Sunteți pe pagina 1din 7

Secrecy Capacity of OFDM Transmissions over Fading Channels

Trevor Allen
Department of Electrical Engineering University of Texas at Dallas, Richardson, TX tx124430@utdallas.edu

Abstract This project considers the transmission of confidential data over wireless channels with multiple communicating parties. Based on an information-theoretic problem formulation in which two legitimate partners communicate over a quasi-static fading channel and an eavesdropper observes their transmissions through another independent quasi-static fading channel, we define the secrecy capacity in terms of instantaneous and average SNR of the channel. In sharp contrast with known results for Gaussian wiretap channels (without feedback), the research shows that in the presence of fading information-theoretic security is achievable even when the eavesdropper has a better average signal-to-noise ratio (SNR) than the legitimate receiver. This project also extends these results to a system consisting of multiple independent parallel channels and shows that the secrecy capacity of the composite system is simply the summation of the secrecy capacities of the individual channels. KeywordsInformation-theoretic security, wireless channels, secrecy capacity, Fading channel, physical-layer security, wiretap channel.

I.

INTRODUCTION

The issues of privacy and security in wireless communication networks have taken on an increasingly important role as these networks continue to flourish worldwide. Traditionally, security is viewed as an independent feature with little or no relation to the remaining data communication tasks and, therefore, state-of-the-art encryption algorithms are insensitive to the physical nature of the wireless medium. The basic principle of information-theoretic security widely accepted as the strictest notion of security calls for the combination of cryptographic schemes with channel coding techniques that exploit the randomness of the communication channels to guarantee that the sent messages cannot be decoded by a third party maliciously eavesdropping on the wireless medium. The theoretical basis for this information-theoretic approach, which builds on Shannons notion of perfect secrecy [1], was laid by Wyner [2] who proved that there exist channel codes guaranteeing both robustness to transmission errors and a prescribed degree of data confidentiality. In the wiretap channel proposed by Wyner, two legitimate users communicate over a main channel and an eavesdropper has access to degraded versions of the channel outputs that reach the

legitimate receiver. In [3] it was shown that if both the main channel and the wiretap channel are additive white Gaussian noise (AWGN) channels, and the latter has less capacity than the former, the secrecy capacity (i.e. the maximum transmission rate at which the eavesdropper is unable to decode any information) is equal to the difference between the two channel capacities. Consequently, confidential communication is not possible unless the Gaussian main channel has a better signal-to-noise ratio (SNR) than the Gaussian wiretap channel. Motivated by the general problem of securing transmissions over wireless channels, this project considers the impact of fading on the secrecy capacity of both a single and multiple independent parallel channels. The contributions of this research are as follows: (a) an information-theoretic formulation of the problem of secure communication over a single-antenna wireless AWGN and Rayleigh fading channels; (b) an information-theoretic formulation of the problem of secure communication over multiple independent parallel channels; (c) a comparison of secrecy capacities of AWGN and fading channels evidencing the benefits of fading towards achieving a higher level of security. Among the different conclusions to be drawn from our results perhaps the most striking one is that, in the presence of fading, information-theoretic security is achievable even when the eavesdroppers channel has a better average SNR than the main channel. The results show that this result has a multiplicative effect when we extend the model to multiple independent, parallel channels.

1|P age

II.

SYSTEM MODEL

The channel is also power limited in the sense that 1

where P corresponds to the average transmit power of each codeword. Furthermore, we denote the power of the noise in the main channel and the eavesdroppers channel as NM and NW, respectively. B. Preliminaries From Figure 1, we see that instantaneous SNR at Bobs receiver is M (i) = P|hM (i)|2 NM NM = = P|hM |2 PE[|hM |2 ] NM P|hW |2 PE[|hW |2 ] NW NW NM

E[|x(i)|2 ]
i=1

and average SNR

M (i) =

PE[|hM (i)|2 ] P|hW (i)|2 NW NW

Likewise, the instantaneous SNR at Eves receiver is W (i) = = =

and average SNR


Fig. 1. Example of a wireless network with potential eavesdropping. Alice and Bob communicate over a wireless medium in the presence of an evesdropper (Eve)

W (i) =

PE[|hW (i)|2 ]

