(An) implementation and analysis of fast discrete cosine transform algorithms on TMS320C80

Abstract

There have been many demands for the real-time image compression. The image compression systems have a wide range of application. MPEG-2, which is the most recent moving picture compression standard, targets for the high quality video and audio communication. But real-time encoding is hard to implement because it needs a large amount of computations. KICS (KAIST Image Computing System) is being developed to realize the real-time MPEG-2 encoding and decoding with parallel architecture of multiple general-purpose DSP, TMS320C80, called multimedia video processor (MVP). The discrete cosine transform (DCT) and motion estimation need many arithmetic operations compared to other algorithms in MPEG-2. In this thesis, the DCT algorithm is implemented and tested through the MVP simulator. The functions of the MVP is analyzed and the DCT algorithm is fully optimized to the MVP environment. This thesis analyzes various fast DCT algorithms and implements two among them. The possibility of real-time system is verified and the required number of the processors, called PP, is decided for real-time MPEG-2 encoding and decoding.