XORSs in the air: Practical wireless network coding

The title says it all: the authors are interested in demonstrating network coding, a topic of much recent interest in network information theory, in a practical wireless testbed. Theoretical results have shown that in a network, allowing coding/mixing in the intermediary nodes achieves the multi-cast capacity in this network (while routing might not). As the authors node, the capacity and optimality of network coding is open for unicast traffic in an arbitrary network. As it turns out, that is also the scenario of interest in this paper. Another key difference is that a broadcast transmission from a node can be heard by many surrounding nodes, whereas in traditional network coding, a transmission is only heard by a node at the other end of the link.

I have to admit that I was expecting something a little different before reading this paper. I thought the paper was going to pick up from the previous paper, and describe a coding approach for exploiting path diversity for higher throughput for a single unicast flow. What the paper seem to be describing is for multiple unicast flows. If the packets from each flow are not shared, its not immediately clear to me why network coding helps. (The butterfly example is when the receivers all want the packets injected by the various sources). COPE seems to assume that the neighbors of a node would want to have all the packets that it has; why that is the case is not immediately obvious to me.