Ph.D. Candidate: Murat Yılmaz
Program: Information Systems
Date: 04.09.2023 / 10:00
Place: A-212
Abstract: This study aims for detection and DOA estimation of muzzle blast and shock wave components of gunshot sound onboard a drone, despite the excessive ego-noise of the vehicle. The method depends on using the whole array data, Array Correlation Map, for improved detection and adaptive usage of Continuous Wavelet Transform scales for tuning to transient events of varying frequency. Although studied specifically for the processing of gunshot sounds on drones, the three novelties this study offers may be generalized to other array processing applications. The first is that low signal-to-noise ratio can be remedied in a detection problem with the help of the directionality of the sound source as compared to the ego noise. This is achieved by introducing the “Array Correlation Map” of the microphone array and using it to emphasize the unanimity among the array sensors. Using a simple mean value of the correlation map revealed successful results for very low signal-to-noise-ratio muzzle and shock wave scenarios, although other geometries and array processing problems may use the correlation map differently. Secondly, the help of CWT analysis is maximized by a self-adaptive selection of CWT scales. Thirdly, the tune-like feature of scales-selection is presented, which is demonstrated by automatically focusing on either muzzle or shock wave scales/frequencies. Results reveal signal-to-noise-ratio enhancement, successful muzzle and shock wave signal detection, and DOA estimation performance improvement.