ComEx Program, April 12, 14.30, room F3.08
- 14.30 : Laurent Decreusefond (Télécom ParisTech)
Abstract : Point processes with dependencies between atoms have been recently shown to be interesting models for radio networks. Determinantal point processes, which were originally defined to represent clouds of particles like electrons, are particularly promising. Unfortunately, performance of network systems under this assumption are much harder to evaluate than it is for Poissonian models. We show how we can compare precisely the value of some indicators, like the coverage probability, obtained for different sorts of point processes.
- 15.00 : Wan Kai (CentraleSupélec)
Abstract : Caching is an efficient way to reduce peak hour network traffic congestion by caching some content locally during off-peak hours. Caching can be divided into two phases: placement phase (each user stores some contents in his cache without knowledge of later requests) and delivery phase (after receiving the connected users' requests and according to their cache contents the central server broadcasts packets to users). If each user directly stores some bits of each file in his cache, the placement phase is called "uncoded". Recently, Maddah-Ali and Niesen proposed a coded caching scheme that utilizes an uncoded combinatorial cache construction in the placement phase, and a linear network code in the delivery phase; the scheme was shown to be optimal to within a factor 12. Our contribution can be divided into two parts. For the systems where the number of files is not less than the number of users, we prove by an index coding outer bound that under the constraint of uncoded cache placement, the Maddah-Ali and Niesen's scheme is indeed optimal. In other words, this result states that further improvement to the Maddah-Ali and Niesen's scheme in this regime can be obtained only by coded cache placement. For the systems where the number of files is less than the number of users, with the same uncoded placement as Maddah-Ali and Niesen's scheme, we propose a novel delivery scheme. The novel scheme can be shown to outperform existing caching strategies and to be optimal in several cases. It is subject of current investigation to determine the optimal delivery scheme in this case too.
Joint work with D. Tuninetti and P.Piantanida. The work of D. Tuninetti is partially funded by NSF.
- 15.30 : Andrea Araldo (Paris Sud)
As predicated by Cisco, in 2019 Internet traffic "will be equivalent to 64 times the volume of the entire global Internet in 2005". On the other hand, it has been observed that the content delivered through Internet shows a high redundancy, since there exist relatively few popular objects that are continuously transmitted. For this reason, distributing replicas of popular objects by means of network caching can help to cope with the traffic deluge.
However, classic caching techniques are not the most appropriate for current content delivery. First, network providers are loosing their capability to cache, since classic caching requires the perfect knowledge of content, which is less and less available, as content is encrypted by the respective content providers. We present a solution based on stochastic subgradient method to overcome this limitation. Second, classic caching methods usually focus on hit ratio maximization and other network-centric metrics. Our research takes a different direction directly considering the final objectives that can be achieved by caching. In particular, we propose caching techniques to i) reduce ISP interdomain traffic cost and ii) maximize the quality of video delivery provided to users. We show that our techniques outperform classic caching in fulfilling these goals. In this presentation, we put particular emphasis on the problem (ii) of video delivery. We observe that classic caching, based on the assumption that each user request can be satisfied by one and only one file, is not suitable for video delivery, where a single video can be served at different representations, i.e. different resolutions and bitrates. For this reason, we add a new dimension to the caching problem, the "representation selection", i.e. choosing which of the available representations must be cached and in which locations. We show the properties of the optimal solution, that maximizes user utility, and we provide a distributed online strategy that approaches the optimum.
- 16.00 : Wang Chien-Yi (Telecom ParisTech)
Abstract :Under the paradigm of caching, two request models have been considered in the literature. One is the quasi-static model studied by Maddah-Ali and Niesen. The other is the ergodic model studied by Wang, Lim, and Gastpar. In this talk, we present a general framework that incorporates the two request models. It is shown that they have a similar behavior for the single-user case but behave differently in general.
- 16.30 : Break
- 17.00 : Fanny Jardel (Télécom ParisTech)
Abstract : Product codes are compound codes built by the Kronecker product of small constituent codes. They are known for their many interesting algebraic properties and their low complexity iterative row-column decoding. Product codes are potential candidates for coding in networks such as distributive storage. In this presentation, we consider Maximum Distance Separable (MDS) constituents. We show the analysis of stopping sets for MDS-based product codes under iterative row-column algebraic decoding. New results on bipartite graphs enumeration, related to algebraic stopping sets, are given. A tight union bound to the performance of iterative decoding is established for the independent symbol erasure channel. We also present the proof that the performance of iterative decoding reaches the performance of Maximum-Likelihood decoding at vanishing channel erasure probability. Finally, in order to enhance our theoretical results, numerical results with MDS components are shown for product codes at different coding rates.
- 17.30 : Jonatan Krolikowski (CentraleSupélec)
Abstract :Caching popular content at the base stations promises to significantly reduce congestion in the backhaul links of future wireless networks. In the recent literature, different placement strategies have been proposed which show benefits with respect to performance measures, such as hit probability or delay.
Our current research is focused on network topologies with multi-coverage, i.e. areas in which user requests have the option to be routed to one of multiple base stations. When edge nodes are equipped by sufficient cache memories, such requests can have access to the union of elements cached at the different base stations, thus allowing the user to take advantage of diversity of cache placement in its neighbourhood.
Our approach takes load-balancing of user requests among base stations into consideration. Then, popular content is placed into caches, targeting two conflicting objectives: On the one hand, the installed cache content should attract as many users as possible, to reduce backhaul traffic. On the other hand, the number of requests should appear balanced among the covering stations, so that resources are fairly utilised.
In this talk, the system model, problem formulation, as well as an outlook onto possible solution strategies of this cache placement problem will be presented.
- 17.45 : Elmehdi Amhoud (Telecom ParisTech)
Abstract : The last two decades have known an exponential growth in the demand for network bandwidth. The optical fiber as a medium of transmission offers many degrees of freedom. Since frequency, time, phase, polarization have already been used to satisfy the demand for bandwidth, space remains the only degree of freedom that can be used in optical transmission systems to increase the capacity. Propagating modes through multimode fibers are affected by a non-unitary crosstalk known as mode dependent loss (MDL). The impact of MDL and its detrimental effect on the channel capacity was reported in many studies. We prove that Space time coding ,already designed for wireless communications, mitigate the non-unitary effects of mode dependent loss, allowing total mitigation of MDL in space division multiplexed transmission systems.