论文部分内容阅读
在医学超声成像算法中,最经典和最广泛使用的是延迟叠加算法。延迟叠加算法在超声成像、雷达信号发射、接收以及天线信号波束形成等方面有着广泛的应用。虽然该算法并不是典型的性能需求型算法,但是在医学成像云计算服务的新需求下,需要提高该算法的计算速度,以克服云计算中网络传输速度相对较慢的约束。然而,以中央处理器作为主要计算资源的传统云计算框架无法满足医学超声图像快速生成的性能需求,因此,本文中使用以现场可编程逻辑门阵列作为异构加速资源的Super Vessel云平台作为并行延迟叠加算法的实现平台。当包括现场可编程逻辑门阵列和中央处理器之间的数据传输时间在内时,Super Vessel云平台上该算法异构实现的运行速度相较于中央处理器中该算法的运行速度提升了约22倍。
In the medical ultrasound imaging algorithm, the most classic and most widely used is the delay overlay algorithm. Delay overlay algorithm in ultrasound imaging, radar signal transmission, receiving and antenna beamforming has a wide range of applications. Although this algorithm is not a typical performance requirement algorithm, the computational speed of this algorithm needs to be improved under the new requirement of medical imaging cloud computing service in order to overcome the constraint that the network transmission speed is relatively slow in cloud computing. However, the traditional cloud computing framework with CPU as the main computing resource can not meet the performance requirements of medical ultrasound image generation. Therefore, in this paper, the Super Vessel cloud platform using field programmable gate array as heterogeneous acceleration resources is used as parallel Delay stack algorithm platform. When including data transfer time between field programmable gate array and CPU, the heterogeneous implementation of this algorithm on SuperFloral cloud platform can achieve a speedup of about 40% faster than that of the CPU in the central processor 22 times.