Re: DRMTICS 2005 Call for Papers

From: Francois Grieu (fgrieu_at_francenet.fr)
Date: 05/27/05 

Date: Fri, 27 May 2005 13:31:33 +0200

In article <1ojdnrwe31sbj$.dlg@news.cis.dfn.de>,
 Sebastian Gottschalk <seppi@seppig.de> wrote:

> > Secure unremovable identification is not bound to be security
> > thru obscurity. Consider this overly simplified scheme
> > - a secret key, known only to provider, is expanded using a
> > good block cipher in CTR mode into a bitstream with as many
> > bits as there are samples in the audio file.
> > - ID is encoded into a bitstream i, with one bit of the ID
> > coded on many (say n=2^16) indentical consecutive bits of i.
> > - audio is combined with bitstream: s' = s+g*((b^i)-0.5) for some
> > approriate gain g.
> >
> > Provably, the ID nor the bitwream can not be recovered without
> > the key, or breaking the cipher. For some approriate g, with
> > knowledge fo the key, the ID can be recovered (modulus
> > synchronisation problems), even without knowledge of the
> > original S, and even with lots of noise digitally added.
> > Small variations of this scheme will survive a low pass-filter
> > or analog recording. Much more complex variations may survive
> > mp3 encoding.
>
> It will not survive applying the same scheme some additional times
> to the audio.

I believe it survives applying the same scheme many times by someone
**without** the key. It all works because the receiver is able to
use its knowledge of the key to cancel-out undesired signal (such
as noise, re-inscriptions, and s if it is unknown). BTW, the
decoder computes
    i' = sign(sum((s'-s)*(b-0.5)))
         with 'sum' over the n samples coding the same bit of ID.

[If s is unknown, make s = 0, or some approximation of s by
approriate processing of s']
The process is reminescent of the correlation in a DPA attack.

The process I outlined does not survive a collusion of a few users
to merge their tagged version of the same bitstream to reconstruct
an un-tagged (or worse, miss-tagged) version. Also, I skipped many
details (for example, the coding used for ID must survive a
polarity change; and the decoder not knowing s in advance will
have a hard time achieving synchronisation).

> As far as we know there can't be any such scheme that would
> survive both neuroacustical encoding and re-embedding.

Any pointer to recent authoritative literature on this ? I stopped
looking at watermarking after 1999 (I then briefly worked on a demo
of a music distribution system with plans to use watermarking when
it would work).

   François Grieu

Relevant Pages

  • Re: I dream of a LISP that...
    ... to see if this post actually survives. ... any Scheme implementation goes far beyond ... please point out any Scheme implementation equally ... I think your best post was asking a German to speak to another German, ...
    (comp.lang.lisp)
  • Re: All Cornwall Form Ws Cancelled!!!
    ... survives until the permit expres or is replaced with the new scheme. ...
    (uk.rec.scouting)