Blind Channel Identification(Equalization) And Its Application

Conventional channel equalization and identification require either training sequence or strict assumptions of the channels (such as, a minimum-phase channel). However, it is not reasonable to model all channels as minimum-phase systems and the transition of training sequence obviously decreases communication throughput. In blind channel identification (equalization), nonminimum-phase channels are considered and training sequences are avoided. Therefore, it is more efficient and more applicable to advanced communication systems. Blind channel identification also has applications in image restoration, geoscience, etc. In this dissertation, we study the problem of blind channel identification and its applications. The contributions are in five folds.(1) Identifiablility is an important topic in blind channel identification. Normally it is discussed associated with certain algorithm. In this dissertation, we build the relation between these identifiablity discussions. We prove that in second-order statistics based blind identification, only a stationary input is necessary, but not a I. I. D. input as mentioned in existing literatures. This conclusion is the key to build such relation.(2) Subspace based methods attract much research attention in blind channel identification. However, it has been proved that these methods can not achieve the least identification error (i.e. the Cramer-Rao bound). To withdraw this difficulty, weighted subspace based algorithms have been proposed. In this dissertation we propose a new weighted subspace algorithm which has better performance and can save computation. Simulation gives some illustrative results.(3) Most of the existing researches deal with FIR channels. In the case the channel has long response it may be not efficient to model the channel as an IIR system. We prove that by oversampling the output of an IIR channel, one can transform this IIR channel into a Single-Input-Multiple-Output model, and each subchannel is an IIR system with same AR order and parameters. Based on this model, two algorithms...