--- begin forwarded text
Status: U Date: Tue, 19 Feb 2002 14:06:58 -0500 From: Richard Lethin <[EMAIL PROTECTED]> Organization: Reservoir Labs, Inc. To: [EMAIL PROTECTED] Subject: theoretical results in watermarking Sender: [EMAIL PROTECTED] Reply-To: Richard Lethin <[EMAIL PROTECTED]> YALE UNIVERSITY DEPARTMENT OF ELECTRICAL ENGINEERING SEMINAR Dr. Aaron S. Cohen Research Laboratory for Electronics Massachusetts Institute of Technology will give a talk on Communication with Side Information: Watermarking and Writing on Dirty Paper Wednesday, February 27 at 11:00 am in Room 500, Watson Abstract We consider two communication scenarios in which a transmitter must make the input to a noisy channel similar to some exogenous side information sequence. The first such scenario is watermark-ing for copyright protection. In watermarking, an original data sequence (the side information) is modified slightly in order to embed some extra information (e.g., the ID number of the owner of the data). The embedded information must be recoverable even if an attacker has further modified the data in order to make an illegal copy. Watermarking has become increasingly important due to the ease with which data can now be reproduced and transmitted around the world. In this talk, we answer some fundamental questions about watermarking system performance. We first show how much information a watermarking system can embed and how to design good watermarking systems. We also investigate whether it is better to have noisier data. We finally compare the optimal attack with lossy compression. In the second scenario, the communication system has to be robust against independent additive noise instead of an attacker as in watermarking. Such a model is useful when digitally enhancing analog systems or when designing broadcast codes. Costa considered the case where both the side information and the additive noise are Gaussian, and dubbed it �writing on dirty paper�. Costa�s surprising result is that the maximum rate of information that can be transmitted does not depend on the variance of the side information. We extend Costa�s result by considering general distributions and showing that the maximum rate does not depend on the statistics of the side information if and only if (under certain conditions) the additive noise is Gaussian. That is, the side information need not be Gaussian but the additive noise must be Gaussian in order for Costa�s result to hold. --------------------------------------------------------------------- The Cryptography Mailing List Unsubscribe by sending "unsubscribe cryptography" to [EMAIL PROTECTED]
