对于空时自适应信号处理(Space-Time Adaptive Processing, STAP)算法的并行处理问题，传统方法以粗粒度的划分方式将STAP算法分配到特定硬件系统中的不同处理器中，利用处理器间的流水计算来提高系统计算吞吐量。该文分析了传统并行处理方法的缺陷：粗粒度的任务划分方式牺牲了STAP算法的并行度；传统处理方法仅能适用于特定的系统环境。针对上述情况，该文提出一种基于细粒度任务分配的STAP并行处理方法，该方法分为以下3个步骤：构建细粒度的DAG(Direct Acyclic Graph)形式的STAP算法任务模型；使用统一拓扑结构模型描述不同结构的目标硬件系统；基于细粒度任务分配算法将任务模型分配到拓扑结构模型中的处理器实现并行计算。实验结果表明该并行处理方法能够达到良好的加速比，并且对于不同的STAP应用系统具有很好的适应性。
In the parallelization of Space-Time Adaptive Processing (STAP) arithmetic, traditional methods schedule the STAP arithmetic to different processors in the specific hardware architecture through coral-granularity division and improve the throughput by pipeline processing between processors. In the paper, its disadvantages are discussed from two perspectives: Coarse-grained scheduling hinders the parallelism; They are only suitable for the specific system parameters and hardware architectures. Thus, a new method based on fine-grained scheduling is put forward, which consists of three steps: Firstly, fine-grained task model in the form of Direct Acyclic Graph (DAG) is constructed; Secondly, the topology model is built to describe the target system; Finally, the established task model in fine-grained manner is assigned to different processors described in model topology. The experiment of the proposed method shows that it achieves better acceleration ratio, and more flexiable adaptation to different STAP applications.