子带自适应滤波技术研究及其应用
发布时间：2018-12-10 07:47
【摘要】：自适应滤波器在通信、声纳、雷达、音频处理、图像处理等领域有着广泛的应用,是自适应信号处理中最受关注的研究课题之一,其性能主要由所使用的自适应算法和滤波器结构决定。子带自适应滤波器具有特殊的结构,能够对输入信号进行频带分割与抽取,从而减小输入信号的相关性、提高收敛速度、降低对信号处理速度的要求,近年来被广泛应用于信号处理领域。本文依据课题需要,针对子带自适应滤波器中的滤波器组设计和自适应滤波算法展开深入研究,主要内容及创新点概况如下:1.研究了均匀子带滤波器组的基本结构以及原型滤波器的设计方法。针对均匀子带余弦调制滤波器组中,采用间接设计法得到的原型滤波器恢复出的信号幅度存在失真较大的问题,首先依据重构条件和过渡带的平方根余弦滚降特性设计代价函数,然后通过选择合适的原型滤波器通带截止频率,使其3dB通带截止频率接近理想滤波器的通带截止频率,最后通过迭代的方式最小化代价函数并获得最佳原型滤波器。仿真实验表明,相比于同类设计方法,改进后的原型滤波器具有更好的重构性能。2.针对系统辨识中的应用,设计了两种变参数子带自适应滤波器。针对仿射投影归一化子带自适应滤波算法收敛速度和稳态失调存在的折中问题,设计了使均方偏差上界最小化的变步长矩阵替代固定步长,该变步长矩阵由各子带的步长参数构成,相对原有算法仅有较低的复杂度增加。实验仿真表明,该方法相比变参数归一化子带自适应滤波类算法具有更好的收敛性能、跟踪性能以及对脉冲干扰的鲁棒性。针对仿射投影符号子带自适应滤波算法收敛速度和稳态失调存在矛盾的问题,提出了一种变正则化参数设计方法。该方法采用随机梯度下降法来更新正则化参数从而使得均方偏差最小化,仿真验证了其收敛性能和对脉冲干扰以及double-talk的鲁棒性好于变参数符号子带自适应滤波类算法。3.针对信道均衡中的应用,设计了两种变参数子带均衡器。通过分析发现多带结构可以有效的克服传统子带均衡器输出端存在混叠分量的问题,在此基础上提出了两种基于多带结构的变参数子带均衡器,来解决定参数均衡器收敛速度和稳态失调之间的矛盾。第一种是在符号子带均衡算法的基础上,将每次迭代中最小化后验误差向量的能量作为代价函数,来设计变步长矩阵,并用于下一次迭代;第二种是在归一化子带均衡算法的基础上,将最小化误差向量的能量作为代价函数,采用随机梯度下降法来更新正则化参数,从而得到该均衡算法的变正则化参数。两种方法都能较好的解决子带均衡器的快收敛速度和低稳态失调之间的矛盾,从而获得良好的整体性能。
[Abstract]:Adaptive filter is widely used in communication, sonar, radar, audio processing, image processing and so on. It is one of the most concerned research topics in adaptive signal processing. Its performance is mainly determined by the adaptive algorithm and filter structure used. The sub-band adaptive filter has a special structure, it can segment and decimate the input signal in the frequency band, thus reducing the correlation of the input signal, improving the convergence speed and reducing the requirement of the signal processing speed. In recent years, it has been widely used in the field of signal processing. In this paper, the design of filter banks and adaptive filtering algorithms in subband adaptive filters are studied. The main contents and innovations are as follows: 1. The basic structure of uniform subband filter bank and the design method of prototype filter are studied. In the uniform subband cosine modulation filter bank, the amplitude of the recovered signal from the prototype filter obtained by indirect design method has the problem of large distortion. First, the cost function is designed according to the reconstruction condition and the square root cosine roll down characteristic of the transition band, and then the 3dB passband cut-off frequency is close to that of the ideal filter by selecting the appropriate passband cut-off frequency of the prototype filter. Finally, the cost function is minimized and the optimal prototype filter is obtained by iterative method. The simulation results show that the improved prototype filter has better reconstruction performance than the similar design method. 2. Aiming at the application of system identification, two kinds of variable parameter subband adaptive filters are designed. Aiming at the convergence rate of affine projection normalized subband adaptive filtering algorithm and the trade-off between steady state misalignment, a variable step matrix is designed to minimize the upper bound of mean square deviation instead of fixed step size. The variable step size matrix is composed of the step size parameters of each subband, and the complexity of the variable step matrix is only lower than that of the original algorithm. Experimental results show that the proposed method has better convergence performance, tracking performance and robustness to impulse interference than the variable parameter normalized subband adaptive filtering algorithm. Aiming at the contradiction between convergence speed and steady state misalignment of affine projection symbol subband adaptive filtering algorithm, a variable regularization parameter design method is proposed. The stochastic gradient descent method is used to update the regularization parameters to minimize the mean square deviation. Simulation results show that its convergence performance and robustness to pulse interference and double-talk are better than those of variable parameter symbol subband adaptive filtering algorithms. Aiming at the application of channel equalization, two kinds of variable parameter subband equalizers are designed. It is found that multi-band structure can effectively overcome the problem of aliasing component in the output of traditional sub-band equalizer. Based on this, two kinds of variable parameter sub-band equalizers based on multi-band structure are proposed. To solve the contradiction between convergence rate and steady state misalignment of constant parameter equalizer. The first is to design the variable step size matrix by minimizing the energy of the posteriori error vector as a cost function on the basis of the symbolic subband equalization algorithm and using it for the next iteration. The second is based on the normalized subband equalization algorithm. The energy of the minimized error vector is taken as the cost function, and the regularization parameters are updated by the stochastic gradient descent method, and the variable regularization parameters of the equalization algorithm are obtained. Both methods can solve the contradiction between the fast convergence speed of sub-band equalizer and the low steady-state misalignment, and obtain good overall performance.
