TY - GEN

T1 - Effect of bit-level correlation in stochastic computing

AU - Parhi, Megha

AU - Riedel, Marc

AU - Parhi, Keshab K

N1 - Publisher Copyright:
© 2015 IEEE.
Copyright:
Copyright 2016 Elsevier B.V., All rights reserved.

PY - 2015/9/9

Y1 - 2015/9/9

N2 - Simple stochastic logic gates can compute complex functions using stochastic computing. A stochastic number is encoded by a unary bit stream where each bit is 0 or 1. The value of the number is represented by the percent of 1's in the number, and is interpreted as a probability. Each bit of the stochastic number can be modeled as a Bernoulli random variable, and each stochastic number can be represented by a binomial random variable. The variance of a stochastic number is given by p(1 - p)/N where N represents the number of bits in the sequence, and p represents the mean value of the number. For long word-lengths, a binomial random variable behaves as a Gaussian random variable. The mean and variance of a two-input stochastic logic gate are dependent on the bit-level correlation of the two inputs. This paper derives closed-form expressions for mean and variance of two-input stochastic logic gates with correlated inputs. An approach to synthesize correlated stochastic bit streams with specified correlation from uncorrelated bit streams is also presented. Using the proposed synthesis method, stochastic logic gates are simulated with correlated inputs. The simulated values of means and variances are shown to be the same as the theoretical values; thus, the closed-form expressions are validated.

AB - Simple stochastic logic gates can compute complex functions using stochastic computing. A stochastic number is encoded by a unary bit stream where each bit is 0 or 1. The value of the number is represented by the percent of 1's in the number, and is interpreted as a probability. Each bit of the stochastic number can be modeled as a Bernoulli random variable, and each stochastic number can be represented by a binomial random variable. The variance of a stochastic number is given by p(1 - p)/N where N represents the number of bits in the sequence, and p represents the mean value of the number. For long word-lengths, a binomial random variable behaves as a Gaussian random variable. The mean and variance of a two-input stochastic logic gate are dependent on the bit-level correlation of the two inputs. This paper derives closed-form expressions for mean and variance of two-input stochastic logic gates with correlated inputs. An approach to synthesize correlated stochastic bit streams with specified correlation from uncorrelated bit streams is also presented. Using the proposed synthesis method, stochastic logic gates are simulated with correlated inputs. The simulated values of means and variances are shown to be the same as the theoretical values; thus, the closed-form expressions are validated.

UR - http://www.scopus.com/inward/record.url?scp=84961315537&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84961315537&partnerID=8YFLogxK

U2 - 10.1109/ICDSP.2015.7251915

DO - 10.1109/ICDSP.2015.7251915

M3 - Conference contribution

AN - SCOPUS:84961315537

T3 - International Conference on Digital Signal Processing, DSP

SP - 463

EP - 467

BT - 2015 IEEE International Conference on Digital Signal Processing, DSP 2015

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - IEEE International Conference on Digital Signal Processing, DSP 2015

Y2 - 21 July 2015 through 24 July 2015

ER -