Reseach On Key Technology Of Compressive Sensing Based Channel Estimation

Compressive sensing(CS) is one of the most important theory in this century, which claims that it is possible to recover a sparse signal from fewer samples than Nyquist rate. The CS arisen from image processing community, is a novel signal processing technique which is a revolution for conventional signal acquisition and compress method.Therefore, it has not only an important role in image processing, but also be of great value in other area, such as channel estimation, signal sampling and data compression, etc.In this thesis, we will introduce the fundamental of CS theory and its application in channel estimation. In wireless communication, the received signal is usually superposition of multiple transmitted signals sending though many paths with different delay and attenuation., the number of transmitting paths is often small with respect to time domain. Therefore, the multi-path wireless channel can be viewed as approximated sparse signal, and the estimation of the channel would be more accuracy with the help of CS technique.In CS technique, the algorithms for the sparse signal reconstruction are divided into three categories:convex relaxation, greedy pursuit and Bayesian learning. Since the greedy pursuit has advantages of low computational complexity and easy to implement, it is more applicable in strict requirement of real-time communication system.In this thesis, the advantage and problem of greedy algorithm as the channel estimation algorithm will be discussed, and the greedy algorithm will be improved from two aspects. First, we analysis the residual vector of OMP, then apply binary hypothesis on the residual vector of OMP at each iteration. The algorithm stops iteration when there is no signal component in the residual vector. Second, the channel estimation problem is considered from Bayesian perspective, taking the prior statistics of the channel into account to achieve better estimation accuracy. Then we give a better criteria for measuring the coherence between residual vector and sensing matrix. The effectiveness of the two improvements are shown in analysis of computational complexity and numerical experiments.