基于粒子滤波的同频数字混合信号单通道盲分离技术研究

发布时间：2018-06-15 08:20

本文选题：同频数字混合信号盲分离 + 粒子滤波　；参考：《解放军信息工程大学》2014年硕士论文

【摘要】：随着通信需求的迅猛增长以及大量复杂通信技术的使用,电磁空间变得复杂和拥挤,第三方非合作接收中出现了同频数字混合信号,给信号分析和信息获取工作带来了一定困难。粒子滤波盲分离算法是解决该类信号盲分离的有效途径之一。本文主要围绕基于粒子滤波的同频数字混合信号单通道盲分离问题展开研究,旨在改进已有算法中所存在的缺点和不足。主要内容如下:1.首先从贝叶斯信号处理角度给出了传统滤波问题的数学描述,对已有的滤波算法进行了简要介绍,引入了解决序贯蒙特卡洛信号处理问题的粒子滤波方法,对其基本原理及优缺点进行了详细分析。其次,建立了单通道接收条件下两路同频数字混合信号的基带模型,从贝叶斯估计的角度,给出了同频数字混合信号盲分离的数学描述,为后续讨论奠定了理论基础。2.针对调制参数非时变情况下,粒子滤波盲分离算法存在参数估计精度低、收敛速度慢等问题,提出了一种改进的粒子滤波盲分离算法。从优化抽样分布的角度出发,将参数粒子的抽样分布建模为Beta分布,有效的提高了抽样效率,改善了算法的参数估计性能和分离性能。为了检验算法的参数估计性能,推导了符号已知条件下的非时变参数联合估计克拉美罗界。3.针对调制参数时变情况下混合信号的盲分离问题,提出了一种时变信道下基于粒子滤波的盲分离算法。通过将时变调制参数建模为一阶AR模型,选取先验分布作为参数粒子的抽样分布,将传统粒子滤波盲分离算法扩展至调制参数时变情况下的盲分离。为了检验算法的参数联合估计性能,推导了符号已知条件下的时变参数联合估计后验克拉美罗界,并给出了性能界的数值计算方法。4.针对粒子滤波盲分离算法计算复杂度高的问题,提出了一种基于部分采样的低复杂度粒子滤波盲分离算法。首先详细分析了算法的计算复杂度,指出算法计算量大的原因:符号粒子采样过程的计算复杂度与平滑长度成指数倍关系。其次,注意到符号粒子采样公式的数值计算过程与经典Viterbi译码算法类似,借鉴M-算法的思想,在搜索分支路径过程中保留部分分支路径,将算法的计算复杂度变为与平滑长度成多项式关系,例如当平滑长度等于4时,部分采样法近似是传统采样法的二十分之一。最后给出了算法正确性的理论证明。5.针对部分采样法在高信噪比条件下性能损失较大的问题,提出了一种基于混合采样法的粒子滤波盲分离算法。首先分析了高信噪比条件下部分采样法性能损失较大的原因。其次,结合传统采样法和部分采样法的优势,提出将平滑区间划分为两部分,各自区间内依据观测值与待抽样符号粒子相关性的大小分别选择传统采样法和部分采样法进行遍历,有效的实现了计算复杂度与性能之间的折中考虑,并给出了算法正确性的理论证明。
[Abstract]:With the rapid growth of communication demand and the use of a large number of complex communication technologies, the electromagnetic space becomes complex and crowded. It brings some difficulties to signal analysis and information acquisition. Particle filter blind separation algorithm is one of the effective ways to solve this kind of signal blind separation. This paper focuses on the single-channel blind separation of the same frequency digital mixed signals based on particle filter, aiming to improve the shortcomings and shortcomings of the existing algorithms. The main content is as follows: 1. Firstly, the mathematical description of the traditional filtering problem is given from the point of view of Bayesian signal processing, the existing filtering algorithms are briefly introduced, and the particle filter method is introduced to solve the sequential Monte Carlo signal processing problem. The basic principle, advantages and disadvantages are analyzed in detail. Secondly, the baseband model of two-channel digital mixed signals with same frequency under the condition of single channel reception is established. From the point of view of Bayesian estimation, the mathematical description of blind separation of the same frequency digital mixed signals is given, which lays a theoretical foundation for further discussion. A modified particle filter blind separation algorithm is proposed to solve the problems of low parameter estimation accuracy and slow convergence rate in the case of time-invariant modulation parameters. From the point of view of optimizing sampling distribution, the sampling distribution of parameter particles is modeled as Beta distribution, which effectively improves the sampling efficiency and improves the performance of parameter estimation and separation of the algorithm. In order to test the parameter estimation performance of the algorithm, the joint estimator of time-invariant parameters under the condition of known symbols is derived. To solve the problem of blind separation of mixed signals with time-varying modulation parameters, a particle filter based blind separation algorithm in time-varying channels is proposed. By modeling the time-varying modulation parameters as a first-order AR model and selecting the prior distribution as the sampling distribution of the parameter particles, the traditional particle filter blind separation algorithm is extended to the blind separation with time-varying modulation parameters. In order to test the joint parameter estimation performance of the algorithm, the posterior Crameiro bound for joint estimation of time-varying parameters is derived under the condition of known symbols, and the numerical calculation method of the performance bound is given. Aiming at the high computational complexity of particle filter blind separation algorithm, a low complexity particle filter blind separation algorithm based on partial sampling is proposed. Firstly, the computational complexity of the algorithm is analyzed in detail, and the reason for the large computational complexity of the algorithm is pointed out: the computational complexity of the symbolic particle sampling process is exponentially related to the smooth length. Secondly, it is noted that the numerical calculation process of symbolic particle sampling formula is similar to that of classical Viterbi decoding algorithm. By using the idea of M- algorithm, some branch paths are preserved in the process of searching branch paths. The computational complexity of the algorithm is changed to a polynomial relation with the smooth length. For example, when the smoothing length is equal to 4, the partial sampling method is approximately 1/20 of the traditional sampling method. Finally, the theoretical proof of the correctness of the algorithm is given. A particle filter blind separation algorithm based on mixed sampling method is proposed to solve the problem that the performance of partial sampling method loses greatly under the condition of high signal-to-noise ratio (SNR). Firstly, the reason for the performance loss of the partial sampling method under the condition of high SNR is analyzed. Secondly, combining the advantages of traditional sampling method and partial sampling method, the smooth interval is divided into two parts. The traditional sampling method and partial sampling method are selected to traverse according to the magnitude of correlation between observed values and the symbol particles to be sampled in each interval. The trade-off between computational complexity and performance is realized effectively. The correctness of the algorithm is proved theoretically.
【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】：2014
【分类号】：TN911.7

【参考文献】