CONSIDER the Two-Way Relay Channel (TWRC) depicted

Transcription

1 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 3, NO. 0, OCTOBER Achievable Rate Regions for Two-Way Relay Channel Using Nested Lattice Coding Sinda Smirani, Mohamed Kamoun, Mireille Sarkiss, Abdellatif Zaidi, and Pierre Duhamel, Fellow, IEEE Abstract This paper studies a Gaussian two-way relay channel where two communication nodes exchange messages with each other via a relay. It is assumed that all nodes operate in half-duplex mode without any direct link between the communication nodes. A compress-and-forward relaying strategy using nested lattice codes is first proposed. Then, the proposed scheme is improved by performing layered coding: A common layer is decoded by both receivers, and a refinement layer is recovered only by the receiver that has the best channel conditions. The achievable rates of the new scheme are characterized and are shown to be higher than those provided by the decode-and-forward strategy in some regions. Index Terms Compress-and-forward, Gaussian channel, lattice codes, physical-layer network coding, side information, twoway relay channel. I. INTRODUCTION CONSIDER the Two-Way Relay Channel TWRC depicted in Fig.. Two wireless terminals T and T, with no direct link between them, exchange individual messages via a relay. Recently, the capacity characterization of this channel has attracted a lot of interest since TWRC is encountered in various wireless communication scenarios, such as ad-hoc networks, range extension for cellular and local networks, or satellite links. While network level routing is the standard option to solve this problem, it has been shown that network coding NC strategies provide better performance by leveraging the side information that is available at each node. In fact, NC [] offers rate improvements by combining raw bits or packets at network layer. The rate performance of the system can be further improved if NC takes place at the physical layer. In this situation, the linear superposition property of the wireless channel is considered as a code and can be exploited appropriately to turn interference into a useful signal []. In this context, we consider a physical-layer network coding PNC architecture in Manuscript received October, 03; revised March 6, 04 and June 6, 04; accepted June 7, 04. Date of publication July 7, 04; date of current version October 8, 04. The associate editor coordinating the review of this paper and approving it for publication was Z. Han. S. Smirani was with the Communicating Systems Laboratory, CEA, LIST, 99 Gif-sur-Yvette, France. She is now with the Orange Labs, Issy-les- Moulineaux, France sinda.smirani@orange.fr. M. Kamoun and M. Sarkiss are with the Communicating Systems Laboratory, CEA, LIST, 99 Gif-sur-Yvette, France mohamed.kamoun@ cea.fr; mireille.sarkiss@cea.fr. A. Zaidi is with the Université Paris-Est Marne-la-Vallée, LIGM, Marne-la-Vallée Cedex, France abdellatif.zaidi@univ-mlv.fr. P. Duhamel is with the CNRS/LSS, Supélec, 99 Gif-sur-Yvette, France pierre.duhamel@lss.supelec.fr. Color versions of one or more of the figures in this paper are available online at Digital Object Identifier 0.09/TWC Fig.. The two-phase transmission of TWRC: MAC and Broadcast phases. which the overall communication requires two phases, namely a Multiple Access MAC phase in which the terminals simultaneously send their messages to the relay and a Broadcast BC phase in which the relay transmits a message that is a function of the signals received in the MAC phase. An outer bound on the capacity region of this model is given in [3], [4]. Several coding strategies have been proposed for PNC by extending classical relaying strategies such as Amplify-and- Forward AF, Decode-and-Forward DF, and Compress-and- Forward CF to TWRC. AF strategy [5] is a linear relaying protocol where the relay only scales the received signal to meet its power constraints. This simple strategy suffers from noise amplification especially at low signal-to-noise ratios SNRs. With DF strategy, the relay jointly decodes both messages, and then re-encodes them before broadcasting the resulting codeword. The authors in [5] derived an achievable rate region for TWRC by using DF strategy and superposition coding in the BC phase. This region has been improved in [6] where the authors propose that the relay sends a modulo sum of the decoded messages, thus mimicking the initial example of XOR NC. These DF relaying based schemes require full decoding of the incoming signals and thus suffer from a multiplexing loss due to the MAC phase limitation [3]. The authors in [], [7] propose PNC schemes based on a partial DF pdf where the relay does not decode completely the incoming signals, but relies on the side information available at each terminal to decode a linear function of the transmitted codewords. The key strategy in these schemes is to design the codes at both transmitting terminals in the MAC phase so that the relay can compute a message which is decodable by both nodes during the BC phase. Nested lattice codes, which have the nice property to ensure that any integer-valued linear IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.