A. Overview A legitimate user named Alice wants to send messages w to another user, say Bob. The message block wk is encoded into the codeword xn = [x(1), . . . , x(i), . . . , x(n)] to be transmitted over a discrete-time channel (the main channel) with output yM (i) = hM (i) xM (i) + nM (i)

For fading channels, hM and hW are random variables and therefore, so are M and W. Since the channel fading coefficients hM and hW are zero-mean complex Gaussian random variables and the instantaneous SNR, |h|2 , it follows that is exponentially distributed, specifically
and
M p(M ) = exp( ) M M

where hM(i) is the channel coefficient at time i and nM(i) denotes the zero-mean circularly symmetric complex Gaussian noise at time i. The coefficient hM(i) is independent from the channel output and assumed to be drawn i.i.d. according to the probability distribution p(hM). For AWGN channels, the coefficients hM(i) are assumed to be constant, hM(i) = hM , and represents link gains normalized by the power spectral density of the additive noise. For Rayleigh fading channels, we assume quasi-static fading (i.e. that fading coefficients are assumed to remain constant over the transmission of an entire codeword, hM(i) = hM, but change independently and randomly from one codeword to another) where the coefficients hM(i) are zero-mean complex Gaussian for Rayleigh fading. A third party, whose name is Eve, is capable of eavesdropping the signals sent by Alice by observing the channel output yW (i) = hW (i) xW (i) + nW (i) where hW(i) and nW(i) are independent from hM(i) and nM(i) but have the same properties.

1 1

In the rest of the paper, we will assume that Alice and Bob have perfect CSI about the main channel, but no CSI about the wiretap channel. Eve in turn has CSI on the wiretap channel.

The transmission rate between Alice and Bob is R = H(Wk)/n (where H(Wk), the binary entropy function of the Bernoulli random process Wk, quantifies the expected value of the information contained in a message Wk). We are interested in maximizing the transmission rate between Alice and Bob while also minimizing the mutual information Eves receives n n about w (i.e., I(YW ; Wk) = H(Wk) H(Wk | YW ) the Mutual n Information measures the amount of information that YW contains about Wk).

W p(W ) = exp( ) W W

2|P age

III.

SECRECY CAPACITY OF SINGLE-ANTENNA WIRELESS


SYSTEMS

Alice to Eve) can be viewed as complex AWGN channels with secrecy capacity is positive when M > W and is zero when M W which can be seen in Fig. 2.
SNR M and W , respectively. Wyner showed [2] that the

We start by deriving the secrecy capacity for the situation where both the main and the wiretap channel are complex AWGN channels, i.e. transmit and receive symbols are complex and both additive noise processes are zero mean circularly symmetric complex Gaussian. The power of the complex input is constrained according to
n 1 2 n i=1 E[|x(i)| ] .

A. Complex AWGN Channels Since each use of the complex AWGN channel can be viewed as two uses of a real-valued AWGN channel, the secrecy capacity of the complex wiretap channel follows is given by CS = CM - CW (per complex dimension), where

C. Quasi-static Rayleigh Fading Channels Now we consider the complex fading coefficients for both the main channel and the eavesdroppers channel, as detailed in Section II. Since in the quasi-static case, hM and hW are random but remain constant over a single codeword, it is reasonable to view these channels (with fading) as a complex AWGN [6, Chapter 5] with instantaneous SNRs M = P|hM |2 /NM (main channel)

CM = log 2 1 + |hM |2 N
is the capacity of the main channel and

= log 2 (1 + M )

Thus, once again based on our results for a complex AWGN channel and the non-negativity of channel capacity, we may write the instantaneous secrecy capacity for one realization of the quasi-static fading scenario as C S = CM - CW

W = P|hW |2 /NW (eavesdropper channel)

CW = log 2 1 + |hW |2 N
is the capacity of the eavesdropper channel.

= log 2 (1 + W )

Since, the channel coefficients are random variables; it is difficult to compare the secrecy capacity of a fading channel to that of a complex AWGN channel. To enable us to compare the two, we will use the average secrecy capacity of the Rayleigh Fading channel. The average secrecy capacity is easily computable numerically. It was shown in [5] that the average secrecy capacity of a channel is given by
(M , W )(M )(W )M W = 0 0

log2 1 + M log2 1 + W

M > W

M W

where

Fig. 2. Secrecy Capacity of a complex AWGN Channel for various combinations of SNR for the Main and Evesdropper channels

1 ( ) = 2 (1 + ) = 1 ( 1 ) log 2 0
and 1 is the exponential-integral function.

