Acoustic noise modeling and identification using neural and fuzzy techniques

This paper presents an identification procedure of a multivariable acoustic noise model, produced by a particular Electro-Mechanical Film (EMFi(TM)) actuator. This model will be integrated in an experimental setup for active noise control purposes. The obtained data represents a nonlinear dynamic system, which led to the use of nonlinear identification techniques. A comparison between neural network models and fuzzy models is presented such that their advantages and range of applications is clearly shown. Neural network models have, among others, the property of learning nonlinear multivariable dynamic mappings without suffering from the curse of dimensionality. Fuzzy models have the advantage of combining accuracy and some transparency. The use of nonlinear modeling techniques allows the possibility of modeling acoustic behaviors in a simpler and faster way than conventional mathematical modeling. Experimental results show the applicability of each modeling paradigm.