Upload
buikien
View
232
Download
4
Embed Size (px)
Citation preview
BANDANABody Area Network Device-to-device Authenticationusing Natural gAitDominik Schürmann*, Arne Brüsch*, Stephan Sigg†, Lars Wolf*, 2017-03-15*Institute of Operating Systems and Computer Networks, TU Braunschweig†Ambient Intelligence, Comnet, Aalto University
Introduction Quantization Protocol Conclusion
Waking up on Hawaii…
Jet lagAwesome sunriseLet’s go jogging
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 2 of 14
Introduction Quantization Protocol Conclusion
Waking up on Hawaii…
Quantified self
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 2 of 14
Introduction Quantization Protocol Conclusion
Waking up on Hawaii…
Putting on your wearables
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 2 of 14
Introduction Quantization Protocol Conclusion
Waking up on Hawaii…
Putting on all your wearables
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 2 of 14
Introduction Quantization Protocol Conclusion
Waking up on Hawaii…
Putting on all your wearables
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 2 of 14
Introduction Quantization Protocol Conclusion
Device-to-Device Authentication
Bluetooth Authentication“Just works” profileStill pressing buttonsDH key exchangeNo MitM protection
BANDANAPerson’s gait (walking pattern)Zero interactionIndependent of on-body locationMitM protection
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 3 of 14
Introduction Quantization Protocol Conclusion
Device-to-Device Authentication
Bluetooth Authentication“Just works” profileStill pressing buttonsDH key exchangeNo MitM protection
BANDANAPerson’s gait (walking pattern)Zero interactionIndependent of on-body locationMitM protection
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 3 of 14
Introduction Quantization Protocol Conclusion
Novelty
Unlock smartphonesMuaaz et al. 2015Hoang et al. 2015
Device2Device AuthenticationNo init procedure with templatesFresh secrets for each D2Dauthentication
Gait Cycle
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 4 of 14
Introduction Quantization Protocol Conclusion
Accelerometer Reading
0 1 2 3 4 5 6 7−5
0
5
Time [s]
Acce
lera
tion
[m/
s2 ]
Accelerometer reading on z-axis only
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 5 of 14
Introduction Quantization Protocol Conclusion
Rotated Signal
0 1 2 3 4 5 6 70
10
20
Time [s]
Acce
lera
tion
[m/
s2 ]
Orientation relative to ground usingMadgwick’s Algorithm
Notice influence of gravity gz
y x
g
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 6 of 14
Introduction Quantization Protocol Conclusion
Noise-Reduced Signal
0 1 2 3 4 5 6 7−5
0
5
Time [s]
Acce
lera
tion
[m/
s2 ]
Apply a bandpass filter to keep frequencies between 0.5 and 12 Hz
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 7 of 14
Introduction Quantization Protocol Conclusion
Gait-Cycle Detection
0 1 2 3 4 5 6−5
0
5
Time [s]
Acce
lera
tion
[m/
s2 ]
Partition data into gait cyclesResample gait cycles to equal lengthCalculate average gait cycle
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 8 of 14
Introduction Quantization Protocol Conclusion
Quantization
−5
0
5
Acce
lera
tion
[m/
s2 ]
−5
0
5
Acce
lera
tion
[m/
s2 ]
−5
0
5
Acce
lera
tion
[m/
s2 ]
Cycle Average Cycle 1 0 0 1
Average gait cycle overlaid on each original gait cycle4 bits per cycle
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 9 of 14
Introduction Quantization Protocol Conclusion
Quantization
−5
0
5
Acce
lera
tion
[m/
s2 ]
a) 1001 0100 1001 1010 1010 1001 0101 0110
b) 1001 0100 1001 1010 1010 1001 0101 0110
c) 0111 1000 1001 0101 1000 1100 1011 1000
Average gait cycle overlaid on each original gait cycle4 bits per cycle
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 10 of 14
Introduction Quantization Protocol Conclusion
Comparison between Locations
−5
0
5
Acce
lera
tion
[m/
s2 ]
forearm: 0111 1000 1001 0101 1000 1100 1011 1000
−5
0
5
Acce
lera
tion
[m/
s2 ]
waist: 0110 1000 1001 0001 1001 1001 1100 1010
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 11 of 14
Introduction Quantization Protocol Conclusion
Evaluation
Intra-body
chest
forear
mhead sh
inthigh
uppera
rmwais
t0
0.2
0.4
0.6
0.8
1
Inter-body
Sim
ilarit
y
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 12 of 14
Introduction Quantization Protocol Conclusion
A
A
A
A
A
A
B
B
B
B
B
B
Authentication Request
Sensor RecordingMadgwick, Bandpass Filter
Gait Cycle DetectionQuantization, Reliability⇒ f̃A, rA
Sensor RecordingMadgwick, Bandpass Filter
Gait Cycle DetectionQuantization, Reliability⇒ f̃B, rB
rA rB
If h(rB) > h(rA): rA = rBfA = Rel(f̃A, rA)fA
Fuzzy Crypto−−−−−−→ k
If h(rA) > h(rB): rB = rAfB = Rel(f̃B, rB)fB
Fuzzy Crypto−−−−−−→ kPassword Authenticated Key Exchange (PAKE)
s = PAKE(k) s = PAKE(k)
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 13 of 14
Introduction Quantization Protocol Conclusion
Conclusion
Device-to-Device authentication for Body Area NetworksZero-interaction based on human gait patternFor 128 bit keys, 192 bit fingerprints are generated (48 cycles),disregarding 64 unreliable bitsWorst-case duration: 96 s80% similarity required for fuzzy cryptography⇒ 103-bit security level for the PAKE password.
