Multilayer network-coded ARQ for multiple unicast flows

The notion of joint network-coded ARQ for multiple unicast flows and superposition coding (multilayering) is introduced. We consider a single sender and multiple receiving users with feedback channels. Much like in a cellular system downlink, each user expects to receive different information. Superposition coding transforms each fading channel into several (or infinite number of) layers of packet erasure channels, and network-coded ARQ is utilized on groups of layers over the users. The expected sum-rate is considered, and optimized. Numerically optimized results indicate roughly additive gains due to layering and network coding. Some analytical results on special cases regarding number of users and layers, are given. For instance, for NC unicast ARQ, one layer and arbitrary number of users, we derive the optimal expected sum-rate in a novel parameterized closed form.