Noise Spectral Estimation Methods and Implementation of an Algorithm in the Frequency Domain for Improving Detection in a Passive Sonar System
Mohammad Zarezadeh Mehrizi, MasoudReza Aghabozorgi
The sonar is actually a radar which is used on or under the surface of the water and uses sound waves instead of radio waves to survey the environment. Detection and classification of marine vessels based on their acoustic radiated noise is of great importance in sonar systems. The non-static nature of the radiated noise of marine vessels and its dependency on the propagation channel, the variability of some of its parameters, and the difficulties in the simulation of this noise, make detection algorithms using a limited number of real data with a low diversity and or simulated training data to be not adequately valid. The present study aims to provide a new algorithm for the separation of target signals received in the form of convolutive mixture. For this purpose, a time-frequency window is designed that reconstructs an approximation of the target signal in the frequency domain. The cosine of the angle between two vectors is equal to the cosine of a parameter, called Hermitian angle between the two vectors. The Hermitian angle between two complex vectors remains unchanged if the two vectors are multiplied in a complex scalar. In the present study, this property is used to design a window for the separation of sources mixed convolutively. The advantage of this window over other detection methods is that there is no need for primary information on the geometric location of sources and hydrophones.