In many communication scenarios, the communicating parties typically have some knowledge or attempt to learn about the time-varying environment or the channel over which communication takes place. To understand these models, it is important to study the fundamental performance limits of state-dependent channels whose probabilistic input-output relationship depends on a time-varying random parameter called channel state. This channel state could be either fading in a wireless environment, interference, or host signal in information embedding, etc. In this thesis, we focus on studying state-dependent network models from an information theoretic perspective when the side information is available at some encoders and state recovery is considered at some decoders. In general, the side information could be either exact channel state, noisy channel state, or channel output feedback. In this thesis, we assume that the side information is exact channel state. We study single-user state-dependent models and multi-user state-dependent models such as multiple access channels (MAC) and broadcast channels (BC).
For the state-dependent MAC with non-causal side information at some encoders and without state recovery at the decoder, we study bounds on the capacity region in the case of independent messages and derive the capacity region in the case of dependent or degraded messages. For the Gaussian case, we develop a coding scheme based on both dirty paper coding and state cancellation to obtain the inner bound on the capacity region.
For the state-dependent MAC with different non-causal side information at different encoders, we derive inner bounds on the capacity region for the case of no state recovery at the decoder and the case of some, but not all, state recovery at the decoder. For this model, we also study bounds on the capacity region for the case of all state recovery at the decoder. It turns out that the inner and outer bounds meet if all state signals are independent. Consequently, we obtain the capacity region for this case.
For the state-dependent BC with non-causal encoder side information, we consider lossless state recovery at some decoders. In the two decoder model, we study bounds on the capacity region in the case of state recovery at only the better decoder. We also study the capacity region for the case of state recovery at all decoders and the case of state recovery at only the worse decoder. Finally, we consider lossy or partial state recovery in single-user state-dependent models without side information and study bounds on the capacity region for a given state-recovery distortion constraint.