Sensor Array Optimization for Multiple Harmonic Sound Source Separation and DOA

Abstract: INTRODUCTION In the last years a lot of researches about source separation have been realized, like extraction of a signal of interest (vocal recognition application), identification of which source gives which sound (motor engine applications) or noise source characterization (environmental application). Most of these techniques for sound source estimation use the signal-subspace approach, where the number of emitting sources is determined by the multiplicity of the lowest eigenvalue of the correlation matrix. The problems arise when the number of microphones is equal to the number of sources radiating, hence the noise subspace could not exist. This Master Thesis investigates how to realize a Goniometer Antenna to record communications, as well as the implementation of an algorithm to optimize the location of the sensors with the intend of separating the different sound sources in the at-worst case(number of sources equal number of sensors). It has been achieve using the eigenvalues of the correlation matrix of the received signals and the delay between microphones. Finally, measurements in the anechoic chamber verified the proposed approach. METHODS An acoustic goniometer is a system that measures the angle between a source and a receptor using the phase delay, thereby obtaining the source direction. The design dwell on two sensors (microphones) collocated in the 2D space in a concrete geometry. The implementation of each algorithm was done in Matlab based on two parts: the time delay estimation used in source localization by computing the azimuth in [2], and also an adaptation of the MPE block carried out in [4]. Likewise different methods based on the properties of the correlation matrix have been studied for delay estimating like in [3]. Apart from that, in [1] is explored the relation between sensor array geometry and eigenvalues to obtain the optimal sound sources separation and detection. This theory has been put into practice in programming in Matlab: minimization of the distance between microphones such that accomplish the condition of sources separation or sources detection. The optimization procedure has been done using two different SQP Methods: Active Set and Interior Point. Moreover, an optimization approach is presented for a system composed by two sensors and three sound sources. Several options based on mathematical theory has been considered for solving the problem. Eventually, taking advantage of the procedure followed in [1] and combined with the circumcenter calculation, the optimal distance for the microphones can be found. RESULTS Afterwards all this work, different simulations with the code in Matlab were tested reaching successful results. Then, a process of validation is required in the anechoic chamber for more realistic measurements. CONCLUSIONS In conclusion is demonstrated by theoretical calculation at first and then by experimental measurements that the optimal array geometry could help to improve the sound source separation approach. Forthcoming works will consist in extending this work for larger bandwidth and much more sound sources. Also, taking into consideration a more realistic model with reflections, interfering signals or noise corrupted.