2 5608 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 3, NO. 0, OCTOBER 04 combination of codewords is a codeword, are used in [7], [8] to implement pdf for Gaussian channels. In [9], the lattice based-scheme proposed in [7] has been extended for TWRC with more than one relay. Although the advantage presented by these schemes in using lattice coding, the problem of pdf schemes is to guarantee phase coherence at the relay during the MAC phase [3]. Another strategy has been proposed where the relay compresses its, utilizing Wyner-Ziv binning. This strategy has attracted our particular attention since it offers a good trade-off between processing complexity at the relay and noise amplification compared to DF and AF strategies. This has motivated us to study CF based techniques. CF for TWRC [0] follows the same approach as CF schemes for the relay channel []. With CF scheme, the relay does not decode any message, but rather compresses the received signal and sends a new message that includes some useful information about the original messages. This technique does not impose decoding rates at the relay as in DF-based schemes. Performance bounds of CF scheme for TWRC have been investigated in [] [4]. It has been shown that for specific channel conditions, namely symmetric channels, CF outperforms the other relaying schemes at high SNR regimes. Random coding tools have been used in the aforementioned references to derive achievable rate regions of CF. Structured codes, on the other hand, have been found to be more advantageous in practical settings thanks to their reduced implementation complexity [5]. CF strategy using lattice coding for three node Gaussian relay channel has been considered in [6], [7]. In [8], we have proposed a CF scheme for TWRC that is based on nested lattice coding. In the MAC phase of this scheme, the communicating nodes simultaneously send their messages and the relay receives a mixture of the transmitted signals. The relay considers this mixture as an analog source which is compressed and transmitted during the BC phase. Taking into account that each terminal has a partial knowledge of this source namely, its own signal that has been transmitted during the MAC phase, now considered as receiver side information, the BC phase is equivalent to a Wyner-Ziv compression setting with two decoders, each one having its own side information. Each user employs lattice decoding technique to retrieve its data based on the available side information. The proposed scheme can be seen as an extension of lattice quantization introduced in [9] to the TWRC model. In this paper, we first generalize this latter scheme and we apply the results to our transmission problem. In the simplest situation, when a single layer of compression is performed, the relay broadcasts a common compressed message to both terminals. Therefore, it is easily understood that the achievable rates in both directions are constrained by the capacity of the worst channel. In this case, the user experiencing better channel conditions and side information is strongly constrained by this restriction on its transmission rate. To overcome this limitation, we propose an improved scheme where the relay also sends an individual description of its output that serves as an enhancement compression layer to be recovered only by the best receiver. Therefore, the new scheme employs three nested lattices. The common information is encoded using two nested lattices while the refinement information is encoded with a finer lattice that contains the other two lattices. The channel codewords corresponding to the two layers are superimposed and sent during the BC phase. Through numerical analysis, we show that this layered scheme outperforms AF and CF strategies in all SNR regimes and DF strategy for specific SNR regions. Layered coding for Wyner-Ziv problem has been addressed in [0] for lossy transmission over broadcast channel with degraded side information. In [4], the authors derive the achievable rate region of layered CF coding for TWRC, based on a random coding approach. The authors in [7], [] and [] proposed schemes for TWRC based on doubly nested lattice coding where different power constraints at all nodes are assumed. In these schemes, each of the two end terminals employs a different code with carefully chosen rate constructed from the lattice partition chain. The relay decodes a modulolattice sum of the transmitted codewords from the received signal. However, in [] full-duplex nodes are considered and in [7] and [], the direct link between both terminals is exploited and the transmission is performed in three phases. In these schemes, the relay follows a DF-lattice coding strategy since it decodes a function of the transmitted lattice codewords. On the other hand, in our proposed enhancement scheme, doubly nested lattice coding is only employed at the relay for CF strategy and half-duplex terminals are considered with no direct link between the two end terminals. Furthermore, the relay does not need to know either the other terminals codebooks or the precise value of the channel. It merely reconstructs its encoder from the channel module and the variances of the transmitted signals. This strategy ensures a small processing load at the relay. To our knowledge, our work is the first that proposes a doubly nested lattice coding for CF relaying in TWRC. The remaining of the paper is organized as follows. Section II introduces the system model. Section III derives the achievable rate region when one layer lattice-based coding scheme is used and Section IV derives the achievable rate region with two layer lattice-based coding. Section V illustrates the performance of the proposed schemes through numerical results. Finally, Section VI concludes the paper. Notations: Random variables r.v. are indicated by capital letters and their realizations are denoted by small letters. Vector of r.v. or a sequence of realizations are indicated by bold fonts. II. SYSTEM MODEL Consider a Gaussian TWRC in which two source nodes T and T exchange two individual messages m and m, with the help of a relay R as shown in Fig.. For this model, we have the following assumptions: a. The relay and the source nodes operate in half-duplex mode; a. The two users are assumed to be synchronized, and due to the half duplex mode, there is no direct link between T and T. a.3 The communication takes n channel uses that are split into two orthogonal phases: MAC phase and BC phase with lengths n = n and n = n, [0, ] respectively.

3 SMIRANI et al.: ACHIEVABLE RATE REGIONS FOR TWO-WAY RELAY CHANNEL USING NESTED LATTICE CODING 5609 During the MAC phase, node T draws uniformly a message m from the set M = {,,..., nr } and sends it to the other terminal T where R denotes the message rate from node T to node T. Similarly, node T draws uniformly a message m from the set M = {,,..., nr } and sends it to the other terminal T where R denotes the message rate from node T to node T.Letx i m i R n be the channel codeword of length n sent by node T i, i =, and P i be the corresponding transmit power constraint that verify the following assumptions a.4 /n n k= x i,k P i. The messages are transmitted via a memoryless Gaussian channel and the relay R receives a signal y R R n given by y R = h x + h x + z R where h i denotes the channel coefficient between T i and R, i =,. We assume that: a.5 The components of the random vector Z R are i.i.d Additive White Gaussian Noise AWGN at the relay with variance σr, i.e., N0,σ R and they are independent from the channel inputs X i, i =,. a.6 The channel coefficients follow a block fading model. Channel reciprocity between MAC and BC channels is assumed, i.e., h i R = h R i = h i. During the BC phase, the relay generates a codeword x R m R R n of dimension n from the received sequence y R. The average power constraint at the relay verifies a.7 /n n k= x R,k. The signal x R is transmitted through a broadcast memoryless channel and the received signal at node T i is y i R n, i =, y i = h i x R + z i, a.8 The components of Z i are i.i.d AWGN at node T i with variance σi, i =, and they are independent from the channel input X R. Perfect CSI is assumed at all nodes. This assumption is further discussed in Remark 3. For the aforementioned TWRC, a rate pair R,R is said to be achievable if there exists a sequence of encoding and decoding functions such that the decoding error probability approaches zero for n sufficiently large. For the sake of completeness, we hereafter outline some preliminaries on lattices [5], [3]. Fundamentals on Lattice Coding A real n -dimensional lattice Λ is a subgroup of the Euclidean space R n,+. λ,λ Λ, λ +λ Λ. We present below some fundamental properties associated with a lattice: The nearest neighbor lattice quantizer of Λ is defined as Q Λ x =argmin λ Λ x λ where x R n and. is the Euclidean norm. The basic Voronoi cell of Λ is the set of points in R n closer to the origin than to any other point of Λ, VΛ = {x Q Λ x =0}. The volume of a lattice V := VolVΛ. The mod-λ operation is defined as x mod Λ=x Q Λ x. It satisfies the distributive law: x mod Λ + y modλ=x + y modλ. The second moment per dimension of Λ is σ Λ := /n./v VΛ x dx. The dimensionless normalized second moment is defined as GΛ := σ Λ/V /n. A sequence of n -dimensional lattices Λ n is said to be good for quantization if GΛ n /πe [4]. n A sequence of n -dimensional lattices Λ n is said to be good for AWGN channel coding if for n -dimensional vector Z N0,σ I n, P {Z VΛ n } vanishes when n goes to. In this case, VolΛ n n nhz, where hz =/ logπeσ is the differential entropy of Z [5]. There exist lattices which are simultaneously good for quantization and channel coding see [6]. Lemma Crypto Lemma [3]. For a dither vector T independent of X and uniformly distributed over VΛ, then Y =X + T modλis uniformly distributed over VΛ and is independent of X. Consider a pair of n -dimensional nested lattices Λ, Λ such as Λ Λ. The fine lattice is Λ with basic Voronoi region V of volume V and second moment per dimension σ Λ.The coarse lattice is Λ with basic Voronoi region V of volume V and second moment σ Λ. The following properties of nested lattices hold: For Λ Λ,wehaveQ Λ Q Λ x = Q Λ Q Λ x = Q Λ x. The points of the set Λ V =Λ mod Λ represent the coset leaders of Λ relative to Λ, where for each λ {Λ mod Λ }, the shifted lattice Λ,λ =Λ + λ is called a coset of Λ relative to Λ. There are V /V distinct cosets. It follows that the coding rate when using nested lattices is R = n log Λ V = n log V V bits per dimension. 3 III. ACHIEVABLE RATE REGION FOR TWRC Theorem : For a Gaussian TWRC, under the assumptions a. to a.8, the convex hull of the following end-to-end rate-pairs R,R is achievable: R log + σ R + h P max h i P i +σ R i {,} h + min i i {,} σ i 4

