No abstract

<p>‎Atmospheric sound waves with frequencies below the human hearing threshold in the range of 0.002 Hz to 20 Hz are generally referred to as infrasound‎. ‎Wind is the main noise in the above-mentioned frequency range‎. ‎The operation of receiving and detecting infrasound are often hampered by wind‎. ‎Therefore‎, ‎high quality detectors are required‎. ‎For this purpose‎, ‎sensor arrays and array signal processing techniques are utilized‎. ‎Fisher ratio-based signal detection is a widely used and powerful method in the field of infrasound‎. ‎The main drawback of this approach is its high computational time due to the repeated computation of test statistics for each element of the slowness grid‎. ‎Thus‎, ‎the researchers use a relatively low-resolution slowness grid in order to save time in processing‎. ‎On the other hand‎, ‎low resolution grid results in an error in the values of estimated parameters of infrasound wave‎. ‎In this study‎, ‎a genetic algorithm based detection method is proposed in order to overcome the fundamental problems of the Fisher method‎. ‎In the proposed method‎, ‎the slowness grid components (px‎,‎py) are defined as the chromosome for the genetic algorithm‎. ‎Despite the previous methods‎, ‎the genetic algorithm has created the advantage that searching could be conducted in a continuous slowness grid‎. ‎Therefore‎, ‎the continuity of the grid and searching only a limited number of slowness vectors reduce error rates and processing time respectively‎. ‎The error of apparent velocity and incoming orientation became 0.5923 and 0.0710 respectively‎, ‎and the processing time decreased considerably from 25835.07 seconds to 533.55 seconds on average.</p>

<p>Special attention is now being paid to the processing of microphone array signals‎. ‎Sensors located in the array receive the source signal at different time intervals‎. ‎In order to find the direction of source‎, ‎the time difference measurement algorithms‎, ‎which are based on the calculation of cross correlation function‎, ‎calculate the correlation between the outputs of the array sensors‎. ‎The maximum peak of the correlation function indicates the time difference of the sensors to each other‎. ‎Then‎, ‎the source signal orientation will be estimated by using the direction of arrival estimation algorithms based on the calculation of the time differences‎. ‎When the signal propagates in a direct path‎, ‎the cross-correlation function is efficient to measure the time differences‎. ‎However‎, ‎in the case the signal propagates in multipath medium‎, ‎the generalized cross-correlation functions will be used due to the presence of multiple peaks in the cross-correlation function‎. ‎The Phase Transform method‎, ‎which is a family of generalized cross-correlation methods‎, ‎yields better results than other methods‎. ‎Nevertheless‎, ‎it performs poorly at low SNRs‎. ‎In this study‎, ‎we add a modification factor to the weighting function of the Phase Transform method‎. ‎In the following‎, ‎the results of time differences obtained from the proposed method are then applied to different DOA methods‎. ‎In the simulations‎, ‎it was observed that the particle swarm optimization method offers the least error among the DOA methods‎, ‎so that the error values of backazimuth and elevation decreased by about 3 degrees and 2 degrees‎, ‎respectively for the factor of 0.65 at low SNRs‎. </p>

<p>Special attention is now being paid to the processing of microphone array signals‎. ‎Sensors located in the array receive the source signal at different time intervals‎. ‎In order to find the direction of source‎, ‎the time difference measurement algorithms‎, ‎which are based on the calculation of cross correlation function‎, ‎calculate the correlation between the outputs of the array sensors‎. ‎The maximum peak of the correlation function indicates the time difference of the sensors to each other‎. ‎Then‎, ‎the source signal orientation will be estimated by using the direction of arrival estimation algorithms based on the calculation of the time differences‎. ‎When the signal propagates in a direct path‎, ‎the cross-correlation function is efficient to measure the time differences‎. ‎However‎, ‎in the case the signal propagates in multipath medium‎, ‎the generalized cross-correlation functions will be used due to the presence of multiple peaks in the cross-correlation function‎. ‎The Phase Transform method‎, ‎which is a family of generalized cross-correlation methods‎, ‎yields better results than other methods‎. ‎Nevertheless‎, ‎it performs poorly at low SNRs‎. ‎In this study‎, ‎we add a modification factor to the weighting function of the Phase Transform method‎. ‎In the following‎, ‎the results of time differences obtained from the proposed method are then applied to different DOA methods‎. ‎In the simulations‎, ‎it was observed that the particle swarm optimization method offers the least error among the DOA methods‎, ‎so that the error values of backazimuth and elevation decreased by about 3 degrees and 2 degrees‎, ‎respectively for the factor of 0.65 at low SNRs‎. </p>

<p>‎Atmospheric sound waves with frequencies below the human hearing threshold in the range of 0.002 Hz to 20 Hz are generally referred to as infrasound‎. ‎Wind is the main noise in the above-mentioned frequency range‎. ‎The operation of receiving and detecting infrasound are often hampered by wind‎. ‎Therefore‎, ‎high quality detectors are required‎. ‎For this purpose‎, ‎sensor arrays and array signal processing techniques are utilized‎. ‎Fisher ratio-based signal detection is a widely used and powerful method in the field of infrasound‎. ‎The main drawback of this approach is its high computational time due to the repeated computation of test statistics for each element of the slowness grid‎. ‎Thus‎, ‎the researchers use a relatively low-resolution slowness grid in order to save time in processing‎. ‎On the other hand‎, ‎low resolution grid results in an error in the values of estimated parameters of infrasound wave‎. ‎In this study‎, ‎a genetic algorithm based detection method is proposed in order to overcome the fundamental problems of the Fisher method‎. ‎In the proposed method‎, ‎the slowness grid components (px‎,‎py) are defined as the chromosome for the genetic algorithm‎. ‎Despite the previous methods‎, ‎the genetic algorithm has created the advantage that searching could be conducted in a continuous slowness grid‎. ‎Therefore‎, ‎the continuity of the grid and searching only a limited number of slowness vectors reduce error rates and processing time respectively‎. ‎The error of apparent velocity and incoming orientation became 0.5923 and 0.0710 respectively‎, ‎and the processing time decreased considerably from 25835.07 seconds to 533.55 seconds on average.</p>