This thesis studies the optimal inputs and capacities of non-coherent correlated multiple-input single-output (MISO) channels in fast Rayleigh fading. We consider two scenarios: channels under per-antenna power constraints and channels under joint per-antenna and sum power constraints. For per-antenna power constraints, we establish the convexity and compactness of the feasible sets, and demonstrate the existence of optimal input distribution. By exploiting the solutions of a quadratic optimization problem, we show that the Kuhn-Tucker condition (KTC) on the optimal inputs can be simplified to a single dimension and prove the discreteness and finiteness of the optimal effective magnitude distribution. Then, we are able to construct a finite and discrete optimal input vector and determine the capacity gain of MISO over SISO. We also extend the results to MISO channels subject to the joint per-antenna and sum power constraints. For this case, the optimal phases and the optimal power allocation among the transmit antennas need to be determined simultaneously via a quadratic optimization subject to inequality constraints. Based on our results, the capacity of considered channels can be obtained and exploited as an upper bound for the operational transmission rate. Further researches can also rely on our analysis of the optimal inputs to construct reliable coding schemes for MISO fading channels. iii ACKNOWLEDGMENT Foremost, I would like to express my thanks and appreciation to my advisor, Dr. Nghi Tran, for his continuous support and guidance during my study and research at the University of Akron. His enthusiasm, motivation and immense knowledge consistently encourage me throughout the research and writing of this thesis. I am very grateful to Dr. Truyen Nguyen from the Department of Mathematics, University of Akron, who has continuously supported me with his patience and knowledge during my study and research. I also wish to express my appreciation to Prof. Hoang Duong Tuan from School of Electrical and Data Engineering, UTS, Australia. His guidance and support contribute significantly to the achievement of this thesis. In addition, I would like to thank my lab mate, Mohammad Ranjbar, whose ideas and comments have enlightened me throughout this research. Partial financial support from National Science Foundation (NSF) is also gratefully acknowledged. Finally, I would like to express my profound gratitude to my parents for providing me consistent support and encouragements. Without their assistances, this thesis would have not been accomplished.