Gradient-Based Variable Step Size Algorithm for Blind Decorrelation
The conflict of convergence rate and misadjustment in steady state is always a big problem for adaptive blind decorrelation, which is considered as a necessary condition for blind source separation. In this paper, a novel variable step size algorithm is proposed for on-line blind decorrelation of the mixed signals. Based the gradient of the cost function, the presented algorithm adaptively updates its step size to match the dynamics of the input signals and decorrelation matrix, then the fast convergence speed is obtained while keeping a low steady-state error. Simulation results show that the convergence and steady state performance of the proposed method outperforms the regular adaptive blind decorrelation algorithm in both stationary and non-stationary environments.
blind decorrelation blind source separation step size adaptive
Chao Deng Hong-min Liu Zhi-heng Wang
College of Computer Science & Technology Henan Polytechnic University Jiaozuo, hina College of Computer Science & Technology Henan Polytechnic University Jiaozuo, China
国际会议
桂林
英文
108-110
2010-11-17（万方平台首次上网日期，不代表论文的发表时间）