Fang Fang Carnegie Mellon University

Efficient Static Analysis of Finite-Precision Effects via Affine Arithmetic Modeling

Modern digital signal processing (DSP) applications are typically prototyped using floating-point arithmetic, which offers both large dynamic range and high precision for each numerical computation. However, for hardware implementations, the final form rarely uses a full-precision floating-point unit, given constraints of silicon area, power, and speed. This creates the common-and still awkward-problem of transforming the application from its initial, for all practical purposes "infinite" precision form, into some final, finite-precision hardware format. A common approach is to run detailed simulation with a finite set of random inputs to capture the necessary numerical ranges and the maximum error.

In this talk, I introduce a static error analysis technique, based on smart interval methods from affine arithmetic, to help designers translate DSP codes from full-precision floating-point to smaller finite-precision formats. The technique gives results for numerical error estimation comparable to detailed simulation, but achieves speedups of four orders of magnitude by avoiding actual bit-level simulation.

Bio
Fang Fang received the B.S.E.E degree from Southeast University, China in 1999 and the M.S. Degree from Carnegie Mellon University, Pittsburgh, PA in 2001. Currently she is a Ph.D. candidate in Electrical and Computer Engineering Department at Carnegie Mellon University with Prof. Rob A. Rutenbar and Prof. Tsuhan Chen as advisors. Her research interest includes numerical performance of signal processing algorithms, VLSI multimedia processing and high level hardware synthesis of DSP algorithms. Her current work focuses on VLSI design of continuous speech recognition system.