Reduction in area and power dissipation of Newton-Raphson (N-R) division module could be achieved by substituting the standard parallel multiplier with truncated multiplier. This is possible since when standard multipliers are employed, their products are
Newton-Raphson division module via truncated multipliers
Alexandar Tzakov
Department of Electrical and Computer Engineering
Illinois Institute of Technology
Chicago ,IL 60616, USA
Abstract
Reduction in area and power dissipation of Newton-Raphson (N-R) division
module could be achieved by substituting the standard parallel multiplier with truncated multiplier. This is possible since when standard multipliers are
employed, their products are rounded to avoid unnecessary growth in word size. Truncated array multipliers, on the other hand, will generate a higher maximum absolute error, and even when limited to one unit in the last place it is not clear how it will affect the performance of the unit. In this paper error analysis of 8-bit N-R module, using Variable Correction Truncated Array and Constant Correction Truncated Dadda multipliers is performed. The maximum and absolute errors are analyzed and contrasted from the implementations of the standard multipliers.
I.Introduction
Typically division in Digital Signal Processing (DSP) Processors is
implemented either with multiplicative algorithms such as Newton-Raphson (N-R) and Goldschmidt's algorithms, or via additive algorithms such as digit-recurrence methods [1] and SRT division. If the N-R method is chosen, the usage of parallel multiplier is inevitable. Parallel multipliers can consume a good portion of the
Processor area, and therefore it is desired to develop techniques to minimize their size.
One such technique is the employment of truncated multipliers [3], which
is especially suitable when N-R division is used, since the product of the multiplication is rounded to avoid unnecessary growth in the word size.
Unfortunately truncated multipliers will introduce additional error, and even when their implementation allows, to limit the error to the Last Significant Bit and N-R approximation is Quadratically convergent, Error Analyses on the division should be performed to determine the accuracy of the processor.
In this paper Error Analysis of 8-bit N-R Module is perfumed for standard
and truncated, Dadda and Array, multiplier implementations. Section 2 gives an overview of N-R division and truncated multipliers, Section 3 present the implementation and data, and Section 4 provides the conclusion.
Newton-Raphson and Goldschmidt division algorithm_IT/计算机_专业资料。Newton-Raphson and Goldschmidt division algorithm Atahar Kamal Mostafa Muhammad Ali Martuza...
Newton-Raphson division ... 16页 免费 Asynchronous Multipliers... 暂无评价... Variable-Correction Truncated Floating Point Multipliers About half the ...
Newton-Raphson and Goldschmidt division algorithm_IT/计算机_专业资料。Newton-Raphson and Goldschmidt division algorithm Atahar Kamal Mostafa Muhammad Ali Martuza...
热门文档