Partial filter duplication: a solution for noise tolerant FIR filter

As a result of the new trends in semiconductor technology using the new scale of integration, the current design tools and methodologies cannot provide a reliable medium for digital circuit implementation due to the existence of deep submicron (DSM) noise. Then, there are two groups of solutions to protect linear operations, such as FIR filters, from technology noise. The first one consists of protecting the data in operations, such as addition or multiplication, using coding theory. The second one treats the problem from an algorithmic point of view, which is known as algorithmic noise tolerant (ANT). The most popular ANT for linear filters is soft-DSP, which is based on a linear predictor to perform a parallel estimation of the filter output, understood as additional information to protect the original filter operation. This paper proposes an alternative algorithm for parallel-estimation ANT (PE-ANT) in which the original filter is partially duplicated. Simulation results show that this approach has better performance under bad noise conditions.