Thursday, September 18, 2008

Splitting up Spectrum and wireless transmitter fingerprinting

Solutions for sharing bandwidth
  • Assign a different channel to each AP
  • Load balance
  • Restrict range of each AP
  • Scheduling
Using Channels
  • Least congested channel search (LCCS)- APs talk to each other
  • LCCS is not enough because clients can end up seeing two APs that can't see each other
  • Solution: get feedback from clients, need to do set coloring (conflict free set coloring)
  • Partially overlapping channels are frequently not used (e.g. ch 1 and ch 2 overlap a LOT so we don't use both of them), this causes wasted space (fragmentation of a sort) in our spectrum
  • Solution: Model interference for partial overlap. I-factor == 0 if non overlapping channel, I-factor == 1 if same channel.
  • I-factor = overlap between transmit spectrum mask and receiver's band pass filter profile

New problem: how to identify who is using our AP more w/ something less spoofable than MAC address. We could build Access Control Lists out of this that would work much better than MAC address white-lists.

PAssive RAdiometric Device Identification Scheme (PARADIS)
  • Imperfections in the signal from each card, which are an artifact of manufacturing imperfections, are unique.
  • compute distance of "measured signal" from "ideal signal" along several axis: phase error, magnitude vector, frequency error, SYNC correlation, I/Q origin offset
  • Provides surprisingly accurate identification. They experimented on the Orbit Testbed which is 130 transmitters all of the same model# with sequential serial numbers, they saw 0.34% error rate, which is very impressive.
  • Audience Question: what can a well funded adversary do? Can he spoof the hardware signature? Answer: the adversary would have to be able to read your signature (like the defender/security system) 
Problems to design against
  • Temperature of NIC
  • Mobility (distance from AP?)
  • Nic aging
So, getting back to how we might use this idea for spectrum sharing, we might want to assign spectrum dynamically, and we would need to identify people more accurately to be fair with our allocation policies (i.e. leases).

Real Time Scheduling
  • Hidden Terminal aware Scheduling (HITS)
  • Hidden terminals can be detected by the higher level router which is sending packets that it can tell ahead of time are conflicting but the routers themselves can't see each other (by definition of hidden terminal problem). When the higher order router that has more information notices that two packets will conflict it can infer that there is a hidden terminal problem and it can delay the packets just right so that packets that would collide are not sent at exactly the same time.
  • Problem: we are not able to be accurate enough with our timings

No comments: