Convergence behavior analysis of FxLMS algorithms with different leaky term

• Main Authors: Wen, Shulin, Gan, Woon-Seng, Shi, Dongyuan
• Other Authors: School of Electrical and Electronic Engineering
• Format: Conference or Workshop Item
• Language: English
• Published: 2023
• Subjects: Science::Physics::Acoustics; Filtered-X LMS (FXLMS) Algorithms; Noise Control
• Online Access: https://hdl.handle.net/10356/169435
• Institution: Nanyang Technological University

In order to improve the robustness of conventional filtered-X LMS (FXLMS) algorithm, different leaky terms are introduced into the cost function. With different leaky term, the convergence behavior differs much. The detailed derivation of LASSO based FXLMS (L1-FXLMS) and elastic-net based FXLMS (L1/2-FXLMS) algorithms are presented. In addition, sufficient conditions for guaranteed convergence are derived for FxLMS algorithms with different leaky term. Furthermore, the convergence behavior of L1-FXLMS and L1/2-FXLMS algorithms are compared with conventional FXLMS and leaky FXLMS (L2-FXLMS) algorithms. It is indicated that L1-FXLMS could have faster convergence speed than L2-FXLMS with the premise of guaranteed convergence. Simulation and experiments are conducted to test the sufficient boundaries for guaranteed convergence of different FXLMS algorithms.