Self-learning active noise control

Abstract

An important step for active noise control (ANC)systems to be practical is to develop model independent ANC (MIANC) systems that tolerate parameter variations in sound fields. Reliabilities and stabilities of many MIANC systems depend on results of online system identifications. Parameter errors due to system identifications may threaten closed-loop stabilities of MIANC systems. A self-learning active noise control (SLANC) system is proposed in this study to stabilize and optimize an ANCsystem in case identified parameters are unreliable. The proposed system uses an objective function to check closed-loop stability. If partial or full value of the objective function exceeds a conservatively preset threshold, a stability threat is detected and the SLANC system will stabilize and optimize the controller without using parameters of sound fields. If the reference signal is available, the SLANC system can be combined with a feedforward controller to generate both destructive interference and active damping in sound fields. The self-learning method is simple and stable for many feedback ANCsystems to deal with a worst case discussed in this study.