Fair rate allocation of scalable multiple description video for many clients

Abstract

Peer-to-peer networks (P2P) form a distributed communication infrastructure that is particularly well matched to video streaming using multiple description coding. We form M descriptions using MDC-FEC building on a scalable version of the "Dirac" video coder. The M descriptions are streamed via M different application layer multicast (ALM) trees embedded in the P2P network. Client nodes (peers in the network) receive a number of descriptions m < M that is dependent on their bandwidth. In this paper we consider the optimization of the received video qualities, taking into account the distribution of the clients' bandwidth. We propose three "fairness" criteria to define the criterion to be optimized. Numerical results illustrate the effects of the different fairness criteria and client bandwidth distributions on the rates allocated to the compressed video layers and multiple descriptions.

The Orchard Algorithm: Building Multicast Trees for P2P Video Multicasting Without Free-Riding

“…The idea of using MDC [7] in ALM has been proposed before [19]- [21]. Pouwelse et al [20] propose the use of a bartering mechanism, but do not provide an algorithm.…”

Joint optimization of the redundancy of multiple-description coders for multicast

“…Taal et al discussed in [10,11] possible ways of designing a fair optimization criterion. In our work, we incorporate fairness by use of the weighting function.…”

“…In the limit of small cell widths, the sum in (10) becomes a Riemann sum, allowing for the approximation J * δ ≈ min π I 1 24 w(p)δ(p, p)fP (p)Δ 2 (p) dp, (11) where the cell widths are assumed to lie on a continuous function of p, such that Δi = Δ(π(i)) ≈ Δ(p) for all p ∈ Ii, and it was used thatδ(π(i), π(i)) ≈δ(p, p) for all p ∈ Ii.…”

“…Taal et al [10,11] proposed using one multicast tree for each description of a multiple-description coder. Besides the advantage that the receiver is able to subscribe to any combination of descriptions without loss of quality, it is obvious that the system is inherently robust to packet-loss.…”

“…Besides the advantage that the receiver is able to subscribe to any combination of descriptions without loss of quality, it is obvious that the system is inherently robust to packet-loss. However, the system optimization in [10,11] was performed only with respect to receiver capacities, assuming no packet-loss.…”