W M = ( M ) M + W

Figure 2 illustrates the secrecy capacity of an AWGN channel using several combinations of SNR values for both the Main and Eavesdropper channel. The above figure confirms our intuition concerning the best way to increase the secrecy capacity is to increase the SNR of main channel relative to the eavesdroppers within the provide power constraint. B. Existence of Secrecy Capacity for AWGN channels As explained in the previous section, the main channel (from Alice to Bob) and the eavesdroppers channel (from 3|P age

setting the secrecy rate R s ., Alice is assuming that the capacity = CM - R s . As long of the wiretap channel is given by CW as Cs > R s , Eves channel will be worse than Alices estimate, , and so the wiretap codes used by Alice will i.e. CW < CW and ensure perfect secrecy. Otherwise, if Cs < R s then CW >CW information theoretic security is compromised. The outage probability can be found by invoking the total probability theorem ( ) = ( < | M > W ) ( M > W ) + ( < | M W ) ( M W ) Using the derivation in [4], we find that

Fig. 3. Average Secrecy Capacity of a Rayleigh Fading Channel for various combinations of SNR for the Main and Evesdropper channels

( ) = 1

M M +2 W

2 1

D. Existence of Secrecy Capacity for Fading Channels We will now consider the existence of the secrecy capacity between Alice and Bob. As explained in Section III-B, for specific fading realizations, the main channel (from Alice to Bob) and the eavesdroppers channel (from Alice to Eve) can be viewed as complex AWGN channels with SNR M and W , respectively. Moreover, the secrecy capacity is positive when M > W and is zero when M W . Invoking independence between the main channel and the eavesdroppers channel and knowing that the random variables M and W are exponentially distributed with probability density functions given in Section II, we may write the probability of existence of a non-zero secrecy capacity as

( > 0) = M > W

It is interesting to note that to guarantee the existence of a non-zero secrecy capacity with probability greater than p0 then it follows that

M = M + W

=
0

(M )(W )M W

Fig. 4. The Outage Probability of a Rayleigh Fading Channel (Rs = 0.5) for various combinations of SNR for the Main and Evesdropper channels

In particular, a non-zero secrecy capacity exists even when M < W , albeit with probability less than 0.5. E. Outage Probability We are now ready to characterize the outage probability

0 M = W 1 0

Fig. 4 depicts the outage probability versus the average M and W for target secrecy rate ( ) equal to 0.5. SNRs M the lower the outage probability, Observe that the higher W the higher the probability of an outage. and the higher M W, the outage probability decays as 1/ M. Moreover, if W M , the outage probability approaches Conversely, if one. It is interesting to examine the asymptotic behavior of the outage probability for the extreme values of the target secrecy rate . We can see that when 0 then

(i.e. the probability that the instantaneous secrecy capacity is less than a target secrecy rate > 0). The operational significance of this definition of outage probability is that when 4|P age

( ) = ( < )

and when R s , Pout (R s ) 1, such that it becomes impossible for Alice and Bob to transmit secret information (at very high rates).

( 0)

W M + W

Fig. 5. The Outage Probability of a Rayleigh Fading Channel (multiple values of Rs) for various combinations of SNR for the Main and Evesdropper channels

Fig. 7. Difference in Secrecy Capacity of Rayleigh Fading Channel vs and complex AWGN channel for various combinations of SNR for the Main and Evesdropper channels

F. Complex AWGN vs. Rayleigh Fading It is important to emphasize that under a fading scenario, in contrast with the AWGN channel, a positive secrecy capacity does not require the average SNR of the main channel to be greater than the average SNR of the eavesdroppers channel [4 & 5]. This is due to the fact that in the presence of fading there is always a finite probability, however small, that the instantaneous SNR of the main channel M is higher than the instantaneous SNR of the eavesdroppers channel W .

