Subliminal channels in signature functions may be unavoidable.

[This is _intense_! 12-Aug-1993 1536 <yerazunis@aidev.enet.dec.com>]

I  was considering the subliminal-channel-in-the-digital-signature question
in the shower, and came to an interesting handwaving proof of the following
statement:

	All signature systems (public key and otherwise) that allow 
	timestamped messages contain subliminal channels of bandwidth 
	proportional to the timestamp resolution.

Fortunately, this mailing list is wide enough to contain the essence of 
the proof.

1) Hypothesize an arbitrary signature system that simply provides 
authentication via a randomish-looking bitstream that is a function
only of the input document and (possibly) a secret or public key
known to the sender and intended recipient, and that an 
"external monitor" exists who will verify that each message is
indeed signed appropriately; 

2) The job of the monitor is to censor communications between the 
sender and recipient; hie does this by examining the contents of
the messages and if 
	1) their visible contents are innoucuous
	2) their signatures do verify
he passes the message; otherwise he refuses the message back to the sender.

3) Assume the signature-generating algorithm is published, and is
a strongly random function of the input stream.

4) Assume any number of messages may be passed.

5) Assume that the sender and intended recipient have previously 
arranged for an unknown-to-the-censor second signature and bit 
count.

6) To send a subliminal-channel message, the sender generates an
innocuous message, time-stamps it, and signs it, then signs the 
signature with the "secret second signature".  If the [bit-count]
low-order bits of the second secret signature match the desired 
first N bits of the desired subliminal channel message, then
the message plus first (authenticating) signature is handed
off to the censor to examine and transport.  If the bits don't
match, the time-stamp of the original message is updated, and
the process repeated until the bits _do_ match.  The loop of
innoucuous-message/first-signature/second-signature/compare is
then repeated again until all the bits of the desired subliminal
channel message have been sent.

Since for any good signature scheme all bits in the output bitstream are
strongly random functions of all bits in the input stream, changing one
bit (in the timestamp) has a chance of 1 in 2^bitcount bits of giving
the desired secret-signature bitset.

Proof of the bandwidth of such a scheme is proportional to 1/2 the
resolution in bits of the time-stamp is left to the student.  (I just
_had_ to say that.  :-)  ).  

Extension: If the number of messages per unit time allowed by the censor 
is limited, then the bandwidth becomes MIN ( [1/2 timestamp resolution] , 
[bitcount * allowed-message-frequency] 

	-Bill