MUSIC算法及其并行实现

发布时间：2018-03-29 13:52

本文选题：多重信号分类算法　切入点：波达方向估计　出处：《西安电子科技大学》2014年硕士论文

【摘要】：如今,在阵列信号处理的整个领域中,波达方向(Direction of Arrival,DOA)估计已经成为其中一个非常重要的问题,它在雷达、无线通讯等非常多的领域都有着及其广泛的应用。波达方向估计是用来确定辐射源辐射方向的一种技术,它能通过阵列接收到的信息,然后按照一定的预处理方法来估计出信源的方位。能实现波达方向估计的算法有很多,其中以数字波束形成为代表的传统算法有个明显的缺陷就是受到瑞利限的制约,即估计精度低、不能分辨出一个波束宽度中的几个信源,而后来出现的以多重信号分类MUSIC算法为代表的超分辨率算法解决了这个问题,这类算法不再受到瑞利限的限制。本文从经典的MUSIC算法出发,分析这个算法在波达方向估计上的应用,然后在不影响信源方位角估计精度的基础上,使用并行处理技术提高其运行计算速度,让该算法在对实时性有重要要求的领域也能得到运用。本文首先介绍了MUSIC算法的有关基本理论,通过实验详细分析了该算法在波达方向估计问题上的处理能力,分析信噪比、阵元间距等等对MUSIC算法在DOA估计过程的影响。鉴于阵列接收到信号后进行相关处理得到的协方差矩阵是Hermite矩阵,我们针对这一矩阵特点提出一种从复数域转到实数域的预处理方法,使待处理的Hermite矩阵(复矩阵)转化成对应的实矩阵,大大节省了处理需要的时间,而且让算法也更适合在硬件DSP上实现。同时由于期间还进行一项前后向平滑处理,因此这种实数化方法不仅针对非相干信号源(本文主要考虑的情况)的DOA估计有效,对相干信号源DOA估计同样有效。从耗时性、硬件适用性、并行性等多个方面进行考虑,本文针对MUSIC算法的核心部分——特征空间分解,在众多常用特征分解算法里选择了QL算法,然后又对MUSIC算法各个模块的串行实现进行详细分析。最后,通过实验证明了本文所提出的并行方法是可行的。又通过实验数据验证了在不影响算法波达方向估计性能的情况下,耗时量得到明显的减少。
[Abstract]:Today, in the whole field of array signal processing, the DOA Direction of Arrivalo estimation has become one of the most important issues in radar. There are many applications in many fields, such as wireless communication. DOA estimation is a technique used to determine the radiation direction of emitter. It can receive information through the array. There are many algorithms to realize DOA estimation, among which the traditional algorithm represented by digital beamforming has an obvious defect, which is restricted by Rayleigh limit. That is, the estimation accuracy is too low to distinguish several sources in a beamwidth, but the super-resolution algorithm, represented by MUSIC algorithm for multi-signal classification, has solved this problem. This kind of algorithm is no longer restricted by Rayleigh limit. Based on the classical MUSIC algorithm, this paper analyzes the application of this algorithm in DOA estimation, and then on the basis of not affecting the accuracy of source azimuth estimation, The parallel processing technology is used to improve the computing speed, so that the algorithm can also be used in the realtime field. This paper first introduces the basic theory of MUSIC algorithm. The processing ability of the algorithm in DOA estimation is analyzed in detail, and the signal-to-noise ratio (SNR) is analyzed. The effect of array spacing on the DOA estimation process of MUSIC algorithm. Since the covariance matrix obtained by correlation processing after the array receives the signal, the covariance matrix is the Hermite matrix. In view of the characteristics of this matrix, we propose a preprocessing method from complex field to real number field, which can transform the Hermite matrix (complex matrix) to the corresponding real matrix, and save the processing time greatly. This method is not only effective for the DOA estimation of incoherent signal sources (mainly considered in this paper), but also suitable for the implementation of hardware DSP. The DOA estimation of coherent signal source is also effective. Considering the time consuming, hardware applicability, parallelism and so on, this paper selects QL algorithm among many commonly used feature decomposition algorithms, aiming at the feature space decomposition, which is the core part of MUSIC algorithm. Then the serial implementation of each module of MUSIC algorithm is analyzed in detail. Finally, the experiment proves that the parallel method proposed in this paper is feasible, and the experimental data verify that the performance of DOA estimation is not affected by the proposed method. The amount of time taken is significantly reduced.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.7

【相似文献】