Rank Metric Convolutional Codes with Applications in Network Streaming.

摘要

We consider end-to-end error correction codes for real-time streaming over a network, where the source and destination are nodes in a graph. Links between nodes are erasure channels and the source-destination relationship a rank-deficient channel matrix.;In an isolated rank loss model, the channel matrix arbitrarily decreases to a minimum rank within each sliding window of network uses. We prove the column sum rank metric determines the effectiveness of a code and construct a family of codes that achieve the maximum column sum rank.;In a burst rank loss model, a fraction of links are simultaneously erased for consecutive network uses. We derive the capacity and prove achievability with a layered construction using the previously introduced codes. Simulations over statistical models reveal our construction to be superior over baseline codes.