Byeong Ho Choi

Implementation of an object audio system based on MPEG-4 audio lossless coding on DSP

In this paper, we propose an object audio system on digital signal processor (DSP) that consists of MPEG-4 audio lossless coding (ALS) to provide high-quality audio. The complexity reduction in the designed object audio system is very critical issue because the system requires several MPEG-4 ALS decoders, as many as the number of objects. A method to efficiently use internal memory on DSP is suggested to overcome the high-complexity situation that happens with the use of external memory. A low-complexity finite impulse response (FIR) filter is also proposed because the short-term prediction filter in the MPEG-4 ALS decoder has the highest complexity in MPEG-4 ALS decoder blocks. A method for efficient use of internal memory is designed so that the critical data of MPEG-4 ALS decoders use internal memory as much as the size of the data for a decoder, and the internal memory is shared with the MPEG-4 ALS decoders. The proposed FIR filter reduces the complexity of the short-term prediction filter by 25% compared to direct convolution. A proposed method for an object audio system is evaluated on DSP; it consists of 12 objects. The proposed audio system has a reduction of complexity by 83% with the application of the two proposed methods, and the audio system operates in real time on DSP. This means that high-quality object audio can be serviced for multimedia products.

Implementation of a fixed-point MPEG-4 scalable lossless coding encoder

In this paper, we propose four fixed-point nonlinear functions to develop a fixed-point MPEG-4 scalable lossless coding (SLS) encoder: the square-root function, the logarithmic function, the exponential function, and the trigonometric function. The accuracy and complexity of the proposed fixed-point functions can be controlled by setting the required number of iterations to process a fixed-point function. A fixed-point square-root function is implemented based on the long-hand decimal division, and the complexity of the function is reduced by scaling the input so that the most significant bit (MSB) of the input can be processed at the first iteration. For the fixed-point logarithmic function, a piece of input is converted into an exponent and a mantissa. Then, the output of the function is obtained by independent operations using the exponent and mantissa. A fixed-point exponential function is implemented through the multiplication of the output range and scale factors that are generated from a characteristic of the exponential function. A fixed-point trigonometric function is proposed by designing the output table that has different step size according to input because the function has a different slope at each input. The fixed-point MPEG-4 SLS encoder is implemented using the proposed fixed-point nonlinear functions so that it can be developed on a chip or a fixed-point digital signal processor (DSP) at low-cost. To evaluate the performance of the implemented fixed-point MPEG-4 SLS encoder, two bit-streams, one encoded by a fixed-point MPEG-4 SLS encoder and the one by a floating-point encoder, are decoded using a floating-point MPEG-4 SLS decoder. Then, the root mean square (RMS) is measured from two decoded sounds. The RMS of the fixed-point MPEG-4 SLS encoder is less than −77.35 dB, and both the decoded sounds are identical by 77.2%.