Professional Documents
Culture Documents
Syed Zain Ali
Syed Zain Ali
NETWORK
Submitted By:
Supervised By:
MAZHAR ISLAM
Department of
Telecommunication
DEPARTMENT OF TELECOMMUNICATION
2021
FINAL APPROVAL
This is to certify that we have read the thesis titled ” REVIEW ON LOCALIZATION
It is our judgment that this project is of sufficient standard to warrant its acceptance by the
Telecommunication Degree.
ii
COMMITTEE
Supervisor:
External Examiner:
Internal Examiner:
Head of Department:
iii
DECLARATION
We Mr. SYED ZAIN ALI SHAH and Mr. BILAL TANVEER hereby declare that we have
produced the work presented in this thesis, during the scheduled period of study. We also
declare that we have not taken any material from any source. It is further declared that we
have developed this project and the accompanied report entirely on the basis of our
personal efforts made under the sincere guidance of our supervisor. No portion of the work
presented in this report has been submitted in the support of any other degree or
qualification of this or any other University or Institute of learning, if found we shall stand
responsible.
Signature:
Signature:
iv
DEDICATION
With profound love and respect, this work is dedicated to my loving parents and
teachers
v
ACKNOWLEDGMENT
All praises and thanks to “AL-MIGHTY ALLAH”, the source of knowledge and wisdom
endowed to mankind, who conferred me with the power of mind and capability. The
completion of this thesis is also one of the ALLAH’S innumerable blessing and
opportunities that Allah has showered upon me throughout my life. All respects are for the
valuable guidance, suggestion and kind behavior that helped me a lot in complete of my
research.
I am proud of and indebted to my parents, who provided me all what I needed and
BILAL TANVEER
vi
ABSTRACT
current location of the sensor nodes. A WSN consist of thousands of nodes that make the
installation of GPS on each sensor node expensive and moreover GPS will not provide
reference on each sensor node isalso not possible in the case of dense network.
This gives rise to a problem where the sensor nodes must identify its current location
without using any special hardware like GPS and without the help of manual configuration.
localization techniques are carried out with the help of anchor node or beacon node, which
knows its present location. Based on the location information provided by the anchor node
or beacon node, other nodes localize themselves. In this paper we present a succinct survey
on the localization techniques used in wireless sensor networks covering its problems and
research gap.
vii
Table of Contents
CHAPTER 2 ............................................................................................................................ 16
sensors.Sensors ................................................................................................................. 16
CHAPTER 3 ............................................................................................................................ 22
3.12 Optical Methods, TOF Indirect and Direct, Distance from Lighthouse Method: ......... 40
viii
3.14 Optical Methods, Triangulation, Active: ...................................................................... 44
3.16 Optical Methods, AOA, CCR, MRR, and Photodiode Receivers for Smart Dust Motes:
.................................................................................................................................... 47
CHAPTER 4 ............................................................................................................................ 51
CHAPTER 5 ............................................................................................................................ 55
CONCLUSION ....................................................................................................................... 55
SUMMARY ............................................................................................................................. 59
REFRENCES .......................................................................................................................... 61
ix
Abstract
Spatial localization (co-location) of nodes in wireless sensor networks (WSNs) is an active area
of research, with many applications in sensing from distributed systems such as micro-aerial
vehicles, smart dust sensors, and mobile robotics. This paper provides a comprehensive review
and comparison of recent implementations (commercial and academic) of physical measurement
techniques used in sensor localization, and of the localization algorithms that apply these
measurement techniques. Physical methods for measuring distances and angles between WSN
nodes are reviewed, followed by a comprehensive comparison of localization accuracy,
applicable ranges, node dimensions, and power consumption of the different implementations. A
summary of advantages and disadvantages of each measurement technique is provided along
with a comparison of co-localization methods in WSNs across multiple algorithms and distance
ranges. A discussion of possible improvements to accuracy, range, and power consumption of
selected self-localization methods is included in the concluding discussion. Although the
preferred implementation depends on the application, required accuracy, and range, passive
optical triangulation is reported as the most energy efficient localization method for low-
cost/low-power miniature sensor nodes. It is capable of providing micron-level resolution,
however the applicable range (inter-node distance) is limited to single centimeters.
10
CHAPTER 1
INTRODUCT
ION
WIRELESS sensor networks (WSNs) are quickly becoming an integrated part of our daily life
[1], [2]. It is predicted that within a decade, the number of sensors per person in developed
countries will exceeds one thousand [3]. Advances in low-power electronics [4], [5], [6], as well
as energy harvesting [7], [8] enable these sensors to operate for years without the need to replace
their batteries. At the same time, advances in MEMS, additive manufacturing, and printed
electronics significantly reduce the cost of deploying such sensor networks. Together, these
advances are expected to make wireless sensors ubiquitous in our daily lives. Spatial localization
(co-location) of sensors is an important aspect of wireless sensor networks, as it allows for the
reconstruction of spatially distributed sensory data. Both centralized and distributed (self-)
localization is reviewed in this paper. Self-localization is generally defined as the ability of each
sensor to estimate its own location relatively to other nodes in the network.In the litterature,
localization is also referred to as colocation, self-localization, distributed localization,
cooperative localization, position estimation, and bearing identification. Manuscript received
November 3rd, 2014. N. Iliev and I. Paprotny are with the Department of Electrical and
Computer Engineering, Micromechatronics Systems Laboratory (MSL), University of Illinois,
Chicago, IL, 60645, USA e-mail: paprotny@uic.edu The individual nodes in a low-power WSN
operate at fixed low power budget, and each sensor node must conserve power to extend its life-
time. Design trade-offs must be made [10], [11], for example an extended communications range
will require less hops between the individual nodes, but may consume more power per node.
Centralized message processing at a single node can make for a simple centralized localization
algorithm, however distributed message processing is more flexible, more accurate, and can be
more fault-tolerant. These choices affect the architecture of the WSN, and its operation, and have
to be tailored in each individual case [10]. Similar tradeoffs need to be made with respect to a
self-localization scheme. For example single mode of operation (RF, acoustic, or optical) enables
a simple microsensor architecture, however multi-mode operation (e.g. RF and acoustic or RF
and optical) can both cover larger distances and provide more accurate localization. A diagram
11
of a generic multi-mode WSN is shown in Figure 1. Each implementation dictates a specific
localization algorithm with inherent limitations.
12
profile, and finally on custom radios for IEEE 802.15.4 build from digital ASIC standard
cells technology [18] with power consumption of 15.6 mW at 0 dB output power. The
communication system is used by a sensor node to perform its message-based self-
localization task as well as for broadcasting the basic data collected by the sensor.
Communication systems based on ultrasound waves have not been attempted due to the
directional dependence of acoustic signals and their dependence on air temperature and
humidity. Optical communication between sensors in WSN has been implemented recently
using low-power laser diode emitters and detectors [19], modulating Corner-Cube Reflectors
(CCR) [20], MEMS CCR arrays [21], MEMS steering mirors (gold coated, reflectivity
greater than 98% over a wide range of wavelengths) [22], MEMS actuated spherical
retroreflectors [23], and free space optics (FSO) [24], [25], and [26]. Optical communication
systems show the largest potential for low-power operation and for scaling up with large
numbers of nodes as recent research in FSO has shown [27], [28], and [29]. An important
and often overlooked aspect of a low-power distributed sensor architectures that rely on self-
localization is how distance and/or angle-of-arrival (AOA) measurements to neighboring
sensors are performed with a limited power budget. Recent advances in CMOS photodiode
pixel arrays, MEMS micro-mirrors, and CCRs, have led to potentially lowpower chip-scale
solutions for distance measurement via Timeof-Flight (TOF) [30], via TOF or Optical
Triangulation [31], and a 10 mm × 10 mm array of photodetectors for AOA measurements as
reported in [27].