4 560 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 3, NO. 0, OCTOBER 04 Fig.. Lattice encoding at the relay and decoding at T i, i =,. R log h P + max h i P i +σ σ R + R i {,} h + min i i {,} σ i for [0, ]. The main idea of the proposed scheme is the following: during the BC phase, the relay sends a quantized version of the signal that was received during the MAC phase. The quantization procedure generates an index which is sent reliably to both users using appropriate channel codes. This index is decoded by both users and, based on their own information sent during the MAC phase, each source recovers the transmitted message. The proof of Theorem is detailed in the next paragraphs: in Section III-A, the lattice coding scheme for the source coding is presented. The end-to-end achievable rates are derived in Section III-B and finally in Section III-C the achievable rate region is obtained by optimizing the lattice parameters. A. Lattice Based Source Coding We suppose that the elements of X i, i =,, aredrawn from an independent identically distributed i.i.d Gaussian distribution with zero mean and variance P i.lets i = h i X i be the side information available at terminal T i, i =,. The signal sent by the relay Y R can be written in two ways as the sum of two independent Gaussian r.v.: the side information S i and the unknown part U i = Y R S i = h ī X ī + Z R, i {, }. From its received signal, each terminal T i, i {, } decodes Û i using S i. The variance per dimension of U i is σu i = VARY R S i = h ī P ī + σr. In the following, we detail the proposed lattice source coding scheme. Encoding: The lattice source encoding LSE operation is performed with four successive operations: first, the input signal y R is scaled with a factor β. Then, a random dither t which is uniformly distributed over V is added. This dither is known by all nodes. The dithered scaled version of y R, βy R + t is quantized to the nearest point in Λ. The outcome of this operation is processed with a modulo-lattice operation in order to generate a vector v R of size n as shown in Fig., and defined by: 5 v R = Q Λ βy R + t mod Λ. 6 The relay sends the index of v R which identifies a coset of Λ relative to Λ that contains Q Λ βy R + t. By construction, the coset leader v R can be represented using log V /V bits. Thus, the rate of the source encoding scheme employed by the relay is R given by Eq. 3. We further assume that Λ is good for quantization and Λ is good for channel coding [9]. For high dimension n and according to the properties of good lattices, we have /n log V i / log πeσ Λ i, i {, }. Thus R reads R = σ log Λ σ. 7 Λ Decoding: For both users, v R is decoded first. Then û i is reconstructed with a lattice source decoder LSD using the side information s i as û i = γ i v R t βs i mod Λ, i =, 8 where γ i, i {, } are the scaling factors at each decoder. B. Rate Analysis At the relay, message m R corresponding to the index of v R is mapped to a codeword x R of size n. We assume that the elements of the r.v. X R are drawn from an i.i.d Gaussian distribution with zero mean and variance.we consider separate source-channel coding. The broadcast rate from the relay to both terminals is bounded by the capacity of the worst individual relay-terminal channel capacity minix R ; Y,IX R ; Y. From Shannon s source-channel separation theorem [7], we have n R n min IX R ; Y,IX R ; Y. 9 Since real Gaussian codebooks are used for all transmissions, we have: IX R ; Y i =/ log + h i /σi, i =,. Finally, by combining Eqs. 7 and 9, we obtain the following constraint on the achievable rates n log σ Λ σ Λ n log + min i {,} h i σi. 0 This constraint ensures that index m R is transmitted reliably to both terminals and v R is available at the input of the LSD of both receivers. At terminal T i, û i in 8 can be written as: û i = γ i βu i + e q modλ = γ i βu i + e q where e q = Q Λ βy R + t βy R + t = βy R + t mod Λ, is the quantization error. By Lemma, E q is independent from Y R, and thus from U i.alsoe q is uniformly distributed over V thus the variance of E q per dimension is

5 SMIRANI et al.: ACHIEVABLE RATE REGIONS FOR TWO-WAY RELAY CHANNEL USING NESTED LATTICE CODING 56 σ Λ. Equation is valid only if βu i + e q V. According to [9], with good channel coding lattices, the probability PrβU i + E q V vanishes asymptotically provided that: E βu i + E q = β σu n i + σ Λ σ Λ 3 With respect to Eq. 3 which is considered as constraint, replacing U i by its value, we conclude that: Û i = γ i βh X + Z R +E q. 4 Let Z eq,i = γ i βz R + E q be the effective additive noise at terminal T i. Under high dimension assumption, n,we can approximate the uniform random variable E q over V by a Gaussian variable Z q with the same variance [4]. Therefore, the communication between terminals T and T resp. T and T is equivalent to an AWGN channel where the Gaussian noise is given by Z eq,i. hence, the achievable rates of both links satisfy nr n + β h P 5 log nr n log β σr + σ Λ + β h P β σr + σ Λ 6 C. Achievable Rate Region The rate region that can be achieved by the proposed scheme is characterized by the constraints 5, 6, 0, and 3. Without loss of generality, we assume that h P h P. With this setting, T is the terminal which experiences the weakest side information. Letting = n /n, from 0 and 3, the lower bound of σ Λ is given by σ Λ + min i {,} β σ U h i σ i 7 The rate region defined in 5 and 6 can be rewritten as R log + SNR 8 R log + SNR 9 where SNR and SNR are the end-to-end SNRs, defined as follows: SNR = β h P β σr + σ Λ SNR = β h P β σr + σ Λ 0 Note that SNR and SNR are maximized when σ Λ is minimal. Thus the optimal choice on the second moment of Λ is σ Λ min = + min i {,} β σ U h i σ i If h P h P, σ U is replaced with σ U in. Finally, replacing σ Λ min in 0 and, Eqs. 4 and 5 are verified and the proof is concluded. Remark : For the transmission problem of the TWRC, the achievable rate region is independent of the choice of the decoders scaling factors γ i. It is also independent of the encoder scaling factor β provided that σ Λ is set to its smallest value σ Λ min in. In the next section, we show that these parameters are involved in the source coding problem that was addressed in [8]. Especially when considering analog signal transmission, the optimisation of these parameters allows to minimize the distortion. D. Analog Signal Transmission Since the relay quantizes an analog source, we can consider an end-to-end analog transmission. In this case, the distortion that affects the reconstructed signals at both terminals becomes the main performance metric. The second moment of this distortion is given by E Y R n ŶRi = D i ; i {, } 3 where Y R = U i + S i and ŶRi = Ûi + S i. By replacing Ûi by its value in, 3 becomes D i = γ i β σ U i + γ i σ Λ ; i {, }. 4 For the analog signal transmission, this distortion has to be minimized to obtain the optimal source coding scheme. For fixed β, the distortion at T i depends only on two parameters namely γ i and σ Λ. The optimal distortion can be obtained by calculating the following derivatives: D i =0 γ βσu i = i γ i β σu i + σλ 5 where γi, i {, } are the optimal decoder scaling factors. Since γ i > 0, then D i / σ Λ > 0. Thus, the function D i is increasing with σ Λ and σ Λ min in is the optimal choice that minimizes the distortion at each terminal. Therefore, γ i = βσ U i β σ U i + σ Λ min, i {, }. 6 By replacing σ Λ and γ i by their optimal values, we obtain the minimal value of Di min, i {, }, given by D min i = σ Λ min σ U i β σ U i + σ Λ min 7 = + min i {,} h i σ i σ U σ U i σ U i + σ U. 8 D min i, i {, }, just like the achievable rates, are independent of β. However, for a fixed β, the optimal lattice parameters and receivers scaling factors depend on that choice.

