Evaluation of Some Exponential Random Number Generators Implemented by FPGA

S. Timarchi, S.G. Miremadi, and A.R. Ejlali (Iran)

Keywords

Random Number Generator, Exponential distributed values, FPGA, Hardware Overhead, Speed, Precision of values.

Abstract

Normally, random number generators produce uniformly distributed values. In some cases, such as Monte Carlo simulation, non-uniform distributed values (e.g. exponential distribution and weibull distribution) are required. One way of producing non-uniformly distributed random values, is to add an extra hardware to a uniformly distributed random number generator. In this paper, three different methods are studied to generate exponential random values. These methods are based on three algorithm of logarithm evaluating namely: CORDIC, Interpolation and Piecewise lookup table. The study is based on an experimental evaluation of the above methods, using FPGA chips, to compare the speed, hardware overhead and the precision of the generated numbers. The results show that the Piecewise method has the highest speed number generation method. The CORDIC hardware overhead will be increased uniformly as the data width varies. For the precision comparison, the CORDIC has the best precision, while the Piecewise method has lower precision than the CORDIC method. For low data width, the Piecewise unit provides the best trade-off in size and performance and for high data width, the CORDIC provides efficient solution.

Important Links:



Go Back