Most WSNs are designed to detect and report on local events such as e.g. temperature,
humidity or chemical content [32] [33] of the surrounding area. Each sensor node is therefore
required to measure the variables associated with the event and report the measured data to a
central processing node. In many cases the location of each sensor is vital for properly
analyzing all measured data at the central processing node and for implementing data fusion-
based improvements of the location estimate [34] [35]. For example, measured data must be
marked with its physical location to allow for energy-efficient routing of the data within the
WSN and to a gateway [36] [37], to allow for distributed source compression techniques
[38], and for position tracking and movement detection techniques [39]. Therefore, one of the
13
first steps during the power-up of a WSN is to establish the initial position for each sensor.
Typically only few nodes in a WSN have a real-time GPS capability or other approaches to
establish their own location with great precision. In this paper we refer to these nodes as
anchors or beacons. Cost, size, and energy-consumption of such precise localization methods
dictate that a WSN has a small number of anchors. The locations of the remaining non-
anchor nodes have to be estimated using centralized or distributed methods. Figure 2 shows
an example WSN in which one node (unknown location) must self-localize using two AOAs
to two known anchors R1 and R2. If the (x,y) coordinates of the anchors and the AOA values
are broadcast to the unknown node, it can compute its own location using trigonometric
relationships and a maximum-likelihood algorithm to account for noisy AOA measurements
[40]. The three nodes can co-operate to measure the AOA values φ1 and φ2. Having
localized itself, the new node broadcasts its coordinates to the rest of the network and to the
server. Alternatively, in a centralized algorithm, the unknown sensor can measure its
distances to three nearest known neighbors (anchors) and broadcast these measurements to
the server. The coordinates will be computed by the server and then relayed back to the node.
14
This project provides a comprehensive review of algorithms and physical implementations
for self-localization for WSNs, and classifies them based on the applicable distance range in
which the self-localization is performed. The localization accuracy and power consumption
of the candidate self-localization schemes is also classified for comparable WSN sizes, as
well as comparable number of anchors (beacons) if applicable. We present a taxonomy of
localization implementations. At the algorithmic level, we distinguish between centralized
and distributed (self-localizing) groups of algorithms. Each group includes range-based or
range-free implementations as well as anchor-free or anchor-based implementations. We
further classify an implementation based on the choice of physical methods (or physical
distance/angle measurement modalities): RF, ultrasound, optical, or hybrid. This taxonomy is
described in the following sections: Section II examines several recent centralized and
distributed (self-localizing) algorithms for WSNs. The mathematical models and resulting
algorithms are summarized as reported from actual implementations or simulations in the
surveyed works. Section III presents physical methods used in recent sensor architectures for
computing range(distance) and AOA metrics between pairs of sensors which are then used as
input to all localization computations. Section IV compares the accuracy of estimated node
locations 1530-437X (c) 2015 IEEE. Personal use is permitted, but
republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more
information. This article has been accepted for publication in a future issue of this journal,
but has not been fully edited. Content may change prior to final publication. Citation
information: DOI 10.1109/JSEN.2015.2450742, IEEE Sensors Journal SUBMITTED TO
THE IEEE SENSORS JOURNAL 3 and applicable region (area) of localization algorithms
used with a physical method in an implementation. Comparisons of power consumption for
recently proposed localization techniques as reported in the literature is also presented.
Section V presents our conclusions based on the reviews of published performance results
and outline areas of future research
15
CHAPTER 2
ALGORITHMS FOR LOCALIZATION IN WSNS
Centralized localization algorithms run on a server (base station) and all sensors in the WSN
must forward their measurement data to the central server. These measurements are then used
by the server to compute the location of each sensor [44],[41],[45]. In the centralized
versions of range-based and range-free algorithms, all pairwise distance (or connectivity)
data are sent to a central node for calculating position estimates, followed by the transmission
of each computed position to its corresponding sensor [46],[47],[48],[49],[50]. Such central
processing has the advantage of ample CPU and memory resources at the central node, which
allows for more complex localization algorithms. However, centralized processing has
typically high communications overhead, therefore high energy consumption. Centralized
localization also scales poorly and is not fault tolerant since all messages required for
centralized localization must flow through the gateway node. Multidimensional scaling
(MDS) is an example of such a centralized algorithm [42], [41], [43]. In particular [41] is an
important first application of MDS to sensor networks that lack anchors and therefore require
relative node localization. In addition the algorithm is range-free and based only on
connectivity information for all nodes in the network. The advantage of this centralized
approach is that one can design an algorithm that has more precision in estimating locations
and contains a comprehensive map of the entire WSN stored in one node. Some of the early
localization algorithms that employ centralized localization are [49], [50], [44], [46].
The following list classifies centralized localization depending on the type of intra-node
measurements:
16
• Convex optimization - uses connectivity constraints : range-free, AOA-free [44]
• Multidimensional scaling (MDS) - uses distances between each pair of nodes [41]
• Maximum likelihood estimation (MLE) - uses time-ofarrival (TOA) and AOA between
anchors and unknown nodes [45] ; can also use RSSI derived distance between anchors and
unknowns [46]
• Linear programming - uses 1) distance between each pair of nodes and anchors, 2)
propagation delays between nodes and anchors, 3) clock biases between nodes and anchors
[47]
• Semi-definite programming (SDP) - uses distance between each anchor-node pair and
upper and lower bounds on these distances
• Isomap - uses distances between all pairs of nodes and an Euclidean metric
• Local linear embedding (LLE) - each sensors coordinates are represented as a weighted
sum of its nearest K neighbors coordinates
• Manifold learning - each sensor records data over a known time interval and sends the data
to its K immediate neighbors. Sensors which are closest have the most similar data
• Supervised learning - uses distances between unknown sensor and K immediate known
neighbors
in distributed localization (self-localization) each sensor node estimates its own location
using available information obtained via message passing, from neighboring nodes (eg.
range, connectivity, and AOA). The computations required for self-localization are typically
performed at the sensor node [53], [54], [55], [56]. T
he inputs to the computations are distances or angles (AOA) as measured between pairs of
neighbor sensors or between sensors and neighbor anchors. These are range-based or AOA-
based self-localization distributed algorithms [56]. When the inputs to the computations are
not distances or angles but only connectivity information about the neighbors (and their
neighbors), the algorithms are considered as range-free and distributed [55]. Both range-
based and rangefree distributed localization algorithms have variations which require three
(or mode) anchors or don’t require anchors at all (anchor free). Distributed processing
17
usually consumes much less energy than centralized processing (or cluster-based centralized
processing) because centralized schemes have to collect relevant information from all nodes
in the WSN network which implies re-transmission in a multi-hop environments. Also, once
the centralized node has computed the location estimate for each node, this information has
to be sent back to the node. Distributed localization algorithms are also called decentralized
[78], pervasive [57] , or collaborative localization algorithms [61]. Most existing WSN
implementations prefer distributed processing so that the computational load is shared, the
computation time is reduced considerably, and the communication bottle-neck is no longer a
problem [69].
1) Classical Multilateration:
Each sensor estimates its multihop range to three (or more) anchor nodes. These ranges
can be estimated via the shortest path between the node and the anchor. The shortest path
algorithm (Dijkstra’s of Floyd’s all pairs) has to be executed in a distributed manner
across the network. When each node has range estimates to three (or more) anchors, its
coordinates can be calculated locally via multilateration [54], [55], [56]. For best results,
the ratio of anchor nodes to the total number of nodes has to be greater than 25% and
even higher. In [55], Niculescu et al. present
18
a seminal work on AOA-based non-Bayesian localiztion. In addition, [56] is an important
early work on non-Bayesian distributed MDS using RSSI and TOA range measurements
2) Successive Refinement:
These non-Bayesian algorithms try to find the maximum (or minimum) of a global cost
function via Least Squares (LS), Weighted LS, or Maximum Likelihood (ML). Each node
estimates its location and then transmits the location to its neighbors. Neighbors must then
recalculate their location and transmit it again, until convergence of the maximization
(minimization) process is achieved. A node starting without any coordinates can begin with
its own local coordinate system and later merge it with neighboring coordinate systems. In
[11], Savvides et al. describe important early work on distributed multilateration. It shows
that collaborating sensors can solve a large non-linear optimization problem using fully
19
distributed computation that none of the nodes can solve individually. Each node requires
distance measurements only to its one-hop neighbors. Other examples include [57] (PPE),
[58] (Bilateration), and [59] (Non-linear non-convex global optimization).
These algorithms are based on the information version of a reduced-order Kalman filter,
running on each sensor. In a Bayesian framework, this information filter (IF) is
20
essentially a Kalman filter expressed in terms of measures of information about the
parameters (state) of interest rather than direct state estimates and associated covariances.
Fisher’s definition of information [65] is used to build the IF with corresponding
information state and information matrix. Gaussian assumptions are made about the
likelihood function and the posterior conditional distribution. Example implementations
are described in [66], in [67] where an extended Kalman filter (EKF) is based on multiple
interacting models and on ultrasound-based localization, and in [68] for an optical
wireless channel. As stated previously, the main shortcoming of this approach is
assumption of Gaussian statistics. In [69], Moore et al. presents important early work on
EKF application in the Cricket Localization system based on TDOA measurements
between ultrasound and RF signals. Unlike previous algorithms, it considers the
possibility of graph flip ambiguities during trilateration due to measurement noise. A flip
ambiguity exists if a vertex can be reflected (fliped) across an edge (formed by two
neighbor nodes) without violating the distance constraints between the vertex and the two
neighbors. Robust quadrilaterals (smallest subgraphs that can be definitely be realized)
are proposed to minimize the probability of flip ambiguities. Distributed Bayesian-based
localization is chosen by most recent implementations due to its relatively low
computational complexity and ability to handle missing intra-node measurements, non-
linearities, and non-Gaussian measurement errors.
21
CHAPTER 3
PHYSICAL METHODS FOR SENSOR LOCALIZATION –
IMPLEMENTATIONS
A sensor localization algorithm can have very different performance characteristics (for example
implementation. This section presents several physical implementation methods used in recent
sensor architectures for computing range(distance) or AOA parameters between pairs of sensors,
An example of architectural layers for a sensor node is shown in Figure 4. The physical layer
(layer 1) is responsible for all communications to nodes which are in radio or acoustic range or in
line-of-sight and for performing all distance and angle measurements to nearby nodes or anchors.
The measurements are used by the location discovery layer (layer 2) along with any available
connectivity data for this node, for example which anchors (or neighbor nodes) with known
coordinates is the node in contact with. In cases when the node is unable to establish contact with
may be instructed to scan the horizon until contact with a neighbor is established. The
application layer (layer 3) ultimately uses a localization algorithm (or several algorithms) to run
at layer 2, for example multilateration [11], or triangulation [34]. In cases when anchors are not
available, layer 2 localizes the sensor-node within a local coordinate system made up of its
neighbors. This local coordinate map is passed to layer 3. Layer 3 then collaborates with other
sensor-nodes to exchange the local coordinate map and stitch a global coordinate map of the
entire network of nodes. Layer 3 typically implements all peer-to-peer protocols in distributed
22
multilateration and coordinate stitching (map reconciliation) algorithms, examples of which are
interferometric and noninterferometric. The accuracy of measured parameters for each physical
23
3.1 Interferometry-Based Methods
Interferometry-based physical methods provide the finest resolution of distance or angle
(AOA) measurements, with distances in the range from sub-micron to a few hundreds of
microns, and angles in the sub-degree range. Such high resolution cannot be achieved
with any other physical methods such as RF, ultrasound, optical triangulation, or time of
flight. The interferometry-based methods include the use of MEMS, MOEMS, and
phase shift analysis. Instead of measuring the distance, the Angle of Arrival (AOA),
reflected light, relative to a reference optical path, as reported in [70], where a MEMS
sensor with sub-nm to 1 micron resolution. It also requires a laser source and has a
measurement range limited to the size of the target device (another MEMS).
24
3.3 AOA via RF interferometry and RF-photonic interferometry:
Angle of Arrival (AOA), between a target and anchor(s) can be measured using a multi-
element antenna array for collecting radio or optical signals. The time difference of
arrival (TDOA) at individual elements of the array is measured and used to estimate the
corresponding AOA
implementation uses Texas Instruments CC1000 radio chips (in XSM motes) to transmit
in the 433 MHz range. Three XSMs form the antenna array. The achieved accuracy of the
interferometer [73] are driven by the microwave signals with various time delayed
25
components, emulating the outputs of a two-element antenna array. The S2 spectral
analyzer monitors the output of the interferometer. The optical power spectra of this
output is sensitive to the time delay of each resolvable frequency component of the input
signal and allows estimation of the emitter’s energy and direction. Using a 3.8 GHz
carrier frequency for the emitters, sub-picosecond time delays have been reported which
An approach which does not use S2 materials is presented in [74]. A two-element array
receives microwave signals at 18 GHz, and each signal modulates a dedicated electro-
optic modulator (Mach Zehnder modulator, MZM). Each MZM outputs an optical
component at the carrier wavelength with the optical power being a function of the phase
shift. A following optical spectrum analyzer is used to find the relationship between
optical power and phase shift. The accuracy in phase shift measurement is reported at +/-
2.5 degree, and the resulting AOA estimation accuracy is +/- 0.07 degree.
26
3.5 RF methods, Distance via RSSI:
Received Signal Strength Indicator (RSSI) methods for estimating the distance between two
nodes (transmitter and receiver) assume a pathloss model [75], [76] between the transmitter
and the receiver. The RSSI value of the received radio packet is used with the path-loss
model to estimate the distance between sender and receiver as reported in [77]. This is easy
to implement in most radio platforms. However RSSI values are strongly affected by noise,
multipath, and line-of-sight obstructions. Fingerprinting is one method to cope with these
issues. A database of RSSI readings is set up beforehand and measured RSSI values are then
compared with database entries for possible matches and thus indicate specific distances,
[78]. The obvious issue with fingerprinting is how to establish the database, especially in ad-
hoc WSN scenarios. Another method to handle the wide fluctuations of RSSI readings is to
27
use stochastic filters [79]. When three reliable distance measurements are obtained to three
can be applied to estimate the sensor’s position as reported in [80]. The reported accuracy is
Recent advances in inertial sensors allow for fusion of inertial data with RSSI-based
localization data to provide more reliable estimates as reported in [79] and in [81]. The
RSSI measurements and centralized or distributed algorithms are typically on the order of a
few meters depending on anchor density and environment conditions. A relatively large area
transmission of an RF pulse and the reception, at the emitter, of an RF echo from the
unknown target. The elapsed time is for a round-trip and is divided by a factor of 2 before
being multiplied by the speed of light (wave propagation velocity). The resulting product is
the measured distance (range) between emitter and target. For accurate results, a narrow RF
beam must be formed at the emitter to concentrate the transmitted energy, the RF transducer
must be matched to the characteristics of the air medium and the receiver (target) must match
the emitter’s characteristics. The emitter is usually an anchor node with a very accurate local
clock used to record the start time of the emission and the reception time of the echo.
