 0.61 bits/s/Hz
0
10

−1
10 −1
10
BER

−2
10

BER
−2
10

−3
10 E, G2,8, 0.73 bits/s/Hz −3
10 E, G4,32
E, G2,16, 0.97 bits/s/Hz
E, P2,32
D, G2,8, 0.73 bits/s/Hz
D, G4,32
D, G2,16, 0.97 bits/s/Hz
D, P2,32

−4
10 −4
0 5 10 15 20 25 10
ρ(dB) 0 2 4 6 8 10 12 14 16 18 20
ρ(dB)

Fig. 3. BER comparison between D and E with higher information bit rate
Fig. 5. BER comparison between diagonal code and nondiagonal code when
using diagonal code.
Γ = 2.

1.23 bits/s/Hz
V. C ONCLUSION
0
10

In this paper, we have proposed two distributed differen-


tially encoded OFDM transmission schemes with deliberate
−1
10 signal randomization to guarantee LPI and achieve the avail-
able spatial and frequency diversities in asynchronous cooper-
ative systems. We use diagonal and nondiagonal unitary codes
to construct the distributed differentially (space frequency and
BER

−2
10
space-time-frequency) encoded OFDM transmission schemes.
The authorized receiver can implement the detection by the
differential decoding without the knowledge of the channels
−3
10 E, G
2,32
nor the timing errors. Meanwhile, the eavesdropper can hardly
E, P2,32
D, G2,32
implement the differential decoding and its blind deconvolu-
D, P2,32 tion capacity is also prevented. Although the proposed scheme
will cause rate loss, it is useful in some cases, such as wireless
−4
10
0 5 10 15 20 25 sensor networks for military use where security is of primary
ρ(dB)
concern.
Note that in this paper we have only considered the low
Fig. 4. BER comparison between diagonal code and nondiagonal code when probability of interception (LPI). The unique property of the
Γ = 1.
proposed schemes is combating the unknown timing errors
existed in the cooperative system as well as keeping the LPI.
Other problems of physical-layer security such as low prob-
Then we compare the performance when utilizing frequency ability of detection (LPD) or malicious relays in cooperative
diversity if Γ = 2. For the diagonal code, we choose the systems are certainly interesting but left for future research.
cyclic group code G4,32 = (4, 32, [1, 7, 9, 15]) in which each
signal matrix represents 5 information bits and this code is the R EFERENCES
optimal cyclic group code at the rate 1.25. For the nondiagonal [1] M. Shin, J. Ma, A. Mishra, and W. A. Arbaugh, “Wireless network
code, we still use the 2 × 2 parametric code V(30, 6, 0) and security and interworking," in Proc. IEEE, vol. 94, pp. 455-466, Feb.
repeat it across the subcarriers to construct a 4 × 4 block 2006.
[2] A. S. K. Pathan, H. W. Lee, and C. S Hong, “Security in wireless sensor
diagonal code to obtain frequency diversity. Thus it has the networks: issues and challenges," in Proc. International Conf. Advanced
same rate 1.25 as the diagonal code. The information bit rate Commun. Technology, 2006, vol. 2, pp. 1043-1048, Feb. 2006.
is 0.61 bits/s/Hz for both cases. As shown in Fig. 5, the [3] M. J. Mihaljevic and R. Kohno, “On wireless communications privacy
and security evaluation of encryption techniques," in Proc. IEEE WCNC
performance is almost the same for the two codes. However, 2002, vol. 2, pp. 865-868, Mar. 2002.
we may use the permutation techniques proposed in [15] to [4] Y. Chen and A. J. H. Vinck, “Wiretap channel with side information,"
further improve the performance of the nondiagonal unitary IEEE Trans. Inform. Theory., vol. 54, pp. 395-402, Jan. 2008.
[5] P. Venkitasubramaniam, T. He, and L. Tong;, “Anonymous networking
codes. amidst eavesdroppers," IEEE Trans. Inform. Theory, vol. 54, pp. 2770-
As a remark, although we have only shown the simulation 2784, June 2008.
[6] X. Li, J. Hwu, and E. P. Ratazzi, “Array redundancy and diversity for
results for two receive antennas for E, similar results hold wireless transmissions with low probability of interception," in Proc.
when there are more than 2 receive antennas for E. IEEE ICASSP 2006, vol. 4, May 2006.

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:38 from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 7, JULY 2009 3379

[7] X. Li, J. Hwu, and E. P. Ratazzi, “Using antenna array redundancy and frequency codes with optimum coding advantage," IEEE Trans. Inform.
channel diversity for secure wireless transmissions," J. Commun., vol. 2, Theory., vol. 51, pp. 229-249, Jan. 2005.
pp. 24-32, May 2007. [16] W. Su and K. J. R. Liu, “Differential space-frequency modulation via
[8] J. Q. Bao and L. Tong, “Protocol-aided channel equalization in wireless smooth logical channel for broadband wireless communications," IEEE
ATM," IEEE J. Select. Areas Commun., vol. 18, pp. 418-435, Mar. 2000. Trans. Commun., vol. 53, pp. 2024-2028, Dec. 2005.
[9] S. Haykin, Blind Deconvolution. Englewood Cliffs, NJ: Prentice Hall, [17] W. Zhang, X.-G. Xia, and K. B. Letaief, “Space-time/frequency coding
1994. for MIMO-OFDM in next generation broadband wireless systems," IEEE
[10] Y. Mei, Y. Hua, A. Swami, and B. Daneshrad, “Combating synchro- Wireless Commun., Special Issue Next Generation CDMA vs. OFDMA
nization errors in cooperative relays," in Proc. IEEE ICASSP 2005, vol. 4G Wireless Applications, vol. 14, no. 3, 2007.
3, pp. 369-372, Mar. 2005. [18] A. Shokrollahi, B. Hassibi, B. M. Hochwald, and W. Sweldens, “Repre-
[11] Z. Li and X.-G. Xia, “A simple Alamouti space-time transmission sentation theory for high-rate multiple-antenna code design," IEEE Trans.
scheme for asynchronous cooperative systems," IEEE Signal Processing Inform. Theory., vol. 47, pp. 2335-2367, Sept. 2001.
Lett., vol. 14, pp. 804-807, Nov. 2007. [19] X.-B. Liang and X.-G. Xia, “Unitary signal constellations for differential
space-time modulation with two transmit antennas: parametric codes,
[12] G. Xu and H. Liu, “An effective transmission beamforming scheme for
optimal designs, and bounds," IEEE Trans. Inform. Theory, vol. 48, pp.
frequency-division-duplex digital wireless communication systems," in
2291-2322, Aug. 2002.
Proc. IEEE ICASSP 1995, vol. 3, pp. 1729-1732, May 1995.
[20] B. M. Hochwald and W. Sweldens, “Differential unitary space-time
[13] J. Proakis, Digital Communications, 4th ed. New York: McGraw-Hill, modulation," IEEE Trans. Inform. Theory, vol. 48, pp. 2041-2052, Dec.
2000 2000.
[14] W. Su, Z. Safar, and K. J. R. Liu, “Systematic design of space-frequency [21] B. L. Hughes, “Optimal space-time constellations from groups," IEEE
codes with full rate and full diversity," in Proc. IEEE WCNC 2004, vol. Trans. Inform. Theory, vol. 49, pp. 401-410, Feb. 2003.
3, pp. 1436-1441, Mar. 2004.
[15] W. Su, Z. Safar, and K. J. R. Liu, “Full-rate full-diversity space-

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:38 from IEEE Xplore. Restrictions apply.

Das könnte Ihnen auch gefallen