Principles of embedded networked systems design

Embedded network systems ENS provide a set of technologies that can link thephysical world to large scale networks in applications such as monitoring of borders,infrastructure, health, t

Embedded network systems (ENS) provide a set of technologies that can link thephysical world to large scale networks in applications such as monitoring of borders,infrastructure, health, the environment, automated production, supply chains, homes,and places of business This book details the fundamentals for this interdisciplinary andfast-moving field The book begins with mathematical foundations and the relevantbackground topics in signal propagation, sensors, detection and estimation theory, andcommunications Key component technologies in ENS are discussed: synchronizationand position localization, energy and data management, actuation, and node architec-ture Ethical, legal, and social implications are addressed The final chapter summarizessome of the lessons learned in producing multiple ENS generations A focus on funda-mental principles together with extensive examples and problem sets make this text idealfor use in senior design and graduate courses in electrical engineering and computerscience It will also appeal to engineers involved in the design of ENS.

G R E G O R Y P O T T I E has been a faculty member of the UCLA ElectricalEngineering Department since 1991, serving in vice-chair roles from 1999 to 2003.Since 2003 he has served as Associate Dean for Research and Physical Resources ofthe Henry Samueli School of Engineering and Applied Science From 1997 to 1999 hewas secretary to the board of governors for the IEEE Information Theory Society In

1998 he was named the faculty researcher of the year for the UCLA School ofEngineering and Applied Science for his pioneering role in wireless sensor networks,and in 2005 was elected as a Fellow of the IEEE Professor Pottie is a deputy director ofthe NSF-sponsored science and technology Center for Embedded Networked Sensorsand is a cofounder of Sensoria Corporation

W I L L I A M K A I S E R joined the UCLA Electrical Engineering Department in

1994 and there, with Professor Pottie, initiated the first wireless networked sensor programs with a vision of linking the Internet to the physical world through

micro-Chairman from 1996 to 2000 He has received the Allied Signal Faculty Award, thePeter Mark Award of the Vacuum Society, the NASA Medal of Exceptional ScientificAchievement, the Arch Colwell Best Paper Award of the Society of AutomotiveEngineers and two R & D 100 Awards He is cofounder of Sensoria Corporation.

Principles of Embedded Networked Systems

Design

Gregory J Pottie and William J Kaiser

University of California, Los Angeles

Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São PauloCambridge University Press

The Edinburgh Building, Cambridge , UK

First published in print format

- ----

- ----

© Cambridge University Press 2005

2005

Information on this title: www.cambridg e.org /9780521840125

This publication is in copyright Subject to statutory exception and to the provision ofrelevant collective licensing agreements, no reproduction of any part may take placewithout the written permission of Cambridge University Press

- ---

- ---

Cambridge University Press has no responsibility for the persistence or accuracy ofsfor external or third-party internet websites referred to in this publication, and does notguarantee that any content on such websites is, or will remain, accurate or appropriate

Published in the United States of America by Cambridge University Press, New York

www.cambridge.org

hardback

eBook (EBL)eBook (EBL)hardback

To Aldo, Cathy, Claire, and Laura for their patient love and support To the dedicatedand creative students and colleagues of the LWIM, AWAIRS, and CENS programs

at UCLA and to the Sensoria Corporation team

Contents

3.7 Further reading 57

4.8 Electronic noise sources and noise reduction in sensor systems 964.9 Reducing sensor system errors by feedback control methods 100

8.5 Interaction of signal processing and networking in

12.5 Further reading 390

16.3 ENS: information technology regulating the physical world 485

Embedded network systems (ENS) provide a set of technologies that can link thephysical world to large scale networks for such purposes as monitoring of borders,infrastructure, health, the environment, automated production, supply chains, homes,and places of business ENS nodes integrate the novel combination of signal proces-sing, communication, sensing, and actuation technology Their composition into largenetworks requires knowledge of networking and distributed software systems Manyexcellent textbooks exist that treat these topics separately, and there are correspondingundergraduate and graduate courses However, these provide both too much informa-tion on some topics and not enough on others for a course specifically devoted to ENS.The purpose of this book is to provide support for senior design courses and intro-ductory graduate courses in ENS without the requirement for students to haveexpertise in all of these areas As such it can also serve as a resource for the practicingprofessional in this rapidly expanding area of research and enterprise Note what thebook is not: a comprehensive and objective treatment of the latest developments insensor networks We do not presume to compete with the varied riches offered on theworldwide web by what is now a large and very creative group of researchers aroundthe world Therefore our focus is consciously on principles and methods which haveproven useful to us in the course of designing multiple generations of ENS (research,commercial products, class projects), with digressions to what in our opinion areinteresting topics for new investigations We encourage our readers to go beyondour necessarily limited and subjective take on the facts, and delve deeper into thisfascinating realm of research

The book begins with an overview of ENS systems, and then discusses tical foundations, signal propagation, sensors, detection and estimation theory, andcommunications Chapters7and8then discuss multiple access and networking issues,including information theoretic results that apply to sensor networks Chapters9 12present other key component technologies in ENS: synchronization and positionlocalization, energy management, data management, and actuation (both articulationand mobility) Chapters13and14discuss architectural principles for the design ofnodes and networks of nodes that are to be remotely accessible, and means to ensure

mathema-xiii

the data are both trusted and secure Chapter15discusses how to design experimentswith ENS for some particular scientific purpose Chapter 16presents a number ofquestions for how ENS fit into our present and future societies The final chaptersummarizes a number of the lessons learned in producing multiple ENS generations.AppendixAconsists of tables of the Gaussian Q function, while Appendix B provides

an introduction to formal optimization techniques

Clearly not all of these issues can be covered in a single quarter or semester, and in adesign team not all of this expertise need reside in every individual However, in totalthis is the set of topics that must be mastered in order to design efficient systems It isexpected that students will have a stronger background in some areas than others, withthe topics selected tailored to their particular expertise, and the specific design con-cerns in the course The material in this book might be supplemented with the basicscience of some particular application such as seismic monitoring, and a systemdesigned for that purpose Chapter16, on ethics and social impact, differs consider-ably in style from the other chapters and could, for example, in conjunction withChapter1serve as a unit in an ethics class or as a stand-alone one-unit seminar course

In a field that is rapidly changing and which spans such a broad set of topics we haveopted to focus more upon basic principles than particular algorithms or hardware Aweb site with reference designs from a senior design course will be maintained toprovide up-to-date examples of software and hardware Other resources available toinstructors include a solution manual which we hope to supplement with new pro-blems suggested by our readers, and equations and figures in electronic form to assist

in the creation of lecture notes

The authors gratefully acknowledge the research contributions of many students andcolleagues who over the years have turned nebulous concepts regarding sensor net-works into real systems, both at UCLA and Sensoria Corporation Particular thanksfor direct contributions to this book go to Dr Hong Chen, Huiyu Luo, Dr AmeeshPandya, and Dr Yung-Szu Tu for the creation of problem sets (and solutions).Professor Andreas Savvides provided the position location Cramer–Rao bound exam-ple in Chapter 9, while Dr Ryan Mukai provided the linearized multilaterationproblem formulation and establishment of a reference coordinate system RichardPon, Jason Gordon, Ryan Speelman, and Dustin McIntire characterized the IEEE802.11b, narrowband radio, and microprocessor systems described in Chapter13 Theauthors would also like to thank Dr Fredric Newberg for much advice over manyyears in ENS system engineering and for the energy storage system analysis in thisbook The genesis of Chapter 16 was a transdisciplinary set of discussions at theUCLA Institute of Pervasive Computing, moderated by Professors Jerry Kang andDana Cuff We are also grateful to both the Defense Advanced Research ProjectsAgency and the National Science Foundation for their support of numerous researchprojects in sensor networks that have enabled both our own work and the creation of avigorous research community Discussions at the NSF Center for EmbeddedNetworked Sensing, under the leadership of Professor Deborah Estrin, have been aconstant source of inspiration The patience of the editorial team at CambridgeUniversity Press for our perpetually pushed-back deadlines is also appreciated.Finally, we also thank our respective spouses, Aldo Cos and Cathy Kaiser for theircheerful tolerance of our many weekend and evening hours spent at work

xv

AES advanced encryption standard

ALU arithmetic logic unit

AODV ad hoc on-demand distance vector

API application program interface

APS active pixel sensor

ARQ automatic repeat request

ASIC application-specific integrated circuit

AWGN additive white Gaussian noise

BFSK binary frequency shift keying

BPSK binary phase shift keying

BSC binary symmetric channel

cdf cumulative distribution function

CDMA code division multiple access

CMOS complementary metal oxide semiconductor

CPU central processing unit

CSMA carrier sense multiple access

CSMA/CD carrier sense multiple access collision detection

DARPA Defense Advanced Research Projects Agency

xvi

DBMS database management system

DCA dynamic channel allocation

DES data encryption standard

DFE decision feedback equalizer

DMT discrete multitone modulation

DPCA dynamic power and channel allocation

DPSK differential phase shift keying

DS-SS direct sequence spread spectrum

DVS dynamic voltage scaling

EDF earliest deadline first

EIRP effective isotropic radiated power

ELSI ethical, legal, and social implications

FDMA frequency division multiple access

FEC forward error correction

FET field effect transistor

FFT fast Fourier transform

FH-SS frequency-hopped spread spectrum

FIR finite impulse response

FPGA field programmable array

FSK frequency shift keying

ftp file transfer protocol

FUSD framework for user space device

GAF geographic adaptive fidelity

GPIO general purpose input/output

GPS global positioning system

HAL hardware abstraction layer

IDE integrated drive electronics

IFFT inverse fast Fourier transform

ISA instruction set architecture

ISI intersymbol interference

ISR interrupt service routine

JTAG Joint Test Action Group

LIA least interference algorithm

LNA low-noise amplifiers

LRT likelihood ratio test

LWIM low-power wireless integrated microsensors

MANET mobile ad hoc networks

MEMS microelectromechanical system

MIMO multiple-input multiple-output

MLSE maximum likelihood sequence estimation

MMSE minimum mean square error

MOSFET metal oxide semiconductor field effect transistor

NACK negative acknowledgement

NES noise equivalent signal

