Expanding-Window BATS code for scalable video multicasting over erasure networks

In this paper, we consider scalable video multicasting over erasure networks with heterogeneous video quality requirements. With random linear network coding (RLNC) applied at the intermediate nodes, the information received by the destinations is determined by the associated channel rank distributions, based on which we obtain the optimal achievable code rate at the source node. We show that although a concatenation of priority encoded transmission (PET) with RLNC achieves the optimal code rate, it incurs prohibitive high coding complexity. On the other hand, batched sparse (BATS) code has been recently proposed for unicast networks, which has low coding complexity with near-optimal overhead. However, the existing BATS code design cannot be applied for multicast networks with heterogeneous channel rank distributions at different destinations. To this end, we propose a novel expanding window BATS (EWBATS) code, where the input symbols are grouped into overlapped windows according to their importance levels. The more important symbols are encoded with lower rate and hence they can be decoded bymore destinations, while the less important symbols are encoded with higher rate and are only decoded by the destinations with high throughput for video quality enhancement. Based on asymptotical performance analysis, we formulate the linear optimization problems to jointly optimize the degree distributions for eachwindowand thewindowselection probabilities. Simulation results show that the proposed EW-BATS code satisfies the decoding requirements with much lower transmission overhead comparedwith separate BATS code, where the degree distributions are separately optimized for each destination.