Research Of Instantaneous Mixtures Blind Source Separation

Blind source separation is derived from "Cocktail Party". It is attractive in the community of signal processing and advances have been developed in recent years.We introduce the basic model of blind source separation, the mathematical principle of them and the latest progress in research. We then analyze the characteristic of typical algorithms. On the basis of them, we propose two new algorithms aimed at the constrained condition of blind source separation. Then, we simulate them using MATLAB software.First, in view of the parameter space, we optimize contrast function within the Stiefel Manifold which stand fororthogonal constrained. It can solve constrained problem without explicitly projecting the parameters back to the constraint space so that decrease the computational complexity.In addition, we introduce the idea of variable step size. In the paper, we use the gradient to adjust the step size and derive the corresponding equation. The computer simulations show it can separate the source signals well. It not only has fast convergence speed and high output SNR compared with other algorithms but also can suppress the interference best.Second, in view of the optimization method, we introduce the idea of penalty function. It can change the constrained optimization problem into the abandoned optimization problem so that get rid of the constrained condition.On the basis of the traditional contrast function, this algorithm increases the penalty term which is made up of constrained condition. Then, we derive the corresponding iteration equation according to the idea of incremental Lagrange penalty function. The computer simulations show it not only can separate the source signals but also has rather fast convergence speed andsmall signal-to-signal interference compared with other algorithms. The separation effect is close to the ideal condition.