Network channel coding: importance of in-network processing and side-information

Network Channel Coding (NCC) generically refers to a framework, under which we embed channel coding functionalities within a network, to improve the throughput in a multi-receiver communication scenario. In this work, we consider two types of NCC problems. The first part is focused on NCC problems related to rateless codes. Rateless codes can be used for asynchronous broadcast applications. We focus our work on Luby-Transform (LT) codes, which are an example of low-complexity rateless codes. Prior designs of LT codes cater to the uniform demand case. In this work we design LT codes for non-uniform demands where the sinks are interested in recovering distinct fractions of the data symbols. Subsequently, we consider the design of Distributed LT (DLT) codes. DLT codes can be used in scenarios where multiple sources communicated to a sink via a common relay. In such a scenario, LT encoded data from various sources can be selectively mixed at the relay to improve the communication efficiency. Prior designs of DLT codes have focused on specific degree distributions, we generalize this work. We also comment upon the analysis of Generalized LT codes using generalized ripples. The second part is focused on problems that study the impact of side-information on NCC. In particular we focus on design of NCC related cross-layer protocols. Typical wireless networks employ a link-abstraction under which packets with corruptions are discarded. While such an abstraction certainly leads to a simpler architecture, the capacity loss on account of the packet drops can be significant. Hence, in this work we argue in favor of developing protocols which relay corrupted packet to higher layers, and across multiple hops. On the basis of theoretical deductions, model and trace based simulations; we analyze the utility of proposed protocols, particularly for multimedia applications. We investigate in detail the impact of receiver-side Channel State Information (CSI) on the performance of these protocols. We identify mechanisms that can provide CSI in commercially available wireless devices. Finally, we extend these protocols to exciting new area of network coding. Network coding saves bandwidth by mixing packets within the network. However, mixture of corrupt packets can lead to error amplification. Capacity loss due to error amplification can nullify the gains of network coding. To circumvent this phenomenon, we design an Adaptive Network Coding protocol that selectively mixes corrupt packets. The decision to mix is determined by the level of bit-corruption in each packet. All of the above proposed protocols exhibit a progressive improvement over comparable conventional protocols.