WP 3

WP3: Channel shaping, and transceiver algorithm design (partners P1, P2, P3, P4, INT1).

In digital communication systems, a number of parameters of different links active at the same time have to be selected and can be optimized according to some performance metric. At the level of individual links, assuming each link can transmit over several physical resources (time slots, frequencies, antennas, conductors of backplane interconnects, twisted pairs, ), the parameters to be considered are, for each resource, the power allocated, the error correction mechanism and the coding rate, and the number of bits per symbol (size of the constellation). At this level, there is also possible room for joint precoding and decoding of the symbols across multiple resources (e. g. space-time coding).

When several links are simultaneously activated, the parameters of the different resources have to be properly determined, because the different links compete to use the same resources (multiple access problem) and/or because interference mechanisms exist between the resources (e.g. frequency used by different links in different cells or twisted pairs). One strategy is resource coordination, meaning that the power assigned to the different resources of the different links is optimized according to a chosen performance metric that takes an interference model into account. This optimization can be conducted by either a central controller, which is provided with all the CSI required (that is between all resources of all links) or in a distributed manner. In the latter case, several strategies are again possible, depending on the information made available to the different links, which have to perform the optimization individually, and the possible exchange mechanism implemented to circulate some messages among the links. Beyond resource coordination, signal coordination can be preferred. In this case, all resources of all (coordinated) links are globally shared and global precoding can then be implemented, be it again in a centralized or in a distributed way with limited channel and/or data symbol information exchange. Moreover, in all cases the CSI can possibly be only partially or imperfectly known (due to quantization or the estimation process ([Meyr1997], [Jacobs2009], [Avram2010]) used). Finally, on top of the design selected for the transmitter(s), the issue of receiver design still remains.

Among the resources to be shared, frequency and space are definitely important. As a matter of fact, most broadband systems considered nowadays use multicarrier-based modulation or multiple access in one way or another (be it OFDM or filter bank-based). Besides this, MIMO is also a basic ingredient of most systems. MIMO can be implemented by means of collocated antennas, or in a distributed manner, by means of nodes building a distributed coalition, or even with the help of one or several relays assisting the other nodes. This raises additional challenges to be taken into account such as the impact of CSI and/or data sharing over imperfect links between nodes, the grouping of nodes or the assignment of the relays to the nodes to name a few.

The SWPs below describe some of the challenges that will receive attention. When appropriate, algorithms will be validated by means of the software radio interfaces available in the BESTCOM network.

SWP3.1: Signal/spectrum coordination and network MIMO.

Robust centralized and distributed resource allocation for multi-cell OFDMA systems.

In order to manage inter-cell cochannel interference (CCI), some margin-adaptive (power minimization under information rate contraints) centralized resource allocation (RA) algorithms have been proposed in ([Pietrzyk004], [Abrardo2007]). Rate-adaptive (sum rate maximization under power contraints) centralized RA algorithms for each subcarrier individually have been reported in ([Kiani2008], [Gjendemsj2008]). However, a joint RA over all subcarriers is preferred in order to better exploit the frequency and multiuser diversity inherent in OFDMA systems. To this end, two rate-adaptive centralized RA algorithms have been proposed in ([Venturino2008], [Wang2011]). Similar rate-adaptive RA allocation algorithms have also been developed for DMT systems (without carrier assignment) in digital subscriber line (DSL) networks ([Cendrillon2006]). All these algorithms are based on a dual decomposition approach, which leads to a high computational complexity. Moreover, perfect channel state information (CSI) is assumed to be available. Therefore, a research effort will be devoted to the design of centralized RA algorithms to optimize the carrier assignment and power allocation in a multi-cell OFDMA system under imperfect CSI, and possibly with cognitive users. Novel robust centralized RA algorithms will therefore be investigated, under performance metrics provided by WP1 and duly accounting for HARQ. Distributed RA algorithms have been proposed for DSL systems and are either cooperative ([Tsiaflakis2008]) or non-cooperative. For multi-cell OFDMA systems, in a cooperative setup each BS optimizes its RA to maximize the weighted sum rate (WSR) of all users of all cells, depending on the local CSI (i.e. the CSI of the cell it serves), as well as certain information sent from other BSs ([Papandriopoulos2006]). On the contrary, non-cooperative distributed algorithms target the optimization by each BS of the WSR of its own users (rather than all users), and no information is passed between BSs ([Shum2007], [Scutari2008]). In multi-cell OFDMA systems, the allocation of both power and subcarriers to multiple users needs to be optimized, which is more complicated than only optimizing the power allocation as in DSL systems. Novel robust distributed RA algorithms will therefore be investigated, under performance metrics provided by WP1 and for systems possibly including cognitive users.

Generalized transceivers for multiuser (network) MIMO systems : robust, centralized and distributed designs.

Uplink-downlink duality has been used to solve SINR (signal to interference plus noise), information rate and MSE precoder/decoder optimization for downlink channels in ([Shi2007], [Hunger2008], [Shi2008b], [Bogale2009], [Bogale2010a], [Bogale2011]). SINR, information rate and MSE uplink-downlink duality is established here under global power constraints and identical noise variances at all antennas. A centralized MSE solution for precoder/decoder optimization in coordinated BS (network MIMO) systems has been reported in ([Shi2008a]). A centralized multiuser MIMO precoder/decoder design procedure for DMT systems in DSL networks, which also includes multiuser power allocation, has been reported in ([Moraes2011]). Two MSE distributed solutions have been reported in ([Bogale2010]). Perfect CSI is always assumed to be available in these works.

