Rayleigh fading channel model versus AWGN channel model in audio watermarking

A well-defined characterization of the distortions in the watermarked audio after mp3 attack would improve channel estimation and increase detection robustness. In previous work in the field of watermarking, noise introduced by attacks was usually modeled as additive white Gaussian noise (AWGN). Spectrum analysis showed that a more appropriate channel model, in the presence of mp3 coding, is the frequency-selective fading model, as it describes more precisely the distortion that appears in audio spectrum. Simulations confirmed that BER curves derived from our model were far closer to the experimentally derived ones, than the BER curves derived from AWGN channel model.