Design of VLSI Architecture of Autocorrelation-Based Lossless Recompression Engine for Memory-Efficient Video Coding Systems

摘要

In this paper, an autocorrelation-based lossless recompression (ABLR) algorithm is proposed. The ABLR can save the memory bandwidth of video coding systems and preserves the visual quality. The ABLR consists of two core techniques: (1) a correlation-based prediction technique and (2) a correlation-adaptive Golomb-Rice code. Furthermore, dual-mode memory addressing (DMMA) is also proposed to provide ABLR with memory random access functionality. The word-length utilization rate (WLUR) of DMMA is as high as 92.34 % on average. The experimental results reveal that the ABLR exhibits a lossless compression ratio of 2.05 on average for 1080p test sequences. This indicates that the memory bandwidth can be saved up to 50 %. The VLSI architecture of ABLR is designed with three-stage pipelining and is realized in 0.18 μm 1P6M CMOS technology with a cell-based design flow. The logic gate count is about 28 K and the core area is 0.69 × 0.68 mm~2. The encoding capability can reach full HD (1920 × 1080)@30 fps at a clock rate of 62.5 MHz. The power dissipation is 9.35 mW at a clock rate of 62.5 MHz.