Research will therefore be conducted on precoder/decoder optimization for multiuser MIMO and network MIMO configurations, with generalized power constraints and noise assumptions, and under new performance metrics provided by WP1. Duality methods will be investigated. Centralized as well as cooperative and non-cooperative distributed solutions will be proposed. Robust solutions for the case where perfect CSI is not available will also be studied, based on worst case scenarios or on stochastic approaches. Extensions to OFDMA setups will eventually be studied.

SWP3.2: Distributed MIMO and cooperative communications.

Multiple access relay channel based on the compute-and-forward strategy.

Recently, the compute-and-forward (CF) strategy, based on lattice codes, has been proposed ([Nazer2011]). The relay decodes a linear combination of the received messages instead of the individual messages, which can be seen as a form of network coding. A receiver can recover the transmitted messages from a sufficient number of linear combinations. The critical point is that the coefficients of the equations must be integer-valued. The CF strategy has been studied for the two-way relay channel ([Wilson2010]), the multiple access channel (MAC) and MIMO channels ([Nazer2007], [Zhan2009]). CF improves the performance of the system compared to other strategies for certain regimes. The multiple access relay channels (MARC) using CF, the encoding/decoding method and the achievable rate have been presented in ([ElSoussi2011]). The optimization of the achievable information rate depends on the integer coefficients and the power allocated to the transmitters and the relay, which makes the problem mixed-integer non-linear non-convex ([ElSoussi2012]).

The objective is to further explore CF, for the WP1 metrics, and investigate it for multicarrier and/or multiple relays setups which will open new research paths. Moreover, the potential of having different lattice codes at different nodes will be studied.

Robust resource allocation (RA) designs for multiple decode-and-forward (DF) relays aided OFDMA systems.

Compared to conventional OFDMA transmission, relay-aided OFDMA transmission raises more complicated RA problems, since it introduces extra tasks such as deciding the transmission mode of each subcarrier (relayed or not), determining assisting relays and the power allocation to them for each relay-aided subcarrier, besides assigning destination and source power at each subcarrier. Therefore, novel efficient RA algorithms are solicited for relay-aided OFDMA transmission systems. Recently, there have been a few related works reported ([Ng2007], [Cui2009], [Salem2010], [Kador2010], [Hsu2011], [Wang2011a], [Wang 2011b]). However, none of these works consider the joint optimization of carrier assignment and pairing, power allocation and transmission mode. Moreover, perfect channel state information is assumed to be available. This motivates us to investigate robust RA for such systems, possibly with cognitive users, and for the metrics that will be exchanged with WP1.

SWP3.3: Channel and transceiver optimization for multi-Gbit/s backplane interconnects.

In order to cope with the increased signal distortion and crosstalk associated with multi-Gbit/s transmission on backplane interconnects, alternatives to uncoded NRZ signaling have been proposed in the literature. A first class of alternatives involves bandwidth-efficient signaling ([Sin2005], [Lee2008], [Kam2009]), such as duobinary signaling and multilevel pulse amplitude modulation (PAM) signaling. For a given bitrate, these signaling formats occupy less bandwidth than NRZ, which makes them less vulnerable to channel dispersion; however, for a given supply voltage the spacing between the levels of the undistorted signal is smaller than for NRZ, thereby increasing the sensitivity to noise and interference. A second class of alternatives involves error correcting coding : in ([Nar2010]) the use of Bose-Chaudhuri- Hocquenghem (BCH) codes with an error correcting capability of 1, 2, or 3 bits is proposed; in ([Far2006], [Far2008]) the use of coded modulation with multi-level PAM signaling is presented. Coding improves the error performance, at the expense of increased power consumption and implementation complexity.

The comparison between NRZ, 4-PAM and duobinary signaling conducted in ([Lee2008], [Kam2009]) seems to favor NRZ signaling in terms of bitrate. However, in this comparison no error correcting coding has been considered. Also, the ranking of these signaling formats might change depending on the considered channel transfer functions and crosstalk spectra, and on whether equalization is present at the receiver. Hence, these results from literature should be interpreted with great care.

The transceiver algorithms proposed in the literature treat crosstalk as additional noise. However, because of the electromagnetic coupling, the set of conductors carrying the transmitted signals should be viewed as a MIMO channel, rather than a set of individual single-input single-output (SISO) channels that suffer from crosstalk. Adopting this MIMO channel view provides the potential for crosstalk cancellation (or at least substantial crosstalk reduction), by allowing cooperation between transmitters and/or receivers (in a similar way as in wireless multi-antenna communication systems ([Kai2005])); this crosstalk reduction would allow a further increase of the bitrate on backplane interconnects. In SWP3.3, we will derive transceiver algorithms that cancel/reduce crosstalk by exploiting the spatial dimension of the MIMO channel model, using cooperation at the transmitting side (linear space-time precoding), at the receiving side (space-time equalization), or a combination of both. These algorithms also involve channel estimation, synchronization and error correction. Transceiver algorithms will be optimized, subject to realistic constraints on power consumption, implementation complexity and chip area, to increase the achievable bitrate per lane well beyond 25 Gbit/s (and aspiring to reach about 100 Gbit/s) for a given target error performance.

This research in SWP3.3 will make use of the channel macro-models of the backplane interconnects derived in SWP1.2 based on the results from SWP2.1. As the values of the parameters of this model reflect the design choices regarding the layout and the materials of the circuit board, the outcome of the transceiver optimization will reveal the effect of the circuit board design on the transceiver performance. In close cooperation with the researchers that provide the channel model, a global optimization of the board and the transceiver will be envisaged for further increasing the bitrate, subject to restrictions regarding both the tranceiver (power consumption, complexity, chip area) and the circuit board (layout, materials).