Slide 1

Speed Dating despite Jammers

Dominik MeierYvonne-Anne PignoletStefan SchmidRoger Wattenhofer

1

Yvonne Anne Pignolet @ DCOSS 2009Wireless Networks

Radio CommunicationFind communication partner (device discovery)Concurrent transmissions disturb each other (Interference)Device discovery under jamming attacks22What kind of problems occur in WN? Consider the smallest of all networks, consisting of two participants.Since the communication medium is shared, other devices emitting radio signals disturb the transmission of messages, even if they dont intend to do so, e.g. microwave

Moreover, radio communication happens to use a lot of energy, which can be problematic, depending on the type of application.In addition, again because any device can potentially communicate with any other device in its reception range, networks can be composed of very varied devices that dont necessary are willing to cooperate.

In this talk, I address these 3 problems by discussing...Yvonne Anne Pignolet @ DCOSS 2009

Adversarial Interference: Jamming

Device Discovery

channelschannels

No discovery!ID2428ID587233Imagine a situation with two devices that have never communicated before. They both have an ID and a number of frequencies available for communication. Now, if they choose different channels, they can send as many messages as they want, they will never receive a replyYvonne Anne Pignolet @ DCOSS 2009

Adversarial Interference: Jamming

Device Discovery

channelschannelsDiscovery!

ID5872ID2428ID5872ID24284

4If they use the same channel, they can exchange messages and agree on a protocol for further communicationYvonne Anne Pignolet @ DCOSS 2009

Adversarial Interference: Jamming

Device Discovery

channelschannels

ID5872

No discovery!ID242855What if there is an adversary that does not want them to meet each other? Jamming devices, are not too difficult to build and if they emit a strong signal in the right channel no communication can happen.Depending on the capabilities of the jammer, the devices can still discover each other...Quality: := maxE[algo discovery time | t unknown ] E[best discovery time | t known ]tModel: Device Discovery Problem2 devices Want to get to know each other m channels Listen/send on 1 channel in each time slot

mt

6Adversary Always blocks t channels t < m Worst caseGoal:Algorithm that lets the devices find each other quickly, regardless of tQuickly?Gracefuldegradation6Lets formalise the problem like this:...We assume that the adversary can always block t channels in a worst case manner. If t>= m communication is impossible and no algorithm can help us here so we exclude this case.

We aim to devise algorithms that... No matter what t is

What do we mean by quickly?Our quality measure rho compares the expected discovery time of the algorithm under scrutiny with the expected discovery time if t is known

We can compute what the expected number of time slots until we meet another node is..

AlgorithmsRandomized AlgorithmsRepresented by probability distribution over channels: choose channel i with probability pi

7

Perfect for sensor nodes because we are rather stupid....p1p2p3p4pmpm-1......Advantages Simple Independent of starting time Stateless Robust against adaptive adversaries

7Lets formalise the problem like this:...We assume that the adversary can always block t channels in a worst case manner. If t>= m communication is impossible and no algorithm can help us here so we exclude this case.

We aim to devise algorithms that... No matter what t is

What do we mean by quickly?Our quality measure rho compares the expected discovery time of the algorithm under scrutiny with the expected discovery time if t is known

We can compute what the expected number of time slots until we meet another node is..

E[best discovery time | t known ]Best AlgorithmIn each time slot if t = 0 choose channel 1 if t < m/2 choose random channel in [1,2t] elsechoose random channel2t

E[discovery time knowing t] = 1 if t=04tif t < m/2m2/(m-t)else

t> 0 : Why uniform distribution? Why channel in [1,2t] ? 2t minimizes discovery timeEasy! What if we dont know t?

8Only Interference can be influenced by protocols... By separation- Frequency: assign different frequencies to different nodes-Time: assign slots to nodesCodes: assign encoding to nodes (orthogonal codes, decoding still possible if more than message arrives)Space: divide space into disjoint cells

Transmitted power level Transmission rate Transmission route (for peer-to-peer connections in multi-hop adhoc networks)

Rewards (utilities) associated with transmission quality Achieved BER (SINR) Energy expenditure Transmission delay ThroughputExample AlgoRandomIn each time slot choose channel uniformly at randomE[discovery time NOT knowing t]

E[time AlgoRandom ] = m2/(m-t)choose t=0Random = mExample Algo3In each time slot with prob 1/3choose channel 1 with prob 1/3 choose randomly in [1,m] with prob 1/3 choose randomly in [1,m] estimate t = 0 estimate t = m/2 estimate t = m/2choose t= m3 = O( m) := maxE[algo discovery time | t unknown ] E[best discovery time | t known ]t9First idea might be to choose channels uniformly at random. In this case the expected discovery time amounts to this expression. What is rho for this algorithm? Well, the adversary would choose

You might wonder whether even better algos exist...The answer is...

Yes...

E[discovery time NOT knowing t]Example Algolog mIn each time slot with prob 1/log m choose channel 1 with prob 1/log m choose randomly in [1,2]... with prob 1/log m choose randomly in [1,2^i]... with prob 1/log m choose randomly in [1,m] estimate t = 0 estimate t = 1

estimate t = 2^(i-1)

estimate t = m/2

:= maxE[algo discovery time | t unknown ] E[best discovery time | t known ]tchoose t= mlog m = O(log^2 m)10First idea might be to choose channels uniformly at random. In this case the expected discovery time amounts to this expression. What is rho for this algorithm? Well, the adversary would choose

You might wonder whether even better algos exist...The answer is...

Yes...

General algorithmGiven probability distribution p, where p1 p2 pm 0In each time slot choose channel i with probability piOptimal Algorithm? := maxE[algo discovery time | t unknown ] E[best discovery time | t known ]t E[algo discovery time | t unknown ] E[best discovery time | t known ] 1/ pi2 if t=0

1/(4t pi2) if t < m/2

(m-t)/(m2 pi2) else=E[algo discovery time | t] = 1/ pi2 m

i=t+1m

i=t+1m

i=t+1m

i=111If we assign a probility to each slot, in other words we select channels i with prob p_i we can derive the following optimization problem.Fortunately, we can solve it analytically and we end up with rho...Note that in this case, the adversary can choose any t, we dont care...How does this perform in practice?

Optimization problem

min * s.t.

t = 01/ * = pi2

1 t m/21/ * = 2 t pi2 t > m/21/ * = m2 pi2 /(m-t)Optimal Algorithm?* = (log2 m)i = 1i = t+1i = t+1mmm

can choose any tGeneral algorithmGiven probability distribution p, where p1 p2 pm 0In each time slot choose channel i with probability pi12If we assign a probility to each slot, in other words we select channels i with prob p_i we can derive the following optimization problem.Fortunately, we can solve it analytically and we end up with rho...Note that in this case, the adversary can choose any t, we dont care...How does this perform in practice?

Simulations: Worst Case Jammer13

13Not assume a worst-case jammer but consider e.g. a microwaveInfluence channels of bt by microwave channels 60-70, happen to be crucial for discovery algo of bt...Simplified bt algo: ...Opt algo...Simulations: Random Jammer14

14Not assume a worst-case jammer but consider e.g. a microwaveInfluence channels of bt by microwave channels 60-70, happen to be crucial for discovery algo of bt...Simplified bt algo: ...Opt algo...Case Study: Bluetooth vs Microwave

Bluetooth ChannelPower (dBm)15

OPT much better than Bluetooth15Not assume a worst-case jammer but consider e.g. a microwaveInfluence channels of bt by microwave channels 60-70, happen to be crucial for discovery algo of bt...Simplified bt algo: ...Opt algo...Yvonne Anne Pignolet @ IMAGINE 2009Lessons Interference can prevent discovery

uniformly random algorithm not always best solution

best expected discovery time

price for NOT knowing t: * = (log2 m) E[Algoopt] = O(log2 m) if t=0O(t log2 m)if t < m/2O(m2 log2 m /(m-t))else

channelschannels

ID5872

ID242816I hope this suffices as evidence that 1) interference can degrade the operation of wireless network2) Easiest is not always best and3) The price for not knowing t is

Yvonne Anne Pignolet @ DCOSS 2009Thats itTHANK YOU!1717