Estimating direction of arrival of long range gunshot signals

Information obtained from recorded gunshot signals can be useful for law enforcement agencies and defense forces. Direction of arrival estimation of a gunshot is an important issue in shooter localization. As the distance between the firing position and the sensor increases, signal to noise ratio decreases and this estimation degrades. In that case, some sort of pre-processing of the firing acoustic signal before the application of direction of arrival estimation algorithms becomes necessary. This paper shows that applying spectral subtraction to enhance the signal improves the results. Moreover, it is shown that further improvement is attained when employing a recently proposed data selection algorithm. The tests were carried out with simulated directions by delaying copies of a real gunshot signal and also with real gunshot signals recorded by a microphone array.