合成孔徑雷達（SAR）具備全天候偵照自然環境的條件，衛星 SAR 回波信號模擬架構可模擬雷達目標物回波訊號，並結合成像處理器來 建立偵照目標物影像資料庫，有效提升 SAR 影像偵照目標物辨識的準 確性。基於合成孔徑雷達影像目標物模擬系統需要耗費較長時間進行 全方位目標物回波信號運算，這裡我們建立負載平衡模式(LBM)的平 行處理架構來降低運算所需時間，此架構藉由高速網路結合資訊傳遞 介面(MPI)及多個圖形處理器(GPU)高速運算卡進行分散運算。 LBM 架構分為 MPI 管理機制及 GPU 管理機制兩種層次步驟，MPI 管理機制主要是結合靜態與動態負載平衡的優點，分成「先期評估階 段」、「工作管理階段」及「調整策略階段」等部分。GPU 管理機制 主要是運用 CUDA 程式呼叫 kernel 函數後，交由 GPU 執行，以便運用 大量 GPU 內的執行緒進行平行處理運算工作。經模擬 TerraSAR-X 和 RADARSAT-2 衛星 SAR 回波信號，並針對麥道 MD-80 飛機目標三維 CAD 圖像雷達散射截面（RCS）進行運算，在入射角固定 35 度、水平角度 是-180 度至 180 度，polygon 26500 時，實驗證明運用 LBM 的平行處 理架構演算法，完成雷達目標物截面積運算的時間大約提升為 4 核心 CPU 電腦的 40 倍。;Synthetic Aperture Radar (SAR) is a powerful tool for studying the natural environment under all weather conditions during the whole day. SAR system design and data-processing algorithm simulation is noted for its controllable parameters. The satellite SAR echo signal simulation framework can incorporate simulated radar target echo signals combined with a graphic processor to create a specific target image database,effectively raising SAR image specific target recognition accuracy. In order to solve the satellite SAR echo signal simulation system taking too long in a full blown SAR image simulation for the raw target echo generation, we developed a “Load-Balancing Model (LBM)” algorithm that uses a Message Passing Interface and Graphics Processing Unit (MPI-GPU) with a parallel processing platform to provide an alternative choice for high performance parallel computing. LBM is divided into two command mechanisms: MPI management and GPU management. MPI management is mainly a combination of static and dynamic loads, with three balanced parts, including a pre-evaluation stage, management-work stage and adjust-strategy stage. GPU management includes feedback and calculation stages. The CUDA program is used to call kernel functions executed by the GPU to perform parallel processing and computing tasks with a large amount of data. The LBM algorithm is used to separate the intensive computing and control number of tasks, exploiting the contemporary GPU computation capability to reduce the inner loop load and improve the computing performance. ix Both TerraSAR-X and RADARSAT-2 satellite SAR echo signals for McDonnell Douglas MD-80 aircraft targets with a spatial resolution of 3m in strip map were simulated and used to evaluate LBM performance. The satellite SAR echo signal simulation system exports a three-dimensional CAD model of the target of interest. The CAD model contains numerous grids or polygons, each associated with computed RCS as functions of incident and aspect angles for a given set of radar parameters. We conducted a relevant experiment on a target radar cross section (RCS) and improved its performance by a factor greater than 40, compared with a 4-core CPU used accelerated program.
