Loading in 5 sec....

Robust Statistical Methods for Securing Wireless Localization in Sensor NetworksPowerPoint Presentation

Robust Statistical Methods for Securing Wireless Localization in Sensor Networks

Download Presentation

Robust Statistical Methods for Securing Wireless Localization in Sensor Networks

Loading in 2 Seconds...

- 325 Views
- Updated On :
- Presentation posted in: Sports / GamesEducation / CareerFashion / BeautyGraphics / DesignNews / Politics

- Zang Li, Wade Trappe, Yanyong Zhang, Badri Nath Presented By : Vipul Gupta Robust Statistical Methods for Securing Wireless Localization in Sensor Networks Outline Introduction and Motivation Related Work Robust Triangulation Robust Fitting: Least Median of Squares

Robust Statistical Methods for Securing Wireless Localization in Sensor Networks

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.

- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -

- Zang Li, Wade Trappe, Yanyong Zhang, Badri Nath

Presented By: Vipul Gupta

- Introduction and Motivation
- Related Work
- Robust Triangulation
- Robust Fitting: Least Median of Squares
- Robust Localization with LMS
- Simulation and Results

- Switched LS-LMS Localization Scheme

- What is Localization (w.r.t. sensor networks)?
- Is the process of estimating the location of a sensor node w.r.t. a known location (also called anchor node)

- Enforcing location aware security policies (e.g. this entity should remain in this building only - laptop), emergencies (e.g. where did the fire alarm go off?)

- Methods of obtaining estimate location information about a sensor node (e.g. DV – Hop, APIT, Cricket)

Anchor Node

d

Sensor Node

Anchor Node

- Threat to Localization Infrastructure
- Purpose of the attacks
- To give false location information.

- Purpose of the attacks

- May be intentional
- Non – cryptographic attacks
- Classical security threats (e.g. Sybil attack)

- Presence of passerby, opening doors of hallway

Sensor Node (True)

Sensor Node

- Single defense mechanism will not work!
- Unforeseen and non-filterable attacks
- Localization should function properly at all times!
- Living with the bad guys!

- Two main localization techniques:
- Range – based localization (more accurate)
- Measurement of absolute point to point distance estimate (or angle)

- Range – based localization (more accurate)
- Range – free localization (no special hardware)

- Time of Flight (e.g. Cricket)
- Angle of Arrival (e.g. APS)

- Hop Count (e.g. DV-Hop)
- Region Inclusion (e.g. APIT)

Anchor Node

d

Sensor Node

Anchor Node

Sensor Node

Anchor Node

Anchor Node

- Cricket
- Time of Flight (Time difference of Arrival)
- Using RF and Ultrasonic Waves
- Utilizes the difference in propagation speeds

- Time of Flight (Time difference of Arrival)
- Pure RF – based system not used! (Why?)
- Difference between the receipt of first bit of RF and ultrasound signals
- Distance = Speed * Time
- For constant speeds, greater the distance, longer the signal takes

Signal 1: T seconds

Signal 2: >T seconds

TRF TUS

- WhereTRF is the time at which the RF signal is received
- TUS is the time at which the Ultrasonic signal is received
- Δ =TRF – TUS ; is the time difference
- Speed * Time = Distance

- Remove direct path & force radio transmission to employ multipath
- Exploit difference in propagation speeds

RF Signal reaches sensor node, nearby adversary hears it

True Ultrasonic signal on its way

Sends ultrasonic signal

Adversary

- Make the signal to pass through another medium
- Speed gets affected and hence the distance estimate

Signal

Sensor node

Another medium

- Ad Hoc Positioning System (APS)
- Uses Angle of Arrival
- Use of directional antennas

Reflective Object

- Use of reflective objects to change the signal arrival angle
- Remove direct path & force radio transmission to employ multipath

Signal

Angle of arrival changes

Reflective Object

- DV – Hop
- Three stages –
- Calculate distance in hops to anchor nodes (using beacons)
- An anchor node calculates distance to other anchor nodes
- Correction (average per hop distance) is calculated for each anchor node and deployed to the nodes
i ≠ j – for all anchor nodes j

- Three stages –

- Example

- Vary hop count:
- Wormhole
- Jamming
- Varying the radio range

- Vary the per-hop distance

Wormhole and Jamming

- APIT (Approximate Point-in-Triangulation Test)
- Uses area-based (Region Inclusion) estimation
- Environment divided into triangular regions
- PIT test narrows the location of the node
- Calculated the Center of Gravity of the narrowed region

