髒紙碼之理論，實現，和對偶性

Abstract

Recently, multi-terminal network equipped with multiple antennas has gained lots of attentions due to its strong potential in performance enhancement. The coding for vector side-information channels plays a fundamental role in this system, since each terminal in the network can provide the others some side-information to help encoding/decoding the data. In this thesis, we will study how to design and implement this new coding scheme. We have developed a lattice-based vector dirty paper coding (DPC) for communication channels with transmitter side-information and perfect transmitter channel state information (CSIT). This coding can completely remove the inference known only at the transmitter (as side-information) but not at the receiver. When only statistics of CSIT is available, good (sometimes optimal) performance can still be obtained by modifying this coding structure. We also show how to achieve the capacity region of the multiple-input multiple-output Gaussian broadcast channel with proposed coding. Motivated by this lattice-based coding scheme, we also develop another vector DPC based on superposition coding which can be implemented with existing channel and source coding schemes. An design example is provided and close-to-capacity rate performance is obtained. Finally, we propose a vector Wyner-Ziv coding (WZC) which is a compression technique using decoder side information to help reconstruction. The key is exploring the duality between the DPC and WZC problems. This vector WZC can be applied to the vector CEO problem in sensor networks, and achieve the Berger-Tung sum rate which is the best known compression rate for this problem.