Effective Learning Rate Adjustment of EASI Algorithm Based on the Fuzzy Neural Network

The behavior of the classic algorithm for blind source separation is reviewed for a fixed step-size. The inherent contradiction of the classic equivariant adaptive source separation via independence algorithm (EASI) make it very difficult to balance convergence speed and the steady-state misadjustment error. In this paper, In response to the above contradiction, a Fuzzy Neural Network(FNN)-based learning rate adjustment method is proposed for EASI algorithm, and a simple decision-making method is introduced for how the learning rate should be applied in the current time slot. Signals separation process adopt the different learning rate according to the characteristics of the separate degree. Therefore it can improve convergence speed and reduce the misadjustment error in the steady state simultaneously. Extensive simulations confirm the theoretical analysis and show the proposed approach is superior to other EASI algorithms.