Liu, Yang, (Lehigh University) “Reliable and Efficient Transmission of Signals: Coding Design, Beamforming Optimization and Multi-Point Cooperation”, (2016)

Liu, Yang, (Lehigh University) “Reliable and Efficient Transmission of Signals: Coding Design, Beamforming Optimization and Multi-Point Cooperation”, (2016), Advisor: Jing Li

This dissertation focuses on reliable and efficient signal transmission strategies in classical two-point and also multi-point communication systems. Although the digital signal processing and communication technologies have been developed very successfully nowadays, the optimal non-digitized communication schemes has always been an attracting and open issue for the past decades. At the same time, enormous amount of communication networks have emerged in a wide range of fields. Its wide applications have also arisen numerous signal transmission design problems to satisfy different requirement and constraints for various system. This dissertation is devoted to some facets of these problems. Specifically, the contribution of this dissertation is as follows.
The first contribution is that the linear analog coding schemes' performance limit and optimal codes have been obtained. Under the general model of additive white Gaussian noise(AWGN) channel and mean square error(MSE) performance metric, the optimal linear analog codes under maximum likelihood(ML) and linear minimum mean square error(LMMSE) criteria are studied. The performance limits using these two decoding schemes have been obtained and they lead to identical optimal linear analog codes—unitary codes.
The second contribution is that the authors propose a novel nonlinear analog coding schemes based on chaotic dynamic systems—baker's dynamic system. Under the general AWGN channel model, various decoding algorithms have been researched, including the minimum mean square error(MMSE) decoding algorithm, maximum likelihood(ML) decoding algorithm and ML-LMMSE algorithms. MMSE algorithm provides optimal decoding performance in MSE, but it also requires prior knowledge and highly nonlinear computation operations. ML and ML-MMSE algorithms are sub-optimal decoding schemes, which do not require knowledge of source's distribution and only involve linear computation. Based on the careful examination of the baker's dynamic system's performance limit, two improving schemes, mirrored baker's dynamic systems and one input baker's system, are proposed. The improvement schemes effectively depress the threshold effect of the original system and outperform the other existing chaotic analog codes systems in literature.
A third contribution is that the authors consider the precoding design for single sensor with single antenna by exploiting signal space diversity. By analyzing pairwise error probability, the authors discuss precoder design criterion. Besides, suboptimal decoding algorithms with low complexity are researched. A kind of partially nulling and canceling (PNC) algorithm is proposed. Extensive numerical results show the proposed PNC algorithm can achieve better bit error rate (BER) performance with even lower computation complexity.
Last but not least, the final contribution is the research on joint transceiver design in centralized wireless sensor networks. A wide range of commonly used performance measures, including MSE, mutual information (MI) and signal to noise ratio (SNR), have been taken into consideration. Under the setup of complex wireless sensor networks involving numerous variables and constraints, the joint transceiver design problems generally have highly non-convex optimization objective and extremely hard. Instead of solving these hard problems in one shot, the authors adopt the methodology of block coordinate descent (BCD) methods, to solve these problems in an iterative manner. By possibility necessary equivalent transformation of the original problems, the authors partition the whole variable space into multiple groups and each time the objective is optimized with respect to only one group of variables with the others being fixed. For the MSE, MI and SNR optimization problems, the authors decompose each them into multiple convex subproblems and by analyzing the optimality conditions, most of these subproblems' closed form solutions are obtained, which significantly decrease the complexity of proposed algorithms. Besides that, convergence characteristics are also of great concerns and carefully examined. Numerical results fully verify our proposed algorithms.