Nous vous rappelons que, afin de garantir l'accès de tous les inscrits aux salles de réunion, l'inscription aux réunions est gratuite mais obligatoire.
Inscriptions closes à cette réunion.
17 personnes membres du GdR ISIS, et 18 personnes non membres du GdR, sont inscrits à cette réunion.
Capacité de la salle : 88 personnes.
Instructions pour une demande de mission par le GdR ISIS
Le GdR ISIS prend en charge les déplacements des organisateurs des réunions et des orateurs. Le GdR prend aussi en charge les déplacements des participants aux réunions membres d'un laboratoire adhérent du GdR dans la limite d'un doctorant et d'un permanent par laboratoire académique et par réunion, ou d'une personne par adhérent du club des partenaires et par réunion.
La plus grande partie du budget du GdR ISIS est consacrée à la prise en charge de ces missions. Pour que le GdR puisse financer le plus grand nombre de réunions, les participants à ces réunions sont vivement incités à choisir les billets les moins chers. Seuls les billets de train ou d'avion en deuxième classe, non échangeables et non remboursables sont pris en charge. Le GdR se réserve le droit de refuser une demande de billet dont le prix excède la moyenne des prix couramment pratiqués pour le trajet de la mission.
Pour le transport et pour l'hébergement, vous êtes priés d'utiliser le portail SIMBAD du CNRS si vous en avez la possibilité. Cela est en particulier obligatoire si vous êtes membre d'une unité CNRS (UPR, UMR, UMI, URA, FRE). Les réservations d'hôtel sont possibles si la réunion dure plus d'une journée ou si le lieu d'habitation le justifie. Dans le cas où le laboratoire n'est pas une unité CNRS, merci d'envoyer votre demande de prise en charge de la mission à l'adresse DR01.SoutienUnites@cnrs.fr en précisant que la mission relève du GdR ISIS. Si vous utilisez votre véhicule personnel pour une distance supérieure à 300 kilomètres (aller+retour), le GdR ISIS ne rembourse pas vos frais de transport.
Les demandes de mission et les réservations sur le site SIMBAD doivent impérativement être effectuées au moins deux semaines avant la date de la mission.
Aucun remboursement de frais de transport ou d'hôtel avancés par l'agent ne peut plus être effectué au retour de la mission.
NB: The afternoon session will take place in Amphi Emeraude, TélécomParisTech.
8th of June 2016, Télécom ParisTech, 46 rue Barrault, 75013 Paris:
Energy efficiency is a major issue in the design of modern communication systems. Information theory has widely investigated fundamental limits of communication over noisy channels under transmit power requirements. These fundamental limits led to the design of very efficient error-correction systems, such as LDPC and Turbo Codes. However, it was also observed empirically that processing power can sometimes dominate transmit power, especially for short distance communications.
One of the main current concerns is to design error-correction systems that minimize the total (transmit + process) energy consumption rather than just the transmit energy consumption. This concern has recently received increased attention from both academia and industrials, and the following two issues are being explored. First, from an information theoretic perspective, there is a need to also take into account processing energy, and to design error-correction systems that guarantee the transmission while minimizing the total energy consumption. Second, as the processing energy highly depends on the technology (voltage supply, 45nm or 65nm CMOS, etc.) and on the circuit architecture, there is a need to construct low energy consumption hardware implementations of error-correction systems. These two issues are highly related, since for example, hardware implementation could provide energy models to information theory, while fundamental limits could give insights for the implementation.
The objectives of the meeting will be to present the recent contributions on error-correction systems with low energy consumption, and to explore the connections between the information theory and hardware implementation issues.
9:30 - 9:45 : Opening talk
9:45 - 10:35: Andreas Burg, EPFL, Switzerland
10:35 - 11:10 : François Leduc-Primeau, LabSTICC, Télécom Bretagne, France
11:10 - 11:30 : Coffee break
11:30 - 12:05 : Valentin Savin, - CEA-LETI, France
12:05 - 12:40 : Chris Winstead, - Utah State University, USA
12:40 - 14:15 : Lunch break
14:15 - 14:50 : Stefan Scholl, - Technische Universität Kaiserslautern, Germany
14:50 - 15:25 : Ingmar Land, Huawei Technologies, French Research Centre, France
15:25 - 15:45 : Coffee break
15:45 - 16:20 : Fakhreddine Ghaffari, ETIS /ENSEA - Université de Cergy-Pontoise - CNRS, France
16:20 - 16:45 : Bertrand Le Gal, INP Bordeaux - CNRS IMS - Université de Bordeaux, France
16:45 - 17:00 : Conclusions
Speaker: Andreas Burg
Authors: A. Balatsoukas-Stimming, R. Ghanaathian, A. Burg
In the first part of this talk, we will have a look at different architectures for LDPC decoders with very different performance requirements. To this end, we will systematically categorize architectures of existing decoder designs and discuss two examples in more detail. We will outline key architectural and implementation issues and show how to solve them and how to achieve low power consumption. In the second part of the talk, we will then consider Polar codes as a competing technology to LDPC codes. We will make an initial attempt to systematically compare the two types of codes (especially with respect to their decoder-implementation complexity), based on various implementations (our own and those in the literature).
