Research On Blind Source Separation And Related Network Security

Blind source separation (BSS) aims at recovering unknown multiple sources from mixtures captured by multiple sensors. BSS has wide applications in various areas such as telecommunications, speech processing, image processing, and bio-medical signal processing.Although it has a rapid development on theories and application in the last decade, there still exists lots of problems as follows. 1) the sources number detection and source separation problem based on the under-determined linear mixing case; 2) the sources number detection and source separation problem based on the convolutive mixing model; 3) the application of BSS in the area such as communication security and smart grid security. For these problems,we try to solve such crucial problems in the BSS from the following aspects:First, the underdetermined mixing matrix identification is introduced based on the sparse component analysis (SCA). Traditionally, the K-HyperLine Clustering (K-HLC) learning algorithm is employed to solve it,but this method is designed under a strict sparse assumption on the source signals. To deal with the blind identification problem of multiple dominant SCA problem, a discriminatory clustering algorithm, K-Weighted HyperLine Clustering (K-WHLC), is developed via proposed weighting scheme. The Gaussian membership function is offered as the weight factor for the hyperline clustering, together with an optimal selection in relation of the involved threshold. This discriminatory clustering scheme is efficient and especially suitable for the hyperline identification against the multiple dominant SCA problem.Also, this developed algorithm can achieve the sources number detection as well as the mixing matrix estimation.Second, the sources number detection problem is discussed based on the convolutive mixing BSS model. Convolutive BSS refers the scenario where sources are recorded by multiple sensors in a reverberant environment, which can be depicted as a convolutive mixing model.A challenging task in convolutive BSS is the model selection problem, i.e., identifying the number of sources. We try to show that this problem can be categorized as the detection of the hyperlines number. Motivated by the observation that only one source component energy is locally dominant in the covariance domain, the features of observed samples are extracted firstly by searching the optimal projection direction based on the locally Second Order of Statistics (SOSs) of mixture data in each frequency bin. A density-based clustering method is then proposed to search the centroids of the feature samples in terms of sorted scores, which is calculated from a product of the local density and the intra-cluster distance of feature sample.Such scores are further utilized to automatically estimate the number of sources by a gap-based detection method. In the proposed method, a successive frequency bins are selected within the sources number detection process.Third, the source separation problem is analyzed based on the convolutive BSS model.Convolutive BSS is usually solved in the time-frequency domain, where source signals are separated independently in each frequency bin. We present a structured tensor framework to address the inherent permutation problem in the time-frequency domain. Anew mixing system model is built by utilizing pairwise frequency bins, i.e., a reference frequency bin and a non-reference frequency bin. Moreover, the mixing matrix is encoded in the proposed mixing system model. Then, the permutation ambiguity can be fully eliminated in theory based on the proposed system model, while the related proof is given. Furthermore, a new time-frequency domain algorithm is provided to show the feasibility of convolutive BSS based on this new system model. It consists of two steps. 1) the mixing matrix in the proposed system model is estimated by a structured tensor decomposition approach; 2) the source signals are separated based on the estimated mixing matrix for both over- and under-determined cases. It is worth noting that only the scaling of separated sources is required to be corrected in the second step.Furthermore, the BSS technique is used for physical layer security of wireless communication.Discriminatory channel estimation (DCE) is a recently developed strategy to enlarge the performance difference between a legitimate receiver (LR) and an unauthorized receiver (UR)in a multiple-input multiple-output (MIMO) wireless system. Specifically, it makes use of properly designed training signals to degrade channel estimation at the UR which in turn limits the UR's eavesdropping capability during data transmission. We propose a new two-way training scheme for DCE through exploiting a whitening-rotation (WR) based semi-blind method. To characterize the performance of DCE, a closed-form expression of the normalized mean squared error (NMSE) of the channel estimation is derived for both the LR and the UR.Furthermore, the developed analytical results on NMSE are utilized to perform optimal power allocation between the training signal and artificial noise. The advantages of our proposed DCE scheme are two folds: 1) compared to the existing DCE scheme, the proposed scheme adopts a semi-blind approach and achieves better DCE performance; 2) the proposed scheme is robust against active eavesdropping with the pilot contamination attack, whereas the existing scheme fails under such an attack.At last, the BSS technique is used for the smart grid security. Meter data collection and management in smart grid has the potential for underlying security risks, e.g., low-sparsity unobservable attacks. Thus, it is crucial to investigate the vulnerability of smart grid through various exposure tests associated with these unobservable attacks. Recently,much attention has been paid to low-sparsity unobservable attacks with the complete knowledge of system matrix. In this paper, the unobservable attack exposure analysis is based on a relaxed condition,i.e., an incomplete knowledge of the system matrix. Furthermore, a data-driven attack scheme is designed to demonstrate that such knowledge can be learned with a two-stage strategy. 1) a sequence of intercepted meter data is utilized to learn about the incomplete system matrix with a blind identification approach. 2) the estimated system matrix at hand is used for the attack vector construction with a sparsity exploiting method. The proposed result reveals the potential risk of meter data leakage to the security of the smart grid.