A Comparison between Different Multiple Input Array Processing Techniques for the Localisation of Correlated Acoustic Sources

Abstract

The localisation of acoustic sources using linear microphone arrays permits source positions and power levels to be evaluated. The identification of the positions of the principal sources is essential before noise reduction methods can be defined. The work presented here concentrates on problems associated with multiple and partially coherent sources. Conventional processing techniques have limited resolution for a restricted number of transducers which can be resolved using high resolution techniques. Experiments showed the limitations of high resolution methods when dealing with correlated sources. Techniques based on the use of reference signals from selected sources were studied to try and solve this problem. This approach consists of pre-processing the cross spectral matrix in a specific way using multiple-input multiple-output (MIMO) methods. This enables the independent contributions of sources to be localised even for relatively high values of inter-source correlation.