基于GPU并行运算的高效雷达成像谱估计算法

Spectral Estimation Acceleration for SAR Imaging Based on the GPU Parallel Computing

Apes和Capon等谱估计算法在SAR成像方面有着广泛的应用。同基于快速傅里叶变换的成像算法相比,谱估计成像算法能够获得更窄的谱峰和更低的旁瓣,但是计算量庞大。本算法基于图形处理器(Graphic Processing Unit,简称GPU)并行计算原理,在Jacket平台上实现了以上两种算法在雷达超分辨成像上的加速。在NVIDIA Tesla C2050和Intel(R)Xeon(R)CPU X5680上的测试表明,与传统基于CPU的SAR成像算法相比,本算法能够使计算速度得到数倍的提升。

Spectral estimation algorithms such as Apes and Capon have been widely used in SAR imaging, which can obtain complex spectral estimation with more narrow spectral peaks and lower sidelobes compared with FFT methods. The major disadvantage is the huge amount of computing which takes too long time. Based on the principle of GPU parallel computing, this paper proposes a technique to achieve the acceleration of the above two algorithms of radar imaging on the platform of Jacket. Tests on NVIDIA Tesla C2050 and Intel （R） Xeon （R） CPU X5680 showed that the GPU-based program can reach several times speed than the traditional algorithms of SAR imaging.