The computational complexitity of inverse qr - rls algorithm and square - root kalman algorithm which had numerical stability were analysized . because the qr - rls needed the squart - root and inverse of matirx operation , which had large computation , was not fit for fixed - point dsp implement of real time , adfe based on square - root kalman algorithm was applied to fixed - point dsp algorithm implement of the equalizer of hf channel 对采用数值稳定的逆qr - rls算法和平方根kalman算法的计算复杂度进行分析，逆qr - rls算法需要平方根和求逆运算，计算量大，不适于定点dsp的实时实现，故采用平方根kalman算法的adfe作为短波信道的定点dsp算法实现方案。
Using the conic function model local approximation , w . cdavidon ( 1980 ) proposed a class of iterative algorithms with modified matrix combining function value , furthermore under the theory d . c . sorensen has used local quadratic approximation method , then applying collinear scaling idea improving on the above algorithm and generalizing it , getting a class of collinear scaling algorithm , unifying former quasi - newton . in the paper , using local quadratic approximation method , the first , constructing the new collinear scaling gene , getting a class of the new collinear scaling algorithm with briefness and numerical stability , . , we discusses some properties of the algorithm and its local linear convergence , q - superlinear convergence and the whole convergence ; secondly we have made numerical experimentation and numerical analysis ; the last , we have done much discussion for collinear scaling idea and given the several new collinear scaling algorithm 本文的工作就是基于局部二次逼近原理，首先通过构造新的共线调比因子，得到了一类新的更简洁，数值稳定性更好的共线调比算法，进而我们给出了本共线调比算法的局部收敛性，全局收敛性以及算法q -超线性速度的理论证明；其次，用经典的无约束优化五大考核函数就本共线调比算法进行了数值试验和数值分析；最后，就局部二次逼近思想，进行共线调比算法思想进行更广泛的讨论，给出了几个新共线调比算法。
First , the difference equation , numerical stability condition , boundary condition , power sources settings and some improved speedup techniques of the finite - difference time - domain yee computational method were introduced briefly in which with the multi - time step method and the frequency - dependent finite - difference time - domain ( ( fd ) 2td ) method emphasized 首先，本文简单介绍了建立电磁模型中所需采用的时域有限差分yee算法的差分方程、稳定性条件、边界条件、激励波源设置等的基本原理和一些用于加速fdtd算法的改进方法。其中，详述了多时间步长( mts )方法和频率相关时域有限差分( ( fd ) ~ 2td )方法。
With numerical experiments , the main work in this study is as follows : ( 1 ) the building of network parallel numerical computing platforms based on pvm on win32 system , the setting of environmental variables , the startup of pvm and the configuration of virtual machine are discussed . ( 2 ) the methods for the design of network parallel numerical computing programs based on pvm on win32 system are relatively detailedly explored in this study , including the basic pvm programming patterns , the decomposition of workload , the design of communication and the numerical stability of pvm parallel numerical algorithms . ( 3 ) the emphases are not only put on the discussion of several important factors which affect the performance of pvm parallel programs , including load - balancing , communication costs , network performance , task granularity , computer ’ s memory capacity and the number of processors , but also on the corresponding strategies for improving the performance of pvm parallel programs 本研究结合具体的数值试验,主要完成了以下工作: ( 1 )探讨了pvmonwin32的网络并行数值计算平台的构架,并行环境变量的设置, pvm的启动及虚拟机的配置等; ( 2 )较详细地讨论了基于pvmonwin32的网络并行数值计算程序设计的方法及注意事项,包括pvm基本编程模式的研究、任务的分解、通信的设计及pvm并行数值算法的数值稳定性等; ( 3 )重点研究了影响pvm并行程序性能的几个重要因素,包括负载平衡、通信开销、网络性能、任务粒度、处理机个数以及处理机内存容量问题等,并提出了相应的策略,以最大限度地提高pvm并行程序的性能; ( 4 )针对基于pvmonwin32的由桌面pc机联网而成的网络并行计算环境中,处理机的运算速度较快而处理机间的通信相对较慢的实际情况,将求解线性方程组的gauss - jordan消去法与gauss列主元消去法结合起来,给出了网上高效并行求解稠密线性方程组的无回代gauss列主元消去法,其算法处理机间的通信开销较少,实现了负载平衡和各处理机间的全并行工作; ( 5 )在考察求解线性方程组的jacobi迭代法和gauss - seidel迭代法并行性能的基础上,将两者的优越结合起来,给出了基于pvm的微机网络并行计算环境下求解线性方程组的一种分组gauss - seidel并行迭代算法。
