A Study On Blind Channel Estimation In Shortwave Communication

In HF communication system, the intersymbol interference caused by nonideal character of channel is the main factor which affects communication quality. In order to reduce the ISI, channel compensation based on channel estimation is needed. This thesis focuses on the blind channel estimation which does not depend on signal's modulation mode.Firstly, aiming at single-input single-output model and single-input multiple-output model, the necessary and sufficient conditions of blind channel identifiability based on second order statistics are analyzed, which is the theoretical basis of algorithms discussed later.Secondly, the subspace method and joint order detection and channel estimation via least squares smoothing method are discussed in deterministic algorithms. In order to denoise, an improved J-LSS algorithm by means of singular value decomposition is presented. The modified algorithm can not only reduce the estimation error but also weaken the dependence on parameters, thus the parameter selections in practice become more convenient. Among statistical blind algorithms, the TXK method and matrix outer-product decomposition method are considered. Targeting at solving the problem of OPD algorithm that the result of equalization would become worse if the channel order is overestimated, a method of amplitude detection is proposed to eliminate the overestimated tail via inverse search. Experiment results demonstrate that new algorithm can reduce the intersymbol interference and be overestimated arbitrarily. Theoretical analysis and mathematical derivations prove that this method makes overestimation of channel order allowed, especially in TXK method.Thirdly, channel order estimation, the kernel problem of blind estimation, is discussed in detail. In light of the fact that the noise power is affected by the change of the channel order, a novel algorithm is proposed. The proposed algorithm, compared with conventional algorithms, has been proved to be of low computation complexity and could work well under low SNR and small set of data. In addition, the new algorithm is independent of the concrete blind channel estimation methods we ever used.Finally, the channel order estimation methods described in the previous sections are applied to the two types of algorithms. Numerous simulations are provided to describe the performance and the application requirements of all algorithms. Some useful conclusions are given.