Practical Considerations for Cognitive Radio NetworkingEnvironments

James Jody Neel james.neel@crtwireless.com j l@ t i l SDR 10 10

Cognitive Radio Technologies 147 Mill Ridge Rd, Ste 119 Lynchburg, VA 24502

Web: www.crtwireless.com Ph: (540) 230-6012 Email: info@crtwireless.com

Material
Interactions of CRs Impact of Hostile Users Security Issues

Cognitive Radio Technologies 147 Mill Ridge Rd, Ste 119 Lynchburg, VA 24502

Web: www.crtwireless.com Ph: (540) 230-6012 Email: info@crtwireless.com

Issues Can Occur When Multiple Intelligences Interact

Flash Crash of May 6, 2010
Not just a fat finger Combination of bad economic news, big bet by Universa, and interactions of traders and computers
http://www.legitreviews.com/images/reviews/news/dow_drop.jpg

Housing Bubble
Bounce up instead of down Slower interactions lead to slower changes Also indicative of the role beliefs play in instability

Cognitive Radio Technologies 147 Mill Ridge Rd, Ste 119 Lynchburg, VA 24502

http://www.nytimes.com/imagepages/20 06/08/26/weekinreview/27leon_graph2. html

3 Web: www.crtwireless.com Ph: (540) 230-6012 Email: info@crtwireless.com

In heavily loaded networks, a single vacation p p process can spawn an infinite adaptation p
Suppose
g31>g21; g12>g32 ; g23>g13
2

Without loss of generality

g31, g12, g23 = 1 g21, g32, g13 = 0.5

Infinite Loop!
4,5,1,3,2,6,4,
1

Interference Characterization
Chan. Interf. ( , , ) (0,0,0) ( , , ) (0,0,1) ( , , ) (0,1,0) ( , , ) (0,1,1)

Phone Image: http://www1.istockphoto.com/file_thumbview_approve/2820949/2/istockph oto_2820949_dect_phone.jpg Cradle Image: http://www.skypejournal.com/blog/archives/images/AVM_7170_D.jpg

( , , ) (1,0,0)

( , , ) (1,0,1)

( , , ) (1,1,0)

( , , ) (1,1,1)

(1.5,1.5,1.5) (0.5,1,0) (1,0,0.5) (0,0.5,1) (0,0.5,1) (1,0,0.5) (0.5,1,0) (1.5,1.5,1.5)

Cognitive Radio Technologies 147 Mill Ridge Rd, Ste 119 Lynchburg, VA 24502

Web: www.crtwireless.com Ph: (540) 230-6012 Email: info@crtwireless.com

Generalized Insights from the DECT Example

If # links / clusters > # channels, decentralized channel choices will have a non-zero looping probability non zero As # links / clusters , looping probability goes to 1 Can be mitigated by increasing # of channels (DECT has 120) or reducing frequency of adaptations (DECT is every 30 minutes)
Both waste spectrum And were talking 100s of ms for vacation times

Centralized solutions become distributed as networks scale Centralized

Rippling in Cisco WiFi Enterprise Networks
www.hubbert.org/labels/Ripple.html

Also shows up in more recent proposals

Aug 2009 White Spaces paper from Microsoft

Major reason most routing algorithms are not load sensitive

Cognitive Radio Technologies 147 Mill Ridge Rd, Ste 119 Lynchburg, VA 24502 Web: www.crtwireless.com Ph: (540) 230-6012 Email: info@crtwireless.com

Potential games yield predictable interactions

Existence of a function (called the potential function, V), that reflects the change in utility seen by a unilaterally deviating player player. Cognitive radio interpretation:

Our use:

Every time a cognitive radio unilaterally adapts in a way that furthers its own goal, goal some real-valued function increases. Predictable, stable emergent behavior Behavior inconsistent with the goals will immediately break the monotonicity

Cognitive Radio Technologies 147 Mill Ridge Rd, Ste 119 Lynchburg, VA 24502

V(a)

time

Web: www.crtwireless.com Ph: (540) 230-6012 Email: info@crtwireless.com

Example behavior
For example, for a collection of 802.11 clusters independently choosing operating frequencies 1. All self-interested adaptations
Scalable resource utilization No synchronization required
0 5

Clusters Frequencies
10 15 20

2. Based only on observations of own performance

No information exchange overhead More responsive network
0

25-30 dB

Observed Interference Levels

10 15 20

3. Decrease aggregate network interference

Cognitive Radio Technologies 147 Mill Ridge Rd, Ste 119 7 Lynchburg, VA 24502

25-30 dB

Self-stable Converges to local-optima CRT Proprietary

Aggregate Network Interference

10 15 20 Web: www.crtwireless.com seconds Ph: (540) 230-6012 Email: info@crtwireless.com

No network is an island
Many TVWS standards
802.22 802 22 (CR for rural ) 802.16h (CR WiMAX) 802.11af (WhiteFi) CogNeA Decisions impact one another
Etiquette

Coordinate i t C di t quiet periods i d

Common time base, scheduling

Share information
Sensing

Merging? M i ?

