Research On Channel Order Estimation And Robust Deterministic Blind Channel Identification Algorithms

In the digital wireless communication system, there is always large inter-symbolinterference（ISI）among recieved signal due to the influence of limited transmission bandwidthand multi-path fading. In order to reduce ISI and then recover transmitted signal symbols,equalizer is essential to compensate the fading caused by channel,which is based on channelestimation. Traditional channel estimation methods need transmit periodical training sequences,which occupies some valuable bandwidth and reduces the use ratio of the frequency band. Withthe aspiration of high capacity and reliability in communication, blind channel identificationwithout training sequence has been widely researched. Blind channel estimaition techniquesbased on second-order statistics (SOS) have been one of the hotspots in blind signal processingfield because of its good property of low computation complexity and fast convergence.Based on the research of Single Input Multiple Output channel model, this paper mainlyfocuses on the deterministic blind channel identification methods based on SOS which is notconstrained by the signal modulation. The main work and contributions are outlined as follows:1. Research from the aspect of SIMO channel models representing the time-invariantchannel. The general identifiablity conditions are induced and summarized in the problem ofSOS blind channel identification, and relative evaluation criterions are introduced, whichprovided the basis of model and theory.2Research from the aspect of SOS deterministic algorithms. Existing three classicaldeterministic algorithms, including the CR、SS and TSML algorithm, are analysed and studied.Aiming at the problem of blind channel estimation for short burst signal, a low computationalcomplexity algorithm is proposed based on FFT. When the over sampling factor is larger than2,the computational complexity of proposed algorithm obviously decrease compared with theoriginal, while the estimation performance is quite approximate.3. Research from the reason of invalid of SOS algorithms when the channel order isoverestimated and the common zeros existed in SIMO channel. Under these two conditions, onlythe characteristic zeros of channel are identified by SOS algorithm, then some extra randomzeros are introduced in real channel zeros which caused the degrading or invalid of the SOSalgorithms’ performance. When the channel order is overestimated, the extra common zeros inLSS and SS algorithm cluster around the unit circle in noisy situation. A blind channelidentification algorithm based on the analysis of the clustering zeros is proposed. The newalgorithm firstly detects the common zeros and then delete them from the estimated zerosthrough the clustering analysis, so the true channel zeros remained. Although the proposed algorithm’s performance is inferior to the alternative in known channel order situation, it ownslow computational complexity and wild adaptability.4. Research from the aspect of channel order algorithms for effective channel estimation.Liavas、NECOE and CMR algorithm are studied and analysed. A blind channel effective orderestimation algorithm based on subspace channel matrix recursion (SS-CMR) is proposed. On thebasis of the special structure of Q matrix and its null space vector equivalently viewed as theconvolution of true channel impulse response and the common-zeros channel, the channelmatrix is obtained by SS method and then the estimated channel order is got by the constructedrecursion cost function. The simulation proves that the performance of SS-CRM is improvedcompared with CMR and obviously better than the other existing algorithm, especially when thechannel impulse response has small head and tail taps; analytical analysis showed that thealgorithm complexity of SS-CMR is obviously reduced compared with CMR.Based on the research of SS-CMR, an extended EM-based joint detection and blind channelestimation algorithm is proposed (EM J-SS). It utilizs the EM iteration to obtain the Q matrixand constructed the joint estimation function. The simulation results prove that its jointestimation performance is obviously superior to J-LSS algorithms.5. Research from the aspect of robust deterministic algorithms which don’t rely on thechannel order estimation due to these algorithms which are sensitive to channel order error. Twomodified CR algorithms robust to order overestimation are summarized, these two algorithmsgive some weight constrains to the estimation solution in order overestimation situation to makeit approximate to ideal solution. On the basis of this thought and combined the FOCUSSalgorithm in compressive sensing, a FOCUSS-based subspace method robust to orderoverestimation was proposed and it transfer the process of search the ideal solution to thesparsity solution when then channel order is overestimated. Under a moderate SNR condition, anappropriate weight factor could make the proposed algorithm adapt to the situation estimatedorder is larger than the effective order. The simulation shows the proposed algorithm owns betterrobustness to order overestimation than other algorithms.