盲源信号分离算法研究及应用

发布时间：2018-09-11 09:25

【摘要】：盲源信号分离(Blind Source/Signal Separation,BSS)是指在未知源信号和传输通道参数的情况下,根据输入源信号的统计特性,从观测信号恢复出源信号各个分量的过程,是目前信号处理中最热门的研究方向之一。由于盲源信号分离技术对混合源和传输通道的先验信息要求不高,能够仅从接收端就能完成信号的估计和恢复,具有其它信号处理技术不能比拟的优势,目前,盲源信号分离技术以其突出的技术优势已经在众多领域得到应用和发展,尤其在生物医学、电子对抗、语音增强、遥感、地震探测、通信系统、地球物理学、计量经济学、机械力学等领域发挥了重要作用。但是,盲源信号分离技术在与实际问题结合中也暴露了分离性能较差、计算复杂度较高、应用条件受限等缺点。研究有效的盲源信号分离理论,提高分离性能、降低计算复杂度、减少先验信息的约束条件是现代通信系统应用的迫切需求,因此,本文开展了盲源信号分离算法及应用方面的研究。本文结合被动雷达系统、跳频信号和邻星干扰等实际应用背景,从理论和实践角度研究盲源信号分离处理算法,力求达到提高通信系统频谱效率、增强通信系统抗干扰和信号检测性能的目的。本文主要研究工作包括被动雷达系统中强干扰弱信号的分离、正交跳频体制下欠定盲源信号的分离、非正交跳频体制下欠定盲源信号的分离和邻星干扰中盲源信号分离四个方面的内容,具体研究内容如下:针对强干扰条件下被动雷达系统中弱信号盲源分离问题,提出了干扰抵消算法(Interference Cancellation Algorithm,IC-Algorithm),达到消除强干扰、提高弱信号检测能力的目的。具体地,本文把强干扰信号分为合作信号和非合作信号两种情况分别开展研究。第一种情况是强干扰信号为合作信号,在强干扰条件下,需要对强干扰信号进行估计和重建,估计和重建的精度直接影响盲源信号分离的效果。根据重建后的强干扰信号提出了干扰抵消算法(IC-Algorithm)消除强干扰信号。由于获取的是弱目标混合信号,本文接着提出了KM-FastICA算法进行弱目标混合信号的分离。干扰抵消后残余信号的强弱对源信号的分离效果也产生很大的影响,本文从信息论的角度分析了残余信号对源混合信号分离的影响。第二种情况是强干扰信号为非合作信号,信号参数未知,不能直接应用本文所提出的干扰抵消算法消除强干扰信号。针对该特殊场景,本文提出了先分离再抵消的方法,降低了对强干扰先验信息的要求,具有更广阔的应用前景。针对正交(指源混合信号两两内积为零)跳频体制下的欠定盲源信号分离问题,提出了基于密度聚类的盲源分离算法(DCBS-algorithm),达到提高盲源信号分离性能、优化使用频谱资源的目的。本文提出的密度聚类盲分离算法(DCBS-algorithm)分为两步:第一步,借助跳频信号的稀疏性,采用短时傅里叶变换(STFT)获得采样信号的时频域信息,根据采样信号的时频域信息构建了代价函数对(?,?)和决策坐标系统,然后利用代价函数对(?,?)对采样信号的时频域值进行密度聚类,找到聚类中心;第二步,根据聚类中心完成采样信号的分类,利用短时傅里叶变换的逆变换完成信号的恢复,实现盲源信号分离。所提的密度聚类盲分离算法(DCBS-algorithm)解决了正交跳频体制下欠定条件盲源信号分离问题,在较低计算复杂度的前提下提升了分离性能。针对非正交(指源混合信号两两之间内积不为零)跳频体制下的盲源信号分离问题,提出了匹配优化盲分离(Matching Optimization Blind Separation,MOBS)算法,达到优化盲源信号分离算法、提高频谱利用效率的目的。将信号分为两类:一类是未发生碰撞的采样信号,对于该类采样信号提出了密度聚类盲分离算法(DCBS-algorithm);另一类为发生碰撞的采样信号,由于该类信号的采样点为多个信号之和,不再满足稀疏性,无法使用聚类的方法完成盲源信号分离。因此,提出了匹配优化盲分离(MOBS)算法,根据信号采样特征提出了代价函数,基于最陡下降法构建了代价函数,实现了盲源信号分离。该算法解决了非正交跳频体制下的盲源信号分离问题,进一步扩展了盲源信号分离算法的应用范围。针对现代卫星通信邻星干扰中存在的盲源信号分离问题,提出了基于粒子群优化的盲源信号分离算法,提高星上处理能力和抗干扰能力。该算法包含三个步骤:首先,通过计算每一个采样点的短时傅里叶变换,获得样本信号的时频域信息;其次,应用K-means聚类算法对样本信号进行预处理,得到更好的分离性能和较低的计算复杂度;最后,根据邻星干扰的特征定义了迭代参数,提出了基于粒子群优化的盲源信号分离方法。该算法具有较好的收敛性和鲁棒性,增强了星上处理能力和抗干扰能力。
[Abstract]:Blind Source/Signal Separation (BSS) refers to the process of recovering each component of the source signal from the observed signal according to the statistical characteristics of the input signal under the condition of unknown source signal and transmission channel parameters. Blind Source/Signal Separation (BSS) is one of the most popular research directions in signal processing. The prior information of source and transmission channel is not high, and it can estimate and restore the signal only from the receiver. It has the advantage that other signal processing technology can not compare. At present, blind source signal separation technology with its outstanding technical advantages has been applied and developed in many fields, especially in biomedicine, electronic countermeasures, and voice. Enhancement, remote sensing, seismic detection, communication systems, geophysics, econometrics, mechanical mechanics and other fields have played an important role. However, blind source signal separation technology in combination with practical problems has also exposed the disadvantages of poor separation performance, high computational complexity and limited application conditions. High separation performance, reduced computational complexity and reduced constraints of prior information are the urgent needs of modern communication systems. Therefore, this paper studies the blind source separation algorithm and its application. Source signal separation and processing algorithms strive to improve the spectrum efficiency of communication systems, enhance the anti-jamming and signal detection performance of communication systems. This paper mainly studies the separation of strong and weak jamming signals in passive radar systems, the separation of underdetermined blind source signals in orthogonal frequency hopping systems, and the underdetermined blind source signals in non-orthogonal frequency hopping systems. Four aspects of separation