基于量化步长线性预测和BFOS算法的MPEG-4AAC量化

MPEG4-AAC quantization based on linear prediction of quantization step size and BFOS algorithm

MPEG-4 AAC的编码性能很大程度上依赖于量化模块的编码效率和收敛速度,但其常用的基于双循环搜索结构的率失真控制器引起编码器性能较差,尤其在低码率时更为突出。提出一种新的量化优化算法。新方案采取单循环结构,用前面数帧的量化信息对当前帧的初始量化步长做线性预测,再用接近最优比特分配的BFOS算法控制量化步长的调节。仿真证明新方案的编码性能明显优于MPEG-4 AAC VM,对比BOFS算法,运算量得到极大降低。

The coding performance of MPEG-4 Advanced Audio Coding （AAC） is highly dependent on the efficiency and convergent speed of the quantization module, but its common rate-distortion controller based on two-loop search algorithm produces poor performance especially at low bit rates. This paper presents a novel quantization optimization algorithm. The new scheme has a one-loop recursive structure, in which the initial quantization step size of current frame is calculated from previous ones through linear prediction, and the adjustment of quantization step size is controlled by BFOS （Breiman, Friedman, Olshen, and Stone） algorithm in which the bit allocation is near optimal. Simulation results show that the performance of the new scheme is significantly better than that of the MPEG-4 AAC Verification Model and the computational complexity is highly reduced compared to BFOS algorithm.