Turbo code based detection for audio watermarking: the generalized Gaussian noise channel model

Additive audio watermarking schemes can be viewed as specific known digital communication channels with very low signal to noise ratio (SNR). The noise is non Gaussian, correlated and piecewise stationary. Even when strong watermark detection schemes are used (blind equalizers, informed coding, etc), the bit rate/detection performances are poor, due to its non robustness faced to the variety of audio situations (MPEG compression, pirate attacks, etc). This "difficult" channel represents then a challenge for turbo code approaches. In this paper, an analysis of a potential contribution of turbo codes to this "watermarking channel" is presented. In particular, we focus on the non Gaussianity and the low SNR of the additive noise. Moreover, we present a modified Maximum A-posteriori Probability (MAP) decoding algorithm of turbo codes in order to include the a-priori knowledge of the noise probability density function (PDF).