6 56 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 3, NO. 0, OCTOBER 04 Comments on the Distortions: At terminal T, the distortion writes: D min = σ U σ U A σ U + σ U = σ U A where A =+ min i {,} h i /σ i /. It can be reformulated as σu D min log σ U D min = + min i {,} h i σ i = log + min i {,} h i σi. 9 We find, in the left hand side of 9, the Wyner-Ziv rate distortion function of the Gaussian source Y R with side information S at the decoder T [8]. It is defined as the minimum rate needed to achieve D min and it is given by: R WZ D min = log σ U D min 30 Note that the source coding rate is no larger than the channel coding rate to the relay. Also, according to 6 the optimal value of γ is given by γ = βσ U β σ U + σ Λ min. With the choice β = γ, we get β = D min /σu.this is in accordance with the optimal scaling factor reported in [8], [8] for the optimum Gaussian forward test channel. For this choice of β, σ Λ min = D min which is consistent with the source coding parameters choices in [8]. At terminal T, the reconstruction distortion is smaller than D min of terminal T. This is compatible with the fact that T has the best side information quality and the proposed achievable scheme is optimal for the worst user. IV. IMPROVED ACHIEVABLE RATE REGION FOR TWRC In the previous section, we presented a PNC scheme in which a common information is sent from the relay to both users. The rates that are achievable by this scheme depend only on the ratio σ Λ min /β. This ratio is determined, as shown by, essentially by the variance σu i of the unknown part of the source at the terminal T i and the lowest channel coefficient amplitude min i {,} h i /σi. Thus, the achievable rates are limited by the user which has the weakest side information and also the worst channel condition. In this case, the best user suffers from this limitation on its achievable rate. In order to improve its rate, an additional refinement information can be sent from the relay, that can be only decoded by the best user. Without loss of generality, assume that terminal T has better channel condition and side information than T i.e., h h and h P h P. The following theorem provides an achievable rate region for the TWRC, obtained using the refinement scheme. Theorem : For a Gaussian TWRC, under the assumptions a. to a.8, the convex hull of the following end-to-end rate-pairs R,R is achievable, for, ν [0, ], as given in 3 and 3, shown at the bottom of the page. As mentioned previously, the main idea of the coding scheme employed for Theorem is that the relay should be sending two descriptions of its received signal, a common layer that is intended to be recovered by both users and an individual or refinement layer that is intended to be recovered only by the best user, i.e., terminal T. The proof of Theorem is detailed below. A. Doubly Nested Lattices for Source Coding We use a doubly nested lattice chain Λ 0, Λ, Λ such as Λ Λ Λ 0. We require that Λ is good for channel coding, Λ is simultaneously good for channel and source coding and Λ 0 is good for source coding. From these lattices, we form three codebooks C c =Λ V C r =Λ 0 V C =Λ 0 V R log + h P σr + R log + σr + h P +σ R ν h + ν h +σ + ν h σ h P h P +σ R [ + ν h ν h +σ ] 3 3

7 SMIRANI et al.: ACHIEVABLE RATE REGIONS FOR TWO-WAY RELAY CHANNEL USING NESTED LATTICE CODING 563 Fig. 3. Layered Lattice encoding at the relay. with the following coding rates: R c = V log n V R r = V log n V 0 R = R c + R r = n log V V 0 n n n σ log Λ σ Λ σ log Λ σ Λ 0 log σ Λ σ Λ Fig. 4. Lattice source decoding at the Terminal. Decoding: v Rc is decoded at terminal T. Then, û is reconstructed with an LSD using the side information s as û = γ v Rc t βs mod Λ. 39 At terminal T, v Rc and v Rr are both decoded correctly. These coset leaders are used to recalculate the total information v R from 38a. Finally, the decoder reconstructs û as defined by 40 and shown in Fig. 4, as û = γ v R t βs mod Λ 40 where R c is the common source rate, R r is the refinement source rate and R is the total source rate at terminal T. Encoding: Fig. 3 shows the LSE operation. The input signal y R is scaled with a factor β. Then, a random dither t which is uniformly distributed over V is added. This dither is known by all nodes. The dithered scaled version of y R, βy R + t, is quantized to the nearest point in Λ 0. The outcome of this operation is then processed to generate two messages. First, the coset leader of Λ relative to Λ 0, v Rr, is generated by a modulo-lattice operation. The index of v Rr identifies the refinement message. Then, another quantization to the nearest point in Λ is performed and processed with another modulolattice operation to generate the coset leader of Λ relative to Λ, v Rc. The index of v Rc identifies the common message. Both messages are defined as: v Rr = Q Λ0 βy R + t mod Λ 36 v Rc = Q Λ Q Λ0 βy R + t mod Λ = Q Λ βy R + t mod Λ. 37 It can easily be seen that v Rr C r and v Rc C c. We obtain the same common information generated in 6. Thus, the total information that is intended to terminal T is such that v R = v Rr + v Rc 38a = Q Λ0 βy R + t mod Λ + Q Λ βy R + t mod Λ 38b = Q Λ0 βy R + t Q Λ Q Λ0 βy R + t + Q Λ βy R + t Q Λ Q Λ βy R + t 38c = Q Λ0 βy R + t Q Λ βy R + t 38d = Q Λ0 βy R + t Q Λ Q Λ0 βy R + t 38e = Q Λ0 βy R + t mod Λ 38f where 38c, 38d, and 38e follow using the properties of the modulo operation as given in Section II. B. Rate Analysis The relay generates the indices of v Rc and v Rr. Then they are mapped to the channel codewords x Rc and x Rr.Therelay sends x R m R which is the superposition of x Rc and x Rr with transmit power ν and ν, ν {0, }, respectively. The refinement codeword x Rr is encoded on top of the common codeword x Rc and it is treated as an interference while decoding the common message. Thus, X Rc X r Y, Y forms a Markov chain. As described in previous single layer PNC scheme, the broadcast rate is bounded by the worst relayterminal channel capacity for the common message, and by the relay-t channel for the refinement message. In addition, the source-channel separation ensures that the codewords x Rc and x Rr are transmitted reliably to the terminals and that v Rc and v Rr are available at the LSD input of corresponding receivers. Therefore, the rates are such that n R c n min {IX Rc ; Y,IX Rc ; Y } 4 n R r n IX Rr ; Y X Rc 4 For real Gaussian codebooks, we have IX Rc ; Y = log ν h + ν h + σ IX Rc ; Y = log ν h + ν h + σ IX Rr ; Y X Rc = log + ν h σ. Since h h, min{ix Rc ; Y,IX Rc ; Y } = IX Rc ; Y. Using 33, 34, 4, and 4, the rates conditions become σ Λ ν h n log σ n log Λ + ν h + σ 43

