基于CUDA的宽带水声信号波束形成算法研究

发布时间：2018-09-09 16:39

【摘要】：海洋具有丰富的自然资源和不可替代的战略作用,海洋探测的急切需求促进了水声工程的飞速发展。波束形成具备空域滤波能力,也是目标方位估计的重要手段,是现在水声系统的关键技术,同时也是水声信号处理中计算量最大的部分之一。近年来,水声探测的信号类型从窄带信号向可搭载更多信息的宽带信号过渡,使得波束形成的计算量进一步增大,系统对于信号处理的速度有更高的要求,因此,将基于图形处理器平台的并行处理技术应用于水声阵列信号的加速中将具有很显著的实用价值。本文主要在研究窄带信号的基础上,利用影响主瓣宽度的多个因素,设计了使宽带信号获得恒定束宽的波束形成方法,并在GPU平台上基于统一计算设备架构(Computer Unified Device Architecture,CUDA)对宽带波束形成中的主要模块及算法流程进行并行优化,从而实现了宽带波束形成算法的应用和加速。首先,研究了阵列信号处理和波束形成理论。建立了基阵的数学模型,推导了接收信号模型和阵列流形。以窄带波束形成为基础讨论了波束形成基本理论、波束偏转控制方法和波束优化方法,并以有用信号无失真响应为约束条件,求解了波束形成最佳权矢量。通过仿真分析验证了波束偏转控制和波束优化的有效性以及波束形成最佳权矢量对于干扰方向的抑制效果。接着,讨论了常规宽带波束形成的结构和特征。仿真发现宽带信号不同频率分量经过同一基阵会有不用的响应,导致波束形成结果时域失真。为解决这一问题,分析了波束优化方法中控制旁瓣级和主瓣宽度的方法,研究了一种基于DolphChebyshev加权的恒定束宽波束形成方法。同时通过分析频率和阵列之间的关系,设计了一种利用高低子阵嵌套的方式对倍频程信号进行恒定束宽波束形成的方法。仿真验证了两种恒定束宽波束形成方法在主瓣区域主瓣方向上对于不同频率具备相同响应,因此通过基阵的信号在时域上不再发生畸变。然后,基于CUDA在GPU上实现了波束形成算法的并行加速。在阐述了CUDA的架构和编程模型之后,将宽带波束形成算法分解为基础计算函数模块,并把最核心也是计算量最大的矩阵乘法和DFT计算根据CUDA特征进行的并行优化最后,通过外场实验验证了波束形成的空域滤波性能和方位估计准确性,并基于各个并行计算模块完善了波束形成并行程序,实现了波束形成算法的并行加速。
[Abstract]:The ocean has abundant natural resources and irreplaceable strategic role. The urgent need of ocean exploration promotes the rapid development of underwater acoustic engineering. Beamforming has the ability of spatial filtering and is an important means of target azimuth estimation. Beamforming is the key technology of underwater acoustic system and one of the most computational parts in underwater acoustic signal processing. In recent years, the signal type of underwater acoustic detection has changed from narrowband signal to wideband signal with more information, which makes the computation of beamforming increase further, and the system requires higher speed of signal processing. The application of parallel processing technology based on graphics processor platform to the acceleration of underwater acoustic array signals will be of great practical value. Based on the study of narrowband signals, a beamforming method is designed to obtain constant beamwidth for wideband signals by using several factors affecting the width of the main lobe. The main modules and algorithm flow of broadband beamforming are optimized in parallel on GPU platform based on unified computing equipment architecture (Computer Unified Device Architecture,CUDA), so that the application and acceleration of wideband beamforming algorithm are realized. Firstly, array signal processing and beamforming theory are studied. The mathematical model of the array is established, and the received signal model and array manifold are derived. On the basis of narrowband beamforming, the basic theory of beamforming, beam deflection control method and beam optimization method are discussed. The optimal weight vector of beamforming is solved under the constraint condition of non-distortion response of useful signal. The effectiveness of beam deflection control and beam optimization is verified by simulation analysis. Then, the structure and characteristics of conventional broadband beamforming are discussed. The simulation results show that different frequency components of wideband signals will have unnecessary response through the same array, which leads to the distortion of beamforming results in time domain. In order to solve this problem, the method of controlling sidelobe level and main lobe width in beam optimization method is analyzed, and a constant beamwidth beamforming method based on DolphChebyshev weighting is studied. At the same time, by analyzing the relationship between frequency and array, a method of constant beamwidth beamforming using high and low subarray nesting method is designed. Simulation results show that the two constant beamwidth beamforming methods have the same response to different frequencies in the direction of the main lobe in the main lobe region, so the signal through the matrix is no longer distorted in time domain. Then, the parallel acceleration of beamforming algorithm is realized on GPU based on CUDA. After expounding the architecture and programming model of CUDA, the wideband beamforming algorithm is decomposed into basic computing function module, and the core is the matrix multiplication with the largest amount of computation and the parallel optimization of DFT computing according to the characteristics of CUDA. The spatial filtering performance and azimuth estimation accuracy of beamforming are verified by field experiments. The parallel program of beamforming is perfected based on each parallel computing module, and the parallel acceleration of beamforming algorithm is realized.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN929.3

【相似文献】