Measuring distances from three anchors to an unknown target enables position estimation via
trilateration. However this requires synchronization of the local clocks between the three
28
anchors and the unknown node. This is illustrated in Figure 7 below [82]. The anchors are
yellow, the unknown node is red. Time intervals tarrive − t1, tarrive − t2, and tarrive − t3 can
be measured since all nodes have synchronized clocks. Each time interval is then multiplied
by the speed of light to compute the corresponding distance. Since this method is based on a
time value, the estimated distance value is affected by the time measurement accuracy, and
therefore by the clock accuracy at the anchors. Clock offset and clock drift can corrupt the
In [83], several techniques are presented for mitigating the errors associated with imprecise
clock synchronization. Code modulus synchronization, (CMS), is used to reduce the effects
of noise, imprecise clock synchronization, and sampling artifacts. In CMS, full duplex
ranging is emulated using half duplex radios such as those used in low-power WSNs. The
29
anchor transmits a code and the target receives the code. Then the transceivers are switched
and the target re-transmits the code it received, synchronized to its local clock, back to the
anchor. The anchor then computes a correlation between the code it sent and the code it
received. The measured code offset is the time of flight. The correlation and code offset
estimation are not done in real time, which allows for processing the signals above Nyquist to
obtain a high resolution limited only by the noise of the system. The related implementation
in [84] is a simple, low-cost, software-defined radio node (Waldo) with off-the-shelf parts,
accuracy needed for many localization problems and does not need specialized base stations,
time synchronization, UWB, or other expensive and complex equipment. The power
consumption of an Waldo node is published as 175 mW (in Radio-off mode), 1145 mW (in
Rx mode), and 1440 mW (in Tx mode, 14 dBm of output power). A recent implementation
of two-way TOF distance measurement [85] also uses standard low-cost RF narrow-band
hardware. TOF is derived from Round Trip Time (RTT) of flight. A major advantage of that
method is that no time synchronization of the communicating partners is necessary and it can
carefully analyzing the error contributions of the microprocessor’s clock jitter and the RF
implementation uses a standard MSB-A2 sensor node, with an LPC2387 microcontroller and
a CC1101 radio transceiver. The microcontroller has an ARM7 core and a 72MHz clock. The
CC1101 is driven with a 26MHz clock and uses the 868MHz ISM radio band. The power
30
To avoid the need for synchronizing clocks at multiple anchors, Symmetrical Double-Sided
Two-Way ranging (SDSTWR) was proposed as part of the IEEE 802.15.4a standard. SDS-
implementation, the Nanotron Protect TOF range-finder, [86], achieves accuracies from 1 m
to 3 m
In [87] a two-way TOF ranging implementation is based on the CC2430 SOC from Texas
Instruments with an IEEE802.15.4 compliant radio transceiver. The local clock frequency at
the anchor is 32 MHz. The resulting mean error in TOF distance measurements is about 1.6
m, over a 10 m2 area. This error is due to clock drift, transmitter loss of synchronization, and
constructive and destructive multi-path effects. For better accuracy, Ultra Wide Band (UWB)
systems have been proposed such as the one in [88]. The achievable time precision is in
nanoseconds, for measured times of flight of UWB pulses. This provides for distance
measurement accuracy within a few centimeters. The range of this system is 100 m indoor
and 2 km outdoor. However, although UWB transmitters are simple to implement and
extremely low power, the power consumption of UWB receivers in a TOF system is high.
an alternative is to indirectly derive this time (or traveled distance) via measuring another
signal parameter, such as phase or frequency, which is changed in the reflected version of the
emitted signal. A two-way TOF ranging scheme is presented in [89] which measures the
frequency difference between the transmitted and received signals using narrowband RF. The
frequency difference is then used to obtain subclock TOF phase offset measurement and a
31
corresponding distance estimate. An implementation with a TI CC2430 development kit [90]
implementation for distance measurement via indirect TOF is described in [91], for ranging
via frequency-modulation. The BOSCH Long Range Radar (LRR), uses the beat frequency
(frequency difference between the transmitted and received RF wave) to estimate the
distance to the target. The reported distance measurement accuracy is from 0.1 m to 0.3 m
over a range of 0.5 m to 250 m. 4) RF methods, TOA: Time of Arrival (TOA) methods for
localization require accurate time synchronization between the unknown node (mote) and
one or more anchor nodes [82], [92]. This is in contrast with direct TOF which does not
require such synchronization. In Figure 8 below, anchors are yellow and the mote is red. In
2D plane, at least three anchors are required. The location of the mote is the intersection of
circles centered at each anchor. Each circle represents the radio coverage of the
corresponding node. In 3D space the location is at the intersection of spheres. The transmitter
(mote or anchor) sends a message and records the time of the transmission. The receiver
(anchor or mote) receives the message and records the time of reception. Both time values
are broadcast to a central processor or used by the anchor (distributed) to compute the
propagation delay for the message. The estimated distance is the product of propagation
delay and the speed of light. Typically, after obtaining distances to three anchors, trilateration
32
TOA algorithms require strict time synchronization. In wireless environments, the cost of
time synchronization is very high, and precise synchronization is difficult to achieve. The
desired precision is influenced by the propagation time of the radio signal (message), which
solutions to these problems are based on Chirp Spread Spectrum (CSS) as reported in [93].
The use of chirp signals in communication has been used in commercial range-finding
clocks between anchors and target nodes is not required. The distance range is 30 m with an
output power of +10 dBm. Localization of over 30 targets creates heavy communications
33
traffic with a resulting increase in power consumption.
Time difference of arrival (TDOA) relaxes the timing synchronization requirement between a
mote and the nearest line-of-sight anchors. In TDOA only the anchors have to be
synchronized in time. The node to be localized transmits the signal, and the mote’s
transmission arrives at different anchors at different times. Each anchor records it’s time of
reception. This information is forwarded to a location engine, which calculates the received
signals time difference between each of the anchors. This difference in reception times at the
anchors is typically used in a nonlinear least squares objective function which also includes
the mote’s coordinates. The minimization of the objective function can be done via the
Levenberg-Marquardt (LM) algorithm [59]. The result of the minimization is the best choice
for the mote’s coordinates given the noisy time difference measurements. Mathematically,
the mote is located at the intersection of 3 hyperbolas in a 2D plane, while the mote is located
at the intersection of 3 hyperboloids in 3D space. Figure 9 shows that the center mote is
located at the intersection of 3 hyperbolas in 2D plane, one centered at each anchor. The
corresponding times of arrival are t1, t2, and t3 at anchor 1, anchor 2, and anchor 3.
The transmitter node emits an ultrasonic signal at a given time for a duration of 1 ms. The
receiving node detects the signal and performs the TOF estimation using the known time of
34
Whereas TOA records the time that a transmitter sends a signal to the receivers (anchors),
TDOA requires that the receivers record when the signals were received. Similar to the TOA
method, TDOA requires the clocks of each of the anchor nodes to be synchronized, which
leads to higher cost. But the mote (transmitter) is not synchronized with the anchors, and this
results in simpler hardware than TOA. The precision of the localization result is correlated to
the accuracy of the clocks used in the anchors. TDOA is also affected by multipath
propagation, noise and interference. Using UWB and CSS techniques together with error
correction algorithms can improve the precision of the algorithms, as proposed by [94].
