Optimization of an AVS Audio Decoder on DSP

The part III of Audio Video Standard of China (AVS) is the first Hi-Fi audio coding standard in China public domain. Through nonlinear quantization and context-dependent bitplane entropy coding (CBC), it offers distinct features of nonlinear quantization noise adaptation and high efficiency fine grain scalability (FGS), but also imposes great challenges on fixed-point implementation with high precision requirement and stringent computational resource restriction. This paper presents tuned optimization strategies for fixed-point arithmetic, buffer manipulation, entropy decoding, and CBC decoding loop reduction to address the challenges. The demonstration real-time implementation on fixed-point DSP TMS320C5509A shows that our design optimization cuts down computational complexity from over 200 MIPs to 60 MIPs while preserving over 15 effective bits precision for 16-bit PCM output