Microfone virtual com base no sinal vibro acústico da fonte emissora

Abstract

The success of an Active Noise Control (ANC) procedure is strongly dependent of the proximity of error sensors and location, where the noise mitigation is desired. One inconvenience of this is the necessity to positioning the sensors in inappropriate places such as: in center of rooms, corridors or near the ears of people. Hence a method based on detection of undesired noise source - with remote physical sensors and the use of transfer functions estimated by Neural Networks is investigated. The use of Neural Network is justified due to changes in the environmental acoustic properties type as studied in this thesis: reverberant room field, walking people, change in furniture s position and etc. The influence of the signals set, to train the Neural Network, number and position of remote sensors were also investigated. The neural network topology was studied and optimized by differential evolution algorithm. The measured and estimated responses were evaluated and compared in terms of RMS level, magnitude and phase. The estimative in a region instead of a single point, shows that it is possible to implement the virtual displacement. As well, it was observed that a Neural Network trained with white noise was able to estimate a harmonic virtual response. These results show that it is possible to obtain good responses in a narrow frequency band. Based on the results it is concluded that the technique has potential to be applied in ANC procedure. Online implementation of ANC and use of the virtual displacement strategy is suggested for future works.