Detecting Malicious Data In Vehicular Networks

1
Detecting Malicious Data In Vehicular Networks

• Jonathan Van Eenwyk
• November 2006

2
Goals and Challenges

• Create adhoc wireless network among vehicles
• Share sensor data
• Access infotainment resources
• Ensure privacy
• Encrypting communications
• Reducing traceability
• Provide auditability
• Recording all events
• Verifying transmitted data

3
Contents

• Research Problem
• Related Work
• Proposed Solution
• Simulation Results
• Future Work
• Conclusion

4
Research Problem

• Scenario
• Vehicles wirelessly share position information
• Aggregated data can construct traffic conditions
• Problem
• Malicious nodes can lie about the position of
  others
• Simulate a traffic jam
• Fake an accident
• Friendly nodes must identify malicious attempts
• No global observer to judge on behalf of others
• Travel speed reduces time to monitor other nodes

5
Related Work

• Detecting and Correcting Malicious Data in VANETs
• VANET 04 Golle, Grenne, Staddon
• Nodes observe each other and share data
• Nodes can bind received communications to
  location
• Nodes can distinguish between neighbors
• Malicious nodes can claim to see anyone anywhere
• Distinguishably limits Sybil (multiple identity)
  attacks
• Colluding attacks expensive due to vehicle
  mobility
• Friendly nodes build models to explain
  observations
• Each observed node can be truthful, malicious,
  or spoof
• If there are 20 nodes, 3203,486,784,401 possible
  explanations
• Favor explanation with fewer malicious and
  spoofed nodes

6
Related Work

• Consistency Check
• All truthful nodes must declare their own
  position
• Self-declaration must be consistent with
  observers
• For all other nodes, all truthful observations
  must agree
• Optimizations
• A node always considers itself truthful
• Search incrementally from zero to N malicious
  nodes
• Select first explanation that works (fewest
  malicious nodes)
• Traverse graph of observations
• Start from current node
• Skip nodes labeled as malicious
• Unreachable nodes are spoofs

7
Related Work

• Example
• Actual situation

8
Related Work

• Example
• Round 1 Zero Malicious Nodes

9
Related Work

• Example
• Round 2 One Malicious Node (Wrong)

10
Related Work

• Example
• Round 3 One Malicious Node (Correct)

11
Proposed Solution

• Goals
• Translate the approach into a real-world
  algorithm
• Simulate with as few assumption as possible
• General Parameters
• Each node maintains private database
• Friendly nodes declare position, speed, and
  heading
• Nodes broadcast database every 0-1 sec (uniform
  dist.)
• Upon receiving a database broadcast
• Nodes merge the contents with their own database
• Add an entry for the message sender
• Assume ability to obtain location but not speed
  or heading
• Nodes attempt to explain database every 5 seconds
• Assume public key encryption

12
Proposed Solution

• Manhattan Grid Mobility Model
• 300 meter (per side) square blocks
• Vehicles turn with 20 probability
• Speed normally distr. (Mean 10 m/s, StdDev 5
  m/s)
• Vehicles can change speed when turning
• No collision detection (vehicles can cross paths)
• Time Parameters
• Recorded events are valid for 15 seconds
• Events usually propagate 15 hops (one per second)
• Vehicles usually travel 150 meters until
  expiration
• Rectangular travel prediction (next slide)

13
Proposed Solution

• Rectangular Travel Prediction
• Compare self-declared information against
  observation
• Observed location must be inside predicted
  rectangle

14
Simulation Results

• Simulation Tool
• NS-2 Sparse documentation for wireless
  applications
• QualNet Non-free, too big for short timeframe
• OMNeT Better documentation, good wireless
  support
• OMNet
• Highly modular C design
• Completely structured around message passing
• GUI allows viewing and tracing messages
• Mobility Framework adds mobile ad-hoc networks

15
Simulation Results

• Stationary Scenario
• Nodes placed in fixed locations as in previous
  example
• 5 friendly nodes
• 1 malicious node
• Broadcasts false data to hide location
• Simulated execution time 30 minutes

16
Simulation Results

• Stationary Scenario
• OMNeT Slow Speed Demonstration

17
Simulation Results

• Stationary Scenario
• Results are exactly as expected

18
Simulation Results

• Manhattan Grid, All Friendly, Straight Line Paths
• Uncertainty due to event lifetimes and direction
  switch

19
Simulation Results

• Manhattan Grid, All Friendly, Speed/Dir Changes
• OMNeT Express Speed Demonstration

20
Simulation Results

• Manhattan Grid, All Friendly, Speed/Dir Changes
• Slightly more uncertainty with speed/direction
  changes

21
Simulation Results

• Manhattan Grid, 2 Malicious, Speed/Dir Changes
• Malicious nodes correctly identified others
  tainted

22
Future Work

• Improve Performance
• Develop heuristics to guess after reasonable
  search time
• Explore methods to scale with more malicious
  nodes
• Enhance Accuracy
• Weight explanations by numeric probability
• Handle event lifetimes better
• Reduce Overhead
• Transmit reactively on speed/heading change
• Broadcast subset of database

23
Conclusion

• Algorithm provides reasonable accuracy
• Not scalable in high density scenarios
• Exciting research area with much more to be done
• Questions?