One also observes, in that, in contrast to the situation of the complex AWGN channel, the average secrecy capacity of the fading channel is nonzero even when the average SNR of the main channel is lower than the average SNR of the eavesdroppers channel. Although the results are not as dramatic as shown in [5] where a real AWGN channel was compared to a Rayleigh fading channel, these observations still underline the potential of fading channels to secure the transmission of information between two legitimate parties against a possible eavesdropper. IV. SECRECY CAPACITY OF K INDEPENDENT, PARALLEL
CHANNELS

The purpose of this section is to extend the prior secrecy results to a system consisting of multiple independent parallel channels. It has been shown in [7] that if we assume that M channels are independent from each other, then the total secrecy capacity of all the channels is simply the summation of the secrecy capacity of the individual subchannels. A. Preliminaries For simplicity, we assume that the noise variance has been evenly distributed over all the subchannels in the system (i.e., the noise variance on the jth subchannel is given by NM (j) = NM /K). In addition, the average transmit power is also evenly distributed of all the subchannels (i.e. P(j) = P/K). As result, the instantaneous subchannel SNRs are given by
Fig. 6. Average Secrecy Capacity of Rayleigh Fading Channel vs and a complex AWGN channel for various combinations of SNR for the Main and Evesdropper channels

Fig. 6 compares the average secrecy capacity of a quasistatic fading channel to the secrecy capacity of a complex AWGN channel. Observe, in Fig 7, that the average secrecy rate of the fading channel is slightly better than that of the complex AWGN channel.

(where M(i,j) and W(i,j) are the instantaneous subchannel SNRs of the jth subchannel at time i).

W (i, j) =

M (i, j) =

P|hM (i,j)|2 NM K

(main)

P|hW (i,j)|2 NW K

(eavesdropper)

5|P age

B. Complex AWGN Channels Extending the concept described in Section III-B, if all subchannels are complex AWGN channels (both hM (i, j) = hM (j) and hW (i, j) = hM (j) because the channel coefficient are constant with respect to time), we can represent the normalized link gains for the Alice-Bob and Alice-Eve subchannels by the vectors M = [M(1) , . . . , M(K) ]T and W = [W(1) , . . . , W(K)]T For the AWGN case, M(j) W(j) implies subchannel j has zero secrecy capacity while M(j) > W(j) implies the jth subchannel has a non-zero secrecy capacity. The total secrecy capacity of the K independent subchannels is then given by CTotal = CM (j) CW (j)
j=1 K

Once again, the independence of the subchannels allows us leverage the result from Section IV-B to compute the total average secrecy capacity of a system with K quasi-static parallel channels by adding the average secrecy capacities from the K subchannels. Therefore CTotal = CS (j)
j=1 K

behave differently (both hM (i, j) and hW (i, j) vary between subchannels as well as between codewords).

= M (j)
j=1

M (j)W (j) M (j) + W (j)

Because each subchannel is AWGN, the transmit power used on each subchannel benefits the secrecy capacity of the system as a whole. A solution for the optimal power allocation requires knowledge of both the main and eavesdropper CSI and has been derived in [7], but not used in this project.

= log 2 1 + M (j) log 2 1 + W (j)


j=1

Fig. 9. Average Secrecy Capacity of a single Rayleigh Fading Channel vs 4 subchannels for various combinations of average SNR for the Main and Evesdropper channels

Fig. 9 illustrates the increase in the secrecy capacity experienced when using 4 independent, parallel subchannels as compared to a single Rayleigh fading channel. As is the case with Fig. 8, figure is somewhat optimistic in the sense that each subchannel is assigned the same instantaneous SNR (to help with the visualization) when in actuality; the average SNRs will vary between subchannels.
Fig. 8. Secrecy Capacity of a single complex AWGN Channel vs 4 subchannels for various combinations of instantaneous SNR for the Main and Evesdropper channels

Fig. 8 illustrates the increase in the secrecy capacity experienced when using 4 independent, parallel subchannels as compared to a single complex AWGN channel. It is important to note that the above figure is somewhat optimistic in the sense that each subchannel is assigned the same instantaneous SNR (to help with the visualization) when in actuality; the instantaneous SNRs will vary between subchannels. C. Quasi-static Rayleigh Fading Channels For Rayleigh fading channels, the instantaneous subchannel SNRs take the same form as defined in Section IV-A, but

D. AWGN vs. Rayleigh Fading From Section III-F, our intuition tells us that the secrecy capacity of K independent, parallel, Rayleigh fading channels will be slightly better than that of K independent, parallel, complex AWGN channels which is indeed the case as shown in Fig 10 and Fig. 11.

