Rate-distortion Optimization for MP3 Audio Coding with Complete Decoder Compatibility

This paper addresses the issue of truly optimizing the rate-distortion (RD) performance for MPEG-1/2 Layer III (MP3) audio coding by presenting a fixed-slope graph-based iterative algorithm to jointly design quantization, Huffman code-book (HCB) selection and region division. Given quantization step sizes and HCB selection, quantization outputs and HCB region division are first adapted to minimize the Lagrangian RD cost function throughout a standard-constrained graph structure. Then quantization step sizes and HCB selection are optimized serially to reduce the RD cost based on given quantization outputs and region division. These three steps are further alternated until convergence occurs. The proposed algorithm has been implemented based on ISO reference codec and LAME 3.96.1 with complete decoder compatibility. Simulation results show that the new MP3 coders resulting from the proposed algorithm indeed improve the RD performance substantially except for extremely low bit rates in both cases