The Lighthouse Location System for Smart Dust Kate Hayes
Dec 18, 2015
The Lighthouse Location System for
Smart DustKate Hayes
Paper’s Topics
What is Smart Dust
How the authors modified it for their own use
What is the Lighthouse Location System
How it works
Problems
Potential fixes and limitations
Conclusion
http://i.imgur.com/1YUEGhl.jpg
Wireless Sensor Networks (WSN)
Designed to fulfill complex modeling tasks
“Consists of a large number of cooperating small-scale nodes capable of limited computation, wireless communication, and sensing.”
Some areas of use:
Geophysical monitoring
Precision agriculture
Habitat monitoring
Military monitoring
Various business processes
WSN rely on emergent behavior which is where the behavior of the group is much more complex than the sum of its parts
Data fusion
WSN rely on emergent behavior and this is facilitated by data fusion
“the process of correlating individual sensor readings originating from various nodes into a high-level sensing result”
Localization of individual nodes is important because of the need to fuse their data
http://ccri.squarespace.com/storage/Data_Fusion_v3.jpg?__SQUARESPACE_CACHEVERSION=1252336688720
Smart Dust Sensors
Modern nodes are not too different from our smartphones with Wi-Fi
Berkeley envisioned a much smaller specific type of data collecting node called Smart Dust
They are roughly a cubic millimeter in size
Inexpensive and easy to deploy
Low complexity of circuits
Require use and line-of-sight of a base station transceiver (BST)
http://www.eecs.berkeley.edu/XRG/Summary/Old.summaries/
03abstracts/warneke.1.fig.4.jpg
Smart Dust components
Small battery
Solar cell
Power capacitor
Sensors
Processing unit
Corner-cube reflector (CCR)
http://farside.ph.utexas.edu/teaching/302l/lectures/img1347.png
Smart Dust Properties
Small size – Current RF and US transducers are too large
Mobility – Move in environment via wind and air currents
Large Scale – Small size and low costs allow many to be made and deployed
Limited Energy – Power consumption of RF and US technology is too large
Limited Computing and Memory Resources – small size limits the amount of circuitry available for processing and storing data
Single Hop Network Topology – nodes do not cooperate with their neighboring nodes unlike multiple hop networks
No external infrastructure besides the base station
http://www.sveosvemu.com/wp-content/uploads/tdomf/16461/vetar.jpg
Localization Challenges for Smart Dust The Base station must know its location exactly and
nodes localize to its coordinate system
The accuracy needed from the network is determined by what is being sensed
Dust nodes do filtering and basic processing of data onboard so as to save energy with communication to other nodes
Instead they communicated only to the base station
The costs involved are special, capitol, and time, all have to be taken into account when designing the system
Nodes must know their own location!
Lighthouse Location System
Because the nodes must know their own location for WSNs, different ways have been suggested
This paper focuses on the lighthouse location system
It uses a cylindrical approach instead of the more traditional spherical scanning pattern
It relies on a parallel beam transceiver that rotates around the cylindrical base (Like a lighthouse, hence, the name)
Nodes are individually calibrated to the beam
Ideal Lighthouse Location
A perfectly parallel beam that sweeps a perfect circle with no errors
Three cylinders are mounted in the corner of a cube perpendicularly to allow a parallel beam to sweep in every direction
Using the relatively simplistic mathematics described in the paper, the location of the node can be determined
Location from measuring the time between sweeps and how long a sweep takes
The node determines its location from solving a equation system of the different distances involved
Realistic Lighthouse location
Perfect parallel beams are difficult to produce, there will usually be spread which can lead to large errors at tens of meters away
Only the edges of a beam are required for the measurement so two lasers mounted parallel are used instead.
