Indexed on: 18 Feb '10Published on: 18 Feb '10Published in: Circuits, Systems, and Signal Processing
The conjugate gradient method is a prominent technique for solving systems of linear equations and unconstrained optimization problems, including adaptive filtering. Since it is an iterative method, it can be particularly applied to solve sparse systems which are too large to be handled by direct methods. The main advantage of the conjugate gradient method is that it employs orthogonal search directions with optimal steps along each direction to arrive at the solution. As a result, it has a much faster convergence speed than the steepest descent method, which often takes steps in the same direction as earlier steps. Furthermore, it has lower computational complexity than Newton’s iteration approach. This unique tradeoff between convergence speed and computational complexity gives the conjugate gradient method desirable properties for application in numerous mathematical optimization problems. In this paper, the conjugate gradient principle is applied to complex adaptive independent component analysis (ICA) for maximization of the kurtosis function, to achieve separation of complex-valued signals. The proposed technique is called the complex block conjugate independent component analysis (CBC-ICA) algorithm. The CBC-ICA derives independent conjugate gradient search directions for the real and imaginary components of the complex coefficients of the adaptive system employed for signal separation. In addition, along each conjugate direction an optimal update is generated separately for the real and imaginary components using the Taylor series approximation. Simulation results confirm that in dynamic flat fading channel conditions, the CBC-ICA demonstrates excellent convergence speed and accuracy, even for large processing block sizes.