针对雷达信号处理中的高精度浮点矩阵运算的需求,设计了一种自适应截位的复数矩阵乘法的新算法,将浮点数据按照指定规则转化为定点格式,从而最大程度地保留计算精度,满足高精度算法的需求,并在国产高性能处理器———“魂芯一号”上编程实现。同时,将新算法分别与“魂芯一号”浮点算法、TS201浮点算法进行对比,新算法从精度和误差分布上都有明显优势,说明新算法是有效可行的,最终在某型号雷达上使用。%To satisfy the accuracy requirement of matrix calculation in radar signal processing,a new adaptive truncation algorithm for complex matrix multiplications is presented.Following the prescribed rules,the algorithm transfers the floating-point data into the fixed-point format,which maintains the calcula-tion accuracy to the largest extent.And it is implemented by applying BWDSP100-based instruction set.Re-sults shown that the proposed algorithm is more robust in both accuracy and error distribution performance compared with BWDSP100 floating algorithm and TS201 floating algorithm.
展开▼