Study On The Key Techniques Of Adaptive Noise Cancellation For Unmanned Underwater Vehicle Array

With the development of unmanned underwater vehicle (UUV) demanding ofincreasing the speed of UUV, the level of the reverberation and interference received bythe sonar array installed in the UUV will lead to be enhanced. Moreover, the acousticwarfare has been improving and widely employed into the practical applications.Therefore, it is significantly urgent needs for ocean explorations and coastal defensethat how to effectively suppress and reduce the self-noise, reverberation and jamming inorder to improve the capability of the detection of long distance weak signal.Firstly, the generation of mechanism and the characteristic of self-noise areinvestigated in detail. In the basis of array signal processing with multi-channel featureand kernel-based method nonlinear adaptive filtering, this dissertation addressed severalimportant topics including the theoretical convergence performance of nonlinearkernel-based method adaptive filtering, the array adaptive noise cancellation of usingsingle-kernel in nonstationary environment, and the combination of multi-kernel basedlinear and nonlinear of array adaptive noise cancellation algorithms. Moreover, thesuppression of reverberation and interference in the space-time two dimension adaptiveprocessing is investigated for MIMO array, and the performances of algorithmspresented in this thesis are validated via simulation experiments. The main contributionsand innovations of this dissertation are summarized as follows:1. Assuming that the dynamic nonstationary signal will result in the obsoleteredundant elements whose statistical distribution mismatch with those of input data inonline dictionary, the transition and steady-state characteristics of the least mean squarenonlinear adaptive filter based on the single Gaussian kernel function are derived in thishypothesis. These theoretical computational methods provide a powerful tool forcomparing the performance and the design of single kernerl adptive filtering. Thenumerical simulation results show that the theoretical convergence cures predicted bythe analysis model consistently agree with the mean-square-error cures averaged by theMonte Carlo simulation tests in the transition and steady-state stages, which verifies thecorrectness and validation of the derived theoretical computational methods. Thetheoretical models provide with the theoretical justification for taking the adaptiveupdating strategy for the dictionary in the nonstationary application of the nonlinear adaptive noise canceller.2. Within the nonstationary noise environment and multi-channel differencingalgorithm, two the forward-backward splitting kernel least-mean-square(FOBOS-KLMS) methods with the1-norm sparsity promoting used as the nonlinearadaptive noise cancellation are seperately proposed. Moreover, the proof of the stabilityof FOBOS-KLMS method with-norm regularization term is also presented. Thesingle Gaussian kernel based nonlinear adaptive filtering is introduced by the-normregularization term yielding to the forward-backward splitting online updating strategyfor dictionary. When the contribution weights of elements in the dictionary for fittingthe function remain below the specified threshold, they will be discarded out of thedictionary. The simulation results using the noise data measured in the lake experiment,when the sonar array is accelerated or decelerated in order to simulate the statisticalnonstationary of the received noise, show that the mean-square-error of the proposedFOBOS-KLMS methods is much lower7dB than the NLMS linear method. And theproposed methods can reduce the “order” of nonlinear adaptive filter or therequirements of computation complexity and memory storage of array nonlinearadaptive noise canceller compared with the existing algorithms, which is a foundationof the practical engineering of online adaptive noise cancellation applications.3. Multi-kernel method outperforms over the single-kernel method due to possessmuch more degrees of freedom and features, which are able to be used to solve therecognition of dynamic system and the problem of parameters of kernel needed to befixed offline. Hence, three types of least-mean-square nonlinear adaptive filtering basedon the K Gaussian kernel functions are presented. Furthermore, the theoreticalconvergence analysis computational methods of first two algorithms are proposed underthe case the determined dictionary elements in a prior. With the derived theoreticalmodels, the respective characteristics of performances of multi-kenrel adaptive filteringcan be compared. The simulation results show that the cures of theoretical modelcoherently match the mean-error-square curves computed by Monte Carlo experimentsin steady-state and transint stages, which indicates that the proposed computationalmethods are correct and effective. The theoretical model can be viewed as an excellenttool of performance analysis and design for multi-kernel adaptive filters.4. Because of the complicate composition of the special and temporal noisereceived by array, two types of bi-kernel normalized least-mean-square (BKNLMS)methods are proposed weighted by linear and nonlinear kernel functions according tothe structures of multi-kernel adaptive filters mentioned. Considering not only the cancellation of linear noise but also the suppression of nonlinear noise, the bi-kernelfunctions are individually weighted to obtain the two BKNLMS methods due to thecomplex relative noise provided by multi-channel differecing algorithm. The continuouswave (CW) signal and linear frequency modulation (LFM) signal are combined with themeasured noise to be the received signals for the atuocorellation detection to show theeffectiveness of ANC. The simulation results show that the two BKNLMS methods canimprove the SNR of input signal due to simultaneously suppress the linear andnonlinear noise. At the identical probabilities of the detection the SNR of two BKNLMSmethods are decreased by5dB and2dB over the conventional linear adaptive noisecanceller and nonlinear adaptive noise canceller based on single KNLMS, separately.Thus the two proposed BKNLMS methods have significant practical value for the sonararray adaptive noise cancellation system.5. Reduced-rank space-time adaptive processing (STAP) method based onestimation of subspace for flank MIMO array is proposed to solve the difficulty ofdetection of low speed moving target when the reverberation and jamming exist. Theproposed method utilized the time-limited and band-limited of characteristics of prolatespheroidal wave function (PSWF) to approximate the clutter subspace. The MIMOsystem space-time weights are obtained by “zero-forcing” method, and the covariancematrix contained jammer and noise is acquired though the matched filter orthogonal toall transmitted independent signal waveforms. The simulation results demonstrated thatthe proposed method can effectively suppress the clutter and jammer with lowercomputational complexity compared with the other methods even existing the influenceof the practical situations.This work has important theoretical significance and practical value to improve thesignal-noise ratio (SNR) and increase the detection performance of the long distanceweak signal for UUV equipped with flank array, and has general significance for theother underwater array systems.