A constrainted least-squares algorithm for solving the objective function was proposed in
35
[95] to further improve the precision. Recent commercial implementations of the TDOA
method include Nanotron [86] which is CSS based with an accuracy of 1 m to 3 m . The
Ubisense Series 7000B [96] is based on a UWB physical layer with an accuracy of 30 cm to
60 cm.
A multi-modal solution, Ubisense Series 7000C, implements both AOA and TDOA methods
Figure 10 below from [82] illustrates the basic AOA measurements needed for localizing the
target node (M) given the locations of two anchors (x1,y1,z1) and (x2,y2,z2). For localization
in a 2D plane AOA angles β1 and β2 have to be measured. This is also shown in Figure 2 in
Sec. I-B above. For localization in 3D space additional angles γ1 and γ2 have to be
1) Ultra-Wideband RF AOA sensors, Ubisense Series 7000A and 7000C, are described in
[82]. The achievable location accuracy is quoted as 30 cm - 2 m (for Series 7000A) and
15 cm - 30 cm (for Series 7000C), the later supplements AOA with TDOA measurements
to achieve better accuracy. The accuracy of the AOA measurements is not listed, only the
resulting location accuracy. Power consumption and applicable ranges are not listed.
2) RSSI is used to estimate AOA angles in [97] using directional antennas such as the
MD24-12 antenna. The angle measurement accuracy is from 0.5 to 3 degrees. The
36
3.10 Ultrasound Methods, TOF Direct:
A direct TOF distance measurement system based on the ultrasonic TELIAMADE node
architecture is reported by [98]. Each node implements the ZigBee protocol (CC2420 radio
chip) for exchanging messages and synchronizing transmission and reception events between
two nodes. Synchronization is based on TDMA scheduled frames between pairs of nodes.
Each TELIAMADE node is also equipped with two low-cost ceramic ultrasonic transducers
show that up to distances of 2 m, the standard deviation of the TOF estimated is 0.92 µsec,
which translates to standard deviation of distance measurements of 0.31 mm. The assumed
speed of sound is 340 m/s. For distances between 2 m and 6 m, the distance measurement has
a standard deviation of 0.64 mm. This sub-cm and sub-mm precision of distance
for the ZigBee module, and a 5 mA for the additional hardware, a total consumption of 30
mA is reported for the TELEMADE node in normal (non-sleep) operation. In sleep mode
37
(not performing measurements) the consumption is reduced to 1.7 mA.
MICAz (MPR2400) board from Crossbow Technology. Nodes exchange messages via an
IEEE 802.15.4 ZigBee radio. Ranging between two nodes consists of the first node sending
an RF packet and emitting an acoustic signal (4 KHz fixed frequency piezoelectric resonator
MICAz MTS300 and MTS310 boards), at the same time, to the second node. When the
second node receives the RF packet, a timer is triggered. When the second node receives the
acoustic signal, it is used to stop the timer. Using the elapsed time and the known
propagation velocity of the acoustic signal, the second node can compute its distance to the
first node. The propagation delay of the RF packet is assumed to be almost instantaneous. A
vary from 20 cm to 2.5 m. Power consumption has not been reported but is assumed to be in
alone [66].
Localization and tracking of mobile nodes using TDOA measurements has been reported by
[67]. An ultrasound localization system estimates the distance between a fixed receiver and a
300 ms, at an ultrasound frequency of 40 KHz. The receiver has a fixed position around the
localization area. The receiver detects the arrival timestamp of the ultrasonic signal and
estimates the distance. A time-synchronization precision, between two nodes, better than 0.1
ms has been achieved using a 802.11 b/g Wi-Fi connection. The resulting maximum error of
distance measurement is 3.46 cm. The applicable rectangular area for measurements is 1.75
38
m by 3.86 m. Power consumption is not reported but is on the order of 10 to 40 mW
depending on the MSP430’s low-power modes and the ultrasound transducers used at each
end. An ultrasound-based TOA measurement system described in [99] uses the derived
distances to localize passive RFID tags. The custom passive RFID tag called a WISP node
implements an 25 KHz acoustic tone detector that receives and multiplies ultrasound signals.
An ultrasound transmitter with known position (beacon) emits a pulse to the WISP node and
the time delay is used to compute the distance. The RFID protocol between transmitter and
receiver (tag) provides the required synchronization. The achieved accuracy of distance
The power consumption of this system is very low since the passive tag implements energy
harvesting from the RFID RF communications link between it and the RFID Reader. The
reported power in [99] is on the order of 100 µW and is sufficient to power the acoustic tone
1) Ultrasound Methods, AOA: Phased microphone arrays have been used to measure the
angle of arrival, from the direction of the target, via beam forming techniques, as
described in [100] and in [101]. These arrays can be used to estimate AOA parameters
(WSSU), signaling is reported in [103]. Three beacons with known locations transmit WSSU
ultrasonic receiver array, associated analog-to-digital converters, and a processor. The array
SPM0204 ultrasonic sensors (microphones based on CMUT technology) are used for the
39
receivers. Three such sensors are fixed in a triangle configuration with a distance of 2.1 cm
between each pair of sensors. The resulting accuracy of the angle measurements is 0.2
degrees (bias), and a standard deviation of 2 degrees. This translates to location accuracy of
14 cm. Better accuracy can be achieved by combining the AOA data with TOF data. This
hybrid algorithm then can achieve an location accuracy of 7.8 cm. The system is unique since
it does not require a global timing reference signal, either wired or RF-derived. Power
3.12 Optical Methods, TOF Indirect and Direct, Distance from Lighthouse
Method:
Distance-based optical methods are based on time-of-flight, (TOF), measurements between a
source beam and a reflected beam of light. They can also be based on the time the node
spends in the anchor’s illuminating beam path, the so-called lighthouse method. TOF
measurements, direct or indirect, and the lighthouse method. Indirect TOF measurements
detect a phase shift, usually via correlation, between the reference and reflected optical
signals. The optical signal source (reference signal) is usually a collimated laser with an eye-
safety-limited illumination power of 1 mW. The reflected optical signal has a power level
typically in the nW range. A phototransistor is used to capture the reflected optical signal, as
reported in [104], and [105]. The error in distance measurements is from 0.15 mm to 2 mm
depending on the amount of background light intensity and on the optical input power. The
applicable range is 1.2 m to 3.2 m. Power consumption of the phototransistor chip alone is
1.5 µW, with an additional 2 mW to 5 mW required by the correlation circuits and processor.
The light source in indirect TOF optical methods can be a diffused-light LED [106] with a
Single-Photon Avalanche Diode (SPAD) detector, for image depth sensors, or a scanned
laser source (collimated laser) using polygonal mirrors [107], [108], or MEMS mirrors [109].
