Integrated Project - EUWB
Contract No 215669
Deliverable
D4.2.1
Initial development of dissemination methods and evaluation
Contractual data:
M12
Actual data:
M12
Authors:
Golaleh Rahmatollahi (LUH), Carlos Héracles Morais de
Lima (CWC), Giuseppe Thadeu Freitas de Abreu (CWC),
Juan Chóliz (UZ), Ángela Hernández (UZ), Christian Kocks
(UDE), Ernest Scheiber (UDE), Alexander Vießmann (UDE),
Shangbo Wang (UDE), Dong Xu (UDE) , Guido Bruck (UDE),
Peter Jung (UDE)
Participants:
UDE (Editor), CWC, LUH, UZ
Work package:
WP4
Security:
PU
Nature:
Report
Version:
1.1
Total number of pages:
97
Abstract
This report will introduce new methods for the dissemination and acquisition of information in a
Location Aware fashion. Routing, relaying and data fusion are the keywords to be addressed in light
of the knowledge of the location of the transceivers in the system.
Keywords
UWB, localization, location tracking, reliability, data communication
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Table of Contents
1 Executive Summary ........................................................................................................................... 11 2 Dissemination and Acquisition of Information .................................................................................. 12 2.1.1 Position-centric solutions in autonomous large-scale multi-hop scenarios .......................... 12 2.1.2 Contention-based Geographic Forwarding Strategies in Multi-hop Scenarios .................... 12 2.1.2.1 Network Model.......................................................................................................... 13 2.1.2.2 Contention-based Geographic Forwarding Strategies ............................................... 13 2.1.2.3 PDF of the distance of the n -th nearest neighbour ................................................... 19 2.1.2.4 Preliminary Results ................................................................................................... 20 2.1.2.5 Final Remarks and Perspectives ................................................................................ 24 2.1.3 Routing Discovery and Maintenance Strategies in Autonomous Multi-hop Networks ....... 24 2.1.3.1 Target Scenarios ........................................................................................................ 25 2.1.3.2 Performance metrics .................................................................................................. 25 2.1.3.3 Routing Protocols for Infrastructureless Networks ................................................... 26 2.1.3.4 Hierarchical beaconless routing using contention-based RSAs ................................ 27 2.1.3.5 Conclusions and Future Directions ........................................................................... 33 3 Co-Ordination Schemes and Information Exchange Protocols .......................................................... 34 3.1 Mechanisms for Information Exchange, Dissemination and Acquisition in distributed,
uncoordinated wireless sensor networks ........................................................................................... 34 3.1.1 Node architecture and fundamental models ......................................................................... 34 3.1.1.1 Packet Error Modeling .............................................................................................. 35 3.1.2 Ranging Information Acquisition ......................................................................................... 36 3.1.3 Recource allocation strategy................................................................................................. 37 3.1.3.1 DLC Layer Description ............................................................................................. 38 3.1.3.2 Game theoretical formulation of the pulse rate adaptation........................................ 40 3.1.3.3 Utility function .......................................................................................................... 40 3.1.3.4 Existence of Nash Equilibrium .................................................................................. 42 3.1.3.5 LPC Algorithm .......................................................................................................... 42 3.1.4 Simulation Scenarios and Evaluation of DLC-MAC Performance ...................................... 44 3.1.4.1 Near-Far Scenario...................................................................................................... 45 3.1.4.2 Performance Comparison with Adaptive Channel Coding (ACC)............................ 48 3.1.4.3 Random Scenario ...................................................................................................... 49 3.1.4.4 Performance Comparison with Adaptive Channel Coding (ACC)............................ 52 3.1.4.5 Conclusion ................................................................................................................. 53 Page 2
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
4 Evaluation of architectures and acquisition & distribution schemes for tracking .............................. 54 4.1 System model .............................................................................................................................. 54 4.1.1 System environment ............................................................................................................. 54 4.1.2 MAC layer description ......................................................................................................... 55 4.1.3 LT application ...................................................................................................................... 57 4.2 Analysis and evaluation............................................................................................................... 59 4.2.1 Simulator description ........................................................................................................... 59 4.2.2 Impact of the tracking architecture ....................................................................................... 60 4.2.3 Impact of acquisition & distribution strategies .................................................................... 64 4.2.4 Impact of number of anchors used for location .................................................................... 69 4.2.5 Impact of the distance between anchors ............................................................................... 72 4.2.6 Effect of target mobility and position update rate ................................................................ 76 5 Network Impact for Localization ....................................................................................................... 79 5.1 Quantity of Information............................................................................................................... 79 5.1.1 System Environment ............................................................................................................ 79 5.1.2 Network Architecture ........................................................................................................... 80 5.1.3 LT MAC Function [39] ........................................................................................................ 82 5.1.4 LT Application ..................................................................................................................... 83 5.1.4.1 Introduction ............................................................................................................... 83 5.1.4.2 Step I: Synchronization of the Fixed Mounted Infrastructure ................................... 84 5.1.4.3 Step II: Localization of a Mobile Device .................................................................. 86 5.1.4.4 Step III: Triangulation and Mapping ......................................................................... 87 5.1.4.5 Quantity of Information Without Packet Loss .......................................................... 87 5.1.4.6 Quantity of Information with Packet Loss ................................................................ 88 5.1.4.7 Simulation results ...................................................................................................... 88 6 Conclusions ........................................................................................................................................ 93 References ............................................................................................................................................. 95 Acknowledgement ................................................................................................................................. 97 Page 3
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
List of Figures
Figure 1: Illustration of the sectoral decision region. ............................................................................ 15 Figure 2: Illustration of the convex lens decision region. ..................................................................... 16 Figure 3: PMF of the CRI for the STA-based RSA. ............................................................................. 20 Figure 4: PMF of the CRI for the auction-based RSA. ......................................................................... 21 Figure 5: Expected distance of the n-th nearest neighbour (radio range of 10m). ................................ 22 Figure 6: Expected distance of the n-th nearest neighbour relative to the expected value of the CRI. . 23 Figure 7: Expected distance of the n-th nearest neighbour relative to the expected value of the CRI. . 23 Figure 8: Illustrative output of the clasterization algorithm. ................................................................. 28 Figure 9: Packet delivery success ration with duty cycle of 10% ......................................................... 32 Figure 10: Excess-hop distribution with duty cycle of 10%. ................................................................ 32 Figure 11: UWB Node architecture and application subsystem............................................................ 34 Figure 12: Packet level collision ........................................................................................................... 35 Figure 13: Air interface frame structure ................................................................................................ 36 Figure 14: Positioning principle using two-way ranging ...................................................................... 37 Figure 15: Positioning estimation using N successive beacon frames .................................................. 37 Figure 16: PRA game utility function for different cost terms.............................................................. 41 Figure 17: Air Frame Format ................................................................................................................ 43 Figure 18: LPC flow-diagram ............................................................................................................... 44 Figure 19: Near-Far Scenario ................................................................................................................ 45 Figure 20: NE for the 2 links near far network ..................................................................................... 46 Figure 21: NE for 4 links in near far network ....................................................................................... 47 Figure 22: NE for 4 links near far network ........................................................................................... 48 Figure 23: Total logarithm utility
∑
|N|
log(ri ( prf )) ............................................................................. 49 i =1
Figure 24: Average BER per link .......................................................................................................... 49 Figure 25: Random Scenario ................................................................................................................. 49 Figure 26: Example of a random network realization ........................................................................... 50 Figure 27: NE in 8 links random example network .............................................................................. 51 Figure 28: NE in 8 links random example network .............................................................................. 52 Figure 29: Total logarithm utility
∑
|N|
log(ri ( prf )) ............................................................................. 53 i =1
Figure 30:Average BER per link ........................................................................................................... 53 Figure 31: UWB tracking of mobile devices in a shopping mall .......................................................... 54 Page 5
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Figure 32: EUWB MAC superframe structure [30] .............................................................................. 56 Figure 33: Mesh centralized topology [30] ........................................................................................... 57 Figure 34: Example of meshed scheduling tree [30] ............................................................................. 57 Figure 35: UWB location simulator scenario ........................................................................................ 60 Figure 36: Resources needed for location for a centralized in the network architecture with 1 location
controller....................................................................................................................................... 62 Figure 37: Resources needed for location for a centralized in the network architecture with 4 location
controllers ..................................................................................................................................... 62 Figure 38: Resources needed for location for a distributed architecture ............................................... 63 Figure 39: Resources needed for location for a centralized in the mobile architecture ........................ 63 Figure 40: Resources needed for location for the different system architectures.................................. 64 Figure 41: Resources needed for location when data aggregation is applied ........................................ 65 Figure 42: Resources needed for location when multicast ranging request is applied .......................... 66 Figure 43: Resources needed for location when broadcast ranging request is applied ......................... 66 Figure 44: Resources needed for location for different acquisition and distribution enhancements ..... 67 Figure 45: Resources needed for location when the ranging procedure is initiated by the anchors ...... 67 Figure 46: Resources needed for location when the ranging procedure is initiated by the anchors and
broadcast ranging response is applied .......................................................................................... 68 Figure 47: Resources needed for location depending on the initiator of the ranging procedure and the
enhancements applied ................................................................................................................... 68 Figure 48: Resources needed for location for different system architectures when data aggregation and
multicast ranging request are applied ........................................................................................... 69 Figure 49: Resources needed for location depending on the number of anchors used for location when
Na=25 anchors .............................................................................................................................. 70 Figure 50: Position estimation error depending on the number of anchors used for location when
Na=25 anchors .............................................................................................................................. 71 Figure 51: Position estimation error variance depending on the number of anchors used for location
when Na=25 anchors .................................................................................................................... 71 Figure 52: Position estimation error for different total number of anchors with trilateration and
σn=0.7m ........................................................................................................................................ 72 Figure 53: Position estimation error for different total number of anchors with trilateration and
σn=0.3m ........................................................................................................................................ 72 Figure 54: Position estimation error for different total number of anchors with Kalman filter and
σn=0.7m ........................................................................................................................................ 73 Figure 55: Position estimation error for different total number of anchors with Kalman filter and
σn=0.3m ........................................................................................................................................ 73 Page 6
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Figure 56: Position estimation error variance for different total number of anchors with trilateration
and σn=0.7m.................................................................................................................................. 74 Figure 57: Position estimation error variance for different total number of anchors with trilateration
and σn=0.3m.................................................................................................................................. 75 Figure 58: Resources needed for location for different total number of anchors .................................. 75 Figure 59: Position estimation error and tracking error depending on the target speed ........................ 76 Figure 60: Position estimation error and tracking error depending on the time between direction
changes ......................................................................................................................................... 77 Figure 61: Position estimation error and tracking error depending on the time between updates ........ 77 Figure 62: Resources needed for location depending on the time between updates ............................. 78 Figure 63: Localization and tracking (LT) of a tag inside the car [38] ................................................. 79 Figure 64: LT in a centralized network architecture [39]...................................................................... 81 Figure 65: Hierarchy in a centralized network architecture .................................................................. 81 Figure 66: Typical super frame structure with three time slots in the ranging period [39] ................... 82 Figure 67: Data format of a ranging frame [39] .................................................................................... 82 Figure 68: Time slot allocation for the synchronization of the fixed mounted infrastructure ............... 83 Figure 69: Generic localization tracking set-up [38] ............................................................................. 84 Figure 70: Synchronization of the fixed mounted infrastructure [38] ................................................... 85 Figure 71: Estimated position of tag with the process delay 10−9 ⋅ a .................................................... 89 Figure 72: Cumulative Density Function (CDF) of the estimation error with the process Delay: 10−9 ⋅ a
...................................................................................................................................................... 89 Figure 73: Estimated position of tag with the process delay 2.5 ⋅10−9 ⋅ a ............................................. 90 Figure 74: Cumulative Density Function (CDF) of the estimation error with the process Delay:
2.5 ⋅10−9 ⋅ a .................................................................................................................................... 91 Figure 75: Estimated position of tag with the process delay 5 ⋅10−9 ⋅ a ................................................. 91 Figure 76: Cumulative Density Function (CDF) of the estimation error with the process Delay:
5 ⋅10−9 ⋅ a ....................................................................................................................................... 92 List of Tables
Table 1: Parameters involved in the LPC initialization phase............................................................... 42 Page 7
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Abbreviations
ABL
Anchor based localization
AFL
Anchor-free localization
A/C
Aircraft
ALBA-R
Adaptive Load-Balanced Algorithm Rainbow
AODV
Ad hoc On-Demand Distance Vector
AP
Access point
APDL
Average Packet Delivery Latency
BPP
Binary Point Process
CAP
Contention Access Period
CDF
Cumulative Distribution Function
CFP
Contention Free Period
CGF
Contention-based Geographic Forwarding
CGSR
Cluster head Gateway Switch Routing protocol
CMS
Cabin management system
CR
Contention Resolution
CRA
Contention Resolution Algorithm
CRI
Contention Resolution Interval
CTS
Clear To Send
CVMS
Cabin video monitoring system
DAL
Design assurance level
DSDV
Destination-Sequenced Distance Vector
DSR
Dynamic Source Routing
E2E
End to End
EADS
European Aeronatic Defense and Space Company
EKF
Extended Kalman Filter
EUWB
CoExisting Short Range Radio by Advanced Ultra-WideBand Radio Technology
FAP
Flight attendant panel
FSB
Fasten seatbelt (sign)
GF
Geographic Forwarding
GTS
Guaranteed Time Slot
HDR
High rata rate
HDTV
Hight definition television
Page 8
EUWB
Initial development of dissemination methods and evaluation
IEEE
Institute of Electrical and Electronics Engineers
IFE
In-flight entertainment
IP
Inactive period
LBS
Location based services
LDR
Low data rate
LOS
Line-of-sight
LT
Localization and tracking
MAC
Medium access control (layer)
MD
Mobile device
MDS
Multi-Dimensional Scaling
MIMO
Multiple input multiple output
NLOS
Non-line-of-sight
NS
Non smoking (sign)
OOP
Object Oriented Programming
PA
Passenger address
PAX
Passenger
PDSR
Packet Delivery Success Ratio
PGF
Probability Generating Function
PHY
Physical layer
PID
Passenger information display
PMF
Probability Mass Function
PMR
Professional Mobile Radio
PPP
Poisson Point Process
PSU
Passenger service unit
QoS
Quality of service
RCA
Random Channel Access
RIBF
Regularized Incomplete Beta Function
R/L
Reading light
RMA
Random Multiple Access
RS
Relay Selection
RSA
Relay Selection Algorithm
RTS
Request To Send
SPP
Spatial Point Process
D4.2.1
Page 9
EUWB
Initial development of dissemination methods and evaluation
STA
Standard Tree Algorithm
TCP/IP
Transmission Control Protocol/Internet Protocol
TDOA
Time difference of arrival
TOA
Time of arrival
TORA
Temporally-Ordered Routing Algorithm routing protocol
UDP
Unity Disk Graph
UMTS
Universal Mobile Telecommunications System
UWB
Ultra wideband
UWB-RT
UWB radio technology
VHDR
Very high data rate
WiMAX
Worldwide Interoperability for Microwave Access
WLAN
Wireless local area network
WP
Work package
WRP
Wireless Routing Protocol
WSN
Wireless Sensor Network
D4.2.1
Page 10
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
1 Executive Summary
In this deliverable, a location aware information dissemination and acquisition method in UWB
communication system is introduced, where novel routing, relaying mechanism are considered. We
introduce the system in networks level, node level and application level independently in the following
chapters.
In chapter 2, an abstract networks model and routing strategy are discussed. Contention-based
geographic forwarding strategies in multi-hop scenarios and routing discovery and maintenance
strategies in Autonomous Multi-hop Networks are focused. The routing solution combines ideas of
network tessellation with simple greedy forwarding, without suffering from the problems afflicting
typical landmark-based alternatives [15]. A clusterized network based on graph-spectral properties
capturing the connectivity among nodes is built. The underlying concept is to exploit the availability
of cooperative communications in relaying networks so as to benefit from spatial diversity gains.
In chapter 3, a distributed mechanism makes use of the framework of game theory to develop a novel
interference management strategy based on the joint combination of adaptive error coding and pulse
rate control is introduced. The system architecture of a communication node that is divided into three
subsystems namely the UWB part, the communication part and the positioning/ranging subsystem is
also introduced. A game has been formulated and a distributed, asynchronous algorithm (LPC) is
proposed to discover the optimal PRF/PL network allocation under several channel and interference
conditions. It has been observed that due to the strong coupling concerning interference management
between PL and PRF adaptation, the game has several feasible steady-state allocations (Nash
equilibria). In general, the different equilibria provide similar aggregated network throughput. Results
suggest that, in order to maximize the cumulative network throughput, it is better that each link starts
with the lowest PL.
Chapter 4 and 5 presents an application of UWB networks for tracking mobile users in indoors
environments, such as a shopping centre. In chapter 4 different system architectures and strategies for
the acquisition and dissemination of location information are discussed and evaluated, as well as other
important parameters in the system design, such as the distance between anchors and the position
update rate. The system performance is analyzed and simulated, with a special focus on the impact of
localization on network traffic, as the limitation of resources is one of the main constraints in this
scenario. Chapter 5 focuses on the network impact for localization and the localization protocol.
Page 11
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
2 Dissemination and Acquisition of Information
2.1.1 Position-centric solutions in autonomous large-scale multi-hop scenarios
When addressable nodes that are separated by distances farther than their own transmission range need
to communicate, multi-hop connections are inevitably established between them. However, the
establishment of such multi-hop interconnections is a key issue in large-scale ad hoc networks. Owing
to the mobility of nodes, network dynamics, and channel impairments among other characteristics, the
wireless links undergo great fluctuation on their availability. Indeed, a major challenge in such
networks is to continuously maintain up-to-date the control information required to properly route
traffic.
Infrastructureless (mobile) networks have no fixed routers, in principle; any node should be capable of
relaying packets to any other node in the network. Since the propagation of both controlling and
payload information shares the same pool of available resources, it is greatly advisable to construct
routes inflicting the minimum overhead on the network. And yet satisfying users' Quality-of-Service
(QoS) requirements, while optimizing the utilization of the meager network resources should be an
underlying assumption. Because of its crucial task, routing procedures for ad hoc networks received
great attention during the last years. Several routing protocols have been defined, while it is possible to
classify them in two main groups (see section 2.1.3 for further details): (1) pro-active solutions, also
known as table-driven routing protocols; and (ii) reactive alternatives corresponding to source-initiated
on-demand routing protocols. Succinctly, table-driven routing protocols pro-actively endeavour to
keep reliable and up-to-date routing information among all nodes in the network. On the other hand,
reactive approaches, which are initiated by the source node on-demand, and are often implemented
relying on distributed and autonomous procedures are seem as promising alternatives for the dynamic
scenarios under investigation.
The transactions related to the selection of adequate relays and that are based on local knowledge of
network topology are addressed in section 2.1.2. Afterwards, the routing discovery and maintenance
protocols are introduced in section 2.1.3. The cost of propagating information in these deleterious
scenarios is evaluated as well. Indeed, a substantial amount of resources is consumed during the MAC
handshake in order to elect suitable relays at each hop along the multi-hop connections. Independent
of the adopted routing strategy - long-hops or short-hops, a negotiation period is inevitably spent to
select the next-hop relay. A considerable amount of resource is also consumed to keep the consistency
of already established multi-hop links. Solutions exploiting the local knowledge of network topology
to ponder operation of network functionalities are particularly interesting. Therefore, solutions that
provide reasonable advancements towards the final destination at each hop, while demanding tolerable
overhead (does not significantly add to packet delivery latency and does not severely compromise
network capacity) are introduced and evaluated on the sequel.
2.1.2 Contention-based Geographic Forwarding Strategies in Multi-hop Scenarios
The applicability of distributed approaches exploiting location awareness to effectively forward and
disseminate information throughput the network are addressed. The effectiveness of such methods are
measured in terms of the achievable progress at each hop towards the final destination and the
corresponding resolution time necessary to find relays. Stochastic geometry and non-parametric order
statistics are conveniently exploited so as to construct suitable network models and appropriately
Page 12
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
characterize the expected progress at each hop. The distributions of the order statistics of the expected
progress - considering specific forwarding decision regions, as described in section 2.1.2.2 - towards
the final destination is further used to characterize the benefits of using more elaborated decision
regions to split candidate relays during the contention resolution procedure.
Whenever nodes get entangled into conflict, several mechanisms are available to deal with the
contention resolution phase. In [6], an auction-based contention resolution algorithm using location as
side-information is introduced. When considering distributed geographic forwarding strategies, the
typical approach is to select the next-hop relay solely based on the progress it may provide towards the
final destination [3]. Based on the single-hop progress, several contention-based geographic
forwarding designs are available in order to select a well-qualified relay [4]. Additionally, a plethora
of different optimality criteria reflecting network dynamics may also be exploited to further augment
the priority mapping functions and increase the effectiveness of the selection procedure, e.g., residual
energy, perceived interference levels, and buffer occupancy among others [5]. Depending on the
consequential priority function used to order potential relays, collisions among packets of candidates
may still occur and a procedure to appropriately resolve the contention is also necessary [6].
The Contention-based Geographic Forwarding (CGF) strategies are characterize by the underlying
geographic forwarding areas: sectoral and convex lens decision regions (see Figure 1 and Figure 2,
respectively) [4]. The influence of the CGF procedures and intrinsic forwarding decision regions on
the effectiveness of the relay selection strategies and, consequently, on the overall performance of the
network is addressed. Two distinct geographic forwarding designs are considered, namely, sectoral
and convex lens decision regions (see Figure 1 and Figure 2, respectively) [7]. Stochastic geometry is
used to model the network deployment as a Spatial Point Process (SPP) - actually, a general twodimensional isotropic Binomial Point Process (BPP) [8]. The distributions of the distance of the n -th
nearest neighbour for each geographic forwarding decision regions are then generated [9]. It is worthy
emphasizing that multiple simultaneous connections are possible in such multi-hop scenarios.
2.1.2.1
Network Model
Nodes are uniformly and randomly deployed over the network area. The distribution of the distance
of the n -th nearest neighbour is derived for the homogeneous Poisson Point Process (PPP) with the
rate parameter λ . A simplified two-state (busy, idle) channel model based on the Unit Disk Graph
(UDG) is employed, i.e., awake neighbours dwelling in source's fixed radio range are considered
eligible relays. Consequently, neighbours located at longer distances than sources' transmission range
cannot operate as relay nodes. The network routing and the Contention Resolution (CR) mechanisms
exploit local knowledge of network topology aiming at improving effectiveness of the proposed
solutions. The next hop relay is selected so as to achieve the maximum expected advancement towards
the final destination. Every node knows its own location (real or virtual position) and the location of
the destination that is propagated in the header of packets.
2.1.2.2
Contention-based Geographic Forwarding Strategies
The sectoral and convex lens Geographic Forwarding (GF) designs are utilized so as to map the
location of eligible relays to the replying priority. In [6], the Convex lens Decision Region (CDR) is
effectively employed to select adequate relays and lead the contention avoidance and resolution stages
of the Medium Access Control (MAC) RMA scheme. Therein, neighbour nodes dwelling in source's
transmission range independently split themselves based solely on their locations and reply in
accordance with the resultant priority ordering. Whether collision occurs, nodes recursively consider
Page 13
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
sub slices of the original decision regions in order to resolve the contention by autonomously
readjusting their priorities of replying.
Two distinct CGF schemes, which properly combine GF designs and RSAs, are introduced and
evaluated in the following sections: (i) SDR combined with Standard Tree Algorithm (STA)-based
RSA; (ii) CDR used in conjunction with the auction-based RSA.
2.1.2.2.1
SDR Combined with the STA-based RSA
The STA is based on the splitting tree algorithm and tailored to RMA communications [10]. The
Contention Resolution Algorithms (CRA) is implemented considering the obvious Blocked Access
Protocol (BAP), which means that no contending relay is allowed in the transactions once the
contention has been already initiated. The performance of the STA-based RSA is addressed by means
of computational simulations in [6]. The source node always initiates the relay selection transactions
by issuing a Request To Send (RTS) packet. Afterwards, neighbour nodes that had listened to the
source's requisition split themselves randomly and independently based solely on the common
probability that dictates the likelihood of accessing the shared channel – totally random approach.
During the contention resolution interval, a neighbour may awake but it is not allowed to take part in
the ongoing transactions [12]. Eventually, if no suitable relay is found in a given Relay Selection (RS)
interaction, the source node goes to sleep and after a predefined interval restarts the connection
addressing probably new players.
Whether eligible relays collide, nodes that have transmitted in the previous slot decide to retransmit or
to refrain tossing a Q -sided coin with fair probabilities. Herein, only binary trees are actually used in
the investigations, though the observations are equally applicable to higher number of splitting groups.
And yet the source node should receive the replies from all the candidate relays so as to select the next
relay greedily - to select the closest node to the destination whether there is one available.
Figure 1 illustrates the sectoral forwarding design and can be interpreted as a snapshot of the network
routing procedure while finding a suitable relay to forward a data packet. As an illustration, after
source node had broadcast the RTS message, candidate relays {21, 55, 87} promptly reply in the same
time slot and a collision unavoidably occurs. Notice that dashed lines defined the forwarding region,
while filled circles identify candidate relays.
Page 14
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Figure 1: Illustration of the sectoral decision region.
The area of the sectoral forwarding decision region represented in Figure 1 is given by equation (2.1).
θ
Asec tor = r 2 .
2
(2.1)
where r corresponds to the source's transmission range and θ spans the forwarding sector.
2.1.2.2.2
CDR Combined with the Auction-based RSA
In this CGF approach, the auction-based RSA deals with the collision avoidance/resolution stages by
employing economic game-theoretical concepts. Typically, when a source node has a packet pending
for transmission, an RTS/CTS handshake is triggered to prevent simultaneous transmissions [2].
However, each suitable relay dwelling in the source's transmission range, which has received a RTS
packet, will reply with a CTS packet and collisions may occur. Thus, an appropriate mechanism has to
be employed in order to cope with the imminent contention.
In [6] Dutch auctions are proposed as an effective alternative to address the RS process in conjunction
with a Random Channel Access (RCA) solution. In such transactions, source node plays the role of an
“auctioneer” and potential relays are the “bidders”. Indeed, Dutch auctions are extremely convenient
to sell goods - assignment of network resources -- quickly. The reasons are two-fold, the auction ends
with the very first bid and the auctioneer may set accordingly the decreasing rate of the artefact value
(depreciation rate) aiming at quickening the auction [13]. The driving idea to use Dutch auctions is
that players interacting in a non-cooperative game will follow the global strategy (Auction-based
RSA) in order to maximize its own pay-off function in such a way that social welfare is obtained
(optimal strategy). The price is derived from the separation distances between source, relays and sink
(similar to the greedy forwarding procedure).
If there is more than one potential relay in a certain forwarding region, a collision will certainly occur.
To deal with this situation two approaches may be employed effectively: either any simple splitting
tree collision resolution algorithm is used [10], [14] (see section 2.1.2.2.1) or, preferably, a recursive
and iterative auction-based strategy is utilized.
When the recursive auction-based strategy is employed, the source node recomputes the forwarding
regions whether a collision is detected. However, since the source node uses the location information
Page 15
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
in a sequentially increased manner, only the colliding region is readdressed. Addressing the time slot
in which the collision occurred (and corresponding forwarding region) indirectly identifies the
colliding nodes. Equally important, potential relays that have not replied in the previous slot but
detected the collision just drop out of the ongoing transaction, since nodes with higher priority have
already replied. After recomputing the new forwarding regions, the source propagates an RTS packet
and a new auction round starts.
Figure 2 illustrates the computation of the forwarding regions. For the first auction round, the relay
candidates are divided into three groups,
{[21, 87] , 55, 11} . Since nodes {[21, 87]} reply at the same
time slot, a collision occurs. The source detects the collision in the first slot and recomputes the
forwarding regions accordingly. Whenever the BAP is considered, the contending nodes can also
recompute the forwarding regions independently. Nodes {55, 11} also detect the collision and, as
nodes of higher priority have already replied, drop out. Afterwards, relays in the first colliding area are
reordered in the sequence
{[ ], 87, 21} . Finally, after the first idle slot, node 87 replies and the auction
stops. The dashed lines define the forwarding region and the filled circles identify candidate relays.
Figure 2: Illustration of the convex lens decision region.
Equation (2.2) gives the area of the convex lens decision region.
Alens = 4r 2 (π − θ ) cos2 (π − θ ) + ⎡⎣ 2φ + sin( 2φ ) − π ⎤⎦ .
(2.2)
Equation (2.3) yields the intersection area of a circle and a half-plane, while the intersection area of
two circles is given by equation (2.4). Equations (2.3) and (2.4) are used to recursively compute the
area of the CDR and the corresponding order statistics of the distance the n -th nearest neighbour.
1
⎛d ⎞
A ( d , r ) = π r 2 − d r 2 − d 2 − r 2 arcsin ⎜ ⎟ .
2
⎝r⎠
(2.3)
where d is the distance of the center of the circle centered at the source node to the edge of the halfplane, which is defined by the chord in the intersection points of the source's radio range and the
dashed arc defining the forwarding region (see Figure 1).
Page 16
EUWB
Initial development of dissemination methods and evaluation
⎧
π r12
⎪
0
⎪
AI = ⎨ 2
⎪π r1 − A ( s , r1 ) + A( s + d , r2 )
⎪⎩
A ( d1 , r1 ) + A ( d 2 , r2 )
, D ≤ r2 − r1 ,
, D ≥ r2 + r1 ,
, D 2 < r22 − r12 ,
, D ≥ r22 − r12 .
D4.2.1
(2.4)
where D is separation distance between the circles centers.
2.1.2.2.3
Contention-Based RSAs
The analysis is driven considering two distinct contention-based Relay Selection (RS) mechanisms: (i)
a purely random solution based solely on the splitting tree algorithm for performing RMA
communications [4]; and (ii) a game-theoretical approach exploiting location information to avoid
collisions and, whether necessary, to improve the contention resolution period [3].
The Contention Resolution Algorithms (CRAs) are characterized in terms of the corresponding CRIs
necessary to select the most suitable relay – in terms of geographic advancement towards the final
destination – among the eligible neighbours. In order to characterize the incurred delay of such
solutions so as to forward packets towards destinations, the PGF is initially used to represent the PMF
of the CRI by means of power series, and therewith the entire distribution is recovered by numerically
inverting the corresponding PGF using the Fourier series method introduced in [5].
2.1.2.2.3.1 Delay analysis of the STA-based RSA
In this section we derive the Probability Generating Function (PGF) of LN – the conditional CRI
length when N nodes initially collide. Equation (2.5) generalizes the conditional CRI length
considering a Q -sided fair coin.
1
⎧
⎪
Q
LN = ⎨
⎪1 + ∑ LI j
⎩ j =1
, N = 1,
(2.5)
, N ≥ 2.
where I j is the discrete random number of candidate relays that tossed the j value on the Q -side
coin.
The PGF of the LN (CRI length) is defined as follows,
Δ ∞
GN ( z ) = ∑ P {LN = k } z k = E {z LN }.
(2.6)
k =0
where LN is a discrete Random Variable (RV) assuming non-negative integer values, and E {•} is the
expectation value. And so taking the conditional expectation on the right-hand side of (2.6) leads to
{
}
E {z LN } = E E {z L | I1 , …, I Q }| N ,
GN ( z ) = z
⎛ N ⎞ Q
ij
⎜ i , …, i ⎟ ∏ GLI j ( z ) Pj .
∑
Q ⎠ j =1
i1 , …, iQ ⎝ 1
(2.7)
N
(2.8)
where PjiJ is the probability of i j nodes flip the j side of the Q -sided fair coin. The summation
iterates over all possible combinations of the splitting groups i1 , …, iQ .
Page 17
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Hereafter we restrict our analysis to the binary case ( Q = 2 ). Whenever a collision occurs, i.e., N ≥ 2 ,
the candidate relays split themselves into two subsets. Each sub-CRI is statically indistinguishable
from a CRI initiated by the same number of contenders.
I 1 = i , I 2 = N − i.
(2.9)
Equation (2.10) yields the probability of exactly i nodes toss 0 (first splitting group), and then
transmit in the very next frame,
⎛N⎞
N −i
BN ,i = ⎜ ⎟ p i (1 − p ) .
⎝i ⎠
(2.10)
where p corresponds to the probability of tossing 0 when using the fair binary coin.
Finally, the PGF of the STA-based strategy is then given by equation (2.11),
N
GN ( z ) = z ∑ BN ,i Gi ( z ) GN −i ( z ).
(2.11)
i =0
where Gi ( z ) addresses the collision among i nodes that flipped 0 (first subset), and GN −i ( z )
corresponds to the additional slots to resolve the collision among N − i nodes that flipped 1 (second
subset).
2.1.2.2.3.2 Delay analysis of the Auction-based RSA
In accordance with the auction-based RSA, potential relays that have not replied in the previous slot
but detected a collision just drop out of the ongoing transaction. Indeed, the “tree pruning” means that
whenever the first subset visited after a collision leads to another collision, the second subset is
dropped. Therefore, for i > 1 , instead of the CRI having length BN ,i Gi ( z ) GN −1 ( z ) , the tree pruning
procedure leads to a shorter length BN ,i Gi ( z ) . Equation (2.12) yields the PGF of the auction-based
RSA [10].
N
GN ( z ) = z 2 BN ,0GN ( z ) + z 2 BN ,1 + z ∑ BN ,i Gi ( z ).
(2.12)
i =2
where the first term accounts for case when no reply is issued in the first slot, and the second term
addresses the cases when there is only one eligible neighbour in the first decision region.
2.1.2.2.3.3 Numerical Inversion of PGFs
The PMF of the CRI length is recovered from the corresponding PGF by means of a numerical
inversion technique introduced in [11] that relays on the Fourier series method. The CRI distribution is
approximated by equation (2.13),
P {LN = k } =
1
2kr k
2k
⎡
⎛ πkji ⎞ ⎤
⎟ ⎥,
N ⎜ re
⎠⎦
⎣ ⎝
∑ ( −1) Re ⎢G
j =1
j
(2.13)
Additionally, a predetermined error bound is derived for 0 < r < 1 and k ≥ 1 as follows,
P {LN = k } − P {LN = k } =
r 2k
.
1 − r 2k
(2.14)
Page 18
EUWB
2.1.2.3
Initial development of dissemination methods and evaluation
D4.2.1
PDF of the distance of the n -th nearest neighbour
A Spatial Poisson Process (SPP) in the two dimensional space is used to derive the distribution of the
Euclidean distance of the n -th nearest neighbour regarding the two distinct forwarding decision
regions. In fact, the distribution on the n -th nearest point from an arbitrary origin (location of the
source node) is given by a two-dimensional isotropic BPP. In other words, given that N (W ) = n , the
conditional distribution of N (W ) for B ⊆ W is binomial [8].
⎛N⎞
N −k
P {N ( B ) = k | N (W ) = n} = ⎜ ⎟ p k (1 − p ) .
⎝k⎠
where p =
(2.15)
λ2 ( B )
. W and B are bounded closed sets in R 2 , such that, B ⊆ W and 0 < λ2 (W ) < ∞ .
λ2 (W )
λ2 ( • ) corresponds to the Lebesgue measure in the plane [8].
Notice that the random position of each neighbour is uniformly distributed within source's
transmission range, therefore the probability of finding a random point in any bounded set B in R 2 is
given by [8],
P { X i ∈ B} = ∫ f X ( x ) dx ,
B
(2.16)
λ ( B ∩W )
P { X i ∈ B} = 2
.
λ2 (W )
where X i are i.i.d (independent and identically distributed) random points which are uniformly
distributed in W and f X ( x ) is the probability density function of each X i .
Therefore, the probability of existing less than n points in the decision region corresponds to the
Complementary Cumulative Distribution Function (CCDF) of the n -th nearest neighbor F Rn ( r ) and
is given as follows [9],
n −1 N
⎛ ⎞
N −k
F Rn ( r ) = ∑ ⎜ ⎟ p k (1 − p ) .
0 ⎝ k ⎠
(2.17)
Whenever a and b are integer values, the Regularized Incomplete Beta Function (RIBF) can be
expressed as in equation (2.18),
I x ( a ,b ) =
( a + b − 1)!
∑ k !( a + b − 1 − k )! x (1 − x )
a + b −1
k
a + b −1− k
(2.18)
k =a
And yet the CCDF can be rewritten in terms of the RIBF as follows,
F Rn ( r ) = I1− p ( N − n + 1, n )
(2.19)
The Probability Distribution Function (PDF) of the n -th nearest eligible relay is then derived as,
f Rn ( r ) =
d
I p ( n , N − n + 1)
dr
(2.20)
where I x ( a ,b ) = 1 − I1− x ( b, a ) .
Page 19
EUWB
2.1.2.4
Initial development of dissemination methods and evaluation
D4.2.1
Preliminary Results
In this section the CGF strategies are analyzed in context of autonomous multi-hop scenarios.
Regarding the CGF schemes, the achievable advancements at each hop are evaluated together with the
corresponding resolution intervals incurred by the conflict resolution algorithms. The network
simulator introduced in [15] is used to perform simulation campaigns and estimate the distributions of
the Contention Resolution Intervals (CRI).
Figure 3 presents the PMF of the CRIs that is achieved when the STA-based Contention Resolution
(CR) mechanism is employed. The distribution of the CRI is generated for increasing values of
contending relays. Computational simulations are also provided so as to corroborate the theoretical
analysis using the PGFs. The network simulator introduced in [15] is used to perform the simulation
campaign. When using the totally random CR approach the CRI significantly lengthens by considering
increasing number of contending relays. In fact, the resolution of the conflict may linger too much
time before the contention is resolved and the next-hop relay is elected. The Number of relays
identifies the number of candidate nodes that initially collide.
Figure 3: PMF of the CRI for the STA-based RSA.
For the Auction-based RSA, the PMF of the CRIs is presented in Figure 4. The impact of the initial
number of colliding nodes on the duration of the contention resolution period is still observed, though
in much lesser extent. As can be seen from the figure, by autonomously using the location information
eligible nodes can independently split themselves into priority groups and effectively quicken the RS
transactions. It is clear that location awareness significantly improve the CR capability of the gametheoretical RS process. By comparing Figure 3 and Figure 4 and considering two eligible relays only,
the probability of resolving the contention in seven slots is nearly 12% for the STA-based procedure,
whereas the same CRI has probability of approximately 7% for the auction-based RS strategy.
Page 20
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Figure 4: PMF of the CRI for the auction-based RSA.
Figure 5 shows the expected value of the distance of the n -th nearest neighbour for the CGF schemes
defined in section 2.1.2.2 (using the sectoral and convex lens decision regions). As can be seen from
this figure, the higher the number of neighbour nodes within source's transmission range, the further is
the expected distance of the furthest eligible relay independent of the forwarding region design.
However, for the evaluated number of relays (from 2 up to 11 nodes), the distance of the furthest
nodes do not increase substantially by considering higher number of relay candidates.
Page 21
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Figure 5: Expected distance of the n-th nearest neighbour (radio range of 10m).
Concerning the expected advancement provided by the furthest neighbour node within source's radio
range, the SDR and CDR schemes are equivalent. Conversely, the expected distance of the nearest
node not only experience higher variance in hop length, but also the expected advancement become
even smaller since the nearest node is found closer to the source when the number of candidate relays
increases.
Figure 6 and Figure 7 relate the expected distance of the n -th nearest eligible relay to the
corresponding CRI, when using the STA-based and auction-based RSA, respectively.
Page 22
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Figure 6: Expected distance of the n-th nearest neighbour relative to the expected value of the CRI.
Figure 7: Expected distance of the n-th nearest neighbour relative to the expected value of the CRI.
Page 23
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
As can be seen from Figure 6 and Figure 7, the CRI militates against the benefit of having higher
number on eligible relays within radio range, since depending on the initial number of colliding nodes
the contention resolution interactions may linger too long and then compromises the overall network
performance. While the two forwarding decision regions provide comparable advancements towards
the final destination at each hop, except for the nearest neighbour case, the CRI undergone by both
RSA is a determinant factor for system performance. Therefore, it is advantageous to keep small the
number of relays that entangle into the election process, since the relay selection procedure adds
substantially to the packet propagation delay in a hop-basis. Note that the eagerness of candidate
relays to take part in the RS transactions can be effectively controlled based on meaningful figures of
the network dynamics, such as, perceived interference and residual energy, so as to appropriately limit
number of contending nodes.
2.1.2.5
Final Remarks and Perspectives
The distribution of the distance of the n -th nearest eligible relay is characterized employing the
different CGF strategies. Two distinct geographic decision regions are considered so as to derive the
corresponding distributions of the expected advancements towards the final destination, namely
sectoral and convex lens decision regions. The CGF strategies are characterized by geographic
forwarding areas, which are used in conjunction with RSAs - a totally random approach and a game
theoretical alternative. The expected advancements each CGF strategy is characterized for increasing
number of contending nodes. The CGF scheme utilizing the convex lens decision region and the
auction-based RSA outperforms the STA approach for both the long- and short-hop localized routing
strategies. Additionally, PGFs are utilized to recover the distribution of the CRI incurred by the
assessed conflict resolution procedures.
It is worth emphasizing that different optimality criteria reflecting network dynamics may also be
exploited to further augment the priority mapping functions and increase the effectiveness of the
selection procedure, e.g., residual energy, perceived interference levels, and buffer occupancy and so
forth. Up to now, such intrinsic heterogeneity of nodes taking part into the network transactions has
not been explicitly addressed, though it is a major issue to be addressed in the forthcoming
investigations. Indeed, the diversity of characteristics and capabilities among network nodes can be
advantageously exploited [16]. Regarding the position-centric routing strategies, the long-hop
approach is less sensitive to the number of contending relays, therefore an favourable trade-off
between the expected advancement and the corresponding CRI is easier to achieve.
During this first period of Task 4.2, theoretical knowledge has been acquired and a suitable evaluation
framework is established resorting to both mathematical tools and computational simulations.
Therefore, the applicability of the solutions is initially appraised throughout computer simulations to
gain insightful comprehension of the topics under investigation; thereafter a theoretical analysis may
be employed to establish pragmatic limits. The investigations were carried out relying on the simple
UDG model, although it is well known that the channel variations severely impact the proper
operation of network functionalities. For that reason, the utilization of more realistic channel models
considering large-scale and small-scale fading manifestations is of great relevance to achieve more
accurate results.
