Source Location

sensors position simuleted source estimates true location 0 1 2 3 4 5 6 7 y 0 1 2 3 4 5 6 7 x 0 1 2 z B C F D E A Figure 3. Distribution of source location estimates around the true position for sources at the six points evidenced in Figure 1, when TDOA estimates are all aaected by gaussian noise with equal distribution. positions using a loudspeaker. It must be observed, however, that the assumptions about the error covariance R are not very realistic in a real environment , where the statistics of the TDOA estimates are aaected by the noise and reverberation conditions, by distance and orientation of the microphone pairs and by direc-tivity and frequency characteristics of the acoustic source. Due to all these elements an assessment of performance obtained in real experiments would require a statistical analysis on a large amount of data. To give an idea of the performance obtained with our preliminary data, Table 1 reports on the average location error measured using data collected by the two array conngurations using the loudspeaker as acoustic source. When using all the microphone pairs available in each square subarray (a total of 12 pairs) and an outlier detection algorithm, average performance is only slightly better than that obtained with only the four diagonal pairs. Besides, the two sensor conngurations do not show a meaningful diierence of location accuracy.