Fixed-Point Processing Optimization of Mpeg Audio Encoder Using Statistical Model

Audio applications for portable devices have two critical restrictions: small size and low power consumption. Therefore, fixed-point implementations are essential for those applications. Even with a fixed-point processor, however, the data width still is an issue because it can affect both the hardware cost and power consumption. In this paper, we propose a statistical model for the MPEG AAC audio encoder that can provide an optimal precision for the implementation. The hardware with the optimal precision, being compared with the floating-point system, is supposed to have perceptually insignificant errors at its output. To have an optimal precision for the AAC encoder, we estimate the maximum allowable amount of fixed-point arithmetic errors in the bit-allocation process using the statistical model. Finally, we present a architecture for the system appropriate for encoding the audio signals with minimum errors by the fixed-point processing. Tests showed that the fixed-point system optimized using the proposed model had sound quality comparable to the floating-point encoding system.