2.1.3 Routing Discovery and Maintenance Strategies in Autonomous Multi-hop Networks
In ad hoc communication systems, both the infrastructureless characteristic and the operational
dynamics impose extra cost on the transmission of information. Typically, information is transferred
Page 24
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
through multi-hop links, which present variable characteristic - propagation gain, reliability, packet
reception probability and so on (see section 2.1.2). The establishment of such multi-hop links is
critical inasmuch as the pool of available resource is rather limited.
This section particularly introduces the problem of routing discovery and maintenance in ad hoc
networks. In order words, different mechanisms and protocols for creating and conserving multi-hop
routes up-to-date are examined in this section. An underlying concept is to exploit the availability of
cooperative communications in relaying networks so as to benefit from spatial diversity gains.
The reminder of this section is organized as follows. Section 2.1.3.1 characterizes the target scenarios
for evaluating the applicability of the investigated solutions. The envisaged performance metrics are
succinctly shown in section 2.1.3.2. Section 2.1.3.3 summarizes widely known solutions to deal with
the network routing, which are broadly divided into proactive and reactive solutions. Following, an
original solution based on hierarchical beaconless routing is presented in section 2.1.3.4. Final remarks
and observations are provided in section 2.1.3.5. Additionally, directions to extend the initial studies
introduced herein are provided along with the next steps.
2.1.3.1
Target Scenarios
The investigations are driven in the context of heterogeneous scenarios following guideline s provided
in [16], where distinct radio access technologies may be advantageously exploited. Indeed, the initial
studies that had hitherto been developed focusing on homogeneous network configuration will be
diversified in forthcoming stages of the work. Large-scale wireless networks deploying hundreds or
even thousand of nodes where information exchange is primarily achieved through multi-hop
interconnections are under investigation. Actually, the initial accomplishments regarding the formation
of links in multi-hop relay networks for transmission of signalling and payload information are
presented in these introductory investigations. In large-scale networks presenting rapidly varying
environments owing to nodes mobility, heterogeneity of traffic and channel impairments, purely
centralized coordination strategies may inflict significant signalling overhead and undesirable
consumption of the already limited resources on the network overall. Therefore, distributed
approaches to control routing operation are implicitly preferred. Not only flat organization of the
network, where all nodes indistinctly undergoes the very same role in the routing procedure, but also
hierarchical organizations with specialized nodes constructively cooperating are plausible scenarios.
The preliminary set of investigations provided here considered the Unit Disk Graph (UDG) model,
consequently only nodes dwelling source’s transmission range are able to communicate. All nodes
have the same transmission power. A fixed packet length is assumed as well. A simple two-state
channel model (idle, busy) is utilized.
2.1.3.2
Performance metrics
The suitability of existing performance metrics depends on intrinsic characteristics of network under
consideration. Commonly used metrics, which are relevant for the studies, are summarized in the
following list [20].
1. Hop count: Regarding a certain routing strategy (long- or short-hops), it simply accounts for
the number of hops between source and final destination.
2. Expected Hop Count (EHC): When considering more realistic channels models, the EHC is
more appropriate once it incorporate the expected number of retransmissions due to packet
reception probability and the corresponding expected number of acknowledgements [18].
Page 25
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
3. Energy consumption: in the nodes there is a limited availability of power supply that may lead
network to path failure due to energy depletion. Then, it is also desirable to address the impact
of routing strategies on the residual energy of network nodes (power consumption).
4. Path reliability: Captures the percentage of time routes are available (node mobility effects can
be incorporated by this metric).
2.1.3.3
Routing Protocols for Infrastructureless Networks
Routing protocols are usually categorized as proactive and reactive. Proactive routing approaches, also
known as table-driven routing protocols, strive to create and maintain routes by acting in advance
rather than responding to meaningful circumstances, such as changes in network topology.
Additionally, each node should acquire and keep routing information regarding every other node in the
network. Of course, there are scenarios where table-driven solutions are fairly applicable and
desirable. However, when considering larger-scale deployments (in number of associated nodes) and
highly dynamic scenarios, the operation of such protocols becomes more demanding and costly, and
their feasibility is definitely compromised. Conversely, on-demand routing protocols only react to
relevant events that are conveniently initiated by the source node. Certainly, hybrid approaches are
possible as well. Furthermore, in order to construct routes each node restlessly assesses the cumulative
cost of incoming multi-hop links from source to itself so as to decide optimally (given the constraints)
the proper routes to participate [19].
The solutions considered herein exploit location awareness to enhance the operation of network
routing procedures. Notice that routing protocols may or may not take advantage of the knowledge of
the network topology (nodes positions or relative distances), either utilizing localized or global
information. All in all, it is evident that the utilization of a specific routing solution is highly
dependent of the network inherent characteristics, namely, network size, mobility profiles of nodes
and traffic patterns. It is worth emphasizing that control overhead is a key aspect for the derivation of
adequate solutions at the risk of compromising overall performance by wasting the scarce radio
resources (see section 2.1.2).
Hereafter, a non-exhaustive list of the most common protocols to attain network routing is presented.
The solutions are enumerated based on their favourable characteristics and applicability in the
scenarios under investigations. Solutions requiring global and up-to-date knowledge of the network
topology and demanding frequent updates are seemed to have limited applicability [20].
Destination-Sequenced Distance-Vector Routing (DSDV) is a table-driven routing protocol based on
the Bellman-Ford protocol, while resolving the problem of routing loops. Routing information is
disseminated throughout the network issuing infrequent full dumps and more frequent incremental
updates aiming at reducing signaling overhead. Every node in the network maintains a routing table
recording all possible destinations in the network and the necessary number of hops to reach them.
Cluster head Gateway Switch Routing protocol (CGSR) is a table-driven hierarchical strategy where
cluster heads are nodes with specialized functionalities in the network. There is the obvious
disadvantage of recurrent election of cluster heads adversely affecting the routing performance. The
CGSR protocol makes use of the DSDV solution as routing strategy.
Wireless Routing Protocol (WRP) is another proactive routing strategy for infrastructureless networks.
This solution also utilizes the Bellman-Ford algorithm to compute paths, whereas being free of routing
loops and precluding the “count-to-infinity” problem. Despite the WRP presents faster convergence
and demands fewer updates than DSDV, a larger memory and greater processing is necessary to keep
Page 26
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
the consistence of multiple tables (distance table, routing table, link-cost table and message
retransmission list table).
Ad hoc On-Demand Distance Vector (AODV) Routing is an active approach to perform unicast and
multicast routing. The primary benefit of using this protocol resides on the fact that routes are only
created when necessary or demanded by a source node. However, periodic beaconing equally leads to
unnecessary bandwidth consumption.
Dynamic Source Routing (DSR) establishes routes on-demand only. This approach is truly sourceinitiated routing, which eliminates the need to flood the network periodically aiming at updating route
tables. The DSR protocol performs well in static and low mobility environments; however, in rapid
varying scenarios this solution presents inadequate performance.
Temporally Ordered Routing Algorithm routing protocol (TORA) is a non-hierarchical routing
approach and tailored to highly dynamic mobile networking environments providing a high degree of
scalability. It is also source-initiated and provides multiple paths between any pair of interconnected
nodes.
Different than the aforementioned typical approaches, where entire routes are established prior to the
actual data transmission, by using the clusterized beaconless solution proposed herein, the routes are
established on-demand and autonomously resorting to a kind of store-and-forward approach. Notice
that such store-and-forward solutions are more applicable to delay-tolerant networks and to deal with
best effort and background services.
Localised position-centric routing protocols aiming at minimizing, for example, the expected hop
count or equivalently the expected E2E packet delivery latency are preferred approaches (see section
2.1.3.2). Hierarchical routing protocols providing hybrid solutions for network routing are also
promising solutions in the heterogeneous radio access technologies that are considered in [16]. Equally
important, power aware routing protocols constitute another interesting research strand.
2.1.3.4
Hierarchical beaconless routing using contention-based RSAs
In Wireless Sensor Networks (WSNs), especially those of larger scale employing hundreds or
thousands of sensors (nodes), information exchange between sensors separated by distances exceeding
radio range requires packets to be routed towards their destinations through sequences of relays [19].
Unlike other large-scale systems such as cellular and computer networks, where routing can be aided
by tables maintained in certain nodes (routers), WSNs often require reactive routing protocols that can
cope with the constant topological changes resulting, e.g., from nodal duty-cycles and variations in the
radio range due to environmental conditions [20].
When nodes have knowledge of their location relative to one another, even if not exact [21], greedy
geographic routing may be an effective technique to autonomously and efficiently route packets
through the network [22]. Unfortunately, a well-know problem of simple greedy geographic routing is
that packets may be lost if anywhere along the multi-hop chain a relay in the forwarding direction
cannot be found. This dead-end problem has prompted the proposal of several clever mitigation
techniques, with approaches ranging from purely autonomous, to centralized1 methods such as graph
planarization [14]. The robustness associated with purely autonomous dead-end bypassing techniques
may, however, come at the expense of efficiency, since the discovery of alternative routes in the
presence of a dead-end often require excessive back-and-forth signalling. Indeed, although an
autonomous method is a sensible choice in networks with volatile topologies, in many cases global
topological features of networks are stable enough to be exploited by routing protocols.
Page 27
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
In this contribution, an alternative to landmark-based geographic routing for infrastructureless
networks is proposed. These two principles: that tiles are made up of strongly interconnected nodes;
and that nodes at the boundaries of adjacent tiles are also interconnected, are what improve the
probability of success of autonomous packet-relaying within tiles, while ensuring global reach via a
succession of “tile-to-tile hops”.
2.1.3.4.1
Solution Description
This section provides a detailed description of the cluster-based geographic routing protocol that is
mainly targeted on large-scale infrestructureless networks. The proposed position-centric routing
protocol is composed of two operational parts: global routing among clusters and localized relayselection inside clusters. The former, which corresponds to the main contribution of this work, exploits
network graph spectral analysis to divide network into clusters composed of sufficiently connected
nodes; while the latter deals with localized relay-selection inside clusters using (not necessarily, but
for the time being) simple greedy forwarding.
2.1.3.4.1.1 Hierarchical Clusterization Procedure
In this subsection we briefly review the clusterization method first introduced in [23]. This
clusterization procedure is a graph-spectrum mathematical tool employed to divide the network into
sub-groups, named clusters, whose constituent nodes are sufficiently connected to one another.
Figure 8: Illustrative output of the clasterization algorithm.
The clusterization method relies on the graph-theoretical spectra analysis of the network graph.
Specifically, clusters are identified by evaluating the eigenvalues and eigenvector components of the
Laplacian matrix associated to the meshed network graph [24]. The method progressively decomposes
Page 28
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
the overall network into sub-clusters using as stop criteria either the minimum completeness or the
compound number of nodes. For details please refer to [23]. Figure 8 exemplifies the output of the
hierarchical clusterization procedure.
Hierarchical Clusterization Procedure: iteration of Simon’s Spectral Bisection
1: Get the Connectivity matrix C1(,k2) of G1,2 ;
2: Compute Fiedler vector b2( k ) ;
3: Sort the vector components and split by the median γ ( k ) ;
4: Identify G1 and G2 from C1( k +1) and C2( k +1) ,
{b( ) } = {b( ) |b( ) ≤ γ ( ) }
{b( ) } = b( ) \ {b( ) }
k
2i
k
2r
k
2j
k
2r
k
2
k
k
2i
C1( k +1) = C ( k ) ({i},{i}) ,
C2( k +1) = C ( k ) ({ j},{ j}).
(2.21)
5: Repeat recursively (divide and conquer approach).
2.1.3.4.1.2 Cluster-based Routing Protocol
The cluster-based routing protocol to be described shortly is designed under the following
assumptions. During a discovery procedure (which is not addressed in this contribution, though it
resides in the mainstream of upcoming investigations), each node discovers its neighbors and forwards
the connectivity information to a central processing unit, based on which the connectivity matrix of the
network is built. The central unit then clusterizes the network, as described in section 2.1.3.4.1.1, and
assigns an identification (ID) number to each cluster and computes their geographical location. For
short, the location of the geographical center of a cluster will hereafter be referred to as the cluster
location.
Notice that a real coordinate system is not strictly necessary. In fact, methods that do not relay on
actual location to route packets may be exploited to build a virtual coordinate system that can be used
as input to the cluster-based routing [21], [25]. Each node is then informed of which cluster it belongs
to and the IDs and geographical locations of adjacent clusters (this is a reasonable assumption since
the central unit has acquired knowledge on the shortest path to all nodes after network discovery).
The routing protocol can be separated into two functional parts, which are described subsequently.
Inter-cluster Routing Procedure
In addition to the cluster adjacency list (cluster IDs and virtual center locations) mentioned above, it is
assumed that each node knows its own position and the position of the sinks (the location of the final
destination is present inside packets header). Actual or virtual positions can be derived distributively
during the discovery procedure [26] or computed by the central unit and passed onto the nodes [27].
When a packet is generated and has to be forwarded to a given final destination (sink), the source node
initially chooses, among the adjacent clusters, an intermediate destination (cluster location), which is
the closest to the final destination. In other words, an adjacent cluster is selected applying a greedy
principle [22] based on real or virtual coordinates (more elaborated strategies are equally applicable as
shown in section 2.1.2.2).
Each packet conveys, besides payload data, the source position, the sink position (final destination),
the location of the next cluster (intermediate destination) and a list of the last visited clusters (the exact
Page 29
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
number of clusters that should be stored in order to avoid loops has not been optimally determined
yet). Consequently, routes are established on demand, but using the clusterization information stored
at the nodes, such that a packet may circumvent a hole (dead-end). The procedure described above is
referred to as inter-cluster routing.
Intra-cluster Routing Procedure
The routing procedure inside each cluster (intra-cluster routing) is largely independent of the intercluster procedure. In this article we considered a simple greedy forwarding mechanism for intracluster routing [22]. In other words, the greedy principle applied in the intra-cluster mechanism is
essentially the same as the one employed in the inter-cluster routing (with the exception that the latter
includes a logic to avoid looping).
The use of (essentially) the same mechanism at inter- and intra-cluster routing is rooted on the notion
that nodes can “re-use” algorithmic resources (and associated hardware/software). Furthermore, the
choice of a simple greedy mechanism does not preclude the use of more sophisticated techniques,
since the overall structure of the solution is essentially modular. For instance, Adaptive LoadBalanced Algorithm Rainbow (ALBA-R), which integrates Medium Access Control (MAC) and
routing in its design, guarantees packet delivery to the sink in the presence of holes (albeit at the
expense of a long “transient” period) [14]. It is evident that the latter technique can be integrated into
the clusterized framework discussed here as well.
The objective, however, is not to establish that clusterized greedy forwarding is superior to any
particular alternative, but rather to show that clusterization can substantially improve the performance
of routing mechanisms, even when such mechanism is as simple as greedy forwarding.
Pseudo-code for the Cluster-based Routing
Require: Cluster adjacency list (descending order), source node, destination node, traveled routing
path, neighbor nodes list (descending order).
1: Inter-cluster routing:
2: for all Adjacent clusters do
3:
if Cluster is NOT in traveled routing path then
4:
Select cluster as intermediate destination;
5:
end if
6: end for
7: Intra-cluster routing:
8: for all Awake Neighbors do
9:
Select the closest relay to intermediate destination.
10: end for
11: if NO suitable Relay then
12:
Select another intermediate destination.
13: end if
Page 30
EUWB
2.1.3.4.2
Initial development of dissemination methods and evaluation
D4.2.1
Preliminary Results
The performance of the proposed cluster-based routing protocol is assessed by means of an eventdriven system-level simulator implemented using C++ Object Oriented Programming (OOP)
language. The main results are presented in terms of the Cumulative Distribution Function (CDF) of
the Packet Delivery Success Ratio (PDSR) as a function of the corresponding Average Packet
Delivery Latency (APDL), and in terms of the distribution of route path length (in number of hops)
[15], [6].
Figure 9 shows the PDSR as a function of the APDL, for different levels of network clusterization
with a duty cycle of 10% . Notice that the network topology is not static, since nodes alternate between
awake and asleep states following their own duty cycle (even though there is no mobility, the network
topology varies with time). It is found that the simple greedy forwarding is severely affected by the
dead-end and fails to route packets originated from the concave cone region towards the sink. This
explains why the PDSR of the greedy algorithm is around 50% , i.e., only the traffic generated on the
convex side of the hole is successfully routed.
Conversely, the proposed clusterized routing solution is capable of delivering a substantially larger
fraction of the traffic to the sink. It is also found that the cluster size has an impact on the effectiveness
of the method. For instance, when the clusterization is such that clusters have at least ( N = 25 ) nodes,
the PDSR is higher. Such a higher delivery ratio, however, comes at the expense of extra latency,
since packets tend to circulate inside the concave region of the hole for longer, before finally being
routed around it. Notice that very small clusters may also degrade PDSR, because this may decrease
the probability of a node to find a relay in a selected adjacent cluster (intermediate destination). The
proposed routing technique proves to be resilient enough to absorb the variable network topology and
to continue delivering a much higher PDSR.
Finally, Figure 10 shows the distributions of the number of hops - in excess of what would take if
packets were to be forwarded in a straight-line, with zero awaiting time (perfect relaying) - required by
the routing protocols under the simulated conditions. The second mode of the distribution
corresponding to the proposed technique with N = 25 shown in Figure 10 illustrates how the proposed
routing algorithm allows more nodes - namely, those at the concave portion of the hole - a path to the
sink.
Page 31
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Figure 9: Packet delivery success ration with duty cycle of 10%
Figure 10: Excess-hop distribution with duty cycle of 10%.
Page 32
EUWB
2.1.3.5
Initial development of dissemination methods and evaluation
D4.2.1
Conclusions and Future Directions
a cross-layer algorithm for geographic routing in infrastructureless networks in introduced. This
solution is robust to topological holes and is resilient to topological variations due to network
dynamics (node duty cycle). The solution combines ideas of network tessellation with simple greedy
forwarding, without suffering from the problems afflicting typical landmark-based alternatives [15].
We first tessellate the network based on connectivity information and then define virtual landmarks to
identify them. The algorithm employed to clusterize the network is based on a recently discovered
graph-spectral property that captures the connectivity among nodes [23]. In addition to not requiring
landmarks to be known a priori, a major advantage of the proposed tessellation approach is that the
clusters are (by construction, as a result of the aforementioned clusterization technique) composed by
groups of nodes that are highly interconnected.
It was shown that cluster sizes may affect the resulting PDSR to APDL. In other words, cluster sizes
can be varied so as to allow for different trade-offs between PDSR to APDL to be reached. It was also
found, however, via extensive simulations, that the dependence of the overall technique on the cluster
sizes is not as strong as the dependence of landmark-based routing mechanisms on the location of
selected landmarks. Alternatively, we may say that while there is room for further improvement via
cluster size optimization, the proposed technique is less sensitive to a bad choice of cluster size than
landmark methods are on the location of landmarks.
All in all, the simulations campaign have confirmed that the technique can substantially improve the
PDSR in networks with large concave holes, with no or little impact on APDL, even if the simple
greedy forwarding mechanisms is employed. There is, however, no reason not to employ more
sophisticated mechanisms combined with the clusterization idea.
It has been assumed that network discovery had been already performed and that the network is in a
steady-state regime of operation, when the routing protocol itself starts to operate. The network
discovery has not been explicitly addressed in the research activities, though it is still an open issue
and it is considered to further investigations.
The concept of clusters may be further augmented to cleverly aggregate information among nearby
nodes, and consequently preserve network resources. The driving idea to perform such data fusion is
that inference resorting to gathered information from multiple sources (for example, co-cluster nodes)
is probably more effective and more accurate than if; alternatively, it is based solely on a unique
source. Techniques to combine and forward data collected from multiple sources are an interesting
topic for the following studies. Besides, the comparative investigations between routing discovery and
maintenance protocols optimizing meaningful network performance metrics are in progress.
Page 33
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
3 Co-Ordination Schemes and Information Exchange Protocols
3.1 Mechanisms for Information Exchange, Dissemination and Acquisition
in distributed, uncoordinated wireless sensor networks
3.1.1 Node architecture and fundamental models
The functional system architecture of a UWB node is depicted in Figure 11 and is divided into three
subsystems namely the UWB, the communication and the positioning/ranging subsystem
communicating via data and control interfaces with each other.
Figure 11: UWB Node architecture and application subsystem
The communication subsystem comprises DLL, networking and transport functionalities and can
configure the UWB subsystem for joint communication and positioning. Whereas the
positioning/ranging subsystem enables the UWB subsystem to obtain ranging measurement data and
in return receive range-related raw measurement data and quality metrics for position calculation. The
positioning subsystem generally consists of three main modules. First, the range management module
which is responsible for gathering of all data which are needed to process ranges/positions. This
implies the management of the own position as well as the neighbour positions and the treatment of
system clock values by forwarding the ranging clock values and their reliability to the range
calculation unit. Besides, it initiates ranging measurements and assists the route management of the
networking.
The second module is the range calculation which has the task to calculate ranges and range reliability
from the system clock information. All range information together with the fix anchor node positions
are sent to the position calculation module which consists of a positioning algorithm based on
Page 34
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
multilateration to calculate the mobile nodes position. The calculated position can be send to other
UWB nodes or to the application subsystem running on a conventional PC for monitoring purposes.
