Research On Matched-field Processing Method For Passive Sonar Broadband Target Detection On Low Frequency And Shallow Water

Passive sonar is currently one of the methods for underwater remote detection while maintaining its own concealment.The development direction of passive sonar is low frequency.One of the important scenes of passive sonar application is shallow sea,and the future development trend of sonar technology will be a combination of underwater acoustic physics and signal processing.This thesis discusses broadband target detection methods on the conditions of low frequency ans shallow sea,and tries to apply the matched-field processing to broadband beamforming.Matched-field processing is mainly used for sound source localization and marine environment parameter inversion.The mainstream coherent processing methods and high-resolution beamforming methods have severe performance degradation when the signal-to-noise ratio decreases,and it is difficult to meet the real-time requirements of the detection algorithm,so there are relatively few studies on the use of matched-field processing for target detection.On the other hand,the conventional beamforming method based on the BTR has unsatisfactory detection performance in the detection of shallow sea targets where the refraction,scattering and multipath effects of sound waves are obvious,and the matched-field processing can effectively use the sound propagation information.In addition,broadband detection has advantages over narrowband detection in terms of stability and signal-to-noise ratio improvement.Considering the above points,this thesis attempts to use the matched-field processing method for passive sonar broadband target detection on the conditions of low frequency and shallow sea.This thesis uses the normal mode method as the sound field calculation method,and compares the detection capabilities of linear incoherent processors,MVDR incoherent processors,linear coherent processors and MVDR coherent processors in low signal-to-noise ratio.Through the simulation experiments,it is concluded that the detection performance of the linear incoherent processor is the best,and the detection performance of the linear incoherent processor decreases with the increase of the fluctuation of frequency spectrum.This thesis derives the output power of the linear incoherent processor,and explains why the detection ability of the linear incoherent processor decreases when the frequency fluctuation of the sound source increases.On this basis,in order to achieve target detection at a lower signal-to-noise ratio,this thesis further proposes a maximum signal-to-noise ratio processor.The maximum signal-to-noise ratio processor no longer tries to eliminate the influence of the sound source spectrum,but uses the estimated value of the sound source spectrum and the replica field to directly match the received data.After that,this thesis analyzes the output power of the maximum signal-to-noise ratio processor,and theoretically proves that the detection performance of the maximum signal-to-noise ratio processor is better than that of the linear incoherent processor.When there is no target,the output power of the two is approximately subject to the chi-square distribution of the same degree of freedom;when there is a target,the output power of the maximum signal-to-noise ratio processor is higher than that of the linear incoherent processor.When the fluctuation of the frequency spectrum increases,the difference is even more obvious.In order to solve the real-time calculation problem of the detection algorithm based on matched-field processing,this thesis proposes a replica field calculation method based on the principle of reciprocity,which reduces the number of model solving times and converts the sound field calculation from multiple small-scale calculations to small-scale large-scale calculations.Large-scale calculation greatly reduces the amount of calculation.To accelerate the large-scale sound field calculation problem generated by the principle of reciprocity,this thesis proposes a parallel algorithm based on the GPU rendering pipeline.In the comparative experiment of HLSL code,Matlab code and C# code,it is found that the running time of HLSL code based on GPU rendering pipeline is significantly lower than that of the other two codes.In order to measure the impact of the sound field mismatch to detection performance,this thesis proposes an equivalent noise model,derives the equivalent received signal-to-noise ratio of the linear incoherent processor and the maximum signal-to-noise ratio processor when the sound field is mismatched,and analyzes the upper bound of the amount of mismatch in the sound field caused by the error of sampling grid and sound velocity measurement error,and the upper bound of the amount of mismatch in the sound field under the influence of multiple factors.This thesis uses the ELBA island sea trial data to verify the detection performance of the maximum signal-to-noise ratio processor and the fast calculation method of the copy field.Target detection experiments of RM5 and RM2 sound sources under Gaussian white noise and colored noise conditions verify that the detection capability of the maximum signal-to-noise ratio processor is better than that of the linear incoherent processor.The effects of different processing bandwidths and frequency sampling intervals on the detection performance are analyzed,and the conclusion that the maximum signal-to-noise ratio processor is not sensitive to spectrum fluctuations is verified.Applying the sound field reciprocity and GPU rendering pipeline to the fast calculation of the copy field in the sea test environment,the calculation speed is significantly improved without affecting the detection results.