- Alter neighborhood
- Wormholes
- Jamming
- Changing the shape of the received radio region
- Placing an absorbing barrier

- According to Wikipedia, is used to model the numerical data obtained from observations by adjusting the parameters of the model so as to get an optimal fit for the data.
- Optimal fit – Sumof squared residuals having least value
- Residue – Difference between the observed value and the value given by the model
- Has its own shortcomings, which we will see soon

(Xa, ya)

da

- Triangulation & Trilateration
- Collecting (x, y, d) values for each node
- (x, y) coordinates of the anchor node
- d is the distance to the anchor node
- Using sufficient (xi, yi, di) solving for (x0, y0) is a simple least squares problem

(x0, y0)

db

dc

(Xc, yc)

(Xb, yb)

- Non-robustness to outliers
- A single incorrect (x, y, d) value may deviate the location estimate significantly away from the true value in spite of other correct values being present
- e.g. altering hop count using wormhole or jamming attacks may deviate d significantly from its original value
- Let 10 samples values of ‘d’ be – 8, 9, 10, 11, 8, 9, 10, 11, 9, 10;
However if an attacker changes one ’10’ to ‘100’, it will significantly affect the location measurement

- Fitting: Finding the best fitting curve for a given set of points
- Cost Function for LS algorithm (in this case) is given by:
- where d is the parameter to be estimated (distance), is the i-th measured distance, xi and yi are the coordinates of the i-th location and x0 and y0 are the coordinates of the true location
- A single outlier may ruin the estimation due to the summation in the cost function

- Under ideal conditions (no attacks), the device location can be estimated by …..(A)
- value of the argument for which the value of the expression attains its minimum value

….(B)

- Equation A is a nonlinear least squares problem and is equivalent to solving:
- Averaging the left and right sides:

- Subtracting the last two equations …
- which is a linear LS problem

- Linear LS has less computational complexity
- Starting with a linear estimate can avoid local minimum
Linear LS and nonlinear LS starting from the linear estimate

- Contamination Ratio Є< 50%, the fraction of distance measurements compromised
- Coordinated corruption of data rather than random perturbations
- Adversary tries to modify NЄ values so that they all “vote” for (xa, ya)
(xa, ya)

(x0,y0)

Greater the da, stronger is the attack

da

- Linear LS used
- mean square error of an estimator (quantity to be measured), according to wikipedia is:
- In simple words, it is the estimation error, i.e. how much the experimental value differs from the mathematical value
- Experiments conducted with different contamination ratio Є and measurement noise level
- Implemented system robust to 30 percent contamination

- Each point represents average over 2000 trials

- Impacts of Є and :
- Severe performance degradation observed at Є = .35

For 50 samples: x = 31… 50 represents outliers

y represents values

- Inliers and outliers well separated – LMS performs good
- Inliers and outliers pretty close, LMS cannot differentiate and messes up – fits partly inlier and partly outlier data giving a worse estimate
- A threshold T is selected and is compared with where is the observed noise level and normal measurement noise level is known
- If T < LMS is used, else LS

- Multiple anchor points deployed
- Signal strengths at each anchor point recorded as {x, y, ss1,…ssN} where ss are the corresponding signal strengths; x,y is the position, N is number of anchor nodes (at least 3)
- Beacons are broadcasted and signal strengths measured at each anchor node
- The signal strengths ss’ (observed) are compared with the ones recorded by the central anchor node
- The closest match is selected as the estimated location (minimum value of )

- A single corrupted signal strength at an anchor node will affect the location. This can be easily done by:
- Using an absorbing barrier between the node and anchor node
- Turning a microwave on

- Finding a correct estimate of the location is important
- Adversaries will always be there, so live in harmony – rather than trying to eliminate all the attacks, tolerate them
- Both LS and LMS have their pros and cons
- Switched LS-LMS does the trick!
- Median based distance metric is good for RF based fingerprinting

- LS-LMS scheme fails when the contamination ratio increases more than 50%
- For large number of compromised nodes, median may be far different from the average value

- Limited attacker capabilities considered. That is, the attacker can compromise only a limited number of percentage of nodes.
- Errors caused by malicious users considered. They have not considered errors caused due to limitations of ranging methods like signal attenuation, multipath signals, etc.

Thank You !!