An efficient macroblock-sync decoding architecture for MPEG-4 FGS video

MPEG-4 amendment 2 provides a new coding technique, Fine Granularity Scalability (FGS), which provides the fine-granular rate adjustment capability. This property meets the growing need of streaming video over bandwidth-varying environment. By coding bitplanes of DCT residues from the most significant bit (MSB) to the least significant bit (LSB), FGS makes the compressed bitstream able to be truncated at any location. The resultant video quality would gradually be refined with the increase of the received bitstream. However, an extra frame buffer for storing bit-values of decoded DCT residues and several times of buffer scanning/accessing for accumulating these residues make FGS difficult to be implemented efficiently, especially in memory-constraint system (for example, PDA, smart phone and TablePC). In this paper, we present a Macroblock-Sync decoding process for FGS decoding, which reduces the extra buffer requirement and decreasing the memory-access frequency at the same time. In our design, all processes of the same macroblock (originally done in the base and enhancement decoders respectively) are operated simultaneously. The proposed decoding architecture provides a distinct data-access order and results in a better cache utilization, which overcomes the two above-mentioned problems when realizing the FGS decoder. In our experiments, up to 80% speedup can be achieved.