Research On Blind Identification Algorithms Of Time-varying Channels

With the rapid development of wireless communication technology, blind channelidentification has been one of the research hotspots in communication signal processing due tothese advantages including estimation without training sequence and improved communicationefficiency.Channel is always assumed time invariant (TI) in the field of blind channelidentification,while it is time-varying (TV) in practice. In slow TV case, the assumption of lineartime invariant (LTI) channel is reasonable in a short time interval.At present, even dozens of datasample are required for blind channel identification algorithm based on TI model.However, theTI algorithms will fail in rapid TV channel situation i.e. the coherence time of channel is smallerthan the required data length.Therefore, the study of blind TV channel identification hasimportant theoretical significance and application value.So far there are fruitful research achievements in the field of blind time-varying channelidentification.However, a large gap between the theoretical research and practical applicationshould be noticed; in order to boost the process of application of these technologies,the mainproblems below still needed be solved:Firstly, how to describe the TV channel exactly, andwhich kind of channel model can not only express the TV channel characteristics but also beeasily handled in blind channel identification processing? Secondly, how to realize the TVchannel identification at low SNR and for short data sample scenarios? This paper mainlyfocuses on some key technologies on blind time-varying channel identification, the main workand contributions are outlined as follows:1. Research from the aspect of models of TV channel, analyzing characteristics andthe scope of application of existing statistical and the deterministic model,finding thatcomplex exponential basis expansion model can not only track the varying of the channel,but also make the TV channel blind estimation convenient. The statistical models suitablydescribing the general features of TV channels are not easy to be applied in estimationprocessing. While TV channel is constructed as the periodic variation in the complex exponentialbasis expansion model (BEM), its parameters have specific physical meaning, and the basisfrequency can be viewed as each path’s Doppler. As the parameters of the model aredeterministic, thus it renders TV channel blind estimation tractable. Subsequent blindidentification of TV channels is based on BEM.2Research from the aspect of the estimation of complex exponentials basis frequency,aiming at that problem that is difficult to set a decision threshold based on cyclic statistics,propose a decision method exploiting constructed test statistics.As the specific distribution ofthe estimation error of the cyclic statistics is unknown, the decision threshold can’t be set directlyas the estimation of cyclic statistics. Applying the constructed test statistics to estimate the basisfrequencies and setting a constant false alarm rate, the decision threshold will be obtained. Thetest statistics makes the decision easier and estimation more accurate.3Research from the aspect of the order estimation of LTI channel,utilizing the innerconnection with TV channel, propose order estimation algorithms based on informationtheoretic criteria, Liavas criteria, subspace projection for blind estimation of the TVchannel ‘s order and path number. The performance of channel estimation is very sensitive tothe exact knowledge of the channel order and path number, especially in deterministic methods.As the performance of these approaches utilizing the rank characteristics of observation matrixdegrade when applied in the presence of noise situation. Therefore, several methods forestimating the order and path number are proposed and the subspace projection shows betterperformance and simulations prove the effectiveness.4. Research from the aspect of the coefficient of BEM, the performance of noise subspace identification algorithm is better than statistical methods due to previousaccurate order estimation. In addition, extend least squares smoothing (LSS) method forblind identification of TI channels to TV channels and it is convenient to joint orderdetection and blind channel estimation. The statistical method such as linear prediction errormethod is not as sensitive to overestimation of channel order, but it is necessary to assume thatinput is white, and require a large number of observation data, which limits its application.Although deterministic subspace method is sensitive to channel order, the first step of orderestimation greatly enhance the identification performance. Moreover, the application of LSSmethod for TV channel identification makes joint order detection and blind channel estimationconvenient due to projection error matrix contained not only the information of the order but alsobasis expansion coefficient.