3.1.1.1
Packet Error Modeling
On its way from the transmitter to the receiver the signal does not only suffer from path loss, but also
from noise and interference. The PER calculation in the simulator consideres the cumulative
interference created during the whole duration of the packet transmission. The cumulative interference
is the weighted sum of the interferences created by simultaneousl transmissions of interfering packets,
that might occur on th other subchannels. Note that the overlap of the wanted packet with interfering
packets generates intervals with equal intereference conditions. The weights correspond with the
duration of these intervals. Figure 12shows this principle.
Figure 12: Packet level collision
A given packet is received in subchannel i and 3 concurrent transmissions of packets on subchannels
k, j and m are shwon in Figure 12. The overlap among the packets generates five different equal
interference intervals.
Given the reception of a wanted packet, in order to keep track of the intervals equal interference and
interference list at the PHY layer has been implemented. For each interval of equal interference the list
saves following information:
•
The interval duration
•
Signal parameters of each interfering packet,
o
THC (subchannel) on which the interfering packet is transmitted
o
PRF of the interfering transmitter
o
Power at which the interfering packet is received
For each interval of equal interference the number of channel bit errors is calculated. We consider a
probabilistic model to determine the number of channel bit errors per interval
N CHBERi = ∑ N CHBERi (t x )
(3.1)
x
Page 35
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
where t x represents the duration of the x -th interval. The packet duration can be calculated as the sum
of the duration of all involved intervals, that is Tpacket = ∑ t x . The resulting channel BER is given by
x
the ratio between the total number of channel bit errors in the packet and the packet payload length.
P ch e ,i =
N CHBERi
Lp
(3.2)
where Lp is the packet payload length. Finally the PER can be expressed by:
PERi = 1 − (1 − Pe ,i )
Lp
(3.3)
For a detailed derivation of the packet error modelling we refer to [42].
Air Interface Format
In order to serve both services namely communication and positioning in parallel, a common air
interface format is needed. Hence, common and private communication channels are provided by the
system.
The common channel will be transmitted in so-called beacon slots using a default set of PHY
parameters like TH codes, modulation and FEC allowing low data rate transmission but highly robust
communication. It usually carries broadcasting and handshaking messages but also short information
such as sensor data. The communication on the common channel is always unidirectional implying
that any data transmitted on the common channel will not be acknowledged.
On so-called TH channels large volume information is transmitted on private communication channels
which are always bi-directional as each packet has to be acknowledged by the destination node. The
PHY parameters of the TH frames are dynamically set as the destination node can inform the
transmitter on the link quality. This kind of medium access control has the goal to maximize the
cumulative communication performance and was first introduced in [43] and enhanced in [44].
For the ranging and positioning purpose we constrain on the beacon frames sent on the common
channel. Figure 13 shows the air interface frame structure with its beacon and TH frames.
Figure 13: Air interface frame structure
3.1.2 Ranging Information Acquisition
Each beacon frame splits into three beacon slots and are used for synchronization, protocol handshake
elements, very low data broadcast and especially for ranging requests and ranging answers.
Ranging is initiated by a UWB node sending a ranging request (RR) to a selected destination node
which usually is an anchor node answering with a ranging answer (RA) back to the initiating node. As
the ranging operation needs as strict timing between the request (RR) and the answer (RA), the third
beacon slot is always reserved for the ranging answer after a ranging request occurred. Thus, three
Page 36
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
beacon slots are gathered in one beacon frame. The second beacon slot can be alternatively used by
any other node to send control information or to broadcast common data channel information.
The duration of a whole superframe structure is 3.5 ms where the beacon frame has a duration of
3*50µs. This implies that one range measurement for one mobile can be done within one superframe
transmission in order to fulfill the requirements of range update time of 10 ms.
In order to carry out position calculation multiple ranging procedures to different anchor nodes are
needed. Figure 14 shows the principle of positioning with two-way ranging where the initiating mobile
node successively addresses three neigbour anchor nodes as an example.
Figure 14: Positioning principle using two-way ranging
Since a single range can be done within one beacon frame, the mobile node reserves the first slots of N
successive beacon frames in order to minimize the latency, see Figure 15. For each two-way ranging
measurement the time in between the transmission of RR and RA has to be corrected by the internal
delay to obtain the round trip time which can be directly translated into a distance between mobile and
anchor node.
Figure 15: Positioning estimation using N successive beacon frames
For detailed simulation results of the ranging performance we refer to [45].
3.1.3 Recource allocation strategy
We implement a novel resource allocation strategy for distributed IR-UWB networks that arise from
the fact that, in contrast to what happen in narrowband networks, bandwidth is no longer the limiting
resource but the number of pulses per second, i.e. the cumulative pulse load. The DLC enables
concurrent transmissions at full power, while allows each source to independently adapt its error
coding rate, and its pulse rate (the number of transmitted pulses per second) to the current channel and
interference conditions. Thus, our strategy combines adaptive channel coding (ACC) with pulse rate
Page 37
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
control (PRC) in a joint link adaptation function, which is denoted hereafter as the Link Parameter
Control (LPC) function.
3.1.3.1
DLC Layer Description
The DLC layer implements two main functional blocks: a medium-sharing block and the LPC block.
Furthermore, each DLC packet is acknowledged by the peer DLC entity so that a stop&wait ARQ
scheme is set up on top of the physical layer.
Regarding medium sharing, THMA has been considered. Besides, to solve collisions between sources
sending to the same destination, a RTS-CTS handshaking combined with a MAC address-based code
assignment protocol has been implemented. The RTS message is sent using a common TH code in the
second BS of a BF, while the CTS and the following data and ACK messages are sent using a private
TH code, which is generated using a random generator with seed the addition of the source and
destination MAC addresses.
Concerning the LPC function, it can be seen as a resource management problem that searches for the
network’s pulse rate and channel coding rate allocation which maximises the aggregated network
throughput while satisfies certain per-link BER constraints ( BERi ). LPC configures the signal
structure of scheduled transmissions at the individual link by optimizing the error protection level (PL)
to the current channel quality and by possibly switching the average pulse repetition frequency (PRF)
in order to relax the channel pulse load. The PL is an adaptable link parameter representing the
ordered pair channel coding rate and modulation order. For the purpose of our analysis, m-ary pulse
position modulation (PPM) is considered. Since it is well-known that in a multipath environment intrasymbol interference degrades the benefits of m larger than 4, only m = 2 or m = 4 are considered next.
Concerning channel coding, it is required that the coding rate is chosen from a finite set.
In order to adapt these parameters, the link’s transmitter must have an estimate of the level of
interference at its intended receiver. In autonomous networks, most approaches make use of feedback
information from the receiver to the transmitter, for example within ACK packets. This information
can take various forms; conventionally it is a function of the SNIR. However, with IR-UWB physical
layer, measuring the SINR is difficult in practice due to the very low transmit power of UWB signals.
Our approach relies on information provided by the channel decoder, i.e. on BER measurements.
It is assumed that the system has fixed chip duration Tc , so that all changes at the PHY layer
transmission parameters are induced by instructions coming from the MAC/DLC layer (cross-layer
character). The selected set of PHY parameters, i.e. modulation, channel coding rate and pulse rate,
remains constant for one DLC packet transmission, but can be changed from packet to packet
according to the time variant channel and interference conditions. The LPC algorithm runs in a MAC
frame basis, over uni-directional or bidirectional links. For bi-directional links where forward and
return link data are transmitted in a serial manner, the same TH code can be used. Nevertheless, the
PHY parameters might be different due to different bit rates and different interference situations.
Note that PL adaptation is a decision local to a sender-receiver pair, so that it does not need
coordination among neighbours not involved in the transmission. In contrast, PRC involves an
interaction among different links since the probability of pulse collision at the i-th receiver Pcoll ,i , does
not depend on its own link's pulse rate, but on the pulse rates of transmitters in its vicinity and their
asynchronism [46]; where low pulse rates imply low probability of collision. Since the collision
probability is an indirect measurement of the BER, we can expect that the BER at the i-th receiver Pe ,i
Page 38
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
does not directly depend on its link's pulse rate, but on the pulse rate of the neighbouring links. Thus,
PRC enforces cooperation among otherwise selfish nodes. The question is here, how can this
cooperation be achieved?
Game theory (GT) has been leveraged in the recent past to model complex interactions among radio
terminals [47]. Indeed, GT is a broad accepted tool for modelling and analyzing resource allocation
problems. Furthermore, GT allows solving such problems in a distributed way.
This subsection describes the physical link parameter control (LPC) functionality of our distributed
MAC mechanism. LPC constitutes a novel MAC strategy which flexibly configures the error coding
rate and the average pulse period per link to control the shared medium utilization. It is based on a low
complexity algorithm, which exploits generally applicable results of game theory.
Generally, an increase in the channel BER points at an increase in the local interference level seen by
the receiver node or/and at degrading channel conditions, e.g. due to node movement or the presence
of obstacles in the line of sight. In order to compensate the degrading link performance, conventional
rate adaptation algorithms just order the PHY layer to decrease the error coding rate (ECR). However,
under uncoordinated access the system load can become so high that pulse collisions may not be
longer tractable with binary error coding schemes. Under such conditions additional mechanisms may
be required to maintain an adequate system throughput.
Note that, with constant average pulse rate a decrease in the error coding rate would induce a decrease
in the link throughput. In order to avoid this throughput degradation, the LPC algorithm may try to
compensate the ECR decrease with an increase in the link’s average pulse repetition frequency (PRF).
LPC combines channel BER estimation with a MUI noise level indicator in its decision, whereas, the
decision on the average PRF adaptation is strongly influenced by the MUI noise level indicator and
regulated as a “game”. LPC doesn’t directly controls the ECR but an abstract parameter called
protection level (PL), which at the PHY layer maps a certain error coding rate or even a pair error
coding rate and modulation order.
LPC acts according following rules:
•
If the increase in channel BER is accompanied by a constant or even decreasing MUI noise level,
LPC tries to compensate the link throughput decrease with a higher link’s average pulse repetition
frequency (PRF). Since a higher PRF increases the local system pulse load, neighbouring receivers
may experience an increase in their channel BER and perform in the same previous elucidated
way, thus originating a feedback loop among the adaptation decisions of neighbouring nodes,
which may lead in a breakdown of the local resource (pulses per second). Such a feedback loop
can be regulated by making use of game theory results.
•
If the increase in channel BER is accompanied by an increasing MUI noise level, the LPC
function doesn’t changes the PL, but it orders the PHY layer to decrease its PRF in a factor
proportional to the MUI noise level increase. Thus, a temporal loss in link throughput has to be
accepted. By reducing its PRF, the reference link contributes to reduce the local system pulse load;
this will translate in a reduction of the channel BER in neighbouring links, which will then
consequently reduce their PRF (as we will see next) and the interference they cause in the
reference link’s receiver. In this way, the feedback loop among adapting links shows a beneficial
effect.
•
Otherwise, if a decrease in channel BER is reported, the LPC function orders the PHY layer to
decrease its PL and to reduce its PRF accordingly so as to maximize the link’s utility leaving
Page 39
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
minimal intervals between packet transmissions, while contributing to keep the local system pulse
load minimal.
3.1.3.2
Game theoretical formulation of the pulse rate adaptation
The pulse rate adaptation within the LPC function can be formulated as a cognitive algorithm using
the framework of game theory. By defining the pulse rate adaptation as a game, we believe it is
possible to motivate each link to reduce its pulse rate fairly and hence avoid resource overload.
The pulse rate adaptation (PRA) game can be modelled as an asynchronous myopic repeated game,
which can be mathematically defined as,
Γ PRA = M , A,{U k }
k∈M
,{d k }k∈M ,T
(3.4)
where M is the finite set of players, A is the strategy space formed as A = × Ak , k ∈ M , with
Ak = ( a1k , a2k ,..., a Nk ) being the set of actions (or strategies) associated with player k, U k : A → ℜ is the set
of utility functions that represent players strategy preferences, d k : A → Ak is the decision process that
guide player k’s action choices, and T represents the decision timings at which player k takes its action
choices.
A player’s action choice ak corresponds with an average pulse repetition frequency choice. The set of
available average pulse repetition frequencies is convex and bounded, on the one hand due to
regulation and technological constraints and on the other hand due to the application requirements. In
general it holds,
ak ∈ PRF = ⎡⎣ prf min , prf max ⎤⎦
(3.5)
As well, the set of possible protection level strategies, S = ×S , k ∈ M , is limited, i.e. it contains a
k
limited number of modulation schemes and error coding rate pairs.
For every player, the utility function uk characterizes the satisfaction level of player k for each
particular action profile a = ( ak , a− k ) ; i.e. uk, is a function of the action selected by player k, ak, and of
the action profile of the rest of players, which is represented as a-k. Thus, players make decisions
independently, but influenced by other players’ decisions. When analyzing the outcome of the game it
is important to determine if, given certain interference and channel conditions, there exists a
convergence point for the pulse rate adaptation, from which no player would deviate anymore, i.e. a
Nash equilibrium (NE). An action profile a , is a NE if and only if
uk ( a ) ≥ uk ( ak* ,a− k ) , ∀k ∈ M ,ak* ∈ Ak ,
(3.6)
The convergence of the adaptation algorithm depends significantly on the choice of the utility
function; concretely on its mathematical properties. Moreover, the utility function must have physical
meaning for the particular application.
3.1.3.3
Utility function
The utility function consists of two terms, a gain and a cost term. The gain term is associated to the
goal of keeping the link throughput as close to the application requirements as possible. The cost term
Page 40
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
is related to the node’s estimated local system pulse load and accounts for the interference that a
particular PRF choice will create to neighbouring nodes.
Mathematically the utility function can be defined as:
(
)
−
uk ( prf k , prf − k ;t ) = − ⎡ ΔPRFt arg et + prf kt − prf kt ⋅ (1 + α k )⎤
⎣
⎦
2
(3.7)
where ΔPRFt arg et represents the variation in pulse repetition frequency which the PHY layer should
accomplish in order to compensate the link throughput degradation associated to a certain PL increase,
and α k is the cost term.
0
alpha=0
alpha=0.09
alpha=0.9
-0.01
-0.02
-0.03
Utility
-0.04
-0.05
-0.06
-0.07
-0.08
-0.09
-0.1
1.2
1.3
1.4
1.5
1.6
PRF
1.7
1.8
1.9
2
5
x 10
Figure 16: PRA game utility function for different cost terms.
Each node evaluates its cost term based on local knowledge; no cooperation between nodes is
foreseen. The cost term consists in a weighted average of the node own MUI noise level estimations.
α k ( t ) = (1 − twin ) MUI kt +
−
1
twin
κ MUI kt
(3.8)
A low value of α k points at a locally low system pulse load and represents for the reference
transmitter none or little drawback when trying to increase its PRF to the target value. In contrast, a
high value of is related to a locally high system pulse load and leads the PRF adaptation to a lower
value than the target one; i.e. the reference link accepts the decrease in its own throughput helping to
keep the local system pulse load low instead.
Figure 16 depicts three realizations of the utility function for different values of the cost term and the
equal ΔPRFt arg et . It can be seen that higher values of the cost term, i.e. higher local system pulse load
conditions, move the maximum of the utility function to lower values of prf, thus helping to keep the
low system pulse load.
Page 41
EUWB
3.1.3.4
Initial development of dissemination methods and evaluation
D4.2.1
Existence of Nash Equilibrium
In order to have good convergence properties for the adaptation algorithm, the mathematical properties
of the utility function are of huge significance. This subsection shows that with the utility function
defined in (3.7) converges to a Nash equilibrium by using a best response decision rule.
Theorem 1: Γ PRA has a unique pure strategy Nash equilibrium.
Proof: First of all from [48] we know that given an asynchronous myopic game modelling a cognitive
radio network, M , A,{U },{d },T , where all players are autonomously rational, if the normal form
k
stage game
M , A,{U k }
k
has a NE a*, then a* is a NE for the cognitive radio network. Thus, in order
to proof the NE existence for Γ PRA
we can concentrate on the properties of the normal form stage
game.
From [49], we know that given a normal form game M , A,{U } , where Ai are nonempty compact
k
convex subsets of ℜ ∀k ∈ M . If ∀k ∈ M uk is continuous in
m
and quasi-concave in
then Г has a
pure strategy NE. If besides, is strictly concave in , then the NE is unique.
In Γ LPC the set of possible strategies is compact and convex, and the utility function uk ( prf ) is
strictly concave in prf, thus we can conclude that Γ PRA has a unique NE in the strategy PRF.
3.1.3.5
LPC Algorithm
LPC can run in a MAC frame basis, over uni-directional or bi-directional links. The link’s
transmitter node acts as the decision maker configuring the signal structure in response to changes in
the interference environment, which are feed back within ACK messages from the link’s receiver
node. The transmitter is then able to adapt the PHY parameters to the actual interference situation and
to send the next data packet with the new parameter set, which is reported in the packet header. For bidirectional links where forward and return link data are transmitted in a serial manner, the same TH
code can be used. Nevertheless, the PHY parameters might be different due to different bit rates and
different interference situations.
The algorithm begins with an initialization phase at which each user chooses some initial
protection level. The initial value of the pulse repetition frequency (PRF) is determined at each node in
order to optimize the link utilization, i.e., such that the total duration of the air frame is approximately
equal to the mean inter arrival time between packets at the application layer.
Table 1: Parameters involved in the LPC initialization phase
Initialization parameters
Appl Payload
500 to 2000 bits
Appl Data rate
10Kbps to 1Mbps
Net HL
Mac HL
Phy HL
128 bits
30 bits
80us
Default PL
Default PRF
Init PL
Init PRF
ECR = ½
500 KP/sec
e.g: 1,1/2,1/3,2/3,3/4
Page 42
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
The airframe duration depends on the init protection level, the application’s target data rate and the
amount of protocol overhead. Concretely, the air frame consists of three parts
•
PHY header, i.e. Pilot&Sync, of duration 80us
•
MAC header of duration 120 us, considering 30 bits, 500KP/sec and ECc=1/2
•
PHYPDU , whose duration depends on the MACPDU length and on the bit rate at the air interface
Pilot
Sync
Mark
Mac
Header
THC Body
5 MP/s
Default parameter set (500KP/s, 1/2)
LPC
Figure 17: Air Frame Format
For optimal link utilization, the bit rate at the air interface should be,
BitRatet arg et =
AirFramePDULength
⎡ ⎛ Payload app ⎞
⎤
⎟ − ( PHYH + MACH ) ⎥
⎢⎜
⎣ ⎝ DataRateapp ⎠
⎦
(3.9)
And thus, the initial PRF calculated by LPC is given by,
PRFInit =
BitRatet arg et
PLInit [bits / pulse]
(3.10)
Each time a node receives a new packet, the PHY layer estimates the channel BER and the MUI
noise level associated to this packet transmission. The channel BER estimate is obtained by reencoding the bit stream at the output of the viterbi decoder and comparing the coded bits with the hard
decision values derived from the decoder input. The MUI indicator is an estimate of the interferer
collision rate and is measured during idle periods, where the node itself neither transmits nor receives
pulses. N samples at the output of the energy detector are analyzed with regards to MUI interference.
The noise level, which is determined based on the median, serves as a basis for the interference (IF)
threshold. Samples exceeding the IF threshold are indicated as shot noise. The MUI indicator is the
relationship of the number of shot noise samples divided by the total number of measurement samples.
The MUI noise level is used to evaluate the node own cost term following equation (3.8). Besides,
the channel BER estimation together with the MUI noise level are attached to the ACK message,
which shall acknowledge the successfully reception of the data packet. Estimated values are certainly
random variables including short-term as well as long-term dynamics. In order to mitigate these
effects, the receiver node doesn’t feedback the current measured values but an average over certain
past observation window. ACK messages are always sent with the default PHY configuration used for
the MAC header. After receiving an ACK message, the link’s transmitter node extracts the attached
link quality information (LQI), evaluates it and decides on the best PHY parameter for the next packet
transmission. The LPC decision block implements following rules: i) if channel BER increases but
Page 43
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
MUI noise level doesn’t, the PL is increased and the game is played to decide on the PRF adaptation;
ii) if channel BER increases but MUI increases as well, PRF is decreased on a factor proportional to
the MUI noise level increase; iii) if channel BER decreases, the PL is reduced and the PRF is reduced
accordingly to maximize the link’s utility.
Within the LPC decision block, the utility function defined (3.7) is used to adapt the link’s average
pulse repetition frequency according to a best response dynamic, i.e. user k selects the average pulse
repetition frequency that maximizes its utility for a fixed a− k ,
(
)
prf k ( t ) = argmax uk ak ( t ) , a− k ( t − ) ;α k ( t − ) ,
ak ∈Ak
(3.11)
start
Init PL
PL0
Init PRF
PRF0:= f (PL0,appDR)
Rcvd
Packet?
yes
Measure:
cBER,MUI
ACK
Extract LQI:
cBER,MUI
LPC
Decision Block
Data
ACKGeneration
(LQI)
To PHY:
PL0,PRF0
Data, ACK
αk = ∫ MUIk
twin
Cost
To PHY:
PL*,PRF*
Figure 18: LPC flow-diagram
3.1.4 Simulation Scenarios and Evaluation of DLC-MAC Performance
The main objective of the simulations presented in this deliverable is to investigate the performance of
the MAC algorithm with its link adaptation for near-far scenarios where worst cases occur and for a
random scenario where the UWB nodes are randomly distributed. Unlike other simulations a statistic
channel propagation model has also been implemented where the receiving energy from the
transmitting node and all interferers follows a Log-Normal Shadowing model. A method for
implementing the probabilistic radio models in OMNeT++ was introduced by [50] and is used for our
simulations.
