Temporal noise source separation by short time coherence output power

The concept of coherent output power has been developed to separate the spectral characteristics of an acoustic source from uncorrelated background noise. The concept has been used successfully in a variety of engineering applications when both the source of interest and ambient are stationary in nature. The data stationarity is crucial for sufficient ensemble averaging to allow convergence of the coherence function and hence its utilization for source separation. When the noise sources are transient or non-stationary (i.e., vehicle pass by) the concept fundamentally breaks down as the averaging obscures the time variant characteristics. The effort described in this work addresses this situation through the application of the time variant coherent output power. All the respective spectra are calculated with the Short Time Fourier Transform. The time variant coherent output power is then formed and used to separate the vibration measured on a structure located near a railroad line during the passage of a freight train. The dominant excitation sources are the ground vibration from the wheel-rail interaction and the acoustic noise emitted from the locomotive. The time variant coherent output power effectively separates and localizes the temporal characteristics of the two sources.