R. Younesi; J. Rastegar; M. Rastgarpour
Abstract
High Efficiency Video Coding (HEVC) is one of the latest coding standards targeting high-resolution video contents. Due to the high complexity of the existing hardware implementation, this paper presents the low-cost and efficient DCT architectures for HEVC, which are able to perform DCT operation of ...
Read More
High Efficiency Video Coding (HEVC) is one of the latest coding standards targeting high-resolution video contents. Due to the high complexity of the existing hardware implementation, this paper presents the low-cost and efficient DCT architectures for HEVC, which are able to perform DCT operation of multiple transform sizes in a single unified architecture. Our objective is to reuse the hardware resources in a DCT architectures using configurable constant multipliers as well as reducing the hardware cost and trading off between hardware complexity and efficiency. We propose three different shift-and-add units with different hardware cost and throughput. The main advantage of the proposed architectures over the existing architectures is a lower hardware and it can also perform DCT transform of different transform units which is available in HEVC standard. The experimental results over 90-nm technology show that the proposed 2D-DCT architecture #1 archives the lowest hardware cost amongst the rest of the architectures with around 57% reduction in gate count, on average. The unfolded 2D-DCT architectures #2 and #3 offer the moderate reduction in gate count around 47%, on average, with a moderate throughput. Apart from architectures #1, #2, and #3, we also develop a reusable architecture by adding an extra ( )-point DCT alongside the main DCT.