Page 44
EUWB
3.1.4.1
Initial development of dissemination methods and evaluation
D4.2.1
Near-Far Scenario
The near-.far scenario is a worst case scenario where the interferers are in a range of 1 m and the target
receiver is in a distance of 20 m. The number of nodes is continuously increased so that the farthest
interferer is in maximum range which corresponds to 20 m. The nodes are static and are distributed in
a 50x50x2 m area. The scenario is depicted in Figure 19.
Figure 19: Near-Far Scenario
In this section, the sensitivity of the game behaviour to different sets of initial conditions is examined.
Extensive simulations have confirmed that the LPC algorithm always converge to a solution in A ,
regardless of its initialization.
In particular, for a 2-link network, simulation results suggest the existence of a unique NE for the
game Γ LPC and that the LPC algorithm converges to it regardless of its initialisation
profile ( prf i max play ,0, 4) , and provided that the pricing parameter has been appropriately chosen.
Figure 20 depicts the PRF and PL steady-state allocation corresponding to the unique NE of Γ LPC in
the case of 2 links with information data rate 1Mbps. It can be observed that both links select the
maximum PL available. Furthermore, the PRF reached is so that in combination with the PL, and
taking into account the protocol overhead, maximises the link utilisation. Since the network exhibits
symmetrical conditions, a symmetric PRF allocation has been expected. The fact that this is not
exactly the case is due to the discretisation of the game’s action space, which is unavoidable in a
practical algorithmic implementation.
In addition, it can be appreciated that the pulse rate allocation reached by the LPC algorithm is not
stable, but fast short-term dynamics can be observed. These are due to the fact that the BER at the
receivers, which is used for the calculation of the pricing factor at the transmitters, is a random
variable.
Page 45
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
PL allocation
PRF allocation
BER per link
Figure 20: NE for the 2 links near far network
For a n-links network, results suggest the existence of several equilibria for the game Γ LPC . The LPC
algorithm discover different feasible PRF/PL steady-state allocations depending on the initial
conditions ( prf i max play ,0, 4) .
In particular for the case of 4 links, we present next two steady-state PRF/PL allocations: the first is
reached when all links start with the lowest PL (2-PPM, CR=1), the second when all links start with
the highest PL (2-PPM, CR=1/3). The information data rate per link is 1Mbps, the initial price and
PRF conditions do not influence the convergence.
Page 46
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
PL allocation
PRF allocation
BER per link
Figure 21: NE for 4 links in near far network
It can be observed that in both cases, the PL allocation is the same for all links and equal to the highest
possible PL (2-PPM, CR=1/3). In contrast, the PRF allocation differs. This behaviour confirms the
fact that, as regards to interference management, there is a strong coupling between PRF and PL
adaptation. Further simulations have confirmed that the steady-state cumulative pulse load, and
therefore the cumulative network throughput, attained with the different steady-state allocations (Nash
equilibria) are very similar, which is desirable. However, the highest cumulative network throughput is
attained if the initial PL profile considers ( prf i max play , 0, 4) .
Page 47
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
PL allocation
PRF allocation
BER per link
Figure 22: NE for 4 links near far network
3.1.4.2
Performance Comparison with Adaptive Channel Coding (ACC)
In this section, the interference compensation functionality of the LPC approach is compared to that of
3 2 1 1
a conventional adaptive channel coding (ACC) strategy with convolutional codes {1, , , , } . For
4 3 2 3
that purpose, we have considered an information data rate per link equal to 1Mbps and we have
increased the number of links from N = 2 − 16 .
The ACC strategy constantly adapts the channel code rate to the level of interference experienced at
the receiver. Initially, the lowest code rate is used, and then ACC always selects the highest code rate
that still ensures the decoding of the data packet. For the ACC strategy the PRF per link is fixed.
Figure 23 show the total logarithmic utility reached with each of the two strategies: LPC and ACC. It
can be observed that the LPC algorithm outperforms the ACC strategy when the number of users in
the network is high, and therefore the cumulative offered pulse load. For low load conditions the LPC
algorithm provides similar performance than the ACC strategy. Note that in the case of N = 2 − 8
Page 48
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
links, the total network utility reached by ACC is higher than the one obtained with the LPC
algorithm, but at the cost of violating the QoS constraint ( BER = 1e −6 ).
Figure 23: Total logarithm utility
∑
Figure 24: Average BER per link
|N|
3.1.4.3
log(ri ( prf ))
i =1
Random Scenario
25 m
The room setup is similar to that described in 4.2.4.1, but this time the room size is 25x25x2 m. Unlike
the near-far scenario, the nodes are randomly distributed, see Figure 25. Again, the nodes are static of
moving slower than the convergence time of the LPC algorithm. This distribution leads to a cluster
formation, where some nodes are in range and others are out of range. The behavior of the MAC
algorithm and the link adaptation within clusters is important and shall be investigated by these
simulations.
25 m
Figure 25: Random Scenario
As in the near far scenario, LPC algorithm always converge to a solution in A , regardless of its
initialization and results suggest that Γ LPC has more than one NE for the game. The LPC algorithm
discover different feasible PRF/PL steady-state allocations depending on the initial conditions
( prf i max play ,0, 4) .
Page 49
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
We present next two steady-state PRF/PL allocations obtained in the example scenario depicted in the
following figures. The first (see Figure 27) is reached when all links start with the lowest PL, the
second (see Figure 28) when all links start with the highest PL (2-PPM, CR=1/3). The information
data rate per link is 1Mbps, the initial price and PRF conditions do not influence the convergence.
9
1
15m
3
5
6
2
8
4
Tx
Rx
15m
Figure 26: Example of a random network realization
Like in the near far scenario,the highest cumulative network throughput is reached when the start PL
profile is .
Page 50
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
PL allocation
PRF allocation
BER per link
Figure 27: NE in 8 links random example network
Page 51
EUWB
Initial development of dissemination methods and evaluation
PRF allocation
D4.2.1
PL allocation
BER per link
Figure 28: NE in 8 links random example network
3.1.4.4
Performance Comparison with Adaptive Channel Coding (ACC)
In this section, the interference compensation functionality of the LPC approach is compared to that of
3 2 1 1
a conventional adaptive channel coding (ACC) strategy with convolutional codes {1, , , , } . For
4 3 2 3
that purpose, we have increased the number of links from N = 2 − 16 . The information data rate per
link and the initial conditions for the LPC algorithm are chosen like in section before.
Figure 29 shows the total logarithmic utility reached with each of the two strategies: LPC and ACC in
a network with random distributed nodes. Similar conclusions can be extracted as in the previous
section. LPC algorithm outperforms the ACC strategy when the number of users in the network, and
therefore the cumulative offered pulse load, is high. For low to moderate offered pulse load the LPC
algorithm provides similar performance than the ACC strategy. Note that in the case of N = 2 − 8
Page 52
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
links, the total network utility reached by ACC is higher than the one obtained with the LPC
−6
algorithm, but at the cost of violating the QoS constraint ( BER = 1e ).
Figure 29: Total logarithm utility
∑
|N|
3.1.4.5
i =1
log(ri ( prf ))
Figure 30:Average BER per link
Conclusion
Our distributed mechanism makes use of the framework of game theory to develop a novel
interference management strategy based on the joint combination of adaptive error coding and pulse
rate control.
A game has been formulated and a distributed, asynchronous algorithm (LPC) has been proposed to
discover the optimal PRF/PL network allocation under several channel and interference conditions.
It has been observed that, due to the strong coupling concerning interference management between PL
and PRF adaptation, the game has several feasible steady-state allocations (Nash equilibria). In
general, the different equilibria provide similar aggregated network throughput. Results suggest that,
in order to maximize the cumulative network throughput, it is better that each link starts with the
lowest PL.
It can be concluded that the joint use of PRC and ACC is an appropriate means to mitigate impulsive
interference in autonomous IR-UWB.
Page 53
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
4 Evaluation of architectures and acquisition & distribution
schemes for tracking
4.1 System model
4.1.1 System environment
Due to its great accuracy and good performance on multipath and NLOS environments, UWB is good
candidate to provide location information to mobile users in indoor environments. A particular
application of LT indoor systems is to provide to wireless/cellular access networks users with location
information. Location information could be used by the wireless/cellular access network to develop
location based services or to improve radio resources management [28].
Some interesting scenarios for this application are relatively wide indoor environments such as
shopping malls, train stations, airports, exhibition centres, sports stadiums, etc. Thanks to the
simultaneous data transmission and location capabilities of UWB, a single UWB network could be
used to interconnect different sensors (e.g. fire detection sensors) and to provide to mobile users with
location information. This would allow the user to position himself on his device in the same way as
car navigation systems and locate the place he wants to go (a shop, check-in desk, his seat...). He could
also get information through the UWB network (special offers, arrival times, match statistics, etc.).
But not only could the user benefit of this localization information. The user position could be sent to
the cellular access network of the device (UMTS, WiMAX, PMR) which would be helpful to improve
Radio Resources Management. And furthermore it could be retrieved by the operator to offer locationbased services. An example of application scenario for tracking mobile devices in a shopping mall is
shown in Figure 31.
Figure 31: UWB tracking of mobile devices in a shopping mall
The application of the UWB technology for the development of combined data transmission and
location networks in relatively wide indoor scenarios has to face important challenges due to the short
range of UWB and the limited data rate of LDR-LT (Low Data Rate with Location and Tracking)
UWB systems.
First of all, the area of interest can be relatively large (length over hundreds of meters) in relation to
UWB range (10-50 meters), which implies that the area of interest will be covered by a high number
of anchor nodes and eventually different UWB piconets. Mobility of users implies that position update
rate must be relatively high in order to keep the error of the estimated position under a reasonable
value, which entails a higher need of resources. Moreover, the duration of the procedure of estimating
Page 54
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
the distances, transmitting the ranging information to the location controller, calculating the position
and transmitting the position to the mobile must be as short as possible in order to minimize the error
due to this delay. Finally, the system should be able to cope with a high number of users, which entails
an increase of the resources needed for location.
Positioning accuracy provided by UWB (in the order of centimeters) meets the requirements of this
kind of application scenario (1-2 meters) as long as the position update rate is enough to accurately
track the mobiles. The main constrain of a UWB system in this application scenario is related with the
availability of resources, especially on simultaneous communication and location systems as an
increase of the resources needed for location entails a reduction of the available data rate for
communication. Therefore, different tracking architectures and strategies for the acquisition and
dissemination of location information must be explored in order to optimize the amount of resources
needed for location. The impact of different system design parameters, such as distance between the
anchors, number of anchors used for location or position update rate, on the positioning error and the
amount of resources needed for location should be also evaluated.
4.1.2 MAC layer description
EUWB MAC layer has been designed for Low Data Rate, Location and Tracking (LDR-LT)
applications. It is based on IEEE 802.15.4 standard [29], with enhancements to meet quality of service
requirements and to provide efficient support for ranging and localization with an ultra-wide-band
(UWB) physical layer. A detailed description of EUWB MAC layer can be found in [30].
A MAC superframe structure is defined, which is divided into timeslots where the different frames
(beacon frames, hello frames, data frames, ranging frames, timeslot requests…) are sent. Superframe
structure is shown in Figure 32. Two main periods are defined in the superframe, namely Control
period and Data period. In the “Control period”, two parts are identified:
•
The Beacon period used for the beacon alignment
•
The Topology Management period used to send the “Hello” frames
The other portion called “Data Period” is used for data frames, ranging frames, GTS request frames
and other command frames. Data Frames, ranging frames and GTS request frames are sent in a
Contention Free Period whereas other command frames are sent in the Contention Access Period
(CAP). Ranging slots are considered as data slots and ranging frames are sent in the CFP period as
other data frames.
The Contention Free Period in the “Data Period” is so composed of:
• Guaranteed time slots (GTS) for data frame transfer, acknowledgments and ranging. The
number of GTS used for ranging can be freely defined. It depends of the number of devices which
are requested to perform ranging as well as the number of ranging measurements needed for a
localisation update.
•
A GTS request period used to send the GTS request frames.
The maximum number of slots (and so sub-slots) for each part of the frame are then the following
ones:
•
12 Timeslots in the Beacon period
•
3 Timeslots in the Topology Management Period (12 sub-slots as each slot is divided in 4
sub-slots)
Page 55
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
•
20 Timeslots in the GTS Period that are used for Data communications and for ranging as in
Phase 2 the ranging slots are considered as data slots. A classical case will be to have 8
Timeslots for data communication and 12 Timeslots for ranging enabling 4 devices using 3
way ranging to perform ranging.
•
6 Timeslots in the GTS request Period
•
12 Timeslots in the CAP period
So, the maximum number of slots in the super-frame is 53. If the maximum number of slots is not
achieved in the super frame, an inactive part is defined before the beginning of the next super-frame.
BP: Beacon Period
TP: Topology Management Period
IP: Inactive Period
Beacon slot
BP
Topology
management
slot (divided
into 4 sub
slots)
TP
Control Period
CFP: Contention Free Period
CAP: Contention Access Perriod
GTS
Request
slot
(divided
into 2 sub
slots)
CFP slot
CFP
CAP slot
(divided
into 4 sub
slots)
CAP
IP
Data Period
Super Frame Duration (SD)
Beacon Interval (BI)
Figure 32: EUWB MAC superframe structure [30]
The piconets are based in a mesh centralized topology, which is defined by a centralised allocation
management and a centralised common timing reference synchronisation. Indeed a coordinator is
elected in the network so as to transmit beacon frames and to handle the scheduling procedures.
Figure 33 describes the construction of a Mesh centralised network. First a scheduling tree is built, it is
used to transport beacon, association request and GTS request command frames. Then it is extended to
a meshed scheduling tree by enabling the transmission out of the tree for the data. Data frames,
ranging frames can so be sent out of the tree as well as hello frames that are broadcasted to the
neighbours. The CFP slots for each link are allocated by the piconet coordinator for the whole route
between the source and the destination.
Page 56
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Figure 33: Mesh centralized topology [30]
In the scheduling tree, the Piconet Coordinator is on top of the tree. Then the different devices are
defined by their scheduling tree level. The typical maximum tree level that must be taken into account
for the application scenario is 4. Then a possible scheduling tree can be represented as follows.
Figure 34: Example of meshed scheduling tree [30]
The scheduling tree is used to transport the beacon frames to each device of the network. It allows also
the transmission of the GTS request frames from the devices to the coordinator, the association request
frames and the disassociation request frames.
Data frames, hello frames, ack frames and ranging frames can be sent on non tree link. Hello frames
are broadcasted and so no destination address is needed. Ranging frames are not relayed and can be
sent only between neighbours. The neighbourhood is known locally for each node of the network
thanks to the hello frames that are periodically sent.
It is to be noticed also that the relaying procedure is performed at the MAC layer level. The source
node requests an allocation to the coordinator for the entire route. Indeed, when a node has data to
transmit, it will send a GTS request on the tree to the coordinator with its address as the source address
and the destination address of the transmission. The request will correspond to the number of GTS
slots needed for the data transmission on a link. The coordinator which has the knowledge of the
whole network will look on its routing table if there are relays between the source and the destination.
If there are relays, the coordinator will allocate the GTS for each link based on the request made by the
source node. The GTS allocation will be available to the source node and relay nodes in the Beacon
frame (Data descriptor in the GTS fields).
4.1.3 LT application
The objective of the LT system is to track the position of mobile users in relatively wide indoor
environments. With that purpose a UWB simultaneous data transmission and location network will be
Page 57
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
deployed on the scenario. The network is composed of Na anchors, fixed and regularly distributed at
known positions, and Nm targets, mobile nodes to be located.
In order to track the position of the target nodes, location information, basically the distances
estimated between the target and anchor nodes, must be acquired and transmitted to the location
controllers, which are the functional units that execute the tracking algorithm to obtain the estimated
position of the targets. Location controllers can be physically implemented in one or more anchor
nodes or in the mobile nodes, depending on the architecture which is defined. In the architecture
centralized in the network, the tracking functionality is implemented in one or more previously
defined anchor nodes. In the distributed architecture, the mobile node picks dynamically which anchor
executes the location functionality. Finally, in the architecture centralized in the mobile nodes the
location controller function is implemented in the mobile nodes.
The acquisition of the location information is done through the ranging procedure. Three procedures
are defined [31]
- One way ranging: The target node transmits a ranging packet and every anchor measures the time of
arrival of the ranging packet. The position of the target can be obtained from the differences between
the times of arrival on each anchor. Only one time slot is needed but it requires synchronization
between all the anchors.
- Two way ranging: The procedure initiator (target or anchor) transmits a ranging request to another
node, which estimates the time of arrival and sends a ranging response after a predefined time. The
initiator measures the time of arrival of the response and can estimate the transmission delay and the
distance between the nodes. Two time slots are needed, but there is no need of synchronization
between the nodes.
-Three way ranging: Similar to the two way ranging scheme, but in this case two ranging responses are
sent in order to compensate the clock drift. The ranging procedure can be initiated by the target nodes
or by the anchor nodes and the distance is estimated by the initiator. It requires three time slots, but
estimated distances are much more accurate.
Once the distances between the target and the anchor nodes are estimated, they must be transmitted to
the location controller in a data packet (measurement report data packet). The location controller
executes the tracking algorithm to calculate the position and transmits the updated position to the
target node (position update data packet).
With respect to the tracking technique itself, parametric and non-parametric approaches can be
distinguished. Parametric approaches compute the location based on the a priori knowledge of a
model, while non-parametric approaches process straightforward the data with the usage, in some
cases, only of some statistic parameters (mean, variance). An example of parametric approach is based
on the Extended Kalman Filter [32], while an example of non-parametric approach is based on
Multidimensional Scaling (MDS) [33]Further information on tracking algorithms for UWB and
proposal of parametric and non-parametric techniques can be found in [34].
In order to reduce the amount of resources needed to acquire and distribute the location information,
different enhanced modes are proposed and explained in detail in [34]. These enhancements include:
- Data aggregation: All the distances estimated can be aggregated and sent in a single measurement
report data packet to the location controller.
- Broadcast/multicast request: In order to reduce the number of slots, a single ranging request can be
broadcasted to all the nodes (broadcast) or to a set of nodes (multicast)
Page 58
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
- Broadcast/multicast response: After receiving the ranging requests of different initiators, a single
ranging response is broadcasted to all the initiators (broadcast) or a set of them (multicast).
4.2 Analysis and evaluation
In order to evaluate different system architectures and strategies for acquisition and dissemination of
location information, and the impact of the different system design parameters, a simulation tool has
been developed, using C++ as programming language and Visual Studio .NET as development
platform. The system performance is evaluated in terms of positioning error (average and variance)
and the amount of resources used for location (number of time slots needed for acquisition and
distribution of location information).
With this purpose, a set of simulations has been performed for the different proposed architectures and
acquisition and dissemination of location information strategies. The influence of parameters such as
the number of anchors used for location, the distance between anchors, the position update rate or
target mobility has also been analyzed.
For these simulations, a common set of parameters has been defined. Simulation duration has been set
to 10000 seconds and the number of targets to be tracked to 10. The area size has been set to 50 m. x
50 m. and the distance between anchors to 10 m., resulting in 25 anchors. Concerning the dynamics of
the mobile nodes, minimum and maximum speeds have been set to 0.1 and 3 m/s respectively, and
direction changes every 20 seconds. Position update rate has been set to 1 update per second and UWB
nodes range to 15 m.
4.2.1 Simulator description
The simulator scenario is the representation of a relatively wide indoors area of dimensions length x
width. On this scenario a UWB network is deployed, as it is shown in Figure 35. The network is
composed of Na anchors, fixed and regularly distributed at known positions, and Nm targets, mobile
nodes to be located. For simplicity, the scenario is covered by a single UWB piconet, so data
transmission between different piconets and handover are not further considered.
In order to track the position of the target nodes, the distances between the target and anchor nodes are
estimated. Estimated distances are sent to location controllers, which are the functional units that
execute the tracking algorithm to obtain the estimated position of the targets. Location controllers can
be physically implemented in one or more anchor nodes or in the mobile nodes, depending on the
architecture which is defined. Finally, the estimated positions are sent to each target.
Page 59
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Y
LC
X
Figure 35: UWB location simulator scenario
A ranging model is used to characterize the ranging error distribution. Range measurements based on
round-trip Time of Flight (ToF) estimation through n-Way Ranging transactions can be modeled as:
d ij = d ij + ε ij + nij = d ij' + nij
(4.1)
where dij is the actual distance between nodes i and j, d′ij is the biased distance (with bias εij) and nij is
a residual noise term. The biased distance is modeled as a weighted sum of Gaussian and Exponential
components conditioned upon the actual distance and the channel configuration (LOS/NLOS/severe
NLOS). The residual noise is modeled as additive and centered, with a variance σn2 that depends on
detection noise terms affecting unitary ToA estimates and involved protocol durations. Further
information about the ranging models can be found in [34] and the specific model and parameters used
can be found in [35][36].
Concerning the tracking technique, a simple geometric trilateration algorithm has been implemented,
which calculates the position according to the distances estimated to the three closest anchors. Also, a
parametric algorithm based on the Extended Kalman filter is implemented, and a non-parametric
algorithm based in Multidimensional Scaling (MDS) is under development.
