Proximity-Based Authentication of Mobile Devices Eyal de Lara Department of Computer Science University of Toronto Alex Varshavsky, Adin Scannel, Anthony LaMarca
Secure Spontaneous Interaction Phone + hotel room TV and keyboard Exchange of private info Phone and hands free Paying for groceries, tickets, cola
Naïve Solution Diffie-Hellman a Alice b Bob
Naïve Solution Diffie-Hellman a Alice b Bob g, g a
Naïve Solution Diffie-Hellman a Alice b K g ab Bob g, g a
Naïve Solution Diffie-Hellman a Alice b K=g ab Bob g, g a gbgb
Naïve Solution Diffie-Hellman a K=g ba Alice b K=g ab Bob g, g a gbgb
Who is my device really communicating with? The Problem
Who is my device really communicating with? Spoofing The Problem a Alice b Bob
Who is my device really communicating with? Spoofing The Problem a Alice b Bob x X
Who is my device really communicating with? Spoofing The Problem a Alice x X
Who is my device really communicating with? Spoofing The Problem a Alice x Bob
Who is my device really communicating with? Spoofing The Problem aK=gxaaK=gxa Alice x K=g ax Bob g, g a gxgx
Who is my device really communicating with? Spoofing Man in the middle The Problem a Alice b Bob x X
Who is my device really communicating with? Spoofing Man in the middle The Problem a K 1 =g xa Alice b K 2 =g xb Bob g, g a gxgx x K 1 =g ax K 2 =g bx X g, g x gbgb
Who is my device really communicating with? Spoofing Man in the middle Solution: Ensure communication with device that is close Assumption: attacker is not between legitimate devices The Problem a K 1 =g xa Alice b K 2 =g xb Bob g, g a gxgx x K 1 =g ax K 2 =g bx X g, g x gbgb
Existing Solutions Use a cable Use short range communication Bluetooth Infrared Laser Ultrasound Near field communication (NFC) Ask user to verify pairing Displaying keys Playing music, images
Existing Solutions Use a cable Use short range communication Bluetooth Infrared Laser Ultrasound Near field communication (NFC) Ask user to verify pairing Displaying keys Playing music, images BlueSniper Rifle by Flexis
Key Idea Secure pairing requires a shared secret Devices in close proximity perceive a similar radio environment Derive shared secret from common radio environment Listen to traffic of ambient radio sources Use knowledge of common radio environment as proof of proximity
Advantages No extra hardware Leverage radio already available on device No user involvement to verify pairing Not subject to eavesdropping Secret derived by listening to ambient sources
Requirements on Radio Environment 1.Temporal variability Signal fluctuates randomly at a single location over time
Requirements on Radio Environment 2.Spatial variability Values at different locations have low correlation
Requirements on Radio Environment 3.Devices in proximity should perceive similar environment 5 cm 10 m 85% common pkts40% common pkts
Potential Authentication Methods Proximity-based authentication token Diffie-Hellman Authenticate using the token Proximity-based encryption keys Directly from the common environment Less CPU intensive?
Amigo: Diffie-Hellman + Proximity Token Devises monitor radio environment following Diffie-Hellman key exchange Send to each other a signature Each device verifies that signature similar to own observation Signature does not have to remain secret after exchange is over
Signature Verification Signature: sequence of hash of packet + RSSI Segment size 1 second
Classifier 2 stage boosted binary stump classifier Stage 1: Filters noisy data Marks as invalid instances with % of common pkts bellow threshold (75% works well) Stage 2: Assigns a score to valid instances Function of differences in signal strength Converts scores into votes based on threshold Tally votes for all instances
Commitment Protocol Reveal man-in-middle attack while exchanging signatures Forces attacker to forge data Break signature S into n blocks Generate nonce Each period exchange K nonce ( Hash (K session_key ),Hash(id),s i ) Send nonce a K 1 =g xa Alice b K 2 =g xb Bob K nA (H(K 1 )H(A)S i ) x K 1 =g ax K 2 =g bx X K nB (H(K 2 )H(B)S i )
Scenario 1 : Simple Attacker 6 laptops Friendly 5cm away Attackers 1,3,5,10 meters WiFi – Orinoco Gold All at same height Line of sight 1m 3m 10m 5m Best case for attacker
Traces 2 traces: training and testing 2 months apart 2 different location in the lab 10 minute trace 30 – 50 thousand pkts per laptop 11 access points 45 – 58 WiFi radio sources
Simple Attacker Can pair within 5 seconds Can detect attacker 3 meters away or more 1 meter is a problem
Local Entropy: Obstacles False Positives Line-of-sight (1m)81% Drywall (10cm)100% Human (1m)12% Concrete wall (30cm)0% Human blocking attacker’s line of sight goes a long way to improve performance
Local Entropy: Movement Hand waving helps!
5 laptops Friendly 1 m away Attackers 3,5,10 meters All at same height Line of sight Stretching Co-Location 1m 3m 10m 5m
Stretching Co-Location
Scenario 2 : Attacker with Site Knowledge Before pairing Attacker samples exact pairing spot Creates RSSI distribution for every wireless source it hears While pairing Pkts from know source assign RSSI from distribution Pkts from unknown source Option 1 Discard Option 2 Leave unchanged(best)
Scenario 2 : Attacker with Site Knowledge With hand waving false rate positives reaches 0 within 5 seconds
Scenario 3: “Omnipotent” Attacker Controls all radio sources Knows which pkts were received by victim Oracle: RSSI from current distribution
Conclusions Possible to use knowledge of radio environment to prove physical proximity Advantages No extra hardware No user involvement to verify pairing Not subject to eavesdropping Two potential methods Location-based authentication token Location-based encryption keys
Future Work System robustness Different cards and antennas Different environments Improve accuracy Software radios Multiple radios Proximity-based encryption keys
Questions? Eyal de Lara Varshavsky, Scannell, LaMarca, de Lara “Amigo: Proximity-based Authentication of Mobile Devices” 9th Int. Conference on Ubiquitous Computing (UbiComp) Innsbruck, Austria, Sep. 2007