Network coding ; Raptor codes ; Video streaming ; Information theory; Optimization Theory ; Overlay networks; Error resiliency
Park Hyunggon and Thomos Nikolaos and Frossard Pascal (2014), Approximate Decoding Approaches for Network Coded Correlated Data, in Elsevier Signal Processing
, 93(1), 109-123.
Tosic Tamara and Thomos Nikolaos and Frossard Pascal (2013), Distributed Sensor Failure Detection in Sensor Networks, in Elsevier Signal Processing
, 93(2), 399-410.
Bourstoulatze Eirina and Thomos Nikolaos and Frossard Pascal (2012), Correlated-aware Reconstruction of Network Coded Sources, in 2012 International Symposium on Network Coding, NetCod 2012
Bourtsoulatze Eirina and Thomos Nikolaos and Frossard Pascal (2012), Distributed Rate Allocation With Intersession Network Coding, in 2012 19th International Packet Video Workshop, PV 2012
Thomos Nikolaos and Frossard Pascal, Toward One Symbol Network Coding Vectors, in IEEE Communicationa Letters
This project focuses on the deployment of low-cost network coding methods for video streaming in overlay networks. It is the follow up work of the Ambizione project with reference number PZ00P2-121906. Here, we plan to continue the exciting work and promising developments of the early part of the project that due to time limitations we were not able to complete. In PZ00P2-121906 project, we have proposed among others low complexity network coding schemes, prioritized network codes to address clients heterogeneity, inter-session network codes, and techniques for approximate network codes decoding. In details, we have already presented a low-cost network coding method based on Raptor codes that first achieves close to linear decoding and encoding times. For decentralized systems, we have proposed another system that employs randomized network coding and restricts the coding operations in selective positions. It is shown that few network coding nodes in large overlay networks are enough to notice large gains in terms of throughput and delay. To keep the computational complexity low all other nodes are store-and-forward. We have defined a game that decides about the network coding positions based on the willingness of network nodes to perform network coding. We have coped with the problem of clients receiving insufficient number of packets to fully recover the transmitted data. Thus, we have developed a method that uses data correlation to enhance data reconstruction. This scheme is the only that provides a systematic framework for data recovery in case of severe losses that is applicable to various types of data. We have also considered the case of multiple concurrent streams that compete for the network resources and first present a general methodology that scales to any arbitrary number of sources. Finally, we have designed a receiver driven UEP protocol based on network coding for video communication. This distributed system solves a simple optimization algorithm to find the optimal coding strategy at nodes. It allows system users to improve their experience and exploit better their resources. The developed randomized network coding method for multiple concurrent streams requires centralized knowledge about network topology and statistics. In this project, we will extend this technique to distributed systems. For low complexity, we will also consider the application of Raptor network coding. Novel source and channel rate allocation algorithm will be devised to take into account the multiple concurrent sources and remove the need for resource allocation algorithms that pre-allocate the bandwidth to the concurrent streams. We have shown that in many cases the sparse application of network coding is very efficient. Here, to further improve resiliency of the developed network coding techniques to network dynamics we will apply online learning methods. This will enable on-the-fly decision about the optimal coding operations and provide maximal quality streams with minimal delay. The designed approximate decoding techniques have made apparent that for rank deficient systems and correlated sources, decoding is possible by taking into account the correlation. To further enhance the performance of systems employing approximate decoding, we propose to benefit from the data correlation at encoding. For example, in wireless communication nodes can exploit overheard data from other nodes and the fact that they interfere with each other.