Design and Implementation of a Doppler Removal Method for Parameters Extraction and Aural Simulation of Traffic Noise

Abstract

Noise effects and traffic noise impact at an urban location can be assessed by means of aural simulation. This simulation could be performed from the knowledge of the noise produced by each vehicle individually, the source movement effects (such as geometric divergence and Doppler shift), and sound reflections from façades. In theory, the noise sample from each vehicle could be obtained by recording it with a microphone that moves along with the source. However, this approach would be impractical due to turbulence noise that would be generated on the microphone. In order to overcome this drawback, a technique based on the recording of the noise sample at a fixed station, followed by a movement-compensation algorithm is proposed and tested with experimental data. In order to carry out the compensation stage, a pilot tone is radiated from the vehicle during the passby in order to estimate the instantaneous frequency shift of the source. In this way, the movement is coded in the phase modulation of the tone, which can be retrieved through a ridge searching algorithm applied to the input spectrogram. Therefore, the distortion affecting the pilot tone can be easily compensated for by reconstructing its original phase from the recorded one, using an interpolation algorithm. The tone frequency is selected so that there is negligible spectral overlap with the spectrum of the source. This stage gives a variable-step temporal sequence which is uniformly interpolated with the input time vector for Doppler compensation of the whole data set. As will be seen, due to the selection of the tone frequency, this technique is quite robust and allows the determination of the source speed over a long time interval. In the data caracterization stage, the compensated noise is separated into tonal and nontonal components locally, i.e., taking into account their energy contribution to the whole noise as a function of the vehicle position. Non-tonal information is simulated by white noise filtered by a reduced set of appropriate Linear Prediction Coefficients (LPC) filters in order to represent accurately the spectral envelope of non-tonal spectrum evolution. This compensation and caracterization method is applied to experimental data and assessed by aural simulation tests over an arbitrary time interval for an open virtual profile urban street.