The new memory-efficient hardware architecture of CAVLD in H.264/AVC for mobile system
Context Adaptive Variable Length Decoding (CAVLD) module takes the lion chair of the H.264/AVC video decoder time due to its complexity. In order to ameliorate decoding speed, a new CAVLD algorithm and an efficient internal memory design were implemented on Digital Signal Processor (DSP). The proposed CAVLD algorithm, Zero Length Prefix (ZLP), was designed to optimize the first syntax element: the CoeffToken. ZLP implementation reduces CAVLD execution time to 21% instead of 41% from decoding time with a throughput of 1.28 MegaMB/s. In addition, the decoder speed was increased from 36 frames per second (fps) to 44 fps for a CIF compressed bitstream.