8 564 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 3, NO. 0, OCTOBER 04 σ Λ n log σ n log Λ 0 + ν h σ. 44 Now, û and û can be obtained using 40 and 39, respectively. At terminal T, û can be written as: û = γ βu + e q, mod Λ 45 = γ βu + e q, 46 where e q, is the quantization error at lattice Λ given by e q, = Q Λ βy R + t βy R + t = βy R + t mod Λ and 46 can be obtained by proceeding as in Section III-B. Note that PrβU + E q, V vanishes asymptotically provided that: E βu + E q, = β σu n + σ Λ σ Λ. 47 In this case, the rate achievable at terminal T is such that nr n log + β h P β σr σ Λ At terminal T, û can be obtained as û = γ βu + e q,0 mod Λ 49 γ βu + e q,0 50 where e q,0 is the modulo-λ 0 quantization error given by e q,0 = Q Λ0 βy R + t βy R + t = βy R + t mod Λ 0 and 50 holds if βu + e q,0 V. Note that, by using Lemma, E q,0 is independent from Y R, and thus from U.Alsothis quantization error is uniformly distributed over V 0. Therefore, VARE q,0 =σ Λ 0. The probability PrβU + E q,0 V vanishes asymptotically provided that: E βu + E q,0 = β σu n + σ Λ 0 σ Λ. 5 Thus, Û = γ βh X + βz R + E q,0. Communication from terminal T to terminal T is equivalent to that over an AWGN channel with noise γ βz R + E q,0. Hence the achievable rate of this link satisfies: nr n log + β h P β σ R + σ Λ 0. 5 C. Achievable Rate Region The rate region that is achievable using the coding scheme that we described so far can be obtained using 43, 44, 47, and 5. Letting n /n =, we get σ Λ σ Λ + σ Λ σ Λ 0 ν h P R ν h +σ + ν h σ σ Λ σ Λ β σu σ Λ 0 σ Λ β σu Since σ Λ σ Λ σ Λ 0, the last constraint in the system is not active. Thus we obtain the following bounds on the second moment of the lattices σ Λ σ Λ 0 + β σ U ν h ν h +σ σ Λ + ν h σ The rate region defined by 48 and 5 can then be rewritten equivalently as R log + SNR 55 R log + SNR 56 where the end-to-end SNRs are given by SNR = β h P β σ R + σ Λ 57 SNR = β h P β σ R + σ Λ It is easily seen that one obtains larger rates if the inequalities in 57 and 58 hold with equality, i.e., the optimal choice on the second moment of Λ is σ Λ min = + β σ U ν h ν h +σ and the optimal choice on the second moment of Λ 0 is σ Λ 0 min = σ Λ min + ν h σ Finally, by substituting σ Λ min and σ Λ 0 min in 57 and 58, we get 3 and 3. This completes the proof of Theorem. Remark : The obtained achievable rates are independent of the choice of the scaling factors β and γ i. The optimal choice of these parameters is explained when considering the source coding problem as explained in the next section.

9 SMIRANI et al.: ACHIEVABLE RATE REGIONS FOR TWO-WAY RELAY CHANNEL USING NESTED LATTICE CODING 565 D. Analog Signal Transmission Proceeding as in the analysis in III-D, it can be easily obtained that the optimal scaling factors γ i that minimize the distortion at each terminal are given by γ = γ = βσ Λ β σ U + σ Λ 6 βσ Λ 0 β σ U + σ Λ 0. 6 Thus, the minimal distortion at terminal T is D min = + σ U ν h ν h +σ 63 and the minimal distortion at terminal T is D min = σ U σ Λ 0 min β σ U + σ Λ 0 min 64 = σ U σ U [ + ν h σ ν h + ν h +σ σ U +σ U ]. 65 Observe that the distortion D min that is allowed by the layered coding scheme described so far is, as expected, smaller than that of the coding scheme of Section III given by 7. To summarize, if we are interested in the distortion problem in addition to the transmission problem addressed in this paper, the choice of β can be left to the designer. On one hand, the optimal lattice parameters and the receivers scaling factors that depend on the chosen value of β are given by and 6 for the first scheme and 59, 60, 6, and 6 for the second scheme. On the other hand, the choice of any value of β does not affect the optimal end-to-end achievable rates and distortions that depend only on the system parameters. V. N UMERICAL RESULTS This section presents numerical results of the achievable rates of our proposed schemes compared to AF and DF protocols and the outer-bound capacity given in [3], [3]. We select the time-division parameter [0, ] that permits to trade among the multiaccess and broadcast phases in a manner that maximizes the users rates. The bounds are determined by maximizing the weighted sum of the rates R and R for each protocol. For example, for the scheme of Section IV, we solve the following problem for all values of η [0, ] max ηr + ηr 66a s.t. R,R satisfy 3 and 3 66b for and ν [0, ] 66c It is worth noting that the time division with AF relaying scheme is set optimally to /. Fig. 5. Achievable rate regions and the outer bound capacity of the Gaussian TWRC. In the left, T has the best transmit power and the worst channel. In the right, T has the best transmit power and the worst channel. a P =5dB, P =0dB, =0dB, h =0.5, h =;bp =0dB, P = 5 db, =0dB, h =, h =0.5. We consider equal noise variances σ =σ =σ R =,different transmit powers and asymmetric channels with h P h P. For convenience, we refer to the achievable rate regions of Theorems and respectively as LCF and LCF. Fig. 5 shows the rates allowed by AF, DF and our proposed scheme LCF for two different setups: i terminal T experiencing better channel conditions and having less power than terminal T in Fig. 5a, and ii terminal T experiencing better channel conditions and having less power than terminal T in Fig. 5b. Note that our scheme LCF is, in essence, a CF relaying strategy that is tailored appropriately for the TWRC. Being based on linear lattice coding, this strategy has been shown in [8] to achieve the same rates as those allowed by random coding [3], [4]. It has been shown in [3], that CF strategy achieves rates that are larger than those by AF for symmetric power and channel configurations. However, this result is not verified for asymmetric channels. This is shown in Fig. 5 where the difference between the rate regions of AF and LCF is negligible for moderate SNR values and asymmetric channels.

