Advancing Wireless Link Signatures for Location Distinction

Advancing Wireless Link Signatures for Location

Distinction

J. Zhang, M. H. Firooz, N. Patwari, S. K. Kasera

MobiCom’ 08Presenter: Yuan Song

Introduction What we want to do: Detecting whether a transmitter is changing its

location or not. Unlike localization or location estimation,

location distinction does not attempt to determine where a transmitter is

Useful in many applications, especially it can enforce physical security by identifying illegal transmitter

Methodology Basic Method Estimating Channel Impulse Response (CIR), called Signature

of the channel, to checking whether multipath channel is changing or not.

Multipath CIR (time-variant channel)

Methodology (cont.) Multipath CIR (time-invariant channel)

Sending s(t), received signal r(t) will be

In frequency domain

Methodology (cont.)Estimating methods Two methods (used in existing papers related to location disti

nction and this paper)

1 2

Both need S(f) to be known in receiver

is nearly constant within the band2( )sP S f

Previous Work in Location Distinction1 Multi Tone Probing

Signature K carrier waves are simultaneously transmitted to the receiver

fk is separated by an amount greater than Channel Coherence Bandwidth, and thus each carrier wave is attenuated by the channel complex gain.

Signature of the Channel (nth recorded signature), in frequency domain

Based on the method

Previous Work in Location Distinction (Cont.)

1 Multi Tone Probing

Metric (The paper also proposes a slightly modified version for enhancing detection stability)

the Nth multiple tone signature h(N) is compared with each previously measured signature in the history Hi,j using a measure called the correlation statistic

T(n) is the correlation of the nth and the Nth measurements

Previous Work in Location Distinction 2 Temporal CIR Signatures

Signature Using sampled CIR, in time domain

Based on the method

Previous Work in Location Distinction (Cont.)

2 Temporal CIR Signatures

Metric the difference between the Nth signature h(N) and those in the histor

y Hi,j is given as the minimum normalized Euclidean distance between the new signature and any signature in the history set.

Comparison Between Two Previous Work Cons of Multi Tone Probing the channel frequency response is sensitive to each multipath. An impulse in

the time domain is a constant in the frequency domain, and thus a change to a single path may change the entire multiple tone link signature.

Temporal CIR signature use a time domain signature, and thus are more robust against small changes of channel.

Cons of Temporal CIR Signature Lack of phase information limits its ability in uniquely identifying links.

This paper addressed these two cons and made improvement. It proposes a new signature called complex temporal link signature

Complex Temporal Signature Signature Slight modification to Temporal CIR Signature, “without taking the m

agnitude of each gain”

Contrast to Signature of Temporal CIR Signature

Special issues: Difficult to discriminate between channel response phase and oscillator drift, and thus some phase changes in the link signature have nothing to do with any changes in the link.

Solution gived.

Complex Temporal Signature Metric (same as Temporal CIR Signature)

Performance EvaluationFramework of Location Distinction

1 For a given transmitter i and a receiver j, a history of N-1 link signatures is measured and stored

2 The Nth signature h(N) at j from an unknown transmitter in the neighborhood of j is then taken, and an evaluation criterion ei,j = sigEval(h(N); Hi,j ) is computed.

3 ei,j is compared to a threshold 4 When ei,j does not satisfy certain condional relationship with threshold, the new signature is determined to be from the same transmitter, i.e. h(N) = h(N)i,j , and we include it in history of H. For constant memory usage, the oldest measurement in H is then discarded. The algorithm returns to step 2 until enough measurements have been collected

Performance EvaluationFramework of Location Distinction (Cont.)

5 Final Step (Principles used to compare the algorithms) We first define the null and alternate hypotheses:

Then we treat ei,j as a random variable and define

Performance EvaluationMultiple Tone and Temporal CIR

Performance Evaluation Multiple Tone, Temporal CIR and Complex Temporal CIR

Temporal Behavior of Link SignaturesIntroduction

The multipath characteristics of a link can change with time. A link can thus be in different distinct states.

A location distinction mechanism that does not consider the temporal changes in link behavior can significantly increase the probability of false alarms.

The paper propose a Markov Model to further decrease the probability of false alarm

Temporal Behavior of Link SignaturesIntuition 1

Using non-linear dimensionality reduction (a method in Statistics to visualize high dimensional data) to reduce the 100 dimension vectors to just 1-2 dimensions.

Below is a 2-D embedding plotted with one set of 333 complex link signature measurements.

Temporal Behavior of Link SignaturesIntuition

Two States With 1D embedding of Isomap algorithm, from the total number of state changes, and

the number of times we are in a state, we calculate the state transition probabilities and the limiting probabilities of the Markov chain

Temporal Behavior of Link SignaturesMarkov Model

Performance EvaluationFurther Definitions of Probability of False Alarm

1. Same-State False Alarm (SSFA): A link signature is measured in state i while there exists in the history some other signatures of state i, however, the new measurement is far enough away from the measurements in the history that they are detected as different, thus a false alarm is raised.

2 Different-State False Alarm (DSFA): A link signature is measured in state i, but no signature previously measured in state i exists in the history. Because link signatures from states j <> i are very different from those measured in state i, this new measurement does not match any in the history, and a false alarm is raised.

Performance EvaluationFurther Definitions of Probability of False Alarm (Cont.)

Caculation of P[DSFA]

Performance EvaluationP[DSFA]

The END

Thank You~~