1900.6 Draft 1900 6 D ft 1

IEEE 802.19-10/0055r3 802 19 10/0055r3

Cognitive Radio Technologies 147 Mill Ridge Rd, Ste 119 Lynchburg, VA 24502

Web: www.crtwireless.com Ph: (540) 230-6012 Email: info@crtwireless.com

Hostile users can create problems from outside your network

What if the environment is unstable?

Suppose another network is compromised in your area Their behavior influences your networks adaptations

Stability impact
Clusters Frequencies
0 5 10 15 20

Performance Impact Jamming Hopping

Tone Jamming Fixed Interferer (Mobile) (Mobile)
Adaptive Interferer (Mobile) (Mobile)

Observed Interference Levels

0 5 10 15 20

~30 dB
pe sloy) ter ilit reamob G (

Aggregate Network Interference

0 5 10 seconds 15 20

Smaller to zero gain

Shifted Left (jammers)

CognitiveCRT Proprietary Radio Technologies 147 Mill Ridge Rd, Ste 119 9 Lynchburg, VA 24502

Need to consider external actors

Detect unexpected behavior, adjust accordingly

Web: www.crtwireless.com Ph: (540) 230-6012 Email: info@crtwireless.com

Hostile users can blend in

Normal CR
Average interference levels for nodes 6-35 5 malicious, 30 normal Given available adaptations and knowledge about network state Maximize system (own) 35 dB performance Given available adaptations and knowledge about network state Minimize system performance

Hostile CR
35 normal

Adapt at inopportune times Simply minimize performance Ensure marginally stable E i ll t bl network goes unstable Plus learning exploits
And spoofing p g
And information corruption

Cognitive Radio Technologies 147 Mill Ridge Rd, Ste 119 Lynchburg, VA 24502

Web: www.crtwireless.com Ph: (540) 230-6012 Email: info@crtwireless.com

Malicious != Selfish
Popular solution to mischievous nodes (selfish nodes that ( f damage network) is to punish nodes
Also implies a way to brainwash learning nodes
From Fig 6 in [MacKenzie_01]

Imperfect information can obfuscate punishment bf t i h t from mischievous behavior and produce catast op c catastrophic cascades
Brittleness

Even with perfect information, malicious node may b masochistic d be hi i

Cognitive Radio Technologies 147 Mill Ridge Rd, Ste 119 Lynchburg, VA 24502

From [S i t [Srivastava_06] 06]

Web: www.crtwireless.com Ph: (540) 230-6012 Email: info@crtwireless.com

Detecting aberrant behavior from p predictable emergent p p g properties

Adaptations
140 120 Channel 100 80 60 40 0 -50 Ii(f) (dBm) ( -60 Ii(f) (dB Bm) -70 70 -80 -90 50 100 150 200 250 300 -40 -50 -60 -70 -80 -90 0 50 100 150 200 250 300 -30 -40 0 50 100 150 200 250 300 Channe el

140 120 100

80 60 40 0 50 100

Adaptations

Policy restricted channels

150 200 250 300

Radio Utilities

Radio Utilities

-50

f) (f (dBm)

-60 -65 -70 -75

Potential Function No Malicious U N M li i User

150 iteration 200 250 300

(f) (dBm)

-55

Malicious User Detected

-50 -60 -70 -80 0

Potential Function
50 100 150 iteration 200 250 300 Web: www.crtwireless.com

Cognitive Radio Technologies 100 0 50 147 Mill Ridge Rd, Ste 119 Lynchburg, VA 24502

Ph: (540) 230-6012 Email: info@crtwireless.com

Implementation Discussion
Implement as monitoring system that evaluates potential (emergent) f nction function
Frequently sum of performance levels Complexity is in the transmission / connectivity t i i ti it No single node / cluster knows / can evaluate emergent function

But B a malicious CR will li li i ill lie

E.g., Claim massive gains to offset others losses

With BSI, a malicious node t S, a c ous ode cant tell a credible lie!
Other relationships exist Need to be WPG / EPG for linear relationships

Cognitive Radio Technologies 147 Mill Ridge Rd, Ste 119 Lynchburg, VA 24502

Web: www.crtwireless.com Ph: (540) 230-6012 Email: info@crtwireless.com

Security Issues
[Clancy_08] Primary user emulation attacks B li f manipulation attacks Belief i l ti tt k A cognitive radio virus Spectrum sensing data falsification [Chen_08a] Quiet period jamming [Bian 08] [Bian_08] Replay sensing attacks [Bian_08] False coexistence information [Bian_08] [Bian 08] Honeypot attacks [Newman_09] Chaff point attacks [Newman_09]

[Brown_08]

Cognitive Radio Technologies 147 Mill Ridge Rd, Ste 119 Lynchburg, VA 24502

Web: www.crtwireless.com Ph: (540) 230-6012 Email: info@crtwireless.com

Questions you should ask before fielding your network

Can you predict what will happen when the network scales and interactions occur? ? How might your measures be turned against you?
S Sensing, l i learning, policy i li enforcement Even when following the rules

How do you accommodate CR networks other than your own?

Can be attacked from outside without jamming

If there are vulnerabilities, how will you detect that they are being exploited?
Cognitive Radio Technologies 147 Mill Ridge Rd, Ste 119 Lynchburg, VA 24502 Web: www.crtwireless.com Ph: (540) 230-6012 Email: info@crtwireless.com