The dynamics of the mobile nodes are modeled by random directions and speeds that are constant
during a certain period of time, after which new random directions and speeds are set. The dynamic
model is characterized by maximum and minimum speeds and the direction change rate.
Finally, the UWB physical layer has been characterized through different models. A coverage model is
used to identify the vicinity relationships between the nodes depending on the distance between them.
A transmission model is used to characterize the packet error rate and the delay depending on the
distance and channel model although, in a first step, an ideal approach has been taken, considering
lossless and instantaneous acquisition and distribution of location data.
4.2.2 Impact of the tracking architecture
As it was previously explained, the tracking functionality is executed by the location controllers,
which can be physically implemented in one or more anchor nodes or in the mobile nodes, depending
on the architecture which is defined.
Page 60
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
In the architecture centralized in the network, the tracking functionality is implemented in one or more
previously defined anchor nodes. To update the position of a target node, the estimated distances
between the target and anchor nodes must be sent to the location controller, which applies the tracking
algorithm to obtain the position. Then, the location controller transmits the updated position to the
target node.
In the distributed architecture, the mobile node picks dynamically which anchor executes the location
functionality. With an appropriate strategy, it is possible to keep the location controller in coverage of
the mobile node. Therefore, the data packets with location information (estimated distances or updated
position) can be transmitted using only one slot as there will be no need of relaying. As a drawback,
the tracking functionality must be implemented in every anchor node.
Finally, in the architecture centralized in the mobile nodes the location controller function is
implemented in the mobile nodes. The mobile nodes perform ranging with their neighbor anchors and
obtain their own position applying the tracking algorithm. Therefore, there is no need of transmitting
the estimated distances and the updated position. Nevertheless, the implementation of the tracking
functionality in the mobile nodes would increase their complexity, and this may not be desirable. The
feasibility of this architecture will depend on the specific application and devices to be tracked
In this section, the impact of each one of the proposed architectures (centralized in the network,
distributed and centralized in the mobile nodes) on the amount of resources needed for location is
evaluated. The resources are measured in terms of time slots used per second. The slots are classified
according to their use in Ranging Requests (RRq), Ranging Responses (RRp), data packets with
measurements reports (DP-MR) and data packets with position update (DP-PU). Also the total number
of slots needed for location is shown.
Figure 36 shows the results of the simulations obtained for an architecture centralized in the network
with one location controller, which will be implemented in the anchor with coordinates (25, 25),
depending on the number of target nodes. As it can be observed, for the ranging request,
approximately 4 slots per second and per target are needed, as we have set to 4 the number of anchors
to be used for location. The number of slots needed for ranging responses is twice the number of
requests, as we are considering three-way ranging and two responses are sent for each request.
Concerning the amount of measurement reports, there will be one per ranging procedure, but it may
need more than one slot, as it has to be relayed to the location controller, which may not be in
coverage of the target node. Finally, only one position update data packet is needed per update
process, but due to relays almost 2 slots per second and per target are needed. This makes a total of
approximately 20 slots per second and per target.
Page 61
Initial development of dissemination methods and evaluation
Number of slots per second
EUWB
220
200
180
160
140
120
100
80
60
40
20
0
D4.2.1
RRq
RRp
DP-MR
DP-PU
Total
1
2
3
4 5 6 7 8
Number of targets
9 10
Figure 36: Resources needed for location for a centralized in the network architecture with 1 location controller
Number of slots per second
In order to reduce the number of times that data packets have to be relayed, the number of location
controllers in the network can be increased. Figure 37 shows the amount of timeslots needed for a
centralized network architecture with 4 location controllers, implemented in the anchors with
coordinates (15,15), (15,35), (35,15) and (35,35). As it can be observed, the number of measurement
reports is almost similar to the number of ranging requests, as now the target will be most of the time
in coverage of one of the location controllers and only few times the packets will have to be relayed.
Due to the same reason, the number of slots needed for position update packets also decrease to
approximately one slot per second per target. Thus, the total number of slots needed is approximately
17 slots per second and per target.
220
200
180
160
140
120
100
80
60
40
20
0
RRq
RRp
DP-MR
DP-PU
Total
1
2
3
4 5 6 7 8
Number of targets
9 10
Figure 37: Resources needed for location for a centralized in the network architecture with 4 location controllers
Another possibility to decrease the amount of time slots needed for location is to use a distributed
architecture. As it can be observed in Figure 38, the number of slots used for measurement reports is
exactly the same as the number of ranging requests, which is expected as with the distributed
Page 62
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Number of slots per second
architecture the target is always in coverage of the location controller. Number of position update
packets is exactly one slot per update.
220
200
180
160
140
120
100
80
60
40
20
0
RRq
RRp
DP-MR
DP-PU
Total
1
2
3
4 5 6 7 8
Number of targets
9 10
Figure 38: Resources needed for location for a distributed architecture
Number of slots per second
Finally, another possibility is an architecture centralized in the mobile. As it can be seen in Figure 39,
the amount of time slots for ranging requests and responses are similar to the architecture centralized
in the network, but measurement report and position update packets are not needed as the tracking
algorithms are executed by the target. The total number of slots needed for location is approximately
12 slots per second per target.
220
200
180
160
140
120
100
80
60
40
20
0
RRq
RRp
DP-MR
DP-PU
Total
1
2
3
4 5 6 7 8
Number of targets
9 10
Figure 39: Resources needed for location for a centralized in the mobile architecture
Figure 40 summarizes the results obtained for the different architectures in terms of total number of
time slots needed for location. Compared to the architecture centralized in the network with one
location manager, the architecture centralized in the mobile nodes presents a reduction of 40% in the
amount of time slots needed. This architecture requires that the location function is implemented in the
mobile nodes, so the feasibility of this architecture depends on the specific devices to be tracked. The
distributed architecture presents a reduction of almost 20% in the amount of time slots needed
Page 63
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Number of slots per second
compared to the architecture centralized in the network with one location controller. Very similar
results are obtained for the architecture centralized in the network with 4 location controllers, which
only requires implementing the location functionality in 4 of the 25 anchor nodes, so this architecture
will be preferred over the distributed one.
220
200
180
160
140
120
100
80
60
40
20
0
Network-1LM
Network-4LM
Distributed
Mobile
1
2
3
4 5 6 7 8
Number of targets
9 10
Figure 40: Resources needed for location for the different system architectures
4.2.3 Impact of acquisition & distribution strategies
In order to track the position of the target nodes, location information, basically the distances
estimated between the target and anchor nodes, must be acquired and transmitted to the location
controller.
The acquisition of the location information is done through the ranging procedure. Ranging procedure
can be initiated either by the target nodes or by the anchor nodes. The procedure initiator transmits a
ranging request to another node, which estimates the time of arrival and sends a ranging response after
a predefined time. The initiator measures the time of arrival of the response and can estimate the
transmission delay and the distance between the nodes (Two Way Ranging). In order to improve the
accuracy of distance estimation, two ranging responses can be sent in order to compensate the clock
drift (Three Way Ranging). One way ranging will not be further considered on this study as it requires
accurate synchronization between all the anchors.
The distribution includes the transmission of the estimated distances to the location controller in a data
packet (measurement report data packet) and the transmission of the updated position the target node
(position update data packet).
In order to reduce the amount of resources needed to acquire and distribute the location information,
the simulator implements different enhancements [34].
- Data aggregation: All the distances estimated can be aggregated and sent in a single measurement
report data packet to the location controller.
- Broadcast/multicast request: In order to reduce the number of slots, a single ranging request can be
broadcasted to all the nodes (broadcast) or to a set of nodes (multicast)
Page 64
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
- Broadcast/multicast response: After receiving the ranging requests of different initiators, a single
ranging response is broadcasted to all the initiators (broadcast) or a set of them (multicast).
If the initiator is the target node, broadcast/multicast response would require a simultaneous update of
the position of all the targets in order to be able to aggregate the ranging responses in a single packet.
This is not very suitable for tracking applications, as the simultaneous update of all the targets would
increase the duration of the position update process, which is critical due to target mobility.
Furthermore, in general each target uses different anchors, so this strategy would not be very effective.
The same applies for data aggregation and broadcast/multicast request when the initiator is the anchor
node.
The amount of resources needed for location have been simulated for each enhancement based on an
architecture centralized in the network with one location controller, so results obtained must be
compared to those in Figure 36.
Number of slots per second
Figure 41 shows the results obtained for the data aggregation enhancement. As all the measurement
reports are grouped into a single data packet, the number of slots needed for the measurement report
packets is reduced. Only one packet will be sent per update, so results are equal to the amount of slots
needed for position update, almost 2 slots per second per target due to relaying.
220
200
180
160
140
120
100
80
60
40
20
0
RRq
RRp
DP-MR
DP-PU
Total
1
2
3
4 5 6 7 8
Number of targets
9 10
Figure 41: Resources needed for location when data aggregation is applied
Concerning the possibility of sending multicast ranging requests, this implies that only one ranging
request is sent in each update, thus minimizing the amount of slots needed for ranging requests to 1
slot per second per target, as it is shown in Figure 42.
Page 65
Initial development of dissemination methods and evaluation
Number of slots per second
EUWB
220
200
180
160
140
120
100
80
60
40
20
0
D4.2.1
RRq
RRp
DP-MR
DP-PU
Total
1
2
3
4 5 6 7 8
Number of targets
9 10
Figure 42: Resources needed for location when multicast ranging request is applied
Number of slots per second
The use of broadcast ranging requests has the same impact on the number of slots needed for ranging
requests, which is reduced to a minimum of 1 slot per second per target, as it is shown in Figure 43.
Furthermore, it is simpler than the multicast option, as no knowledge of the anchors to use is required.
Nevertheless, it also has a drawback, as all the anchors in coverage of the target node will answer, thus
increasing the number of ranging responses.
220
200
180
160
140
120
100
80
60
40
20
0
RRq
RRp
DP-MR
DP-PU
Total
1
2
3
4 5 6 7 8
Number of targets
9 10
Figure 43: Resources needed for location when broadcast ranging request is applied
Figure 44 summarizes the results obtained for each one of the acquisition and distribution
enhancements in terms of total number of time slots needed for location. Data aggregation (DA)
reduces in approximately 25% the amount of time slots needed for location. Concerning multicast
(MRq) and broadcast (BRq) ranging requests, the first one is preferred as it reduces the total amount of
slots needed in a 15%, while the broadcast does not reduce the total amount of slots needed, as the
Page 66
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Number of slots per second
number of ranging responses increases. Finally, combining data aggregation and multicast ranging
request (DA&MRq), a reduction of 40% is reached.
220
200
180
160
140
120
100
80
60
40
20
0
Basic
DA
MRq
BRq
DA&MRq
1
2
3
4 5 6 7 8
Number of targets
9 10
Figure 44: Resources needed for location for different acquisition and distribution enhancements
Number of slots per second
Concerning the initiator of the ranging process, Figure 45 shows the results obtained when the ranging
requests are sent by the anchor nodes instead of the target nodes. Compared to the case when the
ranging requests are sent by the target nodes (Figure 36), there is a small reduction of the number of
slots needed for sending the measurement reports to the location controller. This is due to the
reduction in the number of relays that are needed to send these packets to the location controller.
220
200
180
160
140
120
100
80
60
40
20
0
RRq
RRp
DP-MR
DP-PU
Total
1
2
3
4 5 6 7 8
Number of targets
9 10
Figure 45: Resources needed for location when the ranging procedure is initiated by the anchors
In case the ranging process is initiated by the anchors, data aggregation and multicast ranging requests
are not suitable for tracking applications, as it would mean that all the position update procedures
should be performed at the same time for all the target nodes. On the other hand, the multicast ranging
response enhancement can be applied. As it can be observed in Figure 46, the number of slots needed
Page 67
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Number of slots per second
for ranging responses is reduced to 2 slots per second per target, as only 2 response packets are sent
per update.
220
200
180
160
140
120
100
80
60
40
20
0
RRq
RRp
DP-MR
DP-PU
Total
1
2
3
4 5 6 7 8
Number of targets
9 10
Figure 46: Resources needed for location when the ranging procedure is initiated by the anchors and broadcast
ranging response is applied
Number of slots per second
Figure 47 shows the total number of time slots needed for location depending on the initiator of the
ranging procedure. When no enhancements are applied, the number of slots needed when the anchors
start the ranging procedure is slightly lower than the number of slots needed when the target starts the
ranging procedure. When the target is the initiator and data aggregation and multicast ranging request
(DA&MRq) are applied, a reduction of 40% in the number of slots needed is reached, getting even
better results than when the anchor is the initiator and multicast ranging response (MRp) is applied.
220
200
180
160
140
120
100
80
60
40
20
0
Target
Anchor
Target DA&MRq
Anchor MRp
1
2
3
4 5 6 7 8
Number of targets
9 10
Figure 47: Resources needed for location depending on the initiator of the ranging procedure and the
enhancements applied
Finally, the performance of the different architectures has been evaluated when data aggregation and
multicast ranging response are applied, and results are shown in Figure 48. The architecture
Page 68
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Number of slots per second
centralized in the mobile nodes shows the best results, with a reduction of 30% in the number of slots
needed for location compared to the architecture centralized in the network with one location
controller. The distributed architecture shows similar results to the architecture centralized in the
network with 4 location controllers, with a reduction of 12% compared to the architecture centralized
in the network with one location controller.
220
200
180
160
140
120
100
80
60
40
20
0
Network-1LM
Network-4LM
Distributed
Mobile
1
2
3
4 5 6 7 8
Number of targets
9 10
Figure 48: Resources needed for location for different system architectures when data aggregation and multicast
ranging request are applied
As a reference, the amount of resources available for location with the MAC parameters proposed for
the LDR-LT UWB platforms to be developed within EUWB project is calculated [30]. Slot size is set
to 160 bytes (1280 symbols), which at a physical data rate of 347 kbps entails a slot duration of 3.6864
ms. Number of slots in a superframe is set to 53, which entails a superframe duration of 195.4 ms. As
in each superframe 20 slots are assigned to ranging and data communication, this means that
approximately 100 slots per second are available. Considering the architecture centralized in the
network with one location controller, data aggregation and multicast ranging request, that requires 12
slots per second and per target, approximately 8 mobile nodes could be tracked at the same time in a
picocell.
4.2.4 Impact of number of anchors used for location
In order to reduce the amount of resources needed, a possible strategy is limiting the number of
anchors to be used for location. Therefore, the ranging procedure would not be performed with all the
neighbour anchors, but only with a certain number of them. This would reduce the amount of slots
needed for ranging requests, ranging responses and measurement reports. Of course, this will also
have an impact on the positioning error, as a higher number of anchors allows higher accuracy and
reliability.
Nevertheless, the ranging error depends on the distance, so the ranging error for the closest anchors is
expected to be much lower than for the more distant anchors. Therefore, if only the closest anchors are
used for position calculation, both positioning error and need of resources can be reduced. In order to
select the closest anchors, a procedure to pick the closest anchors must be defined either in the mobile
nodes or in the location controllers, taking advance of the available location information (estimated
position of the mobile and known positions of the anchor nodes).
Page 69
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
In the first version of the simulator, an ideal approach has been taken and the closest anchors are
always picked. But in further versions more realistic approaches, for example using the estimated
position of the mobile and the known positions of the anchors to estimate the distances to the anchors
and select the closest ones, will be implemented and the possible degradation will be assessed.
In this section, the error in the position estimation and the amount of resources needed for location are
evaluated for different number of anchors used for location. Concerning the positioning error, different
tracking algorithms (trilateration and Kalman filter) and levels of ranging residual noise (σn=0.7 m and
0.3 m) are considered. A distance between anchors of 10 meters has been set, which results in 25
anchors.
Number of slots per second
and per target
Figure 49 shows the amount of resources needed depending on the number of anchors used for
location. Results are independent of the tracking algorithm and the ranging noise. As it can be
observed, the amount of timeslots needed increase as a higher number of anchors are used and remains
constant for more than 7 anchors, as for a distance between anchors of 10 meters it is not very likely
that the target is in coverage of more than 7 anchors. Therefore, if the number of anchors used for
location is limited to 4, the amount of resources needed for location would be reduced from 27 to 20
slots per second and per target for this configuration (10 meters between anchors).
50
45
40
35
30
25
20
15
10
5
0
RRq
RRp
DP-MR
DP-PU
Total
3
4
5
6
7
8
9 10
Number of anchors for location
Figure 49: Resources needed for location depending on the number of anchors used for location when Na=25
anchors
Concerning the positioning error, Figure 50 shows the average error depending on the number of
anchors for different tracking algorithms (trilateration and Kalman filter) and residual ranging noise
levels (0.7 m. and 0.3 m.). The trilateration algorithm shows an important increase of the error when
three anchors are used. This is due to the fact that, with the trilateration method, position cannot be
computed when the three reference anchors are aligned, which is likely to happen when the target is
near the area limits. When more than three anchors are used, the performance is independent of the
number of anchors, as only the three closest anchors are used for the trilateration, unless they are
aligned, then the fourth closest anchor is picked. Concerning the Kalman filter, the more anchors are
used, the more information is available, which should improve the accuracy of the filter. But, as the
anchors are picked according to their distance to the target, a higher number of anchors implies that
more distant anchors are picked, which have a higher ranging error due to ranging bias. For high
ranging residual noise, the optimal number of anchors which minimizes the error is 4, and the
Page 70
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Average error (m)
performance is slightly degraded as the number of anchors increase. This degradation is more
noticeable when the ranging residual noise is low, as the importance of the ranging bias, which
depends on the distance, will be higher.
2
1,8
1,6
1,4
1,2
1
0,8
0,6
0,4
0,2
0
σ=0.7 Trilateration
σ=0.7 Kalman filter
σ=0.3 Trilateration
σ=0.3 Kalman filter
3
4
5
6
7
8
9
10
Number of anchors for location
Figure 50: Position estimation error depending on the number of anchors used for location when Na=25 anchors
Error variance (m2)
Figure 51 shows the variance of the positioning error depending on the number of anchors used for
location. As it can be observed, the variance is almost independent of the number of anchor used for
location as long as 4 or more anchors are used. Furthermore, the variance is almost independent of the
ranging residual noise as, for a distance between anchors of 10 m., the variance is mostly due to the
ranging bias. Finally, the variance is much higher for the trilateration algorithm compared to the
Kalman filter, as the Kalman filter uses the previous position to calculate the new one, thus reducing
the variability.
3
2,7
2,4
2,1
1,8
1,5
1,2
0,9
0,6
0,3
0
σ=0.7 Trilateration
σ=0.7 Kalman filter
σ=0.3 Trilateration
σ=0.3 Kalman filter
3
4
5
6
7
8
9
10
Number of anchors for location
Figure 51: Position estimation error variance depending on the number of anchors used for location when Na=25
anchors
Page 71
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
4.2.5 Impact of the distance between anchors
Another important parameter in the system design which has a big impact on the positioning accuracy
and in the amount of resources needed for location is the distance between anchors. In order to assess
this impact, three different configurations, 25 anchors (10 m between anchors), 49 anchors (7.15 m)
and 100 (5 m), have been evaluated depending on the number of anchors used for location, the
tracking algorithm and the ranging residual noise.
Average error (m)
Figure 52 and Figure 53 show the position estimation error for the trilateration algorithm depending on
the number of anchors, with high and low ranging residual noise respectively. For high ranging
residual noise, average error is reduced from 0.8 m for 25 anchors to 0.6 m with 49 anchors. With this
configuration, the error is mostly due to ranging residual noise, as the three anchors used for
trilateration will be relatively close to the target. Therefore, there is only a small reduction when the
number of anchors is increased to 100. For low ranging residual noise, average error is reduced from
0.4 m for 25 anchors to 0.2 m with 49 anchors and to 0.15 m with 100 anchors.
2
1,8
1,6
1,4
1,2
1
0,8
0,6
0,4
0,2
0
σ=0.7 Na=25
σ=0.7 Na=49
σ=0.7 Na=100
3
4
5
6
7
8
9
10
Number of anchors for location
Average error (m)
Figure 52: Position estimation error for different total number of anchors with trilateration and σ n=0.7m
2
1,8
1,6
1,4
1,2
1
0,8
0,6
0,4
0,2
0
σ=0.3 Na=25
σ=0.3 Na=49
σ=0.3 Na=100
1
2
3
4
5
6
7
8
Number of anchors for location
Figure 53: Position estimation error for different total number of anchors with trilateration and σ n=0.3m
Page 72
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Average error (m)
Figure 54 and Figure 55 show the position estimation error for the Kalman filter with high and low
ranging residual noise respectively. As it was stated in the previous subsection, the more anchors are
used, the more information is available, but more distant anchors are picked, which have a higher
ranging error due to ranging bias. This results in an optimal number of anchors which minimizes the
position estimation error. For high residual noise, minimum average error is reduced from 0.64 m for
25 anchors to 0.48 m for 49 anchors and to 0.35 m for 100 anchors. As the distance between the
anchors is reduced, the optimal number of anchors used for position calculation increases, from 4 for
25 anchors to 6 for 49 anchors and 10 for 100 anchors. When the ranging residual noise is low, the
estimated distances are much more accurate and the Kalman filter performance is similar to
trilateration, with an average error of 0.36 m for 25 anchors, 0.19 m for 49 anchors and 0.15 m for 100
anchors. Also, as the residual noise is reduced, the ranging bias becomes more important and the
optimal number of anchors is reduced to 3 for 25 anchors, 4 for 49 anchors and 6 for 100 anchors.
