【作者】 刘庆华；

【导师】 欧阳缮；

【作者基本信息】 西安电子科技大学 ， 信号与信息处理， 2014， 博士

【摘要】 随着经济的发展和社会的进步，噪声污染日益严重，危害人们的身心健康，引起机械设备结构的声疲劳，并降低武器装备的作战性能和声隐身性能。因此噪声控制在民用和军事领域中都具有十分重要的意义。有源噪声控制技术是控制低频噪声的有效方法，根据噪声控制机理可分为声源控制、传播途径控制和接受者保护三类，研究内容和手段具有多样性。论文从接受者保护和声源控制的角度出发，针对基于自适应滤波技术的前馈有源噪声控制方法和基于模态分析的结构声辐射控制方法中存在的一些主要问题开展研究工作。论文所做主要工作概况如下：1.在自适应前馈有源控制技术中，非因果成分会降低噪声控制性能，为了使系统尽量满足因果约束，提出一种参考信号的拾取方法。该方法将参考传声器放置在噪声源来波方向上，并通过距离调整来增大参考传声器和误差传声器的时间延迟，用于抵消串扰和反射等因素引起的非因果成分，从而提高噪声控制性能。在此方法中，为了检测宽带噪声源的来波方向，提出了一种基于KR积的内插聚焦估计算法。该算法将各频率点的协方差矩阵进行KR积扩展后，借助最小二乘内插聚焦矩阵，构造无噪声聚焦后的各频率数据组合矩阵，进而采用MUSIC方法实现过定和欠定两种情况下的来波方向估计。最后，将噪声源的定位结果用于指导参考传声器的布放，通过比较传声器布放位置及数目调整前后的噪声控制效果，验证参考信号拾取方法的有效性。2.在宽带噪声的自适应前馈有源控制技术中，无延迟子带自适应滤波方法由于能避免信号通道的延迟且收敛速度快，成为宽带噪声控制中的主要方法。但该方法的控制效果会受到分析滤波器组的群延迟和带内混叠误差的影响，因此分析滤波组的设计是关键。现有的分析滤波器组是调制线性相位原型低通滤波器得到的，它不能保证滤波器组的低群延迟和小旁瓣效应，致使噪声控制效果不明显。为此，提出迭代二阶锥规划方法设计分析滤波器组。将带内混叠误差和群延迟构造为关于原型滤波器权系数的二次函数，进而通过迭代二阶锥优化得到原型滤波器权系数，最后利用均匀DFT调制为分析滤波器组。与同阶线性相位分析滤波器组相比，该滤波器组具有更低的群延迟和更小的带内混叠误差。在此基础上，采用部分子带权系数更新的归一化最小均方误差算法，构建了频域无延迟噪声控制系统。实验结果表明，该系统的噪声控制效果明显，不仅能有效降低残余噪声功率谱，提高噪声衰减量，而且还具有收敛速度快和计算复杂度低等优点。3．在窄带噪声的自适应前馈有源控制技术中，从理论上分析了频率偏差造成自适应滤波器权系数的均值偏移不能收敛到零矢量，是导致控制性能恶化的根本原因。因此为了提高噪声控制效果，必须对干扰信号的频率进行精确估计。为此，提出了一种用于实值信号的快速子空间跟踪-ESPRIT类频率估计方法。该方法的信号子空间维数与信号频率的个数相等，从而避免了传统方法用于实值频率估计时，信号子空间的维数为频率个数2倍的问题。仿真结果表明算法的精度高，满足实际应用的要求。最后，结合不需要次级通道估计的FXLMS算法，在DSP平台上实现了弹性薄板的基频和谐波分量的振动控制，从而降低了封闭声腔中的结构声辐射的声压级。4．在基于模态分析的弹性结构声辐射的控制技术中，根据输入输出测量数据建立精确受控系统模型是实施该类控制方法的前提。目前，主要采用子空间辨识方法来建立系统状态方程，但该方法计算量大，难以得到有效应用。为此，利用广义能观矩阵的列空间与观测矢量相关矩阵的信号子空间一致的特征，提出了一种在线子空间辨识方法用于系统状态方程的建立。该方法包含了两个递推过程：一是利用矩阵求逆变换计算观测矢量；二是基于多级分解幂迭代子空间跟踪算法估计广义能观矩阵，以保证按指数全局收敛到主子空间。仿真结果表明该方法能提高辨识精度，不仅减小广义能观矩阵估计值与真实值的夹角，还能降低辨识得到的输出数据与真实值之间的均方根误差。最后，利用有限元和声学分析软件，构建了一种基于振动控制的封闭空间降噪系统模拟环境。采用所提的在线子空间辨识法建立了压电结构系统模型，结合线性二次最优控制方法产生压电陶瓷作动器次级电压。通过控制与声模态耦合大的结构模态，有效降低了舱室结构声辐射的声压级，验证了在线子空间辨识方法应用在结构声辐射控制中的有效性。更多还原

【Abstract】 With the economic development and social progress, the noise pollution isbecoming more and more serious problem. Noise harms people’s physical and mentalhealth, causes acoustic fatigue of mechanical equipments. Also, it degrades theoperation and stealth performance of weapon equipments. Consequently, reducing theeffect of noise has important practical significance in civil and military fields. Theactive noise control technique is an effective method for low frequency noise control.According to the mechanism of noise control, the technique can be divided into threeclasses: source control, transmission control and receiver protection. This thesisthoroughly investigates the feedforward active noise control method based on adaptivefiltering, and structural acoustic radiation control method based on modal analysis, torealize the receiver protection and source control. The main content of this thesis can besummarized as follows:1. In the adaptive feedforward active noise control system, the noise attenuation ofnoncausal system is less than that of causal system. In order to ensure the causality andenhance the controlling effectiveness, the pickup method of the reference signal isdiscussed. The reference microphones are demanded to be placed in the directions ofarrival (DOA) of noise sources. The time delay between the reference microphones andthe error microphones are increased by enlarging the distances between them, used tooffset the noncausal components caused by echo and crosstalk in the control paths.Accordingly, it is necessary to estimate the DOA of noise sources in this pickup scheme.Based on Khatri-Rao (KR) product and interpolation focusing matrix, a new algorithmis proposed to estimate the DOA. In this algorithm, the covariance matrix at eachfrequency point is stacked into column vectors through KR product. Then the columnvectors are focused on the center frequency to construct a noise-free data matrix, using aleast square interpolation focusing matrix. Finally, the DOA is estimated by the methodMUSIC in the overdetermined and underdetermined cases. With DOA estimation for theplacement of reference microphones, the experiment results verify the feasibility of theproposed pickup method, by comparing the noise control performance before and afterthe adjustment of location and number for the reference microphones.2. The delayless subband adaptive filtering technique plays a prominent role in thewideband noise active control, which can avoid the signal path delay and accelerate theconverging rate of filter weights. The noise controlling performance of this technique isaffected by the group delay and in-band aliasing of analysis filter banks. Therefore, the design of the analysis filter banks is very crucial. Typically, the analysis filter banks arecomposed of a set of filters created by modulating a prototype FIR filter (PF). Suchfilter banks suffer from a long group delay and a high side-lobe effect, which limits thenoise attenuation. An iterative second-order cone programming (SOCP) method isdeveloped to design the analysis filter banks. The in-band aliasing distortion and groupdelay constraints are expressed as the cost functions of the PF coefficients. In the sequel,the optimization problem of the cost functions can be formulated as an iterative SOCPform. The optimized PF coefficients are uniform-DFT modulated into the desiredanalysis filter banks. The simulation results demonstrate that the well designed filterbanks have a smaller in-band aliasing distortion and lower group delay, compared withthe linear-phase filter banks. Finally, a structure of the delayless subband active noisecontrol in frequency domain is provided, merged with the selective partial updatenormalized LMS algorithm. Simulation results show that the noise controllingperformance is improved with a small residual noise power spectrum, high noiseattenuation level and fast convergence rate.3. In the narrowband active noise control system, the frequency mismatch betweenthe synthesized reference signal and the primary noise will cause that the mean valuedeviation of filter weight coefficients cannot converge to zero vector. As a result, theperformance of narrowband active noise control systems degenerates significantly. Inorder to improve the performance, it is necessary to estimate the frequencies of theprimary noise. An ESPRIT-like algorithm with subspace tracking is proposed toestimate the frequencies of real-valued sinusoidal noise signal. In the proposedalgorithm, the dimension of the signal subspace is equal to the number of thefrequencies present in the observation, and a half of the signal subspace dimension inthe MUSIC method. The simulation results indicate that the proposed algorithmpossesses a high precision to meet the requirement of practical application. Equippedwith the digital signal processing (DSP) platform, the harmonic components ofvibration is controlled for the elastic thin plat through the FXLMS algorithm withoutsecondary path identification. Meanwhile the sound pressure level is reduced for thestructural acoustic radiation noise in the enclosure cavity.4. In the control technique of structural acoustic radiation based on modal analysis, aprecise model is the prerequisite for application of the control method. Recently, thesubspace identification method is popular for establishing state space model. However,it is difficult for practical application because of huge complexity. In the basis of factthat the column space of the extended observability matrix is the same with the signal subspace of autocorrelation matrix of the observation vector, a new recursive subspaceidentification algorithm is proposed for the estimation of state space method. It containstwo recursive processes: one is to compute the observation vectors by using the matrixinversion formula, the other is to estimate the extended observability matrix based onthe multistage power iteration subspace tracking algorithm. Such algorithm canexponentially converge to the principal subspace. With the numerical and practicalmodels, the experiment results show that this method can enhance the identificationprecision. In detail, it manifests a small angle between the estimation and true extendedobservability matrix, and low root mean square error between the identified output andthe actual output. Also we simulate the control environment of the structural acousticradiation by using finite element and acoustic analysis software. The state space modelof the piezoelectric smart structure is identified by the proposed method, and the controlvoltage of the secondary actuator is computed by the linear quadratic optimal controlmethod. Finally, the sound pressure level of the structural acoustic radiation noise isdecreased by controlling the structure modals which have large coupling coefficientswith the acoustic modals.更多还原