10 566 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 3, NO. 0, OCTOBER 04 generally, the SNR threshold required for LCF to outperform DF decreases as the relay power increases. In particular, in middle to high SNR regime at both terminals, if the relay has limited transmit power, LCF can be better than DF. For example, for SNR =5dB and =0dB, CF is better than DF. The achievable rates of DF-Lattice coding scheme [], [] when considering time division optimization are given by 67 and 68. This scheme achieves rates within / bit of the upper bound. This bound becomes tight for high SNR as depicted in Fig. 6. However, for very low SNR, the minimum in the right hand side in both equations 67 and 68 is equal to zero. In this case, the achievable rates of this scheme are equal to zero. All the other schemes, in this case, outperforms DF-Lattice coding Fig. 6. Maximum sum-rate for symmetric channels: SNR = SNR R = SNR R = SNR R = SNR R. LCF outperforms AF and DF for SNR > db. Fig. 6 illustrates the performance of all schemes in the symmetric power and channel conditions case. We consider also for comparison the DF lattice-coding scheme proposed in [] and extended for asymmetric channels in []. End-to-end maximum sum-rate R + R is drawn as a function of SNR for equal channel and power conditions for all nodes. Define SNR ij = h ij P i /σj. It is clearly seen that LCF outperforms DF for SNRs db. This result can be interpreted analytically. In fact, it can be seen easily that for small SNR values, DF rate approaches R DF max min{snr, SNR} = 4 SNR. Also, the rate offered by LCF approaches R LCF SNR + SNR SNR + SNR + + SNR Thus, in such small SNR regime, we have R LCF R DF.On the other hand, for high SNR, DF rate can be approximated by and LCF rate approaches R DF 6 log SNR R LCF 4 log SNR. Therefore, for large SNRs, R LCF R DF that reflects the result in Fig. 6. More generally, for equal channel conditions and transmit power at both terminals, i.e., SNR = SNR R = SNR R, when SNR Ri SNR, R LCF >R DF and when SNR Ri SNR, R LCF <R DF i,. In other words, when the relay power is too high compared to the terminals power, CF is better than DF and vice versa. This result is consistent with the previous comparison in Fig. 6, where we show that LCF is better than DF for high SNR regime. More { [ R min log h P h P + h + h ] + P P σr, log + h } σ 67 { [ R min log h P h P + h + h ] + P P σr, log + h } σ 68 where [x] + =max0,x. In what follows, we consider channel parameters combinations such that h P h P and h h.fig.7 draws the achievable rate regions of LCF and LCF. One can see that the two-layer based scheme LCF enlarges the rate region compared to the basic scheme since the relay sends additional information to the best terminal T. For the setting presented in Fig. 7a, the achievable rate R increases by 60% due to the additional refinement individual description. Fig. 7b illustrates this aspect for a different choice of the channel parameters where R increases by more than 00%. Fig. 8a shows the maximum sum-rate as a function of the transmit SNR for asymmetric channel condition and equal power constraints. At lower SNRs, LCF outperforms DFlattice coding, while at higher SNRs SNR db, DFlattice coding is better. It is important to stress here that with LCF and LCF, the relay uses less information than DF-based schemes to reconstruct its encoder. These schemes have also less complexity since the relay does not have to decode any message. Fig. 8b shows ν, the fraction of relay power allocated to the common message. ν represents the power allocated to the refinement message. For the considered channel settings, although the common message gets more than 90% of the relay power, the remaining power is sufficient to ameliorate the performance of LCF compared to LCF by 0% at high SNR. Fig. 9 illustrates the achievable rate regions of all the schemes for various SNR settings.at low SNR regime, the scheme LCF outperforms the scheme LCF; but they both fall short of attaining the same performance as that offered by DF which is nearly optimal. In fact, in this SNR regime, the rate region obtained

11 SMIRANI et al.: ACHIEVABLE RATE REGIONS FOR TWO-WAY RELAY CHANNEL USING NESTED LATTICE CODING 567 Fig. 7. Achievable rate regions of LCF and LCF. LCF achieves better endto-end rates at T.aP =0dB, P = =5dB, h =, h = 0.5; bp =0dB, P = =5dB, h =6, h =0.5. with DF relaying approaches relatively closely the outer bound as can be seen in Fig. 9d. Note that our observation here is consistent with the results in [3], [9] that showed that DF scheme is better than the other relaying schemes for low SNR region. At very large SNRs, LCF and LCF achieve better sumrates than DF as shown in Fig. 9a. In this case, DF-lattice coding is optimal since it coincides with the outer bound. This scheme approaches the capacity asymptotically as the uplink SNRs increase, i.e., SNR R and SNR R. However, as these SNRs decrease, the achievable rates of this scheme approach zero as depicted in Fig. 9e. At moderate to large SNRs, LCF scheme can achieve sumrates higher than classic DF. For low to moderate SNRs, It can achieve sum-rates higher than DF-lattice coding. Finally, simulations show that LCF scheme outperforms AF in all SNR regimes for symmetric and asymmetric configurations. The proposed schemes present a trade-off between performance and complexity compared to the other schemes. Fig. 0 draws the fraction of power allocated by the relay to the common message in LCF. As the relay transmit power increases, the power allocated to the refinement message decreases. At high SNR regime, with favourable relay Fig. 8. Maximum sum-rate and relay power fraction allocated to the common message in LCF for asymmetric channels and equal powers. Here SNR = P /σr = P /σr = /σ and h =4, h =0.; LCF outperforms DF-lattice coding for SNR < db. a Maximum sum-rate for different schemes; b Relay power fraction ν in LCF. channel conditions, a small power fraction is sufficient for the refinement message to ameliorate the performance of LCF compared to the basic scheme as depicted in Fig. 9a. Remark 3: We have assumed in our system model perfect CSI at all nodes. However, in the proposed two lattice-based coding schemes LCF and LCF, this perfect knowledge of the channel state can be relaxed. In fact, in order to compress its received signal, the relay needs only the module of the channel gains to reconstruct its encoding scheme. For each terminal, the decoder uses the available side information S i = h i X i that depends on its terminal-relay channel. Appropriate training sequences can be employed to estimate the channel of the relay. Furthermore, each decoder estimates only its unknown part of the relay received signals. It is shown in Sections III-B and IV-B that that the link between both terminals is equivalent to a Gaussian channel for both proposed schemes for both proposed schemes. Thus, a training sequence can also be used in order to estimate at each decoder, the channel on the other link.