This allows for less beam spread
Rotating 45 deg. Mirrors are used in order to get the upper half of the cylinder as well as more than one plane
The math to discover the calibration and distances can be found and fairly simply followed in the paper
Images of the Lighthouse System
Nodes in the Lighthouse System
Light is received from the BST via a photodiode
A high pass filter is used to filter out ambient light from the sun and bulbs
An interrupt is triggered when the voltage goes high or low, accomplished by a Schmitt Trigger
There is a Linux processor processing the voltages as spikes over time in order to calculate the different times required to calculate the distance and position
Outliers from nodal movements during the beam passing are rejected from calibrations
Lighthouse Base Transceivers
Bases MUST be mutually perpendicular for the math and angles to work out correctly
The offset in position of the bases must be known
Once placed in the field they will be calibrated
Inherent errors in the system
Mirror vibrations from fast rotation of the base
Time of mirror rotation is limited to a resolution correlating to the speed of rotation
The rotation platform may flutter and cause slight jiggles in the beam
Hardware delays from slow or lagging circuitry and/or processing
Clock resolution limits how fast the beam can rotate without being inaccurate
The clock can drift
Errors are dominated by vibration, time of mirror and flutter.
Conclusion
Base Stations sweep out parallel lasers in all directions in order to allow nodes to calibrate their location and send information back to the system
Smart Dust nodes are tiny cubes a millimeter on each side, tiny compared to other nodes
This size imposes restrictions but also brings about new possibilities
Circuitry and processing on board the nodes is performed simply and quickly to save resources
There are many potential uses to a network of tiny line-of-sight sensors
Can you think of any?
StarDust: A Flexible Architecture for
Passive Localization in Wireless Sensor
NetworksKate Hayes
Paper’s Topics
Wireless Sensor Networks
State of current WSNs StarDust Network Implementation tests Optimization Techniques Results Conclusion http://apod.nasa.gov/apod/image/0711/M45WF_hallas_r800.jpg
Wireless Sensor Networks (WSN)
Envisioned to revolutionize the way humans and machines interact and observe their environments
This paper covers the specific type of WSN that are dropped aerially and sensors are embedded in the environment for recording
Sensor nodes form a network and collaborate to get the sensing job completed
Uses include:
Habitat monitoring
Structural integrity monitoring
Military Surveillance
State of Current WSN Technology
No universally accepted localization problem solution
Range-based technology Uses ranges of different nodes from each other or a base station to determine location
Often accomplished with GPS (expensive, heavy)
Time-of-Flight
Time-Difference-of-Arrival
Or Radio transmitters (large energy expenditure)
Radio Interferometry
RSSI
Range-free technology Sensors use primarily connectivity information to infer proximity to sets of anchors
Centroid localization- distance to center beacon/anchor
APIT- being inside or outside a triangle produced by beacons
DV-hop- uses hop to hop propagation
Spotlight- well controlled events that nodes can use to determine location
Lighthouse Location System- Parallel beams are used to measure distances
StarDust Localization Model
StarDust is a Range-free solution to the localization problem of WSNs
Designed after the universe containing luminous bodies that reflect back light
Rather than having the SensorBalls emit light to be captured they reflect it using a passive optical element
Basically a bunch of SensorBalls are aerially dropped and a flash of light and picture is taken and the light reflected back is analyzed for distance from base
The distance of each individual node is sent to them so they can do their sensing tasks
http://ffden-2.phys.uaf.edu/211_fall2013.web.dir/taylor_hanley/taylor_hanley/Project%20template%202/
project/project%20pics/wide3.jpg
Corner Cube Reflector (CCR)
The angle of incoming light is not important
CCRs reflect light back in exactly the same way it came in
Due to unique design of three mutually perpendicular mirrors
In StarDust model there are many CCRs on a single SensorBall
SensorBalls are designed to be upward orienting with the CCR at the top so as the flashed areal light will always be able to hit the CCRs
https://encrypted-tbn2.gstatic.com/images?q=tbn:ANd9GcSlWPFTsZQXAg1SnMTPSji9Ls20UhCE_pdchP
bvTfFVdpxX9VEy4w
http://www.angelfire.com/moon2/xpascal/MoonHoax/ApolloReflectors/CornerCube.jpg
StarDust System Architecture
Light Emitter- Strobe light produces very intense, non-monochromatic collimated light pulses represented by the spectral density ψ
Transfer Function- A bandpass filter for the incident light on the CCR, is also called the color of the node
Image Processing- Collects light and determines where the nodes are located, but not which node is which
Node ID Matching- Uses locations detected by image processing and uses Probabilistic label relaxation to determine which node is which
Radio Model- Acts as an aid to Node ID matching by providing an estimate of the radio range of nodes within range
Image Processing
“The goal of the Image Processing Algorithm (IPA) is to identify the location of the nodes and their color”
It does not identify which node fell where, but only the location of nodes
Records two pictures, one in which the deployment area is illuminated and one not illuminated
The difference between the dark and light is found to create a filter
This filter image then goes through several transformations to remove features that were present in both the light and dark image
In order to identify the elements in the filter that are the reflected light, an intensity filter is applied to the filtered image creating a grayscale image
From the grayscale it is fairly easy to determine that the brightest objects are the SensorBalls
Once the nodes are identified, an edge detection program on Matlab is run so that the centroid can be computed to get the exact location of the nodes
Node ID Matching
Node ID Matching’s goal is to match each bright spot calculated with the IPA with an actual specific node
Node IDs are also referred to as the node’s labels
The problem of labeling a specific node located on the image processed grid with a label is modeled with a technique called probabilistic label relaxation
The main idea is to iteratively compute the probability a label belongs with a specific node using the support for different labels
StarDust uses four types of label relaxation support constraints
Color constraints
Connectivity Constraints
Time Constraints
Space Constraints
Probabilistic Label Relaxation [1]
Often used for solution of simultaneous nonlinear equations
Features such as edges, points, or surfaces belong to a set of labels and an object
Label schemes tend to be probabilistic in nature
Weights or probabilities are assigned to each label in the set giving an estimate of the likelihood that the particular label is the correct one for that feature
The individual probabilities are then iterated through many times taking using a probabilistic approach taking into account neighboring probabilities until they converge or fail to converge
When they fail to converge the user is left with the probability that the feature has a certain label
Relaxation with Color Constraint
The mapping between a sensor node’s position and a label can be obtained by assigning a unique color © assigned to each node
The IPA can determine color
Obviously this is limited to the number of colored CCRs that can be obtained
If C=0 no specific node can be identified using just this constraint
If C>1 a color is assigned to specific nodes giving them the status of “anchor” node http://www.w3schools.com/tags/colormap.gif
Relaxation with Connectivity Constraint
Connectivity information between the nodes can be obtained through the network through beaconing and assist in labeling the nodes
After deployment there is set of beaconing “Hello” messages sent to each node and from these messages the node builds a table of its neighbor’s information
Each node sends back its neighbor table to the central device
Each node is assigned every possible label with an initial probability
The neighbor tables are used help iterate through every possibility using the relaxation technique
These probabilities are iteratively updated when the consideration of their interaction with radio range is taken into account for large scale networks
Relaxation with Time Constraint
Time constraints can be treated similar to color restraints
The most simplistic case is for one SensorBall to be dropped at a time
The IPA is run and the one new flash of light is obviously the ball just deployed
It is too impractically in terms of time for large or medium scaled networks so it is unlikely just a time constraint can be used as a localization technique
http://www.freerangekids.com/wordpress/wp-content/uploads/
2014/06/clock.jpg
Relaxation with Space Constraint
Information about the space the node is dropped in compared to other nodes is another constraint
There is the location of the node and location of the label (where the node was launched vs. where it landed)
At the exact time of release these two locations are identical
If the most simplistic model of physics was used it would be fairly simple to calculate where to the node landed
Instead wind and other conditions need to be accounted for
It is complex but can be done
The spatial constraint is achieved by recursively assigning the probability a node has a certain label using the distances between the location of a node with multiple nodal locations
The nearest label is not always correct, it is dependent on drop and environment conditions
The more space between drops, the higher the accuracy of the method
http://cache1.asset-cache.net/gc/52154971-philippine-air-force-attack-helicopters-take-
gettyimages.jpg?v=1&c=IWSAsset&k=2&d=gFr7L5pk2CL67N1wgtw
1cP6nYokJhrfw33ewrv68Xtg%3D
Relaxation Techniques Analysis
Energy consumed is the overhead
Network Size is the scale
N = number of nodes
ε_d = energy spent for one areal drop
ε_b = energy spent in the network for collecting and reporting neighbor information
T_d = time taken by a sensor node to reach the ground
Testing Tests were carried out with various
aspects of the entire StarDust localization scheme being tested
Image Processing Test
Node ID testing
Radio Model
Localization error vs. coloring space size
Localization error vs. color uniqueness
Localization error vs. connectivity
Image processing algorithm vs Localization test
Localization Time of different relaxation tests
Image Processing Test
In the pictures in the above slide there are 6 sensor nodes
Different sets of pictures were taken with different angles and zoom of the camera
These images were processed according to the IPA mentioned earlier using Matlab
Nodal ID Radio Model Test
Average number of beacons is procured for low and high connectivity networks
Low connectivity has half the amount of beacons as high connectivity
Results are in good agreement with the predicted radio model
Nodal ID Localization Error vs. Coloring Space Size Test
The effect of the number of colors on localization accuracy is tested
Colors are randomly assigned to the sensor node
The location algorithm is run for three different ranges of distance
Conclusion: A larger number of colors available significantly decreases localization error
Nodal ID Localization Error vs. Color Uniqueness
A unique color gives a node the state of “anchor”
The anchor can easily and accurately be identified throughout the image processing process
Color amounts were fixed (4, 6, or 8) and the number of nodes assigned unique colors varied from 0 - max #nodes
Localization accuracy does increase the more colors are available
The localization accuracy decreases with the amount of specific nodes that are assigned unique colors
Nodal ID Localization Error vs. Connectivity
A low and high connectivity network were once again used
The number of colors available was varied and there were no anchors
In both situations localization error decreased with an increase in the number of colors as expected
Localization Error vs. Image Processing Error from the Nodal ID matching
component was examined and now the error from the image processing module will be examine
Luminous objects (sunlight, reflections, streetlights, cars, etc.) can be mistaken as nodes and are called false positives
The bigger problem is false negatives which is when the sensor nodes fail to reflect back enough to be detected
The localization algorithm was run with a percentage of false negatives induced to see the effect on the localization error
As expected the localization error goes up with the number of false negatives recorded
Localization Time
Duration of the localization of the nodes based on the different techniques or combinations of them
It is assumed 50 unique colored filters can be manufactured
Both the connectivity restrained and time constrained techniques increase linearly with the network size
System Range
The realities of physically dropping and recording the nodes is examined
The range of the localization of the system should obviously decrease in worsening atmospheric conditions
Light scattering limits the visibility range by redirecting the luminance of the source and reducing the apparent contrast (C) between target node and the background (r)
When C reaches its lower limit no increase in source luminance or receiver sensitivity can improve the system range
System performance is drastically reduced in hazy atmospheric conditions
Conclusions
Four primitives for constraint based relaxation algorithms were proposed:
Color, connectivity, time, and space
Interesting research directions could be to implement more than one or two constrain algorithms at a time or employ a voting scheme
Labeling the nodes is not highly accurate, the algorithm sometimes fails to converge
In the future it might be possible to get readings in the environment and use those events to help with labeling
It is possible StarDust can be used for rugged terrain and dense foliage
The location readings would be taken before the sensors disappeared from view under the plants, but they would need self-righting capabilities in the air
StarDust solves the localization problem for areal deployment where passiveness, low cost, small size, and rapid localization is required
References
[1]- http://www.cs.cf.ac.uk/Dave/Vision_lecture/node43.html