The active control of random noise in automotive interiors

Abstract

The problem of actively attenuating random sound inside cars using loudspeakers is considered. A feedforward control approach is adopted, the FIR control filters being adapted to minimise the sum of squared pressures at a number of microphone locations in the enclosure. An expression is developed which defines the optimum coefficients of finite-length, causal filters for the case with multiple reference signals, multiple secondary sources and multiple microphones. In addition it is shown that the maximum possible sound reduction using causally unconstrained filters can be expressed simply in terms of the multiple coherence function. The active control of random sound in the idealised case of a rectangular enclosure bounded by a vibrating plate is examined and it is seen that the degree of attenuation is highly sensitive to delays of a few milliseconds in the control signals. However, lightly-damped low-frequency modes of the plate introduce almost-periodic components into the sound which are more predictable and make the system less susceptible to time-delays. A central part of the work is a prediction of the sound reductions which could be obtained in two widely-used family cars based on measurements of vehicle vibration and noise. It is shown that sound reductions of the order of 5 dB can be expected at a point inside the car (eg at the driver's head) at frequencies in the range 90-1490 Hz using six accelerometers attached to the vehicle structure as reference signals. It is found that accelerometers attached to the vehicle suspension give adequate time-advance for a practical system but many accelerometers are needed to detect all the contributions to interior noise. On the other hand, accelerometers attached to the vehicle floor are more coherent with interior noise (so less are needed) but they give less time-advance and hence poorer sound reduction. A preliminary study was carried out of the possibility of using a neural network to model the nonlinear elements in the transmission path from the suspension to the vehicle interior. Tests with simple idealised dynamic nonlinear elements showed some success, but it appeared that the backpropagation algorithm had the effect of excluding some hidden-layer neurones from the model.

More information

Identifiers

Catalogue record

Export record

ASCII CitationAtomBibTeXData Cite XMLDublin CoreDublin CoreEP3 XMLEndNoteHTML CitationHTML CitationHTML ListJSONMETSMODSMPEG-21 DIDLOpenURL ContextObjectOpenURL ContextObject in SpanRDF+N-TriplesRDF+N3RDF+XMLRIOXX2 XMLReferReference ManagerSimple Metadata