摘要
为了解决传声器阵列用于声场分析时的精确且快速声成像的技术难题,提出了一种利用图像处理器(Graphics Processing Unit,GPU)的通用计算技术实现快速声成像算法。可控响应功率算法是广泛应用的一种声源定位算法,但计算量巨大,阻碍了它在实际场合中的应用。通过将可控响应功率算法进行任务分解及线程映射,实现了利用计算统一设备架构实现的基于GPU的可控响应功率声源成像定位算法,在特定阵元通道数和信号长度情况下,与基于CPU的声源定位计算方法相比,综合计算效率提高了约20倍,并将其成功地应用于平面螺旋阵的声成像应用中,实现了实时声成像定位。
To implement an accurate and rapid acoustic imaging method, an algorithm based on general-purpose computing technology of GPU is used to make fast acoustic imaging calculation, which is a key technical issue for sound field analysis of microphone array. One well-known steered response power(SRP) method is widely used in sound source localization, but due to high computational expense its practical application to real-time processing is limited. The SRP algorithm is implemented in real time based on CUDA(Compute Unified Device Architecture) frame using GPU task decomposition and thread mapping, which gains almost 20 times total efficiency improvement compared to CPU based SRP sound source localization algorithm under condition of certain number of channel and length of signal. And, it has been successfully applied to the acoustic imaging of a plane spiral array to achieve real-time image positioning.
出处
《声学技术》
CSCD
2014年第4期367-371,共5页
Technical Acoustics
基金
国家自然科学基金面上项目(11074278)
国家自然科学基金面上项目(11174320)
关键词
波束形成
可控响应功率
图像处理器
计算统一设备架构
并行计算
beamforming
steered response power
graphics processing unit
compute unified device architecture
par-allel computing