An information-theoretic analysis of dirty paper coding for informed audio watermarking

Upon the introduction a simplified distortion model for a Gaussian audio watermaking scenario, we derive a lower bound for the Gelfand-Pinsker capacity of the watermark channel. Then, we use the capacity bound for designing an efficient practical watermarking scheme based on dirty trellis codes. The proposed information-theoretic framework is then validated, through the use of experimentally acquired audio signals, with the use of a recently proposed frequency-domain audio watermarking. Both an "ideal" (reflection-free) Gaussian acoustic channel and a realistic multipath acoustic channel are considered