【作者】 杨天文；

【导师】 黄大贵；

【作者基本信息】 电子科技大学 ， 机械电子工程， 2013， 博士

【摘要】 参量声源是利用空气中声波的非线性传播效应产生指向性声束的新型声学系统。该系统要获得高质量的可听声源，一方面要充分利用空气的非线性传播效应，另一方面又要最大限度地减少非线性引起的失真。因此，深入认识并合理利用非线性是该领域要解决的关键问题之一。本文从理论和实验两方面研究了参量声源的非线性特性，完成的主要工作如下。根据非线性声学的基础理论，对参量声源的非线性进行了理论分析。结果表明现有理论不足以充分表达参量声源的非线性特性，而基于现有理论发展起来的参量声源信号处理方法在避免可听声非线性失真方面也有待完善。针对现有理论的不足，提出采用人工神经网络对参量声源进行非线性建模。通过神经网络模型，可以在不完全清楚非线性机理的情况下准确地拟合系统的输入输出关系，为参量声源的控制及性能优化提供条件。在解决了参量声源神经网络建模的几个基本问题的基础上，采用BP神经网络和RBF神经网络分别建立参量声源的非线性模型。并通过正弦信号激励和随机信号激励的仿真对模型的有效性进行了评估与对比。阐述了神经网络模型的评估方法，并重点对灵敏度以及泛化能力进行了分析。采用近似函数替代激活函数进行参量声源神经网络全局灵敏度计算的方法，并给出了理论推导。在分析影响神经网络模型泛化能力的因素的基础上，采用了几种方法来提高模型的泛化能力，并通过仿真证明了其效果。将神经网络逆控制的基本思想及方法应用于参量声源的控制中，用系统控制的理论与方法解决参量声源的非线性信号处理问题。设计了参量声源的神经网络直接逆控制系统以及PID复合逆控制系统。并通过仿真验证了神经网络逆控制方法对改善参量声源系统性能的效果。对参量声源自解调信号的非线性进行了实验测试，通过实验数据分析了自解调信号的非线性失真。并在相同条件下，用所建立的神经网络模型及PID复合神经网络逆控制模型对自解调信号的非线性进行了仿真。通过对比验证了基于神经网络的参量声源非线性建模及控制方法的效果，提出后续改进的方向。更多还原

【Abstract】 The parametric sound source is a new type sound system which utilizs the nonlinearpropagation effect of the sound in the air to generate an audible sound beam withdirectivity. To obtain high quality audible sound, this acoustic system needs to make fulluse of nonlinear propagation effect of the sound. And on the other hand, the sounddistortion produced by nonlinear characters should be restrained as far as possible. Sodeep understanding and rational utilizing nonlinear character is a key problerm needs tobe solved in the field of the parametric sound source. This dissertation has studiedtheoretically and experimentally nonlinear characters of the parametric sound source.And the main research works and achievements are summarized as follows:According to the basic nonlinear acoustic theory, the nonlinear characters of theparametric sound source are analyzed theoretically. The analysis results show that thecurrent theories are not enough to understand and explore the nonlinear characters of theparametric sound source. And the signal processing methods developed from the currenttheories remain to be improved on reducing the nonlinear distortion of sound.The artificial neural network(ANN) is proposed to bulid the model of the parametricacoustic system. The mapping relationship between input and output of this system canbe fitted accurately by the ANN even if the nonlinear principle and mechanism is notclear. This provides necessary conditions for system contol and performanceoptimization of the parametric sound source. After solving many basic problems ofANN modeling, the back propagation(BP) neural network model and radial basisfunction(RBF) neural network model are built. These models are demonstrated andcompared through simulations under sinusoidal signal excitation and random signalexcitation.The evaluation methods of ANN model are described. The sensitivity andgeneralization ability of ANN model of the parametric sound source are analyzed in thisdissertation. The activation function substitution method is proposed to calculate theglobal sensitivity of the parametric sound source model, and the theoretical formula isderived. After analyzing the factors influencing generalization ability of ANN model, many methods are used to improve the generalization ability of ANN model of theparametric sound source. And the effects of these methods are demonstrated bysimulations.This dissertation applies the basic idea and method of ANN inverse control to buildthe nonlinear control system of parametric acoustic system. The ANN direct inversecontrol system and PID compound inverse control system are designed. And the systemperformance improvements through the ANN inverse control are verified by simulation.The experiments for testing the self-demodulated signal have been done tounderstand the nonlinear and sound distortion of the parametric sound source. Andunder the same conditions as experiments, with the ANN model and the PID compoundANN inverse control model established in aforementioned work, the MATLABsimulations for analyzing the nonlinear and sound distortion are implemented.Comparisions between experimental result and emulational result confirm that the ANNmodel and the PID compound ANN inverse control model are effective. From thesecomparisions, the improvement direction and breakthrough points are clear in nextresearch work.更多还原