Speaker
Description
Multi-channel road noise cancellation (RNC) systems in electric vehicles increasingly rely on large arrays of reference signals that generate coloured and highly correlated reference signals. These characteristics result in poorly conditioned co-variance matrices, leading to slow convergence when using adaptive feed-forward algorithms such as FX-LMS. This paper investigates the use of a simple, computationally efficient method of preconditioning the reference signals to reduce the eigenvalue spread of the co-variance matrix, thereby improving the convergence properties. The preconditioning is achieved via a set of short multi-channel prediction error filter (PEFs), which whitens the spectral density matrix of the reference signals.Simulation studies using measured vehicle data demonstrate that the preconditioning methods substantially improves convergence, reducing adaptation times from several minutes to tens of seconds when used in conjunction with secondary plant compensation. Importantly, these gains are achieved with only short, inexpensive PEFs, indicating that the approach offers a practical alternative to more computationally demanding whitening techniques such as spectral factorisation.