A modified natural gradient algorithm for blind source separation using generalized constant modulus

This paper proposes a modified natural gradient algorithm (M-NGA) for blind separation of communication signals. In the proposed algorithm, a new cost function is introduced by exploiting both the generalized constant modulus property and the statistical independent property of source signals. Then, the M-NGA is derived by maximizing the new cost function. Simulation results show that the M-NGA has faster convergence than the conventional NGA.