Research On Algorithms For Blind Source Separation, Signal Construction And FSK Detection

In the era, which information technology is now rapidly developing, the authenticity and the reliability on transmitting and receiving information have become kernal issue. The transmission and exchange of information are executed through signal. In other words, signal is carrier and transmitter of information. However, the observed signal is usually contaminated by stochastic noise, or distorted by transmitting systems and receiving equipments. Sometimes, the observed signal is the mixtured source signals and needed to be separated, which is called the blind source separation. Aiming at these issues, we carry on deep-going and particular research, some significant algorithms facing practical engineering applications are proposed. The main research results and innovation points are as follows:1. An algorithm based on the correlation characters for instantaneous linear mixture of the source signal is proposed. The criterion function is established by the Cross-covariance function, and minimized by stochastic gradient descent algorithm. Thus, the separation of mixtured signals is realized.2. An algorithm based on the forword filter for the convolutive mixtured source signals is proposed. First, the model of the mixture is simplified. Then, the separation issue is equal to optimal Wiener filter issue by the theoretical deduction based on the simplified model, and the separated criterion function is established. Last, the separation criterion function is minimized using LMS algorithm, and the source signals are extracted.3. For the Post-nonlinear mixtured source signals, a new algorithm based on the decorrelation is proposed. The conventional idea is broken out and the mixture model is simplified by the differential transform in the paper. The criterion function is proposed and the optimal equation is deduced by using the correlation characters. The LMS algorithm is utilized to make the criterion minimized. The iterative method is used for the equilibrium point of the optimal equation. Source signals are last extracted and whole-blind separation is realized.4. In the noise environment, the criterion function is established by utilizing Cross-correlation and a signal reconstruction algorithm for the distorted signals is last obtained. The relatively precise estimation of source signals is realized in spite of the small distortion.