6|P age

In addition, the secrecy capacity of a system consisting of multiple independent parallel subchannels was also investigated. Multi-carrier systems are becoming increasingly common in modern communication system, as is evidenced by the use of OFDM in WiMax and WiFi systems, to cope with frequency selective fading. It was demonstrated that the extra dimensionality provided in such systems facilitates secret communication and improves the secrecy capacity over that of single channel systems. In particular, the secrecy capacity of the system is achieved when each subchannel achieves its own secrecy capacity. Therefore, we can choose the secrecy capacity achieving codebook for each subchannel independently.
Fig. 10. Average Secrecy Capacity of 4 Rayleigh Fading vs 4 complex AWGN subchannels for various combinations of average SNR for the Main and Evesdropper channels

average SNR than the legitimate partners, and (b) the secrecy capacity of fading channels is slightly better than the secrecy capacity of a complex AWGN channel over the same averaged SNRs M and W.

Previous results [5] comparing the secrecy capacity of a Rayleigh fading channel to that of a real AWGN channel show an increase in secrecy capacity, however this project demonstrates than a more accurate comparison is between Rayleigh fading and complex AWGN (simply due to the same number of dimensions being used in the calculations) where the secrecy capacities are nearly identical. It should be noted that the secrecy capacity of both complex AWGN as well as Rayleigh fading channels can be increase if Alice (having full CSI concerning Bobs channel) implements an optimal power allocation scheme such as the one discussed in [7]. The extent of the benefit could be a topic of a future investigation. REFERENCES
[1] [2] [3] [4] C. E. Shannon, Communication theory of secrecy systems, Bell Syst. Tech. J., vol. 29, pp. 656715, 1949. A. Wyner, The wire-tap channel, Bell Syst. Tech. J., vol. 54, no. 8, pp. 13551367, Oct. 1975. S. K. Leung-Yan-Cheong and M. Hellman. The gaussian wire-tap channel, IEEE Trans. Inf. Theory, 24(4):451456, Jul. 1978 J. Barros and M. R. D. Rodrigues, Secrecy capacity of wireless channels, in Proc. IEEE Int. Symp. Information Theory, Seattle, WA, Jul. 2006, pp. 356360. Bloch, Barros, Rodrigues, McLaughlin, Wireless Information-Theoretic Security, IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 54, NO. 6, JUNE 2008 David Tse and Pramod Viswanath, Fundamentals of Wireless Communication, Cambridge University Press, 2005. Z. Li, R. Yates, and W. Trappe, Secrecy capacity of independent parallel channels, in Proc. 44th Annu. Allerton Conf. Communications, Control and Computing, Monticello, IL, Sep. 2006, pp. 841848. P. K. Gopala, L. Lai, and H. El-Gamal, On the secrecy capacity of fading channels, in Proc. IEEE Int. Symp. Information Theory, Nice, France, Jun. 2007, pp. 13061310. E. Biglieri, J. Proakis, and S. Shamai (Shitz), Fading channels: Information-theoretic and communications aspects, IEEE Trans. Inf. Theory, vol. 44, no. 6, pp. 26192692, Oct. 1998.

Fig. 11. The Difference in Secrecy Capacity of 4 Rayleigh Fading vs 4 complex AWGN subchannels for various combinations of average SNR for the Main and Evesdropper channels

[5]

V.

CONCLUSIONS

[6] [7]

Modern wireless communication systems are being built to exploit the advantages of multiple independent parallel subchannels. It is well known that the advantages of such multi-carrier systems becomes increasingly pronounced as these wireless systems are deployed in environments involving complex channel conditions. In this project, a characterization of the secrecy capacity of both wireless channel with complex AWGN as well as wireless channels with quasi-static fading was provided. The results reveal that (a) perfectly secure communication over wireless channels is possible even when the eavesdropper has a better

[8]

[9]

7|P age

S-ar putea să vă placă și