1. Security in Wireless Sensor Networks
Michael Krishnan

2. Outline Types of Attacks Clusters and Intrusion Detection
Game Theory Approach

3. Characteristics of WSNs
Limited Energy (~6Ah)
Wireless: Intruders can see transmissions and add their own
Traffic is either source to sink (base station) or broadcast

4. Types of Attacks Steal Data – Confidentiality
Alter Data – Data Integrity
Limit Service Availability (DoS)
Consume Energy “Denial of Sleep”
D.o.Sleep – Reduce chances to go to low-power states

5. Confidentiality Public key? Too computationally expensive
Secret key? Bad if node is compromised
Secure Network Encryption Protocol (SNEP)

6. SNEP
Both sides keep (pair-wise) shared key, k, & shared counter, C, to use as IV in DES
Semantic Security
Whole network shares MAC() function for authentication: MAC(k,C|{D}) (8 bytes)
(Weak) Freshness – replay protection and ordering
MAC = msg authentication code

7. Data Integrity
Authentication: Can’t use asymmetric digital signatures – too much overhead
SNEP: two-party
mTESLA: broadcast

8. Data Integrity - mTESLA
One-way function, F(.)
Kn = F(Kn+1)
Keys disclosed periodically, not per packet
Figure from Perrig et al.

9. Service Availability Bogus Routing Information Flooding
Homing – look at traffic to find important nodes
“Black Hole” Attack – compromise neighbors of base-station
De-synchronization (transport layer)

10. Energy – Denial of Sleep Attack
Unique to WSNs – can’t use techniques from wired networks
Sources of Energy Loss
Collision – Frequency Hopping, CDMA, FEC
Message Overhearing – RTS/CTS, NAV
Idle Listening – schedule sleep
Brownfield et al. (2005)

11. Scheduling Sleep – S-MAC
Fixed Sleep Schedule
RTS During Listen Period
If no RTS  sleep
Vulnerable during listen period only
Sensor MAC
Time sync
Broadcast throughout listening period
Can keep one awake with RTS (also DoS)
Figure from Brownfield et al.

12. Scheduling Sleep – T-MAC
Timeout MAC
Sleep Early: wait for timeout period
Longest time hidden node must wait before first bit of CTS (TA = 1.5*(tCW_Max + tRTS + tSIFS)
Saves energy in absence of attacker, but MORE vulnerable to attacks (if never get timeout, stay awake forever)
1.5 is just to keep stable

13. Scheduling Sleep – B-MAC
No fixed listening start time
Periodically wake up and sample channel using low power listening (LPL)
Longer preamble (longer than sleep period)
Just as vulnerable to attack as T-MAC
Figure from Brownfield et al.

14. Scheduling Sleep – G-MAC
Split Frame into Collection and Distribution Period
Gateway Sensor (GS) node schedules traffic for cluster
Rotate being GS to distribute energy use
Gateway can keep misbehaving node in check

15. Scheduling Sleep – G-MAC
Figure from Brownfield et al.

16. Clusters Cluster head (CH) and member nodes (MN)
Popular in routing protocols
Nearby nodes have redundancy, compressed at CH (save energy)
Can also use for intrusion detection
CH monitors MNs, while some subset of MNs monitor CH
X MNs can decommission CH (homing)
CH = GS, rotate

17. Methods of Intrusion Detection
Anomaly Detection – Actions of monitored node are atypical
High probability of false alarm
Signature Detection – Actions of monitored node correspond to a type of attack
Susceptible to new attacks
Typical Attacks:
Drop Packets
Duplicate Packets
Cause Collisions

18. Clusters for Authentification
Everyone watch neighbors? Too much energy
BS checks packet at the end? Waste energy transmitting bad packet whole route – need to discover this sooner
Check packet everywhere? A lot of computation
Check at CH. Send packets first to CH
Also send to CH with some probability p so compromised node can’t bypass CH.

19. Game Theory Approach Agah et al. (2004)
Model: 2-player, non-cooperative, nonzero-sum
Players: IDS, attacker
IDS can choose 1 cluster to defend, Attacker can choose 1 to attack

20. Game Theory Approach - Notation
U = Utility of working WSN
Ck = Cost to defend cluster k
ALk = Average loss for losing cluster k
PI = Attackers profit for intruding
CI = Attackers cost to intrude
CW = Attacker’s cost to wait

21. Game Theory Approach - Assumptions
PI = SAL
CW < PI-CI
Ck ~ gk, where gk = # previous attacks to k

22. Game Theory Approach Payoff Matrix (for cluster k): Attack k
Do Nothing
Attack k”
Defend k
U-Ck
PI-CI CW
U-Ck-ALk”
Defend k’
U-Ck’-ALk
U-Ck’
U-Ck’-ALk”

23. What’s wrong with this?
Attacker benefit is independent of what IDS does…
Intuitively, this should matter
We defend one cluster at a time
Why not more?
How do they coordinate? (Extra transmissions)

24. Modified Game Theory Approach
Uk = Utility of cluster k
Ck = Cost to defend cluster k
We can defend as many clusters as we want
If we defend cluster k, utility of cluster is Uk-Ck
If we don’t and it’s not attacked, utility is Uk
If we don’t and it is attacked, utility is 0
Since attacker always attacks, his utility is proportional to IDS’s loss minus a constant (CI)

25. Modified Game Theory Approach
No Pure NE:
Suppose there were, then attacker always attacks one particular cluster, k. IDS should then only defend k. But then utility of attacker is less than it would be for attacking another cluster.
Requirement for mixed NE:
E[util. of attacker] indep. of k – equally likely to attack any cluster  (1-pk)Uk = const, where pk is probability of defending cluster k

26. Modified Game Theory Approach
Strategy:
each cluster knows its own utility (maybe from G-MAC)
Defend with probability pk=1-X/Uk where X is a constant known to the whole WSN.
Expected utility of cluster k:
pk(Uk-Ck)+(1- pk)(Uk*(m-1)/m) where m = # clusters

27. Modified Game Theory Approach
Total expected utility of WSN:
S[pk(Uk-Ck)+(1- pk)(Uk*(m-1)/m)]
= S[(1-X/Uk )(Uk-Ck)+ X/Uk(Uk*(m-1)/m)]
= S[Uk-Ck-X+XCk/Uk + X*(m-1)/m)]
= m(X*(m-1)/m-X)+S[Uk-Ck+XCk/Uk]
= -X+S[Uk-Ck+XCk/Uk]

28. Modified Game Theory Approach
Total expected utility of WSN always defending (pk = 1 for all k):
S[Uk-Ck ]
= -X+S[Uk-Ck+XCk/Uk ]
Gain for using pk < 1
-X+S[Uk-Ck+XCk/Uk] - S[Uk-Ck ]
= -X+S[XCk/Uk ]
= X(S[Ck/Uk ] –1)

29. Modified Game Theory Approach
Utility gain = X(S[Ck/Uk ] –1)
What does this mean?
Goes to -X As Ck  0
Positive for larger Ck and smaller Uk.
Increases with X (Counter-intuitive)
Conclusion: We can improve our utility by defending less when per cluster utility is low and Ck is relatively high
1-> h for h attackers

30. Review
Classified Attacks: Confidentiality, Authenticity, Service Availability, Energy
Clusters are useful for intrusion detection
Game theory approach