2
1,8
1,6
1,4
1,2
1
0,8
0,6
0,4
0,2
0
σ=0.7 Na=25
σ=0.7 Na=49
σ=0.7 Na=100
3
4
5
6
7
8
9
10
Number of anchors for location
Average error (m)
Figure 54: Position estimation error for different total number of anchors with Kalman filter and σ n=0.7m
2
1,8
1,6
1,4
1,2
1
0,8
0,6
0,4
0,2
0
σ=0.3 Na=25
σ=0.3 Na=49
σ=0.3 Na=100
1
2
3
4
5
6
7
8
Number of anchors for location
Figure 55: Position estimation error for different total number of anchors with Kalman filter and σ n=0.3m
Page 73
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Error variance (m2)
Figure 56 and Figure 57 show the variance of the position estimation error depending on the number
of anchors for the trilateration algorithm with high and low ranging residual noise respectively. In both
cases, increasing the number of anchors from 25 to 49 entails an important reduction in variance, as all
the anchors involved in the position calculation will be relatively close to the target, minimizing the
effect of the ranging bias. Nevertheless, no further reduction is reached when the number of anchors is
increased to 100, as the ranging bias for more than 49 anchors is negligible and the variance is mostly
due to the ranging residual error. Concerning the ranging residual noise, results for 25 anchors are
similar as the error is mostly due to ranging bias, but for 49 and 100 anchors the errors is mostly due to
the ranging residual noise, so a further reduction is reached when the ranging residual noise is low.
Similar results are obtained for the Kalman filter, with the difference that for 25 anchors the error
variance is lower for the Kalman filter as it uses the previous position information, thus reducing the
variability.
3
2,7
2,4
2,1
1,8
1,5
1,2
0,9
0,6
0,3
0
σ=0.7 Na=25
σ=0.7 Na=49
σ=0.7 Na=100
3
4
5
6
7
8
9
10
Number of anchors for location
Figure 56: Position estimation error variance for different total number of anchors with trilateration and σ
n=0.7m
Page 74
Initial development of dissemination methods and evaluation
Error variance (m2)
EUWB
3
2,7
2,4
2,1
1,8
1,5
1,2
0,9
0,6
0,3
0
D4.2.1
σ=0.3 Na=25
σ=0.3 Na=49
σ=0.3 Na=100
1
2
3
4
5
6
7
8
Number of anchors for location
Figure 57: Position estimation error variance for different total number of anchors with trilateration and σ
n=0.3m
Number of slots per second
and per target
Finally, Figure 58 shows the amount of resources needed depending on the number of anchors used
for location for each configuration. As it can be observed, for 100 anchors the number of timeslots
needed increases linearly according to the number of anchors used for location, as the target is always
in coverage of more than 10 anchors. For 49 anchors, the number of timeslots increases linearly for
less than 8 anchors, with a smaller increase for more than 8 anchors. For 25 anchors, the increase is
not linear as the target may not be in coverage of the selected number of anchors, eventually remaining
constant for more than 7 anchors.
50
45
40
35
30
25
20
15
10
5
0
25 Anchors
49 Anchors
100 Anchors
3
4
5
6
7
8
9 10
Number of anchors for location
Figure 58: Resources needed for location for different total number of anchors
Reducing the distance between the anchors leads to an improvement in positioning accuracy, both in
terms of average error and variance, although the number of anchors needed to cover the area and
therefore the cost and complexity of the network increases. As the distance between the anchors is
reduced, the target is in coverage of more anchors and the need of limiting the number of anchors to be
used for location becomes more important in order to avoid an excessive use of resources.
Page 75
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
4.2.6 Effect of target mobility and position update rate
Finally, another important parameter related with both the positioning error and the amount of
resources needed for location is the position update rate, which defines how often the positions of the
mobile nodes are updated. Position update rate is closely related to target mobility, as a higher target
speed requires more frequent position updates in order to accurately track the target.
In order to assess the effect of target mobility and position update rate, a set of simulations has been
carried out. Architecture centralized in the network with 1 location controller, no acquisition and
distribution enhancements, 10 meters between anchors and 4 anchors used for location have been
considered.
First, the effect of target mobility on the positioning error is analyzed. Target mobility is characterized
by target speed and direction change rate. Two parameters are going to be used to measure the error.
Position Estimation Error (PEE) measures the error in the position estimated by the algorithm on each
update. Tracking Error (TE) measures the error in the position available at the mobile at every
simulation step, not only when it is updated. This error is not only due to the error in the position
estimation, but also to the movement of the mobile since the last position update.
Average error (m)
Figure 59 shows the position estimation error and the tracking error depending on the target speed.
Time between updates, which is the inverse of the position update rate, has been set to 1 second.
Concerning position estimation error, it is almost independent of the target speed for the trilateration
algorithm, as the position calculation is independent on each update. On the other hand, the Kalman
filter takes account of the previous position and can get a better accuracy as far as the target speed is
low enough. The performance of the filter degrades as the target speed increases, but even for 3 m/s
the position estimation error is lower than with the trilateration algorithm. Concerning the tracking
error, it increases as the target speed increase, as the distance travelled by the target between each
position update will be greater.
2,4
2,2
2
1,8
1,6
1,4
1,2
1
0,8
0,6
0,4
0,2
0
PEE Trilateration
PEE Kalman
TE Trilateration
TE Kalman
0,3 0,6 0,9 1,2 1,5 1,8 2,1 2,4 2,7 3
Target speed (m/s)
Figure 59: Position estimation error and tracking error depending on the target speed
Concerning the direction change rate, as it can be observed in Figure 60 the error is independent on the
time between direction changes and only for 2 seconds there is a slight increase of the error for the
Kalman filter.
Page 76
Initial development of dissemination methods and evaluation
Average error (m)
EUWB
2,4
2,2
2
1,8
1,6
1,4
1,2
1
0,8
0,6
0,4
0,2
0
D4.2.1
PEE Trilateration
PEE Kalman
TE Trilateration
TE Kalman
2
4
6
8 10 12 14 16 18 20
Time between direction changes(s)
Figure 60: Position estimation error and tracking error depending on the time between direction changes
Average error (m)
Figure 61 shows the position estimation error and the tracking error depending on the time between
updates, considering a random target speed uniformly distributed between 0.1 m/s and 3 m/s, and 20
seconds between direction changes. Concerning the positioning error, the trilateration algorithm
accuracy is independent of the position update rate, as the position calculation is independent on each
update. On the other hand, the position estimation with the Kalman filter is based on the previous
position, resulting in a higher accuracy when time between updates is less than 1.6 seconds, but it
degrades as the time between updates increases. The tracking accuracy degrades as the time between
updates increase in any case, as the distance travelled by the target between each position update will
be greater. This degradation is more important for the Kalman filter, as the degradation in the position
estimation is added.
2,4
2,2
2
1,8
1,6
1,4
1,2
1
0,8
0,6
0,4
0,2
0
PEE Trilateration
PEE Kalman
TE Trilateration
TE Kalman
0,2 0,4 0,6 0,8 1 1,2 1,4 1,6 1,8 2
Time between updates (s)
Figure 61: Position estimation error and tracking error depending on the time between updates
Finally, the impact of position update rate on the amount of resources needed for location has been
measured. As it can be observed in Figure 62, the amount of time slots needed for location is inversely
proportional to the time between updates. Therefore, the position update rate is limited by the
availability of resources for location.
Page 77
EUWB
Initial development of dissemination methods and evaluation
Number of slots per second
and per target
110
100
90
80
70
60
50
40
30
20
10
0
D4.2.1
RRq
RRp
DP-MR
DP-PU
Total
0,2 0,4 0,6 0,8 1 1,2 1,4 1,6 1,8 2
Time between updates (s)
Figure 62: Resources needed for location depending on the time between updates
As it can be observed, there is no optimal value for position update rate, as there is a trade-off between
tracking error and the amount of resources used for location. Position update rate must be selected
according to the required tracking accuracy and the dynamic model of the targets. For tracking people
at walking pace, 1 position update per second is a suitable value in order to keep the tracking error
around 1 meter with a reasonable use of resources.
Page 78
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
5 Network Impact for Localization
5.1 Quantity of Information
5.1.1 System Environment
An important goal of the EUWB project is the deployment of advanced localization and tracking (LT)
schemes. LT solutions have become a major asset for UWB applications [1], applicable to virtually all
industrial sectors, to those considered in the framework of EUWB.
Owing to its vast bandwidth, UWB-RT (ultra-wideband radio technology) enables a high resolution of
space surrounding UWB transceivers. This feature makes UWB-RT particularly attractive for LBS
services in a small scale environment, typically found in indoor or in-cabin scenarios. For instance, LT
can be used to identify whether a particular seat in an airplane has been occupied. In an automotive
environment, the localization of a wireless key, inside or outside a car and its distance from the car,
could be estimated.
Figure 63: Localization and tracking (LT) of a tag inside the car [38]
Figure 63 [38] illustrates a possible scenario for the latter case. Four fixed mounted UWB access
points (APs) are deployed to receive the UWB signals transmitted by a movable UWB tag. The
received UWB signals form the basis for the location estimation and the tracking of the tag. The
necessary signal processing will be done by a controller which could be included in one of the fixed
mounted APs.
The environment in which LT schemes shall be applied can be considered an in-house scenario. In
such an environment, a number of mobile devices (MDs) shall be localized and their position shall be
tracked. The LT scheme is based on a fixed mounted infrastructure, which relies on a network of
Page 79
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
access points (APs). MDs and APs are nodes of the network. It shall be mentioned that because of
reflections and scatterings, the radio channel is a fading multipath channel that can exhibit both LOS
(line-of-sight) as well as NLOS (non-line-of-sight) conditions.
The description of the in-house scenario gives a guideline to the development of LT scheme. To
facilitate a convenient LT application, the required signal and data processing shall consume as little
power as possible and it shall be realized with a least possible computational complexity. In addition,
the LT scheme shall be applicable to already existing UWB-RT devices without modifications of their
hardware. In summary, the following requirements are considered essential:
•
Scalability of the network up to 500 nodes.
•
Low mobility with MD speed below 4m/s.
•
Multi-hops in the network.
•
High localization accuracy in the order of 0.5 m or less in at least 50% of all cases.
•
Anchor-based localization (ABL) or anchor-free localization (AFL).
•
Robust against NLOS conditions and noise.
•
Low computational effort.
In the opinion of the authors, scalability must be considered as a most important factor in the system,
because it has a great impact on the accuracy and the flexibility of the LT scheme.
5.1.2 Network Architecture
The applied LT scheme depends on the network architecture. Both, centralized and decentralized
networks can be assumed [39]. In the case of the home environment a decentralized approach seems
more attractive whereas in public and private transport, a centralized approach is preferred. In this
manuscript we will hence consider a centralized network.
Figure 64 [39] shows a generalized sketch of a centralized network. The network consists of a server,
shown on the left hand side of Figure 64, six APs, shown at the top and at the bottom of Figure 64, and
of three movable tags, i.e. three MDs, shown in the center of Figure 64. The server and the six APs are
connected to each other e.g. by wire or in a wireless fashion. The tags are assumed to be wireless. The
task of the LT application in the network is to localize and track the positions of the tags.
In order to facilitate accurate position estimation, a tailored PHY (physical layer) and MAC (medium
access control) is desirable. It has been well-known that there are various LT procedures candidates,
cf. [39] and the references listed therein. We can distinguish between
•
iterative and non-iterative schemes,
•
LT in distributed architectures, and
•
recursive processing.
With respect to a simple and low-cost variant, however, a single shot scheme for PHY processing,
exploiting either the time of arrival (TOA) or the time difference of arrival (TDOA) is preferred. After
thorough evaluation in the FP6 PULSERS Phase II project [39], a TOA based PHY and MAC scheme
was proposed. According to [39], channel estimation and the exploitation of the channel estimation
results are required.
Page 80
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
UWB devices which do not fully comply with the procedures defined in [39] are initially not capable
of providing LT information. In order to alleviate this threat, a simple and low-cost LT application,
which does not require modifications of the PHY and the MAC, is needed. In this manuscript, the
authors will propose a novel LT application which can be implemented in the aforementioned
scenarios.
Figure 64: LT in a centralized network architecture [39]
The centralized network can also be arranged in a hierarchical manner. This can be seen in Figure 65.
The server is the network master. It controls the subservers. In the case of Figure 65, two subservers
exist. Each subserver handles the traffic by a number of APs, AP1, AP2, ..., AP#n.
Figure 65: Hierarchy in a centralized network architecture
In general, the APs are responsible for the sensing of the tags within their ranging scope, and then the
APs can evaluate the distance between themselves and tags. Finally, the measured data will be
transferred to the central server, which plays a role in computing the coordinates of tags according to
the data, which are collected from each AP. Because the real-time tracking function needs to be
realized, the time needed to acquire the information should be reduced. It assumes that there is only
ranging between anchors and tags.
Page 81
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
5.1.3 LT MAC Function [39]
In this section, the MAC layer of the LT scheme shall be illustrated. We will assume that TDMA (time
division multiple access) is used on the MAC layer. Hence, the server considered as the network
coordinator or master has to allocate one time slot for each node.
Figure 66: Typical super frame structure with three time slots in the ranging period [39]
The time slot architecture is shown in Figure 66. Each super frame consists of sixteen time slots of
equal duration, Tslot . The first time slot is a beacon slot. Then, there are two CAP (contention access
period) time slots. The following eight time slots are CFP (contention free period) time slots. Finally,
there are three ranging period time slots. Following the super frame, there can be an inactive period
(IP) which completes the beacon interval. Then, the next super frame starts. The CFP is used for data
frames, ranging frames, and other command frames. In such system, we consider the ranging frames
as data frames. In this section, we shall concentrate on ranging frames.
Each ranging frame has the data format as shown in Figure 67 [39]. It is assumed to be 64 bytes long.
Figure 67: Data format of a ranging frame [39]
The Frame Control field includes the “100” ranging type, no ACK request and both source and
destination addresses in short format mode; FC=0x8804.The Ranging Sequence Number is generated
by the initiator, and reused by the destination nodes as their response and drift answers. Address Info
should include the correct source and destination address. The first timer value is the fine one, which
occupies the four LSB bits of the two bytes. The second timer value is the coarse one, which occupies
unsigned 32 bits. The request, response and drift ranging frame should contain the time stamp or
processing delay.
Page 82
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Figure 68: Time slot allocation for the synchronization of the fixed mounted infrastructure
The time slot allocation for the synchronization of the fixed mounted infrastructure is shown in Figure
68. It comprises four messages, namely,
•
a ranging request message,
•
a ranging response message,
•
a TX delay message and
•
a synchronization message.
The corresponding MAC procedure will be illustrated in section 5.1.4.2.
5.1.4 LT Application
5.1.4.1
Introduction
In this section, a low-cost LT application which will be used in the UWB open technology platforms
shall be presented. As already pointed out in [38], a cross-layer functionality is needed to accomplish
LT. The required communications protocol, which will be illustrated in what follows, comprises three
steps, namely,
•
the synchronization of the fixed mounted infrastructure (Step I),
•
the localization of a mobile device (Step II) and
•
the triangulation and mapping (Step III).
The LT procedure will be briefly illustrated assuming the deployment of two fixed mounted access
points AP1 and AP2 and a mobile device (MD). Figure 69 shows the schematic set-up. The AP1 also
acts as the master also containing the aforementioned controller whereas the AP2 and potentially
further APs will serve as slaves. Also, the MD is considered as a slave device. The distance between
the AP1 and the AP2 is denoted by d 0 , d1 is the distance between the AP1 and the MD, whereas d 2 is
the distance between the AP2 and the MD. The angle included between the segments (AP2, AP1) and
(AP2, MD) is termed α , β is the angle included between the segments (AP1, AP2) and (AP1, MD),
and γ is the angle included between the segments (MD, AP1) and (MD, AP2). All these six quantities
shall be determined.
Page 83
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Figure 69: Generic localization tracking set-up [38]
5.1.4.2
Step I: Synchronization of the Fixed Mounted Infrastructure
It has been known that synchronized cellular networks provide many benefits, e.g. an increased
spectral efficiency, a dynamic traffic balancing, an easy handover procedure and a simple localization
of mobile devices, all at a protocol effort lower than in non-synchronized cellular environments. A
first solution to synchronization of adjacent cells in mobile radio was introduced in the C 450 network
deployed e.g. in Germany [40]. This concept was however not followed in GSM (Global System for
Mobile Communications) [41] but re-introduced as an option in UMTS (Universal Mobile
Telecommunications System).
Synchronization in small scale networks which occur in environments exploiting UWB-RT hence
seems a feasibly approach. Therefore, the first step, Step I, takes care of the synchronization of the
network.
In what follows we will mainly concentrate on the MAC scheme. Let us assume that the MAC is time
slot based, i.e. using a TDMA (time division multiple access) scheme. Each slot is considered to have
the same duration, Tslot . Furthermore, we will consider AP1 as the master. We will assume that all APs
have the same Tslot , however, that the start and, consequently, the end times of the time slots are not
aligned at the beginning of the procedure. Figure 70 illustrates the procedure which shall be described
in what follows.
Page 84
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Figure 70: Synchronization of the fixed mounted infrastructure [38]
In order to synchronize the fixed mounted infrastructure, AP1 transmits a ranging request to all further
APs, in our simple example to AP2. The ranging request message contains the particular time instant,
t0 , at which it was initiated in the LT application. The ranging request message contains the sequence
number, the source address and destination address as well as the time instant t0 . We assume that t0 is
the beginning of a time slot. The transmitter of AP1 will require a given processing time prior to the
ranging request can be transmitted. This signal processing delay shall be denoted by δ TX,AP1 , the actual
time of transmission is hence
tTX,AP1 = t0 + δ TX,AP1 .
(5.1)
It shall be noted that AP1 measures this delay δ TX,AP1 .
The ranging request message requires the travel time d 0 c to arrive at the antenna of AP2. Hence, the
time of reception is given by
tRX,AP2 = t0 + δ TX,AP1 +
d0
.
c
(5.2)
AP2 then extracts the sequence number, the source address and destination address from the received
ranging request message and checks them. Furthermore, it extracts the message content t0 .
The AP2 initiates a ranging response message at time t1 , the beginning of a time slot following the
reception of the ranging request message. In addition to the sequence number and the source and the
Page 85
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
destination addresses, the ranging response message shall contain tRX,AP2 and the time X that elapsed
between tRX,AP2 and t1 . The transmitter of AP2 will also require a given processing time prior to the
ranging request can be transmitted. This signal processing delay shall be denoted by δ TX,AP2 , the actual
time of transmission is hence
tTX,AP2 = t1 + δ TX,AP2 = t0 +
d0
+ X + δ TX,AP1 + δ TX,AP2 .
c
(5.3)
It shall be noted that AP2 measures δ TX,AP2 .
AP1 receives this ranging response message at
tRX,AP1 = t0 + 2
d0
+ X + δ TX,AP1 + δ TX,AP2
c
(5.4)
and decodes it. Except for δ TX,AP2 and d 0 , all terms in (5.4) are known to AP1.
In the time slot following the transmission of the ranging response message, AP2 transmits a TX delay
message containing the sequence number, the source and the destination addresses and δ TX,AP2 . This
message is received at
tRX,AP1,2 = t0 + 2
d0
+ X + δ TX,AP1 + δ TX,AP2 + Tslot
c
(5.5)
by AP1. After its decoding, AP1 computes
c
d 0 = ( tRX,AP1 − t0 − X − δ TX,AP1 − δ TX,AP2 ) .
2
(5.6)
Next, AP1 transmits the synchronization message containing d 0 , t3 and δ TX,AP1 . After receiving this
message, AP2 can completely synchronize its TDMA frame timing. It is conceivable, that AP2
acknowledges the synchronization to AP1 after completion.
After the synchronization of the fixed mounted infrastructure has been established, the
synchronization of the mobile device MD should be established. This can be done in the same way as
described for the case of AP2.
5.1.4.3
Step II: Localization of a Mobile Device
After the synchronization of the fixed mounted infrastructure and of the MD have been established,
the second step, Step II, can start. In order to improve the readability, the measurement and correction
of processing delays at the transmitters and the receivers shall be neglected. Step II contains six
activities:
1. The AP1 transmits the message containing the time instant T0 at the time instant T0 , .
2. The MD receives this message at the time instant
T1 = T0 + d1 c .
(5.7)
3. The AP2 then transmits the message containing the time difference Tslot at the time instant
(T0 + Tslot ) .
Page 86
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
4. The MD receives this message at the time instant
T2 = T0 + Tslot + d 2 c .
(5.8)
5. The MD then computes
d1 = c (T1 − T0 ) ,
d 2 = c (T2 − T0 − Tslot ).
(5.9)
6. Finally, the MD transmits a message containing the distance values d1 and d 2 to the AP1.
5.1.4.4
Step III: Triangulation and Mapping
In the last step, Step III, the AP1 computes the angles α , β and γ . This is accomplished by applying
the following identities
2
2
2
⎪⎧ ( d 0 ) + ( d 2 ) − ( d1 ) ⎪⎫
α = arccos ⎨
⎬,
2d 0 d 2
⎩⎪
⎭⎪
(5.10)
⎧⎪ ( d 0 )2 + ( d1 ) 2 − ( d 2 ) 2 ⎫⎪
β = arccos ⎨
⎬
2 d 0 d1
⎩⎪
⎭⎪
(5.11)
γ = π −α − β .
(5.12)
and
After completion of Step II and Step III, the location of MD can be computed by AP1 and can be
conveyed to the other UWB-RT devices if required.
It shall be mentioned that this illustration is given to explain the principle of the LT procedure, being
well aware of the fact of the 180 degree ambiguity of the result. This ambiguity can be overcome by
adding further fixed mounted APs.
5.1.4.5
Quantity of Information Without Packet Loss
The quantity of information is the total amount information during the localization process. There are
mainly two types of ranging frames,
•
ranging frames for synchronisation and
•
ranging frames for location tracking
In an ideal world, there is no packet loss during the transmission. In such a situation, the ranging
frames are transmitted only once. The synchronisation procedure for AP2 needs four slots. From
Figure 67, we can find that “Ranging Sequence Number”, “Address Info”, “Time Value (fine)” and
“Time Value (coarse)” should be used. Therefore, for AP2, we can get the quantity of information:
I AP2,syn = 64 ⋅ 4 = 256 bytes .
(5.13)
Let us assume that there are M APs in our system. We can get the quantity of total information for
synchronisation:
I total,syn = (M − 1) ⋅ I AP2,syn .
(5.14)
Page 87
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Because the LT-tracking procedure for each MD needs 11 slots, we can get
I md1,LT = 11⋅ 64 = 704 bytes .
(5.15)
Let us now assume that there are N MDs in our ranging area. Hence, we can get
I LT,total = 704 N bytes .
5.1.4.6
(5.16)
Quantity of Information with Packet Loss
Retransmissions are required in the case of lost packets. The synchronization for one AP needs four
time slots to realize the reliable communication. If AP1 cannot receive the response ranging frame or
the drift ranging frame within the second or third time slot, the communication should be considered
to be unsuccessful, then the retransmission should be done in the next super frame. AP2 should send
AP1 another ranging frame to tell AP1 that it has been synchronized. It assumes that the probability of
request ranging drop is Preq , the probability of response ranging drop is Presp , the probability of drift
ranging drop is Pdrift , the probability of synchronisation ranging is Psyn and the probability of
acknowledgement ranging is Pack . We can get the probability of retransmission
Pdrop = 1 − (1 − Preq )(1 − Presp ) (1 − Pdrift ) (1 − Psyn ) (1 − Pack ) .
(5.17)
We assume that channel coding is used in the retransmission and the code rate is Rcode . We can get the
quantity of information for synchronization:
I AP2,syn,pd = 4 ⋅ 64 + 4 Pdrop
P ⎞
⎛
64
= 256 ⎜ 1 + drop ⎟ bytes .
Rcode
⎝ Rcode ⎠
(5.18)
If there are M APs in our system, we can get the quantity of total information:
P ⎞
⎛
I total,syn = (M − 1)* I AP2,syn,pd = 256 ( M − 1) ⎜ 1 + drop ⎟ bytes .
⎝ Rcode ⎠
(5.19)
It requires 11 time slots for LT. We can get
P ⎞ ⎛
P ⎞ ⎛
P ⎞
⎛
I LT,total = 3⎜ 3 ⋅ 64 + 3 ⋅ 64 ⋅ drop1 ⎟ + ⎜ 64 + 64 ⋅ drop2 ⎟ + ⎜ 64 + 64 ⋅ drop3 ⎟ .
Rcode ⎠ ⎝
Rcode ⎠ ⎝
Rcode ⎠
⎝
5.1.4.7
(5.20)
Simulation results
We give the simulation results of the estimation. The channel uncertainty needs not to be taken into
account because it does not affect the accuracy of estimation, only cause more packet retransmission.
We set a random number a with the normal Gaussian distribution.
Page 88
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Figure 71: Estimated position of tag with the process delay 10−9 ⋅ a
Figure 72: Cumulative Density Function (CDF) of the estimation error with the process Delay: 10−9 ⋅ a
Page 89
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Figure 71 depicts the estimation results, where AP1, AP2, AP3 and AP4 locate at the four corner with
the distance of 2m to each other. We assume that the process delay is 10−9 ⋅ a . Figure 72 shows
cumulative density function of the estimation error. As can be seen from Figure 72, the probability of
estimation error less than 0.1m exceeds 60% and the probability of estimation error less than 0.15m
exceeds 80%.
Figure 73 depicts the estimation results with the process delay: 2.5 ⋅10−9 ⋅ a . Figure 74 shows
cumulative density function of the estimation error. As can be seen from Figure 74, the estimation
becomes more inaccurate due to the larger process delay than that in Figure 71, the probability of
estimation error less than 0.2m just reaches 60% and the probability of estimation error less than 0.3m
locates between 80% and 90%.
Figure 73: Estimated position of tag with the process delay 2.5 ⋅10−9 ⋅ a
Page 90
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Figure 74: Cumulative Density Function (CDF) of the estimation error with the process Delay: 2.5 ⋅10−9 ⋅ a
Figure 75: Estimated position of tag with the process delay 5 ⋅10−9 ⋅ a
Page 91
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
Figure 75 depicts estimation results with the process delay: 5 ⋅10−9 ⋅ a . Figure 76 shows the
corresponding cumulative density function of the estimation, where the probability of estimation error
less than 0.5m just exceeds 60% and the probability of estimation error less than 0.8m just reach 90%.
The estimation becomes more inaccuracy than those in Figure 73 and Figure 75 due to the larger
process delay.
Figure 76: Cumulative Density Function (CDF) of the estimation error with the process Delay: 5 ⋅10−9 ⋅ a
Page 92
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
6 Conclusions
The deliverable mainly considers location aware information dissemination and acquisition in UWB
based networks, where localization and routing strategy and information exchange were mainly
introduced.
A cross-layer algorithm for geographic routing in infrastructureless networks is introduced. This
solution is robust to topological holes and is resilient to topological variations due to network
dynamics (node duty cycle). The solution combines ideas of network tessellation with simple greedy
forwarding, without suffering from the problems afflicting typical landmark-based alternatives. We
tessellate the network based on connectivity information and then define virtual landmarks to identify
them. In addition to not requiring landmarks to be known a priori, a major advantage of the proposed
tessellation approach is that the clusters are composed by groups of nodes that are highly
interconnected. Problems of routing discovery and maintenance in ad hoc networks were also
introduced. The underlying concept is to exploit the availability of cooperative communications in
relaying networks so as to benefit from spatial diversity gains.
It was shown that cluster sizes may affect the resulting PDSR to APDL. In other words, cluster sizes
can be varied so as to allow for different trade-offs between PDSR to APDL to be reached. It was also
found, however, via extensive simulations, that the dependence of the overall technique on the cluster
sizes is not as strong as the dependence of landmark-based routing mechanisms on the location of
selected landmarks. Alternatively, we may say that while there is room for further improvement via
cluster size optimization, the proposed technique is less sensitive to a bad choice of cluster size than
landmark methods are on the location of landmarks. The simulations campaign has confirmed that the
technique can substantially improve the PDSR in networks with large concave holes, with no or little
impact on APDL, even if the simple greedy forwarding mechanism is employed. There is, however,
no reason not to employ more sophisticated mechanisms combined with the clusterization idea. It has
been assumed that network discovery has been already performed and that the network is in a steadystate regime of operation, when the routing protocol itself starts to operate. The network discovery has
not been explicitly addressed in the research activities, though it is still an open issue and it is
considered to further investigations.
We implement a novel resource allocation strategy for distributed IR-UWB networks that arise from
the fact that, in contrast to what happen in narrowband networks, bandwidth is no longer the limiting
resource but the number of pulses per second, i.e. the cumulative pulse load. The DLC enables
concurrent transmissions at full power, while allows each source to independently adapt its error
coding rate, and its pulse rate (the number of transmitted pulses per second) to the current channel and
interference conditions. Thus, our strategy combines adaptive channel coding (ACC) with pulse rate
control (PRC) in a joint link adaptation function, which is denoted hereafter as the Link Parameter
Control (LPC) function.
Our distributed mechanism makes use of the framework of game theory to develop a novel
interference management strategy based on the joint combination of adaptive error coding and pulse
rate control. A game has been formulated and a distributed, asynchronous algorithm (LPC) has been
proposed to discover the optimal PRF/PL network allocation under several channel and interference
conditions.
It has been observed that, due to the strong coupling concerning interference management between PL
and PRF adaptation, the game has several feasible steady-state allocations (Nash equilibria). In
Page 93
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
general, the different equilibria provide similar aggregated network throughput. Results suggest that,
in order to maximize the cumulative network throughput, it is better that each link starts with the
lowest PL. It can be concluded that the joint use of PRC and ACC is an appropriate means to mitigate
impulsive interference in autonomous IR-UWB.
Finally we evaluated the application of simultaneous data transmission and localization UWB
networks for tracking mobile access networks users in indoors environments. Different architectures
and acquisition and distribution schemes for tracking have been analysed and assessed through
simulation. The impact of other parameters such as the number of anchors used for location, the
distance between anchors, target mobility and position update rate have been also evaluated.
Simulation results proved the feasibility of this application scenario, but also showed a limitation in
the number of targets that can be tracked simultaneously in a piconet due to resources limitation.
Therefore we have analysed the quantity of information and total traffic load which is consumed in
localization and tracking system. The enhanced schemes for the acquisition and distribution of
location information and the limitation of the number of anchors used for localization, taking advance
of available location information to pick the closest anchors, are proposed solutions in order to reduce
the amount of resources consumed for localization, and further research will be carried out in this
direction.
Analysis and simulation showed that our novel contention-based geographic forwarding strategies in
multi-hop scenarios as well as routing discovery and maintenance strategies, game theory based
resources allocation schemes, and location and tracking algorithm functions well. The location aware
information dissemination and acquisition prototype was basically outlined. The future task is to
combine all the work together and streamline the combination.
Page 94
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
R. Pabst, B. Walke, D. Schultz, P. Herhold, H. Yanikomeroglu, S. Mukherjee, H. Viswanathan,
M. Lott, W. Zirwas, M. Dohler, H. Aghvami, D. Falconer, and G. Fettweis, “Relay-based
deployment concepts for wireless and mobile broadband radio,” IEEE Communications
Magazine, vol. 42, no. 9, pp. 80–89, September 2004.
J.-W. Lee, A. Tang, J. Huang, M. Chiang, and A. R. Calderbank, “Reverse-engineering MAC:
A non-cooperative game model,” IEEE Journal on Selected Areas in Communications, vol. 25,
no. 6, pp. 1135–1147, August 2007.
P. Casari, A. Marcucci, M. Nati, C. Petrioli, and M. Zorzi, “A detailed simulation study of
geographic random forwarding (GeRaF) in wireless sensor networks,” in IEEE Military
Communications Conference, vol. 1, October 2005, pp. 59–68.
D. Chen, J. Deng, and P. K. Varshney, “On the forwarding area of contention-based geographic
forwarding for ad hoc and sensor networks,” in Second Annual IEEE Sensor and Ad Hoc
Communications and Networks, 2005, September 2005, pp. 130–141.
F. Davoli, S. Palazzo, and S. Zappatore, Distributed Cooperative Laboratories: Networking,
Instrumentation and Measurements. Springer, 2006.
C. H. M. de Lima and G. T. F. de Abreu, “Game-theoretical relay selection strategy for
geographic routing in multi-hop WSNs,” in 5th Workshop on Positioning, Navigation and
Communication 2008 (WPNC’08), Leibniz University of Hannover, Germany, March 27 2008.
D. Chen, J. Deng, and P. K. Varshney, “Selection of a forwarding area for contention-based
geographic forwarding in wireless multi-hop networks,” IEEE Transactions on Vehicular
Technology, vol. 56, no. 5, pp. 3111–3122, September 2007.
A. Baddeley, I. Bárány, R. Schneider, and W. Weil, Stochastic Geometry, 1st ed. Springer,
November 2006.
S. Srinivasa and M. Haenggi, “Distance distributions in finite uniformly random networks:
theory and applications,” IEEE Transactions on Vehicular Technology, 2008, submitted.
P. Mathys and P. Flajolet, “Q-ary collision resolution algorithms in random-access systems with
free or blocked channel access,” IEEE Transactions on Information Theory, vol. 31, no. 2, pp.
217–243, March 1985.
J. Abate and W. Whitt, “Numerical inversion of probability generating functions,” Operations
Research Letters, vol. 12, pp. 245–251, 1992.
G. Fayolle, P. Flajolet, M. Hofri, and P. Jacquet, “Analysis of a stack algorithm for random
multiple-access communication,” IEEE Transactions on Information Theory, vol. 31, no. 2, pp.
244–254, March 1985.
S. Biswas and Y. Narahari, “Iterative dutch combinatorial auctions,” Annals of Mathematics
and Artificial Intelligence, vol. 44, no. 3, pp. 185–205, 2005.
P. Casari, M. Nati, C. Petrioli, and M. Zorzi, “Efficient non-planar routing around dead ends in
sparse topologies using random forwarding,” in IEEE International Conference on
Communications, 2007, 24–28 June 2007, pp. 3122–3129.
C. H. M. de Lima and G. T. F. de Abreu, “Clusterization for robust geographic routing in
wireless sensor networks,” in Wireless Communications & Networking Conference (WCNC),
Las Vegas, Nevada, USA, 31 March–3 April 2008.
Bruno Selva, Ana Sierra, Amir Krause and Juan Chóliz, “Definition of application scenarios
and definition of requirements”, D 6.1.1 version 1.0, July 2008, EUWB Project.
Elizabeth M. Royer and Chai-Keong Toh, “A review of current routing protocols for ad hoc
mobile wireless networks”, IEEE Personal Communications, April 1999.
Page 95
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
[18] Ivan Stojmenovic, Amiya Nayak and Johnson Kuruvila, “Design guidelines for routing
protocols in ad hoc and sensor networks with realistic physical layer”, IEEE Communications
Magazine, March 2005.
[19] J. N. Al-Karaki and A. E. Kamal, “Routing techniques in wireless sensor networks: a survey,”
IEEE Wireless Communications, vol. 11, no. 6, pp. 6–28, December 2004, see also IEEE
Personal Communications.
[20] S. Waharte, R. Boutaba, Y. Iraqi, and B. Ishibashi, “Routing protocols in wireless mesh
networks: challenges and design considerations,” Multimedia Tools Appl., vol. 29, no. 3, pp.
285–303, 2006.
[21] A. Rao, S. Ratnasamy, C. Papadimitriou, S. Shenker, and I. Stoica, “Geographic routing without
location information,” in ACM MobiCom Conference, September 2003, pp. 96–108.
[22] B. Karp and H. T. Kung, “Greedy perimeter stateless routing for wireless networks,” in
Proceedings of the Sixth Annual ACM/IEEE International Conference on Mobile Computing
and Networking, 2000. MobiCom 2000, August 2000, pp. 243–254.
[23] G. Destino, D. Macagnano, and G. Abreu, “A clusterized WLS localization algorithm for large
scale WSNs,” in 4th Workshop on Positioning, Navigation and Communication - WPNC’07,
March 2007, pp. 261–265.
[24] O. Krishnadev, K. Brinda, and S. Vishveshwara, “A graph spectral analysis of the structural
similarity network of protein chains,” PROTEINS: Structure, Function, and Bioinformatics, vol.
61, pp. 152–163, 2005.
[25] A. Jadbabaie, “On geographic routing without location information,” in 43rd IEEE Conference
on Decision and Control, vol. 5, December 2004, pp. 4764–4769.
[26] Q. Fang, J. Gao, L. J. Guibas, V. de Silva, and L. Zhang, “GLIDER: Gradient Landmark-Based
Distributed Routing for sensor networks,” in 24th Annual Joint Conference of the IEEE
Computer and Communications Societies, 2005. INFOCOM 2005, vol. 1, March 2005, pp. 339–
350.
[27] G. Destino and G. Abreu, “Sensor localization from WLS optimization with closed-form
gradient and Hessian,” in IEEE Global Conference on Telecommunications - Wireless
Communications & Networking Symposium (Globecom 2006), San Francisco, USA, November
27th - December 1st 2006.
[28] A. Sierra, J. Chóliz, I. Bucaille, A. Villanúa, A. Hernández, I. Alastruey, “UWB in
Heterogeneous Access Networks”, IEEE ICUWB’08, Hanover, Sept. 2008.
[29] IEEE 802.15.4 standard, “Part 15.4: Wireless Medium Access Control (MAC) and Physical
Layer (PHY) Specifications for Low-Rate Wireless Personal Area Networks (LR-WPANs)”
[30] “LDR-LT platform requirements, feasibility analysis and specification” D7.1.1a final version,
September 2008, EUWB (IST FP7-215669), public deliverable.
[31] D. Macagnano, G. Destino, F. Esposito, and G. T. F. de Abreu, “MAC Performances for
Localization and Tracking in Wireless Sensor Networks”, In Proc. WPNC’07, Hannover.
[32] F. Daum, ‘’Nonlinear filters: beyond the Kalman filter’’, IEEE Aerospace and Electronic
Systems Mag., vol. 20, no. 8, pp. 57-69, Aug. 2005.
[33] D. Macagnano and G. T. F. de Abreu, “Tracking Multiple Targets with Multidimensional
Scaling,” in Wireless Personal Multimedia Communications, Sep. 2006
[34] “Enhanced LT system- PHY, MAC and Network layers (final)” D3a-2.3 final version, June
2008, PULSERS Phase II (IST FP6-027142), public deliverable.
[35] B. Denis and N. Daniele, “NLOS Ranging Error Mitigation in a Distributed Positioning
Algorithm for Indoor UWB Ad Hoc Networks”, in Proc. IEEE IWWAN’04,, pp. 356-360 Oulu,
May-June 2004
Page 96
EUWB
Initial development of dissemination methods and evaluation
D4.2.1
[36] B. Denis, J.B. Pierrot and C. Abou-Rjeily, “Joint Distributed Time Synchronization and
Positioning in UWB Ad Hoc Networks Using TOA”, IEEE Trans. on MTT, Special Issue on
Ultra Wideband, Vol. 54, Is. 4, Part 2, pp. 1896-1911, April 2006.
[37] S. Zeisberg, V. Schreiber: “EUWB - Coexisting short range radio by advanced ultra-wideband
radiotechnology,” ICT-MobileSummit 2008 Conference Proceedings, Paul Cunningham and
Miriam Cunningham (Eds), IIMC International Information Management Corporation, 2008,
ISBN: 978-1-905824-08-3.
[38] C. Kocks, E. Scheiber, A. Vießmann, D. Xu, S. Wang, G. H. Bruck, P. Jung: “ A low-cost
protocol and application for UWB localization, exploiting cross-layer design and cognitive
radio aspects,” Invited Paper at the Second International Workshop on Cognitive Radio &
Advanced Spectrum Management (CogART 2009), May 18-19, 2009, Aalborg, Denmark.
[39] -, “LT system – PHY, MAC and network layers,” D3a-2.1 final version, June 2006, PULSERS
Phase II (IST FP6-027142), public deliverable.
[40] K. Kammerlander: “Factors influencing the design of economical cellular radio networks,”
Siemens Telecom Report, vol. 9 (1986), pp. 9-16.
[41] M. Mouly, M. B. Pautet: “The GSM-system for mobile communications,” 1992.
[42] G. Rahmatollahi, M.D. Perez Guirao, P. Nardelli, G. Abreu, “Aggregate Information Efficiency
in IR-UWB ad hoc single-hop WSNs, WPNC 2009, Hannover, 2009.
[43] M.D. Pérez Guirao, K. Jobmann, Cognitive resource access scheme for IR-UWB autonomous
networks, , 4th Workshop on Positioning, Navigation and Communication 2007, Hannover,
2007
[44] M.D. Pérez Guirao, K. Jobmann, CA (UWB)-MAC: Cognitive Autonomous MAC for IR-UWB
Networks, Second International Conference on Cognitive Radio Oriented Wireless Networks
and Communications, CrownCom’07, 1-3 August 2007, Orlando, FO
[45] G. Rahmatollahi, M.D. Perez Guirao, S. Galler, T. Kaiser, “Position Estimation in IR-UWB
autonomous WSNs”, WPNC 2008, Hannover, 2008.
[46] Weisenhorn, M. and Hirt W., “Uncoordinated rae division mutliple access scheme for pulsed
UWB signals”, IEEE Trans. on Vehicular Technology, vol. 54, pp. 1646-1662, Sept. 2005
[47] Srivasta, V. et. al, “Using game theory to analyze wireless ad hoc networks”, IEEE
Communications Survey & Tutorials, vol.7, no.4, pp 46-56, Fourth Quarter 2005.
[48] J. Neel, "PhD Dissertation: Analysis and Design of Cognitive Radio Networks and Distributed
Radio Resource Management Algorithms," in Faculty of the Virginia Polytechnic Institute and
State University. Blacksburg, VA: Virginia Tech, 2006.
[49] D. Fudenberg and J. Tirole, Game Theory: MIT Press, 1991.
[50] PULSERS II D3b.2.1b, System Architecture.
Acknowledgement
The EUWB consortium would like to acknowledge the support of the European Commission partly
funding the EUWB project under Grant Agreement FP7-ICT-215669.
Page 97