40
The error in distance measurement, for the four references above (image depth sensors), is
Power consumption is not published but the 0.18 µm CMOS SPAD detector chip consumes
about 2 mW alone. Methods for direct measurement of travel time of light, direct TOF, are
presented in [110]. These methods are based on time-to-digital CMOS converter circuits and
can achieve picosecond resolution for a single TOF measurement, [111]. The error in
distance measurement can range from 1 cm to 4 cm, over a range of 2.5 m to 1500 m. Power
consumption is listed for a complete 3D depth-sensor camera system (120 × 120 to 320 ×
240 pixels), and not for the TOF sub-system alone. A recent chip from STMicroelectoronics
[30], the FlightSense VL6181, performs a direct TOF measurement using an infrared light
emitter and a SPAD photon detector. These methods have yet to be implemented for self-
localization in WSNs. The lighthouse method, first proposed and implemented in [112], was
devised to allow smart dust motes, (SDMs), to self-localize in 2D or 3D space using two or
three rotating anchors (lighthouses) only in one-hop topologies (direct line of sight between
each SDM and the anchors). Each anchor illuminates the SDM with a collimated rotating
laser beam. Once illuminated, the SDM measures the amount of time it remains in the
beam’s path. An SDM closer to an anchor spends more time in the beam’s path than an SDM
further away from the anchor. Figure 11 below illustrates this method: w is the constant
width of the rotating beam, angle β is computed from the sweep time and rotation time of the
beam, and distance d is derived from β and w. The time measurements, known beam width,
known rotation rate, and known anchor locations are used by each SDM to estimate its own
location. Range (distance between anchor and SDM) is from 1 m to 10 m (room-size) and the
41
accuracy is about 2% of the range 2 cm to 20 cm. This accuracy has been validated with data
laptop (emulating an SDM), using off-the-shelf parts. This method is also called cylindrical
lateration, in contrast to spherical lateration [42]. It does not have a wide baseline
and anchors) to achieve this accuracy. The accuracy may be improved by using deflecting
MEMS mirrors in the anchors to build a truly parallel illuminating beam and by
implementing a more precise in-beam-path time measurement in the SDM. The desired
dense concentrations of SDM sensors. Power consumption for the 650 nm semiconductor
laser in each lighthouse is 1 mW alone. The SDM power consumption is not listed since it’s
emulated by a laptop.
location of an unknown point within a triangle by means of a known optical basis and the
related side angles pointing to the unknown point. A popular method for 2D applications is
optical parallax implemented in may commercial sensors, such as [113], [114], [115]. Figure
12 below illustrates the basic measurements used in this method. The value h is a fixed
known distance between the center points of the laser diode and the camera. When
the distance to the target object D changes, so do both the angle θ and the value for PFC
(pixels from center of camera axis). Note that PFC measures the vertical distance from the
focal plane center to the reflected laser spot (blob) formed on the focal plane. Image
processing in the camera is used to detect the reflected blob (spot) and determines the PFC
42
value. From PFC, angle θ can be extrapolated (via table lookup) and then distance D can be
computed. The design uses a laser diode and a CMOS camera and calculates distance to a
target using simple trigonometry between the centroid of laser light, camera, and object. For
[113] the error in distance measurement is 0.45 cm to 3.66 cm over a range of 15 cm to 122
43
A two-dimensional localization system, based only on optical triangulation was presented in
[116]. Coordinates are computed by low-power analog image processing circuits and
transmitted to a basestation via an RF link. The reported accuracy of planar (x,y) coordinates
improved implementation has been disclosed in [117], using only low-cost off-the-shelf
components. The accuracy in localizing static x and y coordinates is 0.1 mm to 1.7 mm over
an area of 1.8 m (x) and 1.2 m (y). The unknown mote has a rotating laser transmitter which
illuminates three arrays of photodetectors fixed at known positions. The currents generated in
the illuminated photodetector arrays are used to compute the angles to the unknown mote and
44
illumination, is presented in [118]. A setup analogous to stereo cameras is used, with one
camera replaced by a binary-modulated light source. The modulator uses a binary signal to
temporally modulate the radiant flux of the lighting beam. The application is 3-D imaging
and depthrecovery via projection of a series of fringe patterns with different wavelengths. A
3D localization system for a robotic drill-fastener is described in [119]. The system uses
active illumination beacons (three rotating laser sources) and four photodetectors embedded
in the unknown target. This is similar to the system described in [120] where two active
illumination beacons are used and the target has a CCR with three photodiodes. In [119] a
triangulation algorithm is used to measure the angles between the beacons and the target. At
least one of the photodetectors in the target is always in line of sight of the rotating lasers.
The system achieves sub-millimeter precision in localizing the target’s 3D coordinates. The
range for localization is from 25 cm to 1 m. These are typical dimensions for the inside of a
small aircraft wing. In this application, the robotic drill-fastener performs tasks inside the
aircraft wing and has to be localized in 3D space. A recent 134-pixel CMOS sensor is
described in [31] with a dual-mode architecture for optical triangulation or TOF. For short
ranges, 0.4 m to 1.0 m, optical triangulation is used to measure the distance, with a precision
sensors are proposed in [24] to compute the sensor’s optical AOA parameters and derived
coordinates in three dimensions using two anchors. This method can be applied to two
dimensions as well. Architectures for similar optical antennas and spherical retroreflectors
45
are described in [23] and [121]. Using a spherical retro-reflector gives a constant optical
power over a large range of incidence angles (AOAs) when compared to using CCRs.
Tessellated spherical optical antennas have been proposed in [26] for both localization and
data communication in 3D ad-hoc optical wireless networks. Figure 13 below shows two
anchors ( #1 and #2 ) and a third unknown node #3 which is localized. Each node has a
spherical optical antenna, and localization can be done via triangulation using the anchor’s
known positions and measured angles φ13, φ23, φ31, θ13, θ31, and θ23. Each hexagonal
panel has a laser LED source and a photo-detector to maximize angular diversity and
maintain constant optical power in any direction. However, the reported localization
per node. A recently reported indoor localization system, [22], uses free-space optics (FSO)
to localize a subscriber unit using a centralized laser transceiver. Both the subscriber unit and
transceiver have MEMS steering mirrors, lenses, and uplink - downlink optical
(CPC), fiber collimator, and pigtailed photodiode. The optical transmitter has VCSEL laser,
lens, and a steering MEMS mirror. The localization accuracy is 5.26 cm to 8.58 cm, over a 1
m to 2 m range. The optical link is based on on-off-keying (OOK) at a bit rate of 50 Mbps.
46
3.16 Optical Methods, AOA, CCR, MRR, and Photodiode Receivers for
Smart Dust Motes:
A recent photosensor has been reported in [27]. It consists of three photodiodes, each one
embedded in one side of a corner-cube reflector, CCR. The photosensor can estimate the
optical AOA of light emitted by an optical beacon given a direct line-of-sigh. The inherent
measurement error is less than +/- 5 degrees in either azimuthal or polar angles. A complete
sensor localization system has been built using this type of photosensor, as reported in [120].
Figure 14 below illustrates the 3D localization of a node equipped with this photosensor and
using location data from two known anchors (beacons A and B). Triangulation is used to
compute the node’s coordinates based on the measured AOA data. The reported localization
47
A Modulated Retro-Reflector, (MRR), has been proposed and implemented in [122]. The
main application is not localization of the unknown sensor or Smart Dust Mote, (SDM), but
optical wireless communication between the SDM and the anchor (base station) node. The
SDM includes an MRR, while the anchor has the laser source, the beam steering controls,
and a CMOS image sensor for target SDM acquisition. Figure 15 below illustrates the one-
hop, direct line-of-sight topology of the WSN. The anchor actually has to locate (acquire and
track) the SDM before communicating with it. This centralized SDM localization is done at
the anchor using image processing algorithms and data from the CMOS imaging camera. The
48
communications link supports up to 32 bits/sec (uplink) and 100 bits/sec (downlink) over a
range of 30 m. The total power consumption of the SDM is 40 nW which is the lowest
achieved value in our survey. This is achieved thanks to builtin energy harvesting of the
optical signal received from the anchor, using two photodiodes in series. The application of
an MRR-based architecture to the localization problem should achieve the desired low power
consumption since optical wireless can be used for both data communications and for
localization via distance and AOA measurements. Distributed localization can be achieved if
each SDM has the ability to compute its own AOA relative to two anchors as shown in
Figure 15.
