【作者】 苏理云；

【导师】 马洪；

【作者基本信息】 四川大学 ， 概率论与数理统计， 2007， 博士

【摘要】 随着通信、计算机、电子学科的不断发展,其工程应用日趋广泛和深入,所采用的数学方法也随之不断地丰富,在大量的工程应用中由于信息处理的需求,软计算理论和方法得到大量的应用,其自身理论也在不断的发展,软计算与其它方法的结合应用也在不断出现,使之在信息处理应用中显现出极大的重要性,本文主要对软计算方法(包括:小波理论、人工神经网络、分形理论、混沌理论、模糊数学等)在信息处理中的应用进行了深入的研究,探讨了软计算理论及其它相关数学理论的融合进行信息处理的方法,并进行计算机仿真实验,仿真实验结果表明,该文提出的几种方法是有效的。本文详细而具体地讨论和研究了小波理论、人工神经网络、分形理论、混沌理论、模糊数学在超宽带通信和微弱信号检测的应用研究。综合运用这些技术以构建全新的超宽带多用户检测、微弱谐波信号检测方法和多传感器多目标跟踪系统。首先,在通信信号处理领域,对超宽带通信中的多用户检测问题进行了深入分析,提出了两种多用户检测算法。在超宽带通信领域,提出了基于隐马尔可夫模型和Kalman滤波的多用户检测算法。算法1:该算法将跳时二进制相移键控超宽带通信系统建模为隐马尔可夫系统模型。考虑到发射序列状态之间可能存在时序的非继承性,而搜索与前向序列对应的具有最大转移概率的后续序列,提出了最大后验概率(MAP)多用户检测算法,模拟结果表明了该算法的有效性。算法2:该算法将跳时二进制相移键控超宽带通信系统建模为动态系统模型,然后利用Kalman滤波算法,设计了一种新的超宽带盲自适应多用户检测算法。数值模拟结果表明,该方法比传统的检测器有更低的误码率,说明该算法是一种有效的检测算法。其次,对微弱谐波信号难于检测的问题,针对强混沌背景和强分形噪声背景干扰下的微弱谐波信号检测问题,提出了相应的解决方案。对强混沌背景噪声干扰下的微弱谐波信号检测问题。利用RBF神经网络预测混沌时间序列的能力建立预测模型,用模糊聚类算法来获得RBF神经网络的径向中心,将观测信号减去预测出的混沌信号,得到误差信号,而误差信号中可能含有有用微弱信号及普通的白噪声。然后把误差信号送入Duffing混沌振子,充分利用Duffing混沌振子对噪声的免疫性,来检测是否含有微弱信号。该方法充分利用了模糊聚类,神经网络,混沌时间序列的短期可预测性,从而大大提高检测微弱信号的性能。仿真实验结果表明:该方法能够在信噪比-120dB时检测出微弱谐波信号。而对强分形噪声干扰下的微弱谐波信号检测问题。本文提出了小波域多尺度模糊自适应Kalman滤波和Duffing混沌振子相结合的方法来解决强分形背景干扰下的谐波信号检测问题。先对淹没在强分形随机背景干扰中的观测信号进行多尺度小波变换,根据分形随机信号小波系数的平稳性,建立状态方程和观测方程,利用Kalman滤波,对每一尺度进行进行滤波最后估计出分形随机信号,并对噪声方差未知时,利用模糊推理系统来动态逼近,然后将估计信号与观测信号作差得到误差信号,把误差信号送入Duffing混沌振子,充分利用Duffing混沌振子对噪声的免疫性,来检测是否含有微弱信号。仿真实验结果表明:该方法能在低信噪比和低信干比下有效地检测出淹没在强分形噪声中的微弱谐波信号。同时,文中也给出了Duffing混沌振子的免疫性一种新的统计解释。最后,对多传感器多目标跟踪问题提出了基于模糊聚类和JPDA的多传感器多目标跟踪算法。传统的联合概率数据关联算法(JPDA)是适合于密集杂波条件下的一种良好的多目标跟踪算法,但它是针对单传感器对多目标跟踪的情况下使用的,不能直接用于多传感器多目标的跟踪,并且随着目标数和回波数的增加,其运算量出现了组合爆炸现象,为此提出了一种新简化的JPDA算法,该算法适用于多传感器多目标跟踪的情况。该算法首先利用极大似然估计对多传感器的测量集合进行同源划分,然后对来自同一目标的测量进行融合,然后采用改进的模糊C-均值(FCM)聚类算法计算关联概率,采用JPDA算法的类似结构对目标航迹进行更新和修正。仿真实验结果说明了该方法能有效地进行多传感器多目标的跟踪,并且算法简单、跟踪精度高、计算量小、易于工程实现。更多还原

