A novel approximate constant multiplier and HEVC discrete cosine transform case study

Official URL: https://dx.doi.org/10.1109/ICCE-Berlin50680.2020.9352166

Approximate computing is used for error tolerant applications to design faster, smaller area and lower power consuming hardware than exact optimized hardware designs by trading off speed, area and power consumption with quality. Constant multiplication is used in error tolerant applications with high computational complexity such as video processing, video compression and machine learning. Therefore, in this paper, a novel approximate constant multiplication technique is proposed. Approximate constant multiplier hardware implementing the proposed approximation technique reduces constant multiplication to multiplication with a smaller constant. The proposed approximate constant multiplier causes negligible quality loss when it is used to implement the constant multiplications in High Efficiency Video Coding (HEVC) discrete cosine transform (DCT). It reduces area, reduces power consumption and increases performance of HEVC DCT hardware.