Research On Blind Signal Separation And Its Application In Non-Cooperative Communication

Blind signal separation(BSS) is one of the most important researches in modern signal processing field. It has been utilized in many fields due to the fact that it does not require too much prior knowledge for the source signals and transmission process. Independent component analysis(ICA) is the main method to solve the BSS problems. As it is easy to implement and can have reliable separation performance, it has been attached more and more importance.Multiple Input Multiple Output(MIMO) is a very important breakthrough in modern wireless communication systems. It can improve transmission rate and spectrum efficiency without increasing the available bandwidth and the transmission power. Thus, it has become an important research topic for the field of wireless communications. Space-Time Block Code(STBC) has become the main coding technology for MIMO systems due to it is easy to encode and decode and can realize reliable transmission. It is very important for military and civil applications to blindly estimate the parameters and separate the signals of MIMO-STBC communication systems from the non-cooperative aspect. However, the research on this topic has been scarcely reported in literature. This thesis studies the blind source separation technique and its application in the non-cooperative communication of MIMO-STBC systems. The main contributions of this thesis are listed as follows.1、The research on the performance of noncircular complex Fast ICA(nc-Fast ICA) algorithm and its improvement on the special communication signals are addressed in this thesis. Firstly, the performance analysis of the nc-FastICA algorithm is studied based on the fixed-point iteration and the cost function. Through the analysis, this thesis shows that(1) there exist undesirable solutions for nc-FastICA algorithm and the undesirable solutions are related to the initial unmixing matrices;(2) nc-Fast ICA algorithm can work well even when the sources fall into the area of instability. A simple check of undesirable solutions is also proposed based on the kurtoses of the estimated sources. Computer simulations confirm the theoretical results. Secondly, by analyzing the estimation error of the nc-Fast ICA algorithm, an approximate optimal nonlinearity for the separation of digital communication signals is derived. As the modulation types are rarely known in the practical case, an alternative scheme to choose the nonlinear function is proposed based on the features of different modulation types. Thus, based on the approximate optimal choice of the nonlinearity, an efficient variant of the nc-FastICA(E-nc-FastICA) algorithm for the separation of digital communication sources is proposed. Finally, simulations have shown that the proposed E-nc-Fast ICA algorithm can exhibit better performance than the traditional ICA algorithms in separating the digital communication signals.2、The research on the performance of complex maximization of non-Gaussianity(CMN) algorithm and its improvement on the noisy ICA model are addressed in this thesis. Firstly, the performance analysis of the CMN algorithm is studied based on the fixed-point iteration and the cost function. Through the analysis, this thesis shows that(1) there exist undesirable solutions for CMN algorithm and the undesirable solutions are also related to the initial unmixing matrices;(2) CMN algorithm can work well even when the sources fall into the area of instability. Computer simulations confirm the theoretical results and show that the CMN algorithm using symmetric orthogonalization can have slightly better results than using the deflationary mode. Thus, this thesis suggests using symmetric orthogonalization rather than deflation one except when extracting only partial sources. Secondly, a noisy Complex Maximization of Non-Gaussianity(noisy CMN) algorithm is proposed for the complex noisy ICA model. The new algorithm can mitigate the influence of the noise by pseudo-whitening and adding the noise information into the fixed point iteration of original CMN algorithm. Theoretical analysis illustrates that the fixed point iteration of new CMN algorithm in the noisy ICA model is equivalent to the fixed point iteration of original CMN algorithm in the noiseless ICA model. This conclusion illustrates the asymptotic unbiasedness of the proposed algorithm. Simulations show the superiority of the proposed noisy CMN algorithm in the presence of the noise.3、The source separability of MIMO-STBC systems using the nc-FastICA and the Joint Approximative Diagonalization of Eigenmatrices(JADE) algorithms are studied in the thesis. For the nc-FastICA algorithm, this thesis proves that the optimal solutions for most STBC sources can produce the extrema of the cost function for the algorithm to converge to. It shows that the classical nc-Fast ICA algorithm can be applied to a wider range of situations rather than strictly independent sources. For the JADE algorithm, this thesis proves that the significant eigenmatrices constructed by the fourth-order cumulant tensor of whitened receiver for most MIMO-STBC systems can satisfy the essential uniqueness condition of joint diagonalization. It shows that the classical JADE algorithm can be applied to a wider range of situations rather than strictly independent sources. Thus, on one hand, it extends the application of classical nc-FastICA and JADE algorithms, and on the other, it provides a good blind source separation method for most MIMO-STBC systems. Computer simulations confirm the theoretical results.4 、 A low complexity modulation classification algorithm for the MIMO-Orthogonal STBC(MIMO-OSTBC) system is proposed in the thesis. Both the cases of perfect and unknown channel state information(CSI) are considered. In the case of perfect CSI, the MIMO-OSTBC system is firstly decoupled in to several Single Input Single Output(SISO) systems by utilizing the orthogonal property of OSTBC. Then, these SISO systems are devided into several Two Input Two Output(TITO) systems with each TITO system consisting of the real and imaginary part of the independent modulated symbols. At last, the Maximum Likelihood(ML) based approach is used to classify the modulation type. In the case of unknown CSI, the second-order statistic(SOS) method is firstly used to obtain an initial channel estimate. Then, the channel ambiguity is removed for unidentifiable OSTBCs by utilizing the independence of the original symbols. Next, partial phase ambiguity of the estimated channel is removed by using the modulation assumption. Finally, the LF function is proved to be not sensitive to the remaining ambiguity and the estimated channel can be used to classify the modulation. Simulations show that, compared with the Multidimensional ML algorithm,(1) the proposed algorithm can have significantly lower complexity with slightly performance loss;(2) the proposed algorithm can be applied to a wider range of modulation classification.