and blind source separation in adjacent satellite jamming are studied as follows: Aiming at the problem of weak signal blind source separation in passive radar system with strong jamming, an interference Cancellation Algorithm (IC-Algorithm) is proposed to eliminate strong jamming and improve the ability of weak signal detection. Specifically, this paper divides the strong interference signal into cooperative signal and non-cooperative signal. The first case is that the strong interference signal is cooperative signal. Under the strong interference condition, it is necessary to estimate and reconstruct the strong interference signal. The accuracy of estimation and reconstruction directly affects the effect of blind source signal separation. After the strong interference signal, the IC-Algorithm is proposed to eliminate the strong interference signal. Because the weak target mixed signal is obtained, the KM-FastICA algorithm is proposed to separate the weak target mixed signal. The residual signal after interference cancellation also has a great influence on the separation effect of the source signal. This paper presents a new method to separate the weak target mixed signal. From the angle of information theory, the influence of residual signal on the separation of mixed source signals is analyzed. The second case is that the strong interference signal is a non-cooperative signal and the signal parameters are unknown. A density clustering based blind source separation algorithm (DCBS-algorithm) is proposed to improve the performance of blind source separation and optimize the use of spectrum resources in orthogonal frequency hopping system. The density clustering blind separation algorithm (DCBS-algorithm) proposed in this paper is divided into two steps. The first step is to obtain the time-frequency domain information of the sampled signal by means of the sparsity of the frequency hopping signal. The cost function pair (???) and the decision coordinate system are constructed according to the time-frequency domain information of the sampled signal. The second step is to classify the sampled signal according to the clustering center and restore the signal using the inverse transform of short-time Fourier transform to realize blind source separation. Conditional Blind Source Separation (CBS) improves the separation performance with lower computational complexity. Matching Optimization Blind Separation (MOBS) algorithm is proposed to optimize the Blind Source Signal Separation (BSS) in non-orthogonal frequency hopping (FH) systems. The signal is divided into two types: one is the sample signal without collision, and the other is the sample signal with collision, which can no longer satisfy the sparse requirement. Therefore, a matching optimization blind source separation (MOBS) algorithm is proposed. According to the characteristics of signal sampling, a cost function is proposed. Based on the steepest descent method, a cost function is constructed to realize blind source separation. In order to solve the problem of Blind Source Separation in the neighboring satellite jamming of modern satellite communication, a Blind Source Separation algorithm based on particle swarm optimization is proposed to improve the on-board processing ability and anti-jamming ability. The algorithm consists of three steps: firstly, the short of each sampling point is calculated. Secondly, K-means clustering algorithm is used to preprocess the sample signal to obtain better separation performance and lower computational complexity. Finally, iteration parameters are defined according to the characteristics of neighboring star interference, and a blind source separation method based on particle swarm optimization is proposed. It has good convergence and robustness, and enhances the processing capability and anti-jamming capability on the satellite.
【学位授予单位】：电子科技大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TN911.7

【参考文献】