Speaker: François Leduc-Primeau
Authors: François Leduc-Primeau, Frank R. Kschischang, and Warren J. Gross
Quasi-synchronous (QS) systems are systems based on synchronous circuits that experience occasional timing violations, and expose the resulting deviations to the system level. Optimizing the energy consumption of a QS system requires constructing a deviation model that represents the effect of timing violations on the system?s performance, and an energy model that captures the energy gains obtained by permitting these violations. We present a memoryless deviation model for QS LDPC decoders that is compatible with the assumption of codeword-independent errors while remaining accurate, and show how instances of the deviation-energy model can be constructed from a small number of Monte-Carlo simulations of a small test circuit. We can then use this model to find operating conditions that minimize the energy consumption of the decoder, subject to performance constraints and without increasing the size of the circuit. We describe an approximate optimization approach that provides a good sequence of operating conditions to be used in each decoding iteration, while guaranteeing that the solutions are feasible. Finally, we show how the finite-length performance and energy consumption of a QS decoder can be predicted for a specific average channel quality and maximum number of decoding iterations, and compare these predictions with the performance measured on the gate-level model of a decoder.
Speaker: Valentin Savin
In this talk we introduce a new theoretical framework, akin to the use of imprecise message storage in Low Density Parity Check (LDPC) decoders, which is seen as an enabler for cost-effective hardware designs. The proposed framework is the one of Non-Surjective Finite Alphabet Iterative Decoders (NS-FAIDs), and it is shown to provide a unified approach for several designs previously proposed in the literature. NS-FAIDs are optimized by density evolution for WiMAX irregular LDPC codes and we show they provide different trade-offs between hardware complexity and decoding performance. In particular, we derive a set of 27 NS-FAIDs that provide decoding gains up to 0.33 dB, while yielding a memory / interconnect reduction up to 25% / 30% compared to the Min-Sum decoder.
Speaker: Chris Winstead
Recent advances in LDPC decoding have revealed unexpectedly good performance for bit-flipping algorithms. Three important algorithm features are examined: noise enhancement, message memory and decoder rewinding (i.e. re-decoding). While the theory behind these features is still developing, empirical research has shown these methods to be important both for improved performance and reduced complexity. We discuss the architectural advantages and tradeoffs associated with these methods, and provide a review of recent ASIC results showing that bit-flipping decoders offer extremely low energy-per-bit and implementation area compared to traditional architectures.
Speaker: Stefan Scholl
LDPC codes are usually decoded by iterative belief propagation. However, especially for small block lengths it exhibits significant losses in coding gain compared to maximum likelihood decoding. An advanced decoding scheme is proposed, that combines several runs of belief propagation to improve the coding gain by up to 1.6 dB compared to sum-product decoding. Hardware architectures and implementation results for a 28 nm ASIC technology are presented. Besides the additional coding gain, the new decoder implementation copes with the great challenge of "dark silicon" currently arising in microelectronic systems.
Speaker: Ingmar Land
Authors: Ingmar Land, Valerio Bioglio, Frédéric Gabry, Loïg Godard
Polar coding has attracted lots of attention in research and industry in recent years. Originally developed as a low complexity scheme to provably achieve capacity, polar coding also represents a flexible high-performance scheme for practical code lengths. Rate adaptation is naturally possibly by choosing frozen sets of different sizes; length adaptation, one the other hand, requires careful application of puncturing or shortening in order to ensure performance. Decoding is possible by simple and fast successive-cancellation (SC) decoding or by advanced SC list decoding, where list length and thus complexity may be traded against performance.
Due to its flexibility in terms of code rate, code length, performance and decoding complexity, polar coding is currently being discussed as a candidate coding scheme for 5G. The presentation will review the options for encoding and decoding of polar codes and relate them to complexity and energy efficiency of decoding algorithms and decoding implementations.
Speaker: Fakhreddine Ghaffari
Authors: Fakhreddine Ghaffari, Khoa Le, David Declercq and Bane Vasic
In this presentation, we introduce a new and original method for the random generation of bits in the Probabilistic Gradient-Descent Bit-Flipping (PGDBF) decoder, which provides a highly efficient hardware implementation compared to others conventional methods while preserving the outstanding decoding performance of PGDBF. This new random generator method called Intrinsic-Valued Random Generator (IVRG) avoids generating the required random bits by using a sequence of bits from the existing decoder memory. Moreover, we analyze through statistical analysis the behavior of the PGDBF and identify the important features of the random generators, which are important to obtain large coding gains compared to the deterministic GDBF. Our results show that the high PGDBF performance can be obtained with a very small extra hardware complexity, without any performance loss.
Speaker: Bertrand Le Gal
Authors: Bertrand Le Gal, Camille Leroux, Christophe Jégo
During the last decade, many works on software implementations of decoders for different error correction code families were proposed. Indeed, from an designer point of view, programmable architectures offer some flexibility during development cycle and also at runtime. However, their throughput and energy efficiencies were far from real time application constraints. Since about four years, multiple efficient throughput implementations for ECC decoding were proposed (LDPC, Turbocodes, Polar codes ...). Associated, first, with the drastic reduction of the power consumption of the programmable architectures (x86, GPU, ARM) and secondly, with the Software Defined Radio or Cloud-Ran systems, the interest of such solutions becomes a reality. Whereas their efficiency is always lower than dedicated hardware architectures (e.g. for handled devices), their real time usages in modern communication systems where short development time and flexibility matter (e.g. base stations or embedded systems) is possible. The presentation will review the progresses in the field of ECC decoding on programmable architectures by providing a state of the art of the current throughputs and energy consumptions. Then, future challenges and improvement ways will be discussed.
Date : 2016-06-08
Lieu : Télécom ParisTech (amphi Estaunié / amphi Emeraude)
Thèmes scientifiques :
D - Télécommunications : compression, protection, transmission
Inscriptions closes à cette réunion.
Accéder au compte-rendu de cette réunion.
(c) GdR 720 ISIS - CNRS - 2011-2018.