Simulation And Performance Of The Blind Channel Identification Algorithm Based On Second-order Statistics

In recent years, digital communication systems have become increasingly common in commercial applications. Consequently, advanced signal processing techniques, such as blind equalization and blind channel identification have been adopted in a wide range of communication systems, because these techniques could improve system performance and capacity. Blind equalization and blind channel identification mean that estimate the unknown sending signals and unknown channel respectively when there is no training serials.In most communication systems, channel is unknown and time-changed. So when designing the equalization, we should use training serials to adjust the coefficients of equalization. But this will induce some problems. First, adding training serials in the sending signals will increase spending and reduce system efficiency; secondly, some time we couldn't send the training serials.From the reasons above, it's necessary for us to study the equalization not depend on the training serials, but on the received signals only, which called blind equalization technique. Similarly, the methods needing no training serials are called blind methods.Traditional equalization, including MLSE, ZF and MMSE equalizations all need that the channel character should be known first. It's same as the blind equalization, estimating the unknown and time-changed channel only from the received signals is called blind channel identification.Traditionally, blind channel identification and equalization are all based on high order statistics, which are known to suffer from many drawbacks. In the 1990's, the method of blind channel identificationusing only second order statistics has been proposed, and it's a major breakthrough. Now it includes five kind of basic algorithms. First is linear prediction algorithm(LPA), which is based on only on the estimate of the first few columns of the channel parameter outer-product matrix, which depends critically on the leading coeffcients of the unknown multi-channel impulse responses,so the estimation error can be very large if the channel has a weak pre-cursor impulse response. To improve the performance of the original LPA, several linear estimation approaches have been pursued: outer product decomposition algorithm(OPDA), multi-step linear prediction(MSLP), constrained minimum output energy algorithm(CMOE), least square smoothing algorithm(LSS).These algorithms are all based on the channel parameter outer-product matrix which deduced from the received signals, then we can estimate the channel by eigen-decomposition or singular decomposition. But every algorithm has it's own method. This paper will analyze and simulate these algorithms, so we can compare them in each condition.Chaper 1 will introduce the background and development of this technique.Chaper 2 will analyze all this algorithms and the relationship of them, and find the simplest one by comparing.Chaper 3 will simulate LPA, OPDA and CMOE by matlab in different conditions, and find the best one by comparing.