【Abstract】 With the development of communication, computer and electronics, the en-gineering applications have become more widely and more deeply, the number ofmathematical methods increases rapidly in abundant engineering applications. Be-cause of the need of information processing, soft computing methods and othermathematical methods are used widely. The theory of soft computing developedrapidly, and some other relative theories appear constantly, which increases its im-portance in the application of information processing. This thesis investigates theuse of soft computing methods (including: wavelet theory, artificial neural network,fractional theory, chaos theory and fuzzy mathematics, etc) in information process-ing, reveals the problems in engineering applications of soft computing methods andcombines different mathematics methods to realize information processing.This thesis investigates and discusses the use of soft computing methods inultra-wide band(UWB) communications, weak signal detection and multi-sensormulti-target tracking. These techniques are used to construct novel ultra-wide bandcommunication multi-user detectors, weak signal detection systems and multi-senormulti-target tracking system.Firstly, in digital communication signal processing fields, two multi-user de-tection methods are proposed. For impulse radio UWB systems based on the proba-bility and statistics, the first novel multi-user detection is presented. HMM systemsof time-hopping BPSK in UWB are developed separately in the paper. Consideringseeking for the following sequence of maximal transition probability correspond-ing to previous sequence, a maximum a posteriori (MAP) multi-user detector hasbeen constructed. The second novel blind multi-user adaptive detection algorithm ispresented. Dynamical systems of time-hopping BPSK in UWB are developed sepa-rately in the paper. multi-user detector has been constructed using the recently pro-posed canonical representation of multi-user receivers and Kalman filtering. The-oretical analysis and numerical results show that the new scheme has better BERperformance comparing with the conventional detector.Secondly, in digital signal processing, for strong chaotic noise and fractionalnoise, two weak signal detection schemes are investigated respectively. A novel approach of weak sinusoid signal detection from a chaotic noise background is pro-posed. The RBF neural networks are used to predict the chaotic background, and theradial center is gained by fuzzy c-means algorithm. The noisy error signal extractedfrom the detected signal, then the error signal is added into the Duffing chaotic os-cillator using the sensitivities to the initial conditions and immunity to noise of thesystem. Based on the motion transition of a chaotic system, new schemes to de-tect weak sinusoidal signal berried in a chaotic background are presented forward.Simulation measured data demonstrate the effectiveness of the proposed algorithm.Weak signal detection of strong fractional Brownian motion (fBm)using Duffingoscillator and fuzzy adaptive Kalman filter is proposed in this paper. A dynamicsystems based on the orthonormal wavelet decomposition coefficients of the fBm isformulated, fuzzy adaptive Kalman filter is used to estimate the fBm when the vari-ance of measurement noise is unknown. The noisy error signal extracted from thedetected signal, then the error signal is added into Duffing chaotic oscillator usingthe sensitivities to the initial conditions and immunity to noise of the system. Basedon the motion transition of a chaotic system and the estimation of fuzzy adaptiveKalman filtering, new schemes to detect weak sinusoidal signal berried in a strongfractional background noise are presented forward. Simulation results demonstratethe effectiveness of the proposed algorithm.A new statistical understanding of immunity to noise of Duffing systems is alsopresented.Finally, the Joint Probabilistic Data Association (JPDA) solves single sensormulti-target tracking in clutter, but it can not be used directly in multi-sensor multi-target tracking (MMT) and has high computational complexity with the numbers oftargets and the number of returns. The novel algorithm was presented by combiningMaximum Likelihood Estimation (MLE) with Fuzzy C-Means (FCM) clustering.The MLE is used to classify the same source observations at one time into the sameset, then the FCM approach is used to calculate the data association probability, andthe similarity structure of JPDA algorithm is used to realize the MMT. The computersimulations indicate that the scheme achieve MMT perfectly with low computationalcomplexity, higher precision and easy realization.更多还原