Clustering source/channel rate allocations for receiver-driven multicast under a limited number of streams

We propose a clustering based algorithm to address the problem of receiver-driven multicast with error control under the constraint of a given limited number of streams, where each stream represents the same audio or video source having the same transmission rate but a different redundancy. Our algorithm formulates a design methodology for this set of packetized streams such that it best covers, in a global expected distortion sense, a collection of possible packet erasure channels with respect to a probability distribution over the collection. The resulting set of streams is suitable for Internet multicast to a collection of receivers with various packet loss characteristics. The algorithm is essentially a K-means algorithm, and in one simulation, reduces the expected distortion by over 4 dB with only 4 streams