New approach to multichannel linear prediction problems

Abstract

This paper provides a new approach to the multichannel linear prediction problem. To solve the consequently redefined problem in an order-recursive manner, a new multichannel Levinson algorithm is presented by virtue of the time-varying Wiener filter theory. The new algorithm is computationally efficient relative to the Levinson-Wiggins-Robinson algorithm and its variants since it avoids any matrix operations and needs scalar operations only. Furthermore, the new algorithm is optimum in that it produces the exact minimum prediction error power.