THIRD WORKSHOP ON POWER LINE COMMUNICATIONS, OCTOBER 1-2, 2009, UDINE, ITALY 1 Some (Known) Facts on Multicarrier Communication R. Raheli (1) , M. Franceschini (2) , R. Pighi (3) , and G. Ferrari (4) (1) University of Parma, Italy, e-mail: raheli@unipr.it (2) IBM T. J. Watson Research Center, New York, USA, e-mail: franceschini@us.ibm.com (*) (3) Selta SpA, Piacenza, Italy, e-mail: r.pighi@selta.it (*) (4) University of Parma, Italy, e-mail: gianluigi.ferrari@unipr.it Abstract It is widely known and accepted that multicarrier communication is effective in frequency selective channels if resource allocation strategies are used, namely per-subcarrier adaptive modulation and coding. We discuss its efficiency for uniform resource allocation in terms of fundamental information-theoretic measures. These results are relevant to broadcast systems and, more generally, systems which cannot rely on a feedback channel. Index Terms Multicarrier Communication, Orthogonal Frequency Division Multiplexing (OFDM), Information Rate, Broadcast Systems. I. I NTRODUCTION M ULTICARRIER (MC) communication, typically based on orthogonal frequency division multiplexing (OFDM), has established itself as an efficient and convenient method in most scenarios, including wireline, powerline and wireless systems. MC systems are currently being intensively investigated and experimented in optical communication as well. The efficiency of the MC approach in terms of both performance and complexity is out of discussion in several communication scenarios, so the common belief tends to view it as a panacea. However, a closer look at some fundamental features of the MC approach reveals that it may exhibit inefficiencies in specific scenarios. In this respect, a recently investigated communication scenario is a channel impaired by impulse noise, which can be considered as a limiting factor in many applications, such as, for example, powerline, digital subscriber line (DSL), and wireless communication systems. Impulse noise typically originates from electrical and electromagnetic equipments and affects the transmission in the form of random bursts of relatively short duration and very high instantaneous power. Among a number of aspects, an element which played a role in the establishment of MC modulations was their improved robustness in impulse noise-limited communications, with respect to single-carrier (SC) schemes [1]. However, it has recently been shown that standard MC systems, employing interleaving and channel coding, are indeed less robust to impulse noise than corresponding SC schemes [2]. Interestingly, this result does not contradict previous findings obtained for uncoded systems, which show that MC schemes may outperform SC ones [1], because the loss manifests itself at rates of typical coded systems. The results in [2] are obtained analyzing the ultimate performance limits of SC and MC communication systems in terms of the achievable information rate (IR). From an engineering viewpoint, this performance measure is appealing because it provides a realistic information-theoretic benchmark in the fact that it takes inherently into account specific design constraints, such as the modulation format or the transmitted power spectrum. Following this approach to performance evaluation in terms of fundamental measures, this paper uses the achievable IR to analyze and compare SC and MC schemes in frequency selective additive white gaussian noise (AWGN) channels. II. SYSTEM MODEL AND PERFORMANCE MEASURE We consider the SC and MC system models shown in Fig. 1, parts (a) and (b), respectively. The information sequence A is input to quadrature amplitude modulation (QAM) devices. In the SC system, the sequence of QAM independent symbols X SC is directly sent over a frequency selective (FS) channel, which adds a sequence W of independent identically distributed thermal noise samples. In the MC system, the QAM sequence is input to an inverse discrete Fourier transform (IDFT) device, whose output sequence X MC is sent over the FS channel. The received sequences are denoted as Y SC and Y MC . For a general channel with discrete input X and continuous output Y , the IR is defined as [3] I (X ; Y )= h(Y ) - h(Y|X ) (1) where h(Y ) denotes the differential entropy rate of the channel output process Y , and h(Y|X ) denotes the differential entropy rate of the channel output given the channel input. In particular, h(Y|X ) equals the differential entropy rate of the noise (*) Work performed while M. Franceschini and R. Pighi were at the University of Parma. ISBN: 978-88-900984-8-2