Implementation of a real-time audio watermark extractor on ARM940T

Standardization of audio watermarking systems by SDMI has been a major step towards digital distribution of music. In this paper, we discuss the implementation of Verance-4C (SDMI Phase-I) audio watermarking scheme on ARM940T processor. It uses several multiply intensive signal processing algorithms such as IIR filtering and autocorrelation. Since multiplications take many cycles in ARM processor, it is challenging to optimize the extractor with minimal number of multiplications. Also, data loading, storing and cache miss issues are important since the target system doesn\´t use internal memory. We also discuss a "Minimum Load and Store using Parallel Computation" (MLS-PC) algorithm for reducing data load and store operations while implementing a signal processing algorithm. The optimized watermark detector can extract audio watermarks from 192kHz, 24 bit audio at 50 MIPS.