Although there are the attractions due to the simplicity, low-cost
design, faut-tolerance come from this novel computing, SC equally faces
some drawbacks which cause it to be viewed a an impractical method. This
is the difficulty in generating the real random sequences, because up to
now, remains an open issue. However, there are lately a lot of
applications concerning this kind of computation: design of digital
filter (FIR, IRR) [6], design
of circuits for real-time image processing applications
[3],
[14],
[16],
[15], as well as using of
strochastic computational elements for neural networks
[4],
[5]. Those applications have
partly proved the capacity and perspectives of SC in practice and future
research.
Chapter
2
Stochastic Number Generator
One of the major problems in the design of stochastic computing system
is the synthesis of hard- ware system for the generation of sequences of
independent, uniformly distributed, random numbers. Early forms of
random number generators for SC used physical noise sources. However,
there are the disavantages of such random sources
[19]:
Hence, the alternative approaches to random number generation, which
ensure that the generated numbers must be uniformly distributed across
the interval [0, 1], have been proposed. In our work, LFSR
and CA will be considered as suitable solutions for generating random
numbers.
About SNG, we will investigate two group of methods: SNG using
comparator and SNG without using comparator. The evaluation and
comparison of these approaches will be carried out to provide a deep
understanding on this important problem in SC.