双正交叠式变换的定点实现算法与DSP并行计算  

Fixed Point Algorithm and Parallel Computing in DSP of Lapped Biorthogonal Transform

作　　者：王继东[1] 罗武胜[1] 鲁琴[1] 肖学敏[1] 

机构地区：[1]国防科技大学机电工程与自动化学院,长沙410073

出　　处：《光电工程》2008年第4期74-78,共5页Opto-Electronic Engineering

基　　金：国家高技术研究发展计划863基金资助项目(2006AA701121)

摘　　要：针对硬件实现过程中双正交叠式变换(LBT)的浮点算法复杂度较高、二进整数算法的精度较低的问题,提出了LBT的定点实现算法。该算法根据Q格式数对浮点数的近似,采用定点逼近旋转矩阵中的三角函数值。并对该算法在TMS320C6000系列DSP实现中的并行计算问题进行了研究,通过任务划分、复合指令、软件流水等方式提高算法实现速度。该算法已在某侦查卫星数据压缩样机中实现,实验结果表明,它是一种有效的图像压缩变换,压缩质量接近CDF9/7小波,计算复杂度非常低。The computation complexity of Lapped Biorthogonal Transform （LBT） is high and the precision of its integer algorithm is lower. To overcome these shortcuts, a fixed point algorithm of LBT is proposed, Based on the approximation of numbers in Q format, fixed point is used to approximate the value of trigonometric function of the rotation matrix in the algorithm. And the parallel computing of the algorithm in TMS320C6000 DSP is studied by using assignment partition, complex instruction, software pipeline and optimization method to improve the processing speed. The algorithm has been used in prototype data compression equipment of a spy satellite. Image coding experiments show that coding performance of the algorithm is very competitive with CDF9/7 wavelet and the complexity of the algorithm is far low .

关 键 词：图像压缩 双正交叠式变换 三角函数 流水线 

分 类 号：TN919.8[电子电信—通信与信息系统]