Optimizing Stochastic Computing-Based FIR Filters

Abstract

Stochastic computing (SC) is an unconventional computing paradigm based on digital computation on stochastic bit streams. It is promising for many applications, one of which is the digital filter design. A previous work proposed an area-efficient SC-based finite impulse response (FIR) filter by optimizing the stochastic number generators (SNGs), which are used to produce stochastic bit streams of the desired probabilities. An SNG is composed of a random number source (RNS) and a probability conversion circuit (PCC). The previous technique is based on reducing the total number of RNSs within all the SNGs. In this work, we exploit two techniques that reduce the area of PCCs to further reduce the area of the SC-based FIR filters. They optimize the PCCs for variable and constant input probabilities, respectively. With the PCC area significantly reduced, it allows us to add few RNSs back to reduce the computation error due to correlation. Our experimental results showed that our proposed design can further improve both the area and computation accuracy of stochastic implementations of FIR filters.