12 568 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 3, NO. 0, OCTOBER 04 Fig. 9. Achievable rate regions for different channel and power settings. a P =0dB, P =0dB, =5dB, h =4, h =0.5;bP =0dB, P =0dB, =0dB, h =4, h =0.; cp =5dB, P =4dB, =4dB, h =4and h =0.5; dp =5dB, P =3dB, =3dB, h =4and h =0.5; ep =db, P =db, =5dB, h =and h =0.5. VI. CONCLUSION In this paper, we studied the problem of exchanging messages over a Gaussian two-way relay channel. We derived two achievable rate regions based on compress and forward lattice coding. In the proposed schemes, the relay uses a lattice based Wyner-Ziv encoding by taking into account the presence of the side information at each node. i.e., the signal broadcasted by the relay includes also the signal that has been transmitted by each user to the relay during the first MAC transmission phase. First, we developed a coding scheme in which the relay broadcasts the same signal to both terminals. We showed that this scheme offers the same performance as random coding based compress-and-forward protocol [8]. Then, we

13 SMIRANI et al.: ACHIEVABLE RATE REGIONS FOR TWO-WAY RELAY CHANNEL USING NESTED LATTICE CODING 569 Fig. 0. Relay power allocation fraction ν in LCF for different fixed relay transmit power and equal SNR = P /σr = P /σr at both terminals, h =4and h =0.5. proposeed, and analyzeed the performance of, an improved coding scheme in which the relay sends not only a common description of its output, but also an individual description that is destined to be recovered by only the user who experiences better channel conditions and better side information. We showed that this results in substantial gains in rates. Numerical results demonstrate an enhancement of the achievable rate region over the basic scheme up to 00% for moderate SNR regime and asymmetric channel conditions. Also, the improved scheme outperforms classic amplify-and-forward at all SNR values, and classic and lattice coding decode-and-forward for certain SNR regimes. This scheme can achieve higher performance than DF strategies with less complexity at the relay without use of full CSI. Finally, it is worth mentioning that our schemes are based on structured codes that have low complexity compared to random coding from practical viewpoints. However, in these schemes, lattices codewords are used only at the relay while Gaussian codewords are used at the terminals nodes. Considering lattice codes at all the nodes can be even more appropriate for practical systems. REFERENCES [] R. Ahlswede, N. Cai, S.-Y. R. Li, and R. W. Yeung, Network information flow, IEEE Trans. Inform. Theory, vol. 46, no. 4, pp. 04 6, Jul [] S. Zhang, S. Liew, and P. Lam, Physical layer network coding, in Proc. ACM MOBICOM, Los Angeles, CA, USA, 006, pp [3] R. Knopp, Two-way wireless communication via a relay station, in GDR-ISIS Meet., Paris, France, Mar. 007, pp [4] S. J. Kim, P. Mitran, and V. Tarokh, Performance bounds for bidirectional coded cooperation protocols, IEEE Trans. Inform. Theory, vol. 54, no., pp , Nov [5] B. Rankov and A. Wittneben, Spectral efficient signaling for half-duplex relay channels, in Proc. ACSSC, Pacific Grove, CA, USA, Nov. 005, pp [6] R. Knopp, Two-way radio networks with a star topology, in Proc. Int. Zurich Semin. Commun., Zurich, Switzerland, Feb [7] M. P. Wilson, K. Narayanan, H. D. Pfister, and A. Sprintson, Joint physical layer coding and network coding for bidirectional relaying, IEEE Trans. Inform. Theory, vol. 56, no., pp , Nov. 00. [8] B. Nazer and M. Gastpar, Compute-and-forward: Harnessing interference through structured codes, IEEE Trans. Inform. Theory, vol. 57, no. 0, pp , Oct. 0. [9] Y. Song, N. Devroye, H.-R. Shao, and C. Ngo, Lattice coding for the two-way two-relay channel, in Proc. IEEE ISIT, 03, pp [0] B. Rankov and A. Wittneben, Achievable rate regions for the two-way relay channel, in Proc. IEEE Int. Symp. Inf. Theory, Seattle, WA, USA, Jul. 006, pp [] T. M. Cover and A. E. Gamal, Capacity theorems for the relay channel, IEEE Trans. Inform. Theory, vol. 5, no. 5, pp , Sep [] C. Schnurr, T. J. Oechtering, and S. Stanczak, Achievable rates for the restricted half-duplex two-way relay channel, in Proc. 4st ACSSC, Pacific Grove, CA, USA, Nov [3] S. J. Kim, N. Devroye, P. Mitran, and V. Tarokh, Comparison of bidirectional relaying protocols, in IEEE Sarnoff Symp., Princeton, NJ, USA, Apr. 008, pp. 5. [4] D. Gunduz, E. Tuncel, and J. Nayak, Rate regions for the separated twoway relay channel, in Proc. 46th Annu. Allerton Conf. Commun., Control, Comput., Urbana-Champaign, IL, USA, Sep. 008, pp [5] J. H. Conway and N. J. Sloane, Sphere Packings, Lattices and Groups, 3rd ed. New York, NY, USA: Springer-Verlag, 998. [6] Y. Song and N. Devroye, A lattice compress-and-forward scheme, in Proc. IEEE ITW, 0, pp [7] Y. Song and N. Devroye, Lattice codes for the Gaussian relay channel: Decode-and-forward and compress-and-forward, IEEE Trans. Inform. Theory, vol. 59, no. 8, pp , Aug. 03. [8] S. Smirani, M. Kamoun, M. Sarkiss, A. Zaidi, and P. Duhamel, Wyner- Ziv lattice coding for two-way relay channel, in Proc. Int. Conf. ATC, Hanoi, Vietnam, Oct. 0. [9] R. Zamir, S. Shamai, and U. Erez, Nested linear/lattice codes for structured multiterminal binning, IEEE Trans. Inform. Theory, vol. 48, no. 6, pp , Jun. 00. [0] J. Nayak, E. Tuncel, and D. Gunduz, Wyner-Ziv coding over broadcast channels: Digital schemes, IEEE Trans. Inform. Theory, vol. 56, no. 4, pp , Apr. 00. [] W. Nam, S.-Y. Chung, and Y. H. Lee, Capacity of the Gaussian twoway relay channel to within / bit, IEEE Trans. Inform. Theory,vol.56, no., pp , Nov. 00. [] Y. Tian, D. Wu, C. Yang, and A. F. Molisch, Asymmetric two-way relay with doubly nested lattice codes, IEEE Trans. Wireless Commun., vol., no., pp , Feb. 0. [3] U. Erez and R. Zamir, Achieving / log + SNR on the AWGN channel with lattice encoding and decoding, IEEE Trans. Inform. Theory, vol. 50, no. 0, pp , Oct [4] R. Zamir and M. Feder, On lattice quantization noise, IEEE Trans. Inform. Theory, vol. 4, no. 4, pp. 5 59, Jul [5] G. Poltyrev, On coding without restrictions for the AWGN channel, IEEE Trans. Inform. Theory, vol. 40, no. 5, pp , Mar [6] U. Erez, S. Litsyn, and R. Zamir, Lattices which are good for almost everything, IEEE Trans. Inform. Theory, vol. 5, no. 0, pp , Oct [7] T. Cover and J. Thomas, Elements of Information Theory. New York, NY, USA: Wiley, 99. [8] A. Wyner, The rate-distortion function for source coding with side information at the decoder II: General sources, Inf. Control, vol. 38, no., pp , Jul [9] S. J. Kim, N. Devroye, P. Mitran, and V. Tarokh, Achievable rate regions and performance comparison of half duplex bi-directional relaying protocols, IEEE Trans. Inform. Theory, vol. 57, no. 0, pp , Oct. 0. Sinda Smirani received the B.Eng. degree in signals and systems and the M.Sc. degree in communication and electronic systems from Tunisia Polytechnic School, La Marsa, Tunisia, in 007 and 008, respectively, and the Ph.D. degree in physics from the Université Paris-Sud, Orsay, France, in 04. From April 00 to 03, she pursued her Ph.D. degree in the Communicating Systems Laboratory, CEA, LIST, Gif-sur-Yvette, France. From September 008 to January 00, she was a Research and Development Engineer with kopileft, Tunis, Tunisia. She is currently a Research Engineer with Orange Labs, Issy-les-Moulineaux, France. Her research interests include cooperative wireless communications, network coding, and, recently, interference management for future networks.

