A High Throughput Processor Chip for Transform and Quantization Coding in H.264/AVC

摘要

This paper presents an ASIC processor chip for real-time implementation of the computing of the complete process of forward transform, quantization, inverse transform, dequantization, and reconstruction of a 16 × 16 macro-block in full compliance with the H.264/AVC video coding standard. This processor is capable of processing 4 × 4 blocks without interruption, with a parallelism in the datapath of 16 data/cycle, in a pipeline architecture with the twofold aim of achieving high operation frequency and high throughput. To implement the four 4 × 4 transforms and two 2 × 2 transforms required in the H.264/AVC coding system, two configurable multitransform direct 2-D architectures are used, one for forward and another for inverse. Moreover, a reduction in hardware is achieved by reformulating of quantization and dequantization equations and appropriately adjusting the datapath bus widths. A prototype of this processor chip was fabricated in the HCMOS9 STMicroelec-tronics 130 nm standard cell technology. The latency for 16 × 16 macroblocks is 26 clock cycles in normal mode and 42 in Intra 16 × 16 mode with a maximum operating frequency of 280 MHz and a throughput of 4,480 Mpixels/s. As a result, our processor chip is able to support the UHDTV 7680 × 4320@60 Hz (3 G sample/s) format requirement.