Parallel Computation of Doppler Spectral from Dynamic Sea Surfaces at Microwave Bands

摘要

The Doppler spectrum of the electromagnetic (EM) scattering field from the two-dimensional dynamic sea surface is calculated based on the composite scattering model. The two-dimensional dynamic sea surfaces are generally simulated as a superposition of large-scale gravity waves and small-scale capillary ripples. On this basis, the Doppler spectrum of the EM scattering field from the two-dimensional dynamic sea surface can be calculated based on the composite scattering model, which takes both the quasi-specular scattering and Bragg scattering mechanism into account. However, due to the high resolution and real-time dynamic complexity of the dynamic sea surfaces, the calculation of the Doppler spectrum will be computationally expensive and very time-consuming. In this paper, a GPU-based algorithm of Doppler spectrum was proposed by utilizing the Tesla K80 GPUs with diverse CUDA optimization techniques. The GPU-based Doppler spectrum implementation includes five optimization strategies: first, the temporary arrays are utilized to reduce the repeat float-points operations in the loop; then the device memory was effectively exploited to reduce the data transfer time between the CPU and GPU; the fast math compiler option was also utilized to further improve the computation performance of the Doppler spectrum calculation; finally the data transfer time between the device and host memories can be effectively hide by using the asynchronous data transfer (ADT). Compared to the CPU serial program executed on Intel(R) Core(TM) i5-3450 CPU, the GPU-based Doppler spectrum implementation can achieve a significant speedup of1200x.