Multi-feature Fusion Based Weak Underwater Acoustic Signal Detection

Detection of underwater acoustic signals is one of the most important aspects in oceanic information technology and has an irreplaceable place in national economic development and national security.The complexity of ocean,namely the oceanic impulsive noise,is the reason for the differences between underwater signal detection and territorial signal detection.The analysis contained herein is the underwater acoustic signal detection in oceanic impulsive noise.The main contributions are as follows.1.We studied the statistical behavior of the S?S distribution and analyzed each parameter's influence on the probability density function.Then the S?S distribution was used to model the impulsive noise,whose correctness is verified by the real marine noise.2.Two kinds of single feature detection algorithms accustomed to impulsive noise were proposed,including the fractile piecewise processing based signal detection and the maximum/minimum covariance matrix eigenvalue based signal detection,and the kernelized energy detection was studied.Each of the three methods has their own strengths and weaknesses.The first one needs precise estimation of noise characteristic exponent,with which it significantly outperforms the Cauchy detector.However,it only works in the range of 1~2 for characteristic exponent and performs best in around 1.6 and 1.9.The second one does not require any noise statics and is immune to noise uncertainty.It,however,needs a strong correlation among the detected signals.The third one has a robust performance and works under all values of characteristic exponents,especially for small values.Nevertheless,it is sensitive to noise uncertainty.3.By integrating the three signal characteristics,a multi-feature fusion robust underwater acoustic signal detection algorithm was devised and it is capable of transforming the problem of signal detection into a pattern recognition one.It can also transform the conventional threshold judging into the function judging where the optimal judging function is obtained through quantum memetic algor ithm based on the minimum mistake criterion.The multi-feature fusion based signal detection has the benefits of the three algorithms and is robust under different conditions.4.Two kinds of underwater DSSS signal detection algorithms are studied,namely the fractional mean square correlation and principle component analysis.When the precise information about the spread spectrum sequence period is available,the latter one achieves its best performance and outperforms the former one,and the verse visa.A multi-feature fusion detection algorithm integrates the benefits of the two methods.The results demonstrate its robustness under parameter matching and mismatching.