Development of a robust and computationally-efficient active sound profiling algorithm in a passenger car

Active noise control is a technique to cancel unwanted sound using adjustable secondary sound. In active sound profiling, the target is to obtain a certain sound field or profile and the power over specific frequencies can be altered in a desired way, even by amplifying it. Active sound profiling can be used for increasing the sound quality in a passenger car, for example, by modifying the engine noise inside the car cabin. A fundamental algorithm in active sound profiling is the command-FXLMS (CFXLMS) algorithm, which is an extension of the famous FXLMS algorithm widely used in active noise control. The computational demand of the C-FXLMS algorithm becomes excessive in multiple-channel systems with engine noise components to be controlled using several loudspeakers and microphones. The most time-consuming part of the C-FXLMS algorithm is the filtering of the reference signals. In order to reduce the computational burden, a new way to modify the reference signals in narrowband systems has been developed in this work. Instead of conventional filtering operations, the new method is based on delaying the sinusoidal reference signals and modifying their amplitude. The algorithm should work reliably and maintain stability in all operating points. In this work, an adaptive leakage has been developed for the C-FXLMS algorithm to increase its robustness. The objective is to limit the adaptive filter coefficients at frequencies where the phase shift of the plant is large. Such phase shifts occur at resonances, for example, and the performance of the algorithm is drastically degraded. In this work, the C-FXLMS algorithm has also been combined with the EEFXLMS algorithm so that frequency-independent step sizes can be used. This increases robustness and enables faster tuning of the algorithm. The developed algorithm has been tested in a simulation model and in an experimental active sound profiling system installed in a car. The results prove that the algorithm works with sufficient accuracy. The convergence is fast and stability is maintained in all operating points.