Optimizing Power Consumption in Signal Processing through Approximate Compressors

Abstract

The majority of computer arithmetic applications are highly precise and reliable because they are implemented utilising digital logic circuits. However, a lot of applications, like those in voice processing, multimedia, image processing, and image enhancement, segmentation, compression, and so on, can withstand computational mistakes and imprecision and still yield meaningful and practical results. It is not always appropriate or efficient to utilise exact and accurate models and algorithms in these kinds of applications. When constructing energy-efficient systems, for example, the paradigm of inexact computing focuses on relaxing absolutely accurate and perfectly predictable building components. This enables imprecise computing to take advantage of a possible boost in performance and power efficiency together with a drop in complexity and cost, rerouting the current design process of digital circuits and systems. Using this feature, approximation (or inexact) computing focuses on designing simpler, yet approximate circuits that perform better and/or use less power than precise (exact) logic circuits. In certain digital signal processing applications, approximate computing is used in a hesitant manner because it naturally tolerates inaccurate computing outputs. They use the approximation arithmetic blocks in them to enhance these circuits' electrical performance. This enables imprecise computing to take advantage of a possible boost in performance and power efficiency together with a drop in complexity and cost, rerouting the current design process of digital circuits and systems. Using this feature, approximation (or inexact) computing focuses on designing simpler, yet approximate circuits that perform better and/or use less power than precise (exact) logic circuits. In certain digital signal processing applications, approximate computing is used in a hesitant manner because it naturally tolerates inaccurate computing outputs. They use the approximation arithmetic blocks in them to enhance these circuits' electrical performance. The approach we developed in this research may greatly lower the cost and power consumption of signal processing jobs by employing approximate compressors to create low-power approximate multipliers and minimising the mistakes resulting from the approximation. Our primary focus in this was using compressors to approximate partial product reductions.