【摘要】 通过模拟数据和实际资料处理的结果来看,时间S变换作为一种近似计算具有良好的时频特性,它不仅能提供付里叶变换所不具备的时间分辨率,而且还能克服频率S变换在滤波时的边缘效应.为了提高工作效率,缩短计算周期,本文研究实现了基于GPU的时间S变换,并对炮集记录使用时间S变换来压制面波.实验表明,时间S变换的GPU/CPU协同计算模式具有很好的加速效果,正变换使用全局内存加速比峰值可达一百三十多倍,使用零拷贝内存时达一百八十多倍,反变换使用全局内存达十四倍,而零拷贝内存达一百三十五倍.时间S变换优良的时频特性借助GPU的高性能加速,使得时间S变换在地震资料的时频分析中将有广阔的应用前景.更多还原

【Abstract】 Through the simulated data and real data processing results,the time Stransform as an approximate calculation has good time-frequency features.It not only provide the time resolution that Fourier transform do not have,but also overcome the side effects of frequency Stransform.In order to improve efficiency,shorten the computing cycles.We study the time Stransform based on GPU,and use it to suppress the surface wave on shot gather records.The experiments show that,GPU/CPU collaborative parallel computing model of time Stransform has a perfect acceleration effect.The speed-up ratio of forward transform use global memory reaches more than 130 times at peak time and zero-copy memory reaches 180 times.The speed-up ratio of inverse transform use global memory reaches 14 times,while zero-copy memory reaches 135 times.The good time-frequency features and accelerated by GPU high performance computing makes time Stransform a broad prospect in seismic data processing in time-frequency analysis.更多还原