Any questions?Dominik Schürmann <[email protected]>
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 14 of 14
Introduction Quantization Protocol Conclusion
Conclusion
Device-to-Device authentication for Body Area NetworksZero-interaction based on human gait patternFor 128 bit keys, 192 bit fingerprints are generated (48 cycles),disregarding 64 unreliable bitsWorst-case duration: 96 s80% similarity required for fuzzy cryptography⇒ 103-bit security level for the PAKE password.
Any questions?Dominik Schürmann <[email protected]>
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 14 of 14
Backup Slides
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 15 of 14
Spectral Coherence
0 5 10 15 20 250
0.2
0.4
0.6
0.8
1
Frequency
γ2
Same subjects, different locationsDifferent subjects, same locationsDifferent subjects, different locations
Figure: Average spectral coherence over full sensor readings of the Mannheimdataset for same and different subject.
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 16 of 14
Reliability
N+0
N+16
N+32
N+48
N+64
N+128
0.6
0.7
0.8
0.9
1
Total Fingerprint Size M
Sim
ilarit
y[%
]
Figure: Fingerprint similarity of different sizes M with cutoff at N = 128 toevaluate the influence of Rel().
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 17 of 14
Fingerprint Similarity
Table: Fingerprint similarity between locations on the same body(intra-body). Shown is the mean over all 15 subjects.
chest
forear
m
head shin
thighupp
erarm
waist
chest 1.0 0.82 0.74 0.78 0.78 0.88 0.81forearm 0.82 1.0 0.8 0.81 0.88 0.89 0.89head 0.74 0.8 1.0 0.8 0.76 0.77 0.78shin 0.78 0.81 0.8 1.0 0.77 0.78 0.8thigh 0.78 0.88 0.76 0.77 1.0 0.85 0.84upperarm 0.88 0.89 0.77 0.78 0.85 1.0 0.88waist 0.81 0.89 0.78 0.8 0.84 0.88 1.0
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 18 of 14
Entropy
1-19 20-35 36-47 48-51 52-55 56-88 89-950
0.2
0.4
0.6
0.8
1
p-va
lue
1:birthdays 5:bitsream 9:count1sstr 13:3dsphere 17:marsagliatsangcd 36-47:rgb-bitdistribution (1-12) 90:dab-bytedistrib2:operm5 6:opso 10:count1sbyt 14:squeeze 18:sts-monobit 48-51:rgb-minimum-distance (2-5) 91:dab-dct3:rank32x32 7:oqso 11:parkinglot 15:runs 19:sts-runs 52-55:rgb-permutations (2-5) 92-93:dab-filltree (20-21)4:rank6x8 8:dna 12:2dsphere 16:craps 20-35:sts-serial (1-16) 56-88:rgb-lagged-sum (0-32) 94:dab-filltree (32)
89:rgb-kstest-test 95:dab-monobit2 (12)
Figure: Distribution of p-values achieved for 128 bit keys (fingerprint lengthM = 192, 64 unreliable bits removed) in 21 runs of the various statistical tests ofthe dieHarder set of statistical tests.
2017-03-15 Dominik Schürmann, Arne Brüsch, Stephan Sigg, Lars Wolf BANDANA Page 19 of 14