【学位授予单位】：解放军信息工程大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN713
本文编号：2370230
[Abstract]:Adaptive filter is widely used in communication, sonar, radar, audio processing, image processing and so on. It is one of the most concerned research topics in adaptive signal processing. Its performance is mainly determined by the adaptive algorithm and filter structure used. The sub-band adaptive filter has a special structure, it can segment and decimate the input signal in the frequency band, thus reducing the correlation of the input signal, improving the convergence speed and reducing the requirement of the signal processing speed. In recent years, it has been widely used in the field of signal processing. In this paper, the design of filter banks and adaptive filtering algorithms in subband adaptive filters are studied. The main contents and innovations are as follows: 1. The basic structure of uniform subband filter bank and the design method of prototype filter are studied. In the uniform subband cosine modulation filter bank, the amplitude of the recovered signal from the prototype filter obtained by indirect design method has the problem of large distortion. First, the cost function is designed according to the reconstruction condition and the square root cosine roll down characteristic of the transition band, and then the 3dB passband cut-off frequency is close to that of the ideal filter by selecting the appropriate passband cut-off frequency of the prototype filter. Finally, the cost function is minimized and the optimal prototype filter is obtained by iterative method. The simulation results show that the improved prototype filter has better reconstruction performance than the similar design method. 2. Aiming at the application of system identification, two kinds of variable parameter subband adaptive filters are designed. Aiming at the convergence rate of affine projection normalized subband adaptive filtering algorithm and the trade-off between steady state misalignment, a variable step matrix is designed to minimize the upper bound of mean square deviation instead of fixed step size. The variable step size matrix is composed of the step size parameters of each subband, and the complexity of the variable step matrix is only lower than that of the original algorithm. Experimental results show that the proposed method has better convergence performance, tracking performance and robustness to impulse interference than the variable parameter normalized subband adaptive filtering algorithm. Aiming at the contradiction between convergence speed and steady state misalignment of affine projection symbol subband adaptive filtering algorithm, a variable regularization parameter design method is proposed. The stochastic gradient descent method is used to update the regularization parameters to minimize the mean square deviation. Simulation results show that its convergence performance and robustness to pulse interference and double-talk are better than those of variable parameter symbol subband adaptive filtering algorithms. Aiming at the application of channel equalization, two kinds of variable parameter subband equalizers are designed. It is found that multi-band structure can effectively overcome the problem of aliasing component in the output of traditional sub-band equalizer. Based on this, two kinds of variable parameter sub-band equalizers based on multi-band structure are proposed. To solve the contradiction between convergence rate and steady state misalignment of constant parameter equalizer. The first is to design the variable step size matrix by minimizing the energy of the posteriori error vector as a cost function on the basis of the symbolic subband equalization algorithm and using it for the next iteration. The second is based on the normalized subband equalization algorithm. The energy of the minimized error vector is taken as the cost function, and the regularization parameters are updated by the stochastic gradient descent method, and the variable regularization parameters of the equalization algorithm are obtained. Both methods can solve the contradiction between the fast convergence speed of sub-band equalizer and the low steady-state misalignment, and obtain good overall performance.
【学位授予单位】：解放军信息工程大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN713
【参考文献】
相关期刊论文 前8条
1 赵艳;罗胜恩;万坚;;余弦调制子带滤波器组迭代设计的改进算法[J];数据采集与处理;2013年01期
2 张天骐;谭方青;高春霞;黄烈超;;一种新的窗函数法设计余弦调制滤波器组系统[J];系统工程与电子技术;2011年12期
3 蒋俊正;水鹏朗;周芳;;M带余弦调制滤波器组的迭代设计算法[J];电路与系统学报;2011年06期
4 赵艳;万坚;;一种单载波宽带信号非线性均衡技术[J];计算机工程与应用;2012年05期
5 张炜;戴旭初;许小东;;基于非均匀子带分解的宽带线性盲均衡器[J];电子学报;2010年04期
6 张炜;戴旭初;;一种分数间隔恒模算法的宽带盲均衡器[J];中国科学技术大学学报;2009年09期
7 张炜;金光明;戴旭初;;一种基于子带分解技术的宽带盲均衡器结构和算法[J];电子与信息学报;2009年03期
8 张子敬,焦李成;M带余弦调制滤波器组的设计[J];电子学报;2001年01期
，本文编号：2370230
本文链接：https://www.wllwen.com/kejilunwen/dianzigongchenglunwen/2370230.html