14 560 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 3, NO. 0, OCTOBER 04 Mohamed Kamoun received the B.Eng. degree from the École National Supérieure de Techniques Avancées, Paris, France, in 00; the Master s degree in semiconductor physics from the Université Paris- Sud, Orsay, France, in 00; the Master s degree in digital communication from the École Nationale Supérieure des Télécommunications, Paris, in 00; and the Ph.D. degree from the Université Paris-Sud in 006. From 00 to 008, he was a Research Engineer with Motorola Laboratories, Paris. Since 009, he has been with the Communicating Systems Laboratory, CEA, LIST, Gif-sur-Yvette, France. His research interests include cooperative and energy-efficient wireless communication, distributed processing, and storage. Mireille Sarkiss received the degree in electrical engineering from the Faculty of Engineering, Lebanese University, Beirut, Lebanon, in 003 and the M.S. and Ph.D. degrees in communications and electronics from TELECOM ParisTech, Paris, France, in 004 and 009, respectively. From February 009 to June 00, she pursued postdoctoral research in the Department of Communications and Electronics, TELECOM ParisTech. Since October 00, she has been a Research Engineer with the Communicating Systems Laboratory, CEA, LIST, Gif-sur-Yvette, France. Her research interests include wireless communications, particularly energy-efficient and cooperative communications, resource management techniques, lattice coding and decoding, network coding, and, recently, distributed coding and security for storage systems. Abdellatif Zaidi received the B.S. degree from the École Nationale Supérieure de Techniques Avancées, ENSTA ParisTech, Paris, France, in 00 and the M.Sc. and Ph.D. degrees from the École Nationale Supérieure des Télécommunications, TELECOM ParisTech, Paris, in 00 and 005, respectively, all in electrical engineering. From December 00 to December 005, he was with the Department of Communications and Electronics, TELECOM ParisTech and the Signals and Systems Laboratory, CNRS/Supélec, France, pursuing his Ph.D. degree. From May 006 to September 00, he was a Research Assistant with the École Polytechnique de Louvain, Université Catholique de Louvain, Belgium. In the Fall of 007 and Spring of 008, he was a Research Visitor with the University of Notre Dame, Notre Dame, IN, USA. He is currently an Associate Professor with the Université Paris-Est Marne-la- Vallée, France, where he is also a member of the Laboratoire d Informatique Gaspard-Monge LIGM. His research interests cover a broad range of topics from network information theory and signal processing for communication. Of particular interest are the problems of multiterminal information theory, Shannon theory, relaying and cooperation, network coding, physical-layer security, source coding, and interference mitigation in multiuser channels. Dr. Zaidi serves as an Editor for the EURASIP Journal on Wireless Communications and Networking. Pierre Duhamel F 98 was born in France in 953. He received the B.Eng. degree in electrical engineering from the National Institute for Applied Sciences INSA, Rennes, France, in 975 and the Dr.Eng. and D.Sc. degrees from Orsay University, Orsay, France, in 978 and 986, respectively. From 975 to 980, he was with Thomson-CSF now Thales, Paris, France, where his research interests were in circuit theory and signal processing, including digital filtering and analog fault diagnosis. In 980, he joined the National Research Center in Telecommunications CNET, now Orange Labs, Issy-les-Moulineaux, France, where his research activities were first concerned with the design of recursive CCD filters. Later, he worked on fast algorithms for computing Fourier transforms and convolutions and applied similar techniques to adaptive filtering, spectral analysis, and wavelet transforms. From 993 to September 000, he was a Professor with ENST, Paris National School of Engineering in Telecommunications with research activities focused on signal processing for communications. From 997 to 000, he was the Head of the Signal and Image Processing Department. He is currently with CNRS/LS Laboratoire de Signaux et Systemes, Gif-sur-Yvette, France, where he is developing studies in signal processing for communications including equalization, iterative decoding, multicarrier systems, and cooperation and signal/image processing for multimedia applications, including joint source-protocol-channel coding. He is currently investigating the connections between communication theory and networking. Dr. Duhamel was the Chairman of the DSP Committee from 996 to 998, a member of the SP for Com Committee until 00, and a member of the IEEE Signal Processing for Communications and Networking Technical Committee from 004 to 00. He was an Associate Editor of the IEEE TRANSACTIONS ON SIGNAL PROCESSING from 989 to 99, an Associate Editor of the IEEE SIGNAL PROCESSING LETTERS, and a Guest Editor of the IEEE TRANSACTIONS ON SIGNAL PROCESSING Special Issue on Wavelets.