49
MRR-based optical communications and the lighthouse method for localization are the least
researched methods for distributed self-localization, since they represent relatively new
localization architecture, combining multi-element FSO antennas and MRR structures for
cylindrical lateration to improve distance estimates using the AOA data, may represent a
approach the millimeter scale, optical methods for localization are able to deliver the precise
distance and angle measurements required by the higher layer algorithms of Section II.
50
CHAPTER 4
POWER AND ACCURACY COMPARISON
The relative distribution of localization accuracy between the nodes, achieved with
Sec. II and Sec. III, is shown in Figure 16 below. The data presented in Figure 16 has been
If an implementation can estimate distances in 2D and 3D, the dimension with the best
triangulation and TOF via active optics, TOF via RF and ultrasound signaling, etc.). Range-
free and angle-free methods [41], [16], [44] can be less accurate than range-/angle-based
methods [56], [53], [69]. Note that millimeter and sub-millimeter precision has been
achieved with mostly optical methods and with an ultrasonic method in one case [98].
Triangulation using TOF or AOA derived distances is used in all cases and the scale of the
unknown node is also typically in the millimeter range [124], which includes smart dust
nodes [123].
Theoretical analysis and experimental results from [31],[124], and [120] indicate that
AOA-based optical triangulation has the highest precision at shorter distances, while TOF
maintains highest precision at longer distances as shown by [31], [98], [83], and [105].
Optical triangulation (active or passive) and image processing based algorithms for
localization [122] continue to deliver the most accurate results in the sub-millimeter to
51
[96] methods using distances derived by TOF, TOA, and TDOA measurements. The most
common among these are RF narrow-band TOF ranging implementations, due to the simple
design and available low-cost hardware. Standard off-the-shelf solutions such as [84] and
[85] provide meter-level accuracy without the need for specialized base-stations or UWB.
derived in [67].
The least accurate localization methods use RF RSSI, however these methods are
applicable in a wide sub-meter to tens of meters range (up to 10 m in the reported work),
and can extend beyond that range depending on the signal strength of the transmitter.
Theoretical analysis presented in [80] and [77] shows that even complex non-linear RSSI
models based on hundreds of measurements can generate large errors in the estimated
distance. This is especially true in an ad-hoc RF environment in which path-loss has never
been measured and characterized, ie. without a training phase.. The relative average power
consumed by each node while running all or most of its sub-systems required for
The data in this figure has been compiled from published implementation results. If the
power needed explicitly to perform the specific localization task is published, the reference
is marked with a star, otherwise the total published power consumption of the node is
the ultra-low power end, 40 nW is the smart dust millimeter-cubed node from [122] which
harvests optical power for all of its circuits and is capable of two-way free space optical
communication. Range-free and RF-ID methods with RF energy-harvesting [15], [16], [22],
52
[68], [99], [104], and [105] fall in the 1 µW to 10 mW range as well as a solar-powered
Fig. 16. Localization range for different implementations of different colocation hardware methods. Grey
inset shows the applicable distance ranges for each of the physical implementation methods (optical,
ultrasound, and RF) described in the surveyed papers. A star next to a reference indicates a hybrid
implementations (eg. triangulation and TOF via active optics, TOF via RF and ultrasound signaling, etc.)
Finder methods [17], [18], [67], [98], and [113] fall in the 10 mW to 100 mW category. The most power,
in the 100 mW to 1 W category, is consumed by methods based on RF RSSI and commercial protocols
such as IEEE802.15.4, Bluetooth, and Wi-Fi WLAN [112], [13], [14], and [124]. This is due to the fact
that commercial RF protocols typically cover a large geographical area without repeaters and make use of
retransmissions in order to maintain an acceptable bit-error rate at high data rates. The use of MIMO,
relaying, and diversity techniques by these protocols also results in a high duty-cycle operation of the
53
power-hungry RF transceivers. Active (laserdiode) optical localization systems can also consume power
in the 100 mW range, as reported in [124]. The implementation, MEMS-Eye, is a LOS (one-hop) 3D
localization system. It’s based on a centralized, optical-triangulation algorithm with two anchors, each
equipped with a laser diode source, a MEMS mirror, and a photo-detector. The target contains a CCR for
passively retro-reflecting the illuminating beams to the anchors. The reported distance accuracy is 1 mm
or better in each of the 3 dimensions over a volume of 1 m3 . The operating power for the MEMS mirrors
is very low, less than 1 mW, however the amplifier circuits alone consume 50 mW - 100 mW in each
anchor. Note that power consumption numbers have been published only for a subset of the
54
CHAPTER 5
CONCLUSION
WSNs. A detailed review of the algorithms and physical implementations of node localization
methods is presented. With the onset of lowpower wireless sensor networks, this paper also
classifies the localization algorithms (both centralized and distributed) in terms of power usage.
desired applicable rangeand desired accuracy of localization with the proper physical method for
obtaining the required intra-node measurements (distance or angle). For example, optical
interferometric physical methods allow for nanometer accuracy but only over a range of a few
provide for accuracies from microns to millimeters, over a much larger range spanning
centimeters to a few meters [113], [114], [115]. Ultrasound techniques provide accuracy of a few
centimeters, over an applicable range of a few meters [98]. Radio frequency techniques have
even lower accuracy, from tens of centimeters to meters, however providing a much larger
applicable range from tens to hundreds of meters, depending on the power levels of the
network and by increasing the number of nodes. The advantages and disadvantages of the three
main physical implementations of distance measurement techniques (RF, optical, and ultrasound)
are summarized in Figures A.18 - A.20 in the Appendix. For small-sized SDM type WSNs, the
stacked-die smartdust sensor platform such as the one reported in [123] may be applicable as the
55
backbone of the self-localization hardware. A distributed self-localization algorithm can be
implemented with this platform since it contains a processor, DSP, and memory resources of it
own. Based on the accuracy and power data from Sec. IV, passive optical triangulation is the
preferred method for distance or AOA measurements to neighboring nodes and anchors. Free
possible with neighboring nodes when implementing some of the distributed message-passing
algorithms from Sec. II. When combined with optical energy harvesting, the resulting power
consumption of a self-localizing SDM network covering a few squared meters outperforms any
area for future research into multi-hop very lowpower optical methods for self-localization of
example RF RSSI and RF TOF or ultrasound AOA and ultrasound TOF, can significantly
improve the localization accuracy of the multi-mode implementation and increase the effective
range, when compared to a single-mode implementation. At the same time the multi-mode
implementation can reduce the power required for the localization task when compared to the
single-mode implementation. For example, RSSI alone has low power (few message exchanges)
but low accuracy in mutli-path interference environments with non line-of-sight (NLOS)
conditions. When combined with RF TOF, as proposed in [87], localization accuracy can be
improved. Power can be reduced since some nodes are localized only by RSSI and others by
TOF. Localization of all nodes by TOF alone would require higher power since two-way TOF
results in a large amount of messages between nodes and consumes a large amount of channel
capacity. A combined ultrasound AOA and ultrasound TOF localization architeture has been
56
implementation. Improved timing synchronization between all nodes can improve the accuracy
of TOF measurements and resulting localization. One strategy to achieve this could be to use
synchronous CDMA instead of TDMA at each node. Optical triangulation (OT) can achieve
lost at ranges typically longer than tens of cm, and TOF-based distance measurement can be used
in this case. A combined optical triangulation and optical TOF measurement architecture has
been implemented in [31]. This multi-mode implementation increases the effective range of
accurate localization from 40 cm to 5 m. The accuracy of the OT mode can be further improved
by enhanced algorithms for finding the reflected laser spot on the receiver’s pixel array and by
Fig. 17. Power consumption for different implementations of co-location hardware methods. If
the power needed to perform the specific localization task is published, the reference is marked
57
with a star, otherwise the total published power consumption of the node is listed. Total power
58
SUMMARY
Summary of advantages and disadvantages of using RF, acoustic, and optical physical co-location
59
Fig. A.18. Summary of advantages and disadvantages of RF physical methods.
60
REFRENCES
http://www.freescale.com.
energy scavenging from current-carrying conductors,” IEEE Sensors J., vol. 13, no.
R. White, and P. Wright, “A design methodology for energy harvesting: With a case
61
unit,” in Energy Harvesting and Systems: Materials, Mechanisms, Circuits and
[11] A. Savvides, H. Park, and M. Srivastava, “The bits and flops of the nhop
September 2002.
September 2010.
[15] Conexant and Corp, “Prism chipset solutions from conexant,” Conexant, Tech.
Rep., 2013.
WCOM, 2009.
[19] M. Davidovic, “Time-of-filight range finding sensor using an integrated pnp pin
62
phototransistor in 180 nm cmos,” Institute of Electrodynamics, Microwave and
[21] J. Jeon, H. Jeon, and C. Park, “Theoretical analysis of reflection process using
mems ccr array for the localization of optics-based sensor node.” 13th International
Conference on Control, Automation and Systems ICCAS 2013 , Oct. 20-23, 2013,
Korea.
Conference.
scheme using optical wireless.” IFIP IEEE International Conference on Wireless and
63
[26] M. Bilgi, M. Yuksel, and N. Pala, “3-d optical wireless localization.” IEEE
applications.” IEEE Sensors Journal, vol. 10 , no. 12, pp.1887-83 , Dec 2010.
ccrs.” International Journal of Advanced Science and Technology, vol. 13, Dec
2009.
2013, www.st.com.
triangulation 3-d imaging.” IEEE Journal of Solid State Circuits, vol. 44 , Dec. 2009.
[33] N. Barrocaa, L. M. Borgesa, and al., “Wireless sensor networks for temperature
algorithm for 2d.” IEEE 2011 International Conference on Computer Science and
64
Network Technology.
[35] P. Hebert, N. Hudson, J. Ma, and J. Burdick, “Fusion of stereo vision, force-
torque, and joint sensors for estimation of in-hand object location.” IEEE 2011
2011.
517.
Demand Network Systems and Services, 2009. WONS 2009. Sixth International
65
connectivity - mds-map.” ACM Proceeedings of the 4th ACM International
[42] T. Cox and M. Cox, Multidimensional Scaling. Chapman and Hall , London,
1994.
[43] X. Ji and H. Zha, “Sensor positioning in wireless ad-hoc sensor networks using
networks.” IEEE Signal Processing Letters, vol. 11 , no.3 , pp.334-337 , March 2004.
[48] P. Biswas and Y. Ye, “Semidefinite programming for ad hoc wireless sensor
Sept 2003.
66
[52] S. Simic, “A learning theory approach to sensor networks,” University of
distances,” Kestrel Institute Technical Report, Kestrel Institute, 3260 Hillview Ave.
[55] D. Niculescu and B. Nath, “Ad hoc positioning system (aps).” IEEE Global
scaling for node localization in sensor networks.” ACM vol V , No. N , June 2005
pages 1-26.
[57] V. Dang, V. Le, Y. Lee, and S. Lee, “Distributed push-pull estimation for node
constrained local problems.” IEEE Sensors Journal, vol. 13 , no. 6 , June 2013.
67
[60] S. Basagni and M. Conti, “Mobile ad-hoc networking.” IEEE Press , Wiley
Interscience 2004.
ID 281592.
distributed sensor self-localization and target tracking using belief propagation and
[66] A. Gasparri and F. Pascucci, “An interlaced extended information filter for self-
positioning based on tdoa with extended kalman filter.” IEEE 37th Annual Computer
68
Software and Applications Conference, COMPSAC 2013.
[74] X. Zou, W. Li, W. Pan, B. Luo, L. Yan, and J. Yai, “Photonic approach to the
wsn.” Proc. Int. Indoor Positioning and Indoor Navigation (IPIN) Conference 2010.
69
Cybernetics, Part C: Applications and Reviews vol.37 2007.
[77] X. Li, “Rss-based location estimation with unknown pathloss model.” Wireless
based location tracking.” Personal and Ubiquitous Computing, vol.11, no. 6, pp489-
503, 2007.
[79] A. Paul and E. Wan, “Rssi-based indoor localization and tracking using sigma-
inertial data for signal strength localization.” Proc of the 8th Workshop on
[82] J. Liu, Q. Wang, J. Wan, and J. Xiong, “Towards real-time indoor localization
[83] S. Lanzisera, “Rf ranging for location awareness,” Ph.D. dissertation, University
measurement for low-cost wireless sensor localization.” IEEE Sensors Journal, vol
70
[85] H. Will, S. Pfeiffer, S. Adler, T. Hillebrandt, and J. Schiller, “Distance
[87] T. Gadeke, “A bi-modal ad-hoc localization scheme for wireless network based
on rss and tof fusion,” ITIV Karlsruhe Institute of Technology, Germany. IEEE
Conference, 2013.
[91] B. Engineering, Long Range Radar LRR3 EN V05 Final Report, Bosch Corp.,
www.bosch-engineering.de, 2014.
[93] IEEE, “Css technology white paper ieeep802.15.” IEEE 802.15 Working Group
[94] Y. Chan and K. Ho, “A simple and efficient estimate for hyperbolic location.”
[95] Y. Chao, J. Fang, and X. Luo, “Constrained least-squares and wireless location
71
based on tdoa measurements.” Chinese Journal of Radio Science, vol. 23, no.5 ,
pp.841-846 , 2008.
[97] J. Ash and L. Potter, “Sensor network localization via received signal strength
[98] C. Medina, “A synchronous tdma ultrasonic tof measurement system for low-
[100] H. C. Chen, A.M.Ali, “Design and testing of robust acoustic arrays for
[102] P. Stoica and R. Moses, Spectral Analysis of Signals. Prentice Hall, 2005.
6, June 2012.
[105] M. Davidovic and J. Seiter, “A background light resistant tof range finder with
72
integrated pin photodiode in 0.35 um cmos.” SPIE Opt. Metr. 2013 , 13-16 , May
2013.
cmos soc for a 100m-range 10fps 200x96-pixel time-offlight depth sensor.” IEEE
mems-based laser scanning time-of-flight sensor.” Opt. Express, vol. 20, pp.11863-
11881, 2012.
di Torino, 2010.
[111] C. Veerappan, “A 160x128 single-photon image sensor with on-pixel 55ps 10b
[112] K. Romer, “The lighthouse location system for smart dust.” MobiSys
[113] P. Inc., Laser Range Finder 28044 ; Laser Diode and CMOS camera (detector),
73
[114] S. Inc., OD Mini Prime Laser Distance Sensor product datasheet, SICK
localization system for mobile objects without image processing.” IEEE Conference
[119] S. Linga, “An optical external localization system and applications to indoor
angles or arrival with an inertial measurement unit.” IEEE GNSS Conference 2012.
74
[123] Y. Lee, S. Bang, I. Lee, Y. Kim, G. Kim, M. H. Ghaed, P. Pannuto, P. Dutta,
low power i2c inter-die communication and multi-modal energy harvesing.” IEEE
75