Finite wordlength effects analysis and wordlength optimization of Dolby digital audio decoder

The finite wordlength effects of a Dolby digital audio decoder are studied using both analytical and simulation-based methods for the wordlength optimization. The wordlength optimization was conducted to minimize the hardware cost while keeping the quantization noise below the human perceptual masking curve. In order to derive an accurate fixed-point error expression using the analytical approach, two kinds of errors, the signal and the coefficient quantizations, are considered separately. In the simulation-based method, 24 sine-wave streams in the center of human ear's 24 critical bands are used, and the frequency regions which are the most critical for fixed-point optimization are determined considering the masking curve. The optimized data-path saves about 14% of the chip area, and 20% of the ROM size when compared with a general purpose DSP data path.