"Traffic Analysis of Wireless Networks Using Signal Processing"

[Lucky Green]

Chris wrote:
> I'm sitting in NE43-518 at LCS, where a talk by Craig 
> Partridge that I thought might interest you just finished.  
> He described how his group at BBN has implemented 
> successfully several techniques for traffic analysis using 
> methods from signal processing.  Essentially, they convert 
> series of samples consisting of a message originator and 
> timestamp into signals appropriate for signal analysis, and 
> perform correlation analyses on these signals.  From this, 
> they derived network topologies and conversation traces, as 
> well as some other interesting pieces of information about 
> the network.
> 
> Partridge concludes that the techniques of delay dithering 
> and adding random packets are inneffective, since these just 
> look like white noise, easily dealt with by standard signal 
> processing.  He agrees that simply shaping traffic to a 
> constant packet rate works, of course, although it's 
> expensive.  Perhaps those of you working on onion routing may 
> be interested in considering or refuting his conclusions.

The traffic shaping does not necessarily have to use a constant packet
rate as long as the packet rate is data independent, but other than that
Partidge's claims would be difficult to refute since they are
unfortunately correct and have been known for close to a decade.

--Lucky Green