Efficient Implementation of Kernel Regularization Based on ADMM and Its Application to Room Impulse Response Estimation

Abstract

This work discusses an efficient implementation of the kernel regularization method. In particular, this work focuses on the Alternating Direction Method of Multipliers (ADMM) which is one of the convex optimization methods. This work employs two assumptions; (1) the identification input is periodic, and (2) the kernel matrix is tridiagonal. It is shown that ADMM updates can be implemented efficiently under these assumptions. A numerical example is shown to demonstrate the effectiveness of the proposed method. In addition, a practical experiment to estimate the room impulse response of a classroom at the University of Kitakyushu is shown.

Keywords

• Acoustic impulse response
• alternating direction method of multiplier
• impulse response
• kernel regularization
• non-parametric identification
• regularization