High speed decoding of context-adaptive binary arithmetic codes based on most orobable symbol prediction

Abstract

The Context Adaptive Binary Arithmetic Coding is the key entropy coding algorithm of H.264/AVC which is adopted as next generation moving picture compression. Although the performance gain of H.264/AVC is mostly resulted from Context-Adaptive Binary Arithmetic Coding (CABAC), the decoding complexity of CABAC makes it difficult to implement the decoding system at a moderate operating frequency. Despite the multiplication-free algorithm, CABAC still needs a lot of computations to calculate range, offset and context variables. This paper presents a prediction scheme to decode maximally two bits at a time and reduce overall decoding cycles in CABAC. With the proposed prediction scheme, we implemented a CABAC decoder, which reduces total cycles by 23~24% compared to conventional CABAC decoding schemes. With a little modification to the proposed BAC decoder, it can operate at 300MHz and logic area of the decoder is about 8740 gates without memory cell. The proposed architecture is implemented in Verilog HDL and synthesized in Samsung 0.18㎛ technology.