Low-Energy FIR Filter Realisations on Hardware and Software Programmable Platforms
Authors
Gullaksen, Stine Martine ; Nielsen, Morten Kramer Danmark
Term
10. term
Education
Publication year
2012
Submitted on
2012-05-31
Pages
178
Abstract
Mange hverdagssystemer afhænger af digitale filtre. Dette projekt undersøger, hvordan energiforbrug kan minimeres i digitale FIR‑filtre på både software og programmerbar hardware. Vi analyserer, implementerer og evaluerer flere optimeringsmetoder og sammenligner dem i praksis ved hjælp af fungerende realiseringer. På software blev to tilgange afprøvet: multirate‑filtrering (behandler dele af signalet ved lavere hastighed, når det er muligt) og Hamming‑distance‑optimering (mindsker antallet af bit‑skift og dermed omskiftningsaktivitet). Sammenlignet med en reference gav multirate‑filteret op til 20 % besparelse, mens Hamming‑distance‑metoden gav op til 2 %. På hardware blev tre multiplikatorfrie metoder sammenlignet med to referencefiltre. Et shift‑baseret filter klarede sig bedst, derefter et parallelt direkte‑form‑filter, og som nummer tre et multirate‑filter. De mindst energieffektive var et generisk direkte‑form‑filter og et filter baseret på distributed arithmetic. Evalueringen viser, at flere metoder kan bruges som energieffektive løsninger i forhold til en generisk reference. På baggrund af resultaterne præsenteres retningslinjer til designere, der vil udvikle lavenergifiltre.
Many everyday systems rely on digital filters. This project examines how to minimize energy use in digital finite impulse response (FIR) filters on both software and programmable hardware. We analyze, implement, and evaluate several optimization methods and compare them in practice using working realizations. On software, two approaches were tested: multirate filtering (processing parts of the signal at lower rates when possible) and Hamming‑distance optimization (reducing the number of bit changes to lower switching activity). Relative to a reference, the multirate filter achieved up to 20% savings, while the Hamming‑distance method achieved up to 2%. On hardware, three multiplierless designs were compared with two reference filters. A shift‑based filter performed best, followed by a parallel direct‑form filter, and third a multirate filter. The least energy‑efficient were a generic direct‑form filter and a design using distributed arithmetic. Overall, several methods offer energy‑efficient alternatives compared with a generic baseline. Based on the findings, the project provides guidelines for designers aiming to build low‑energy filters.
[This summary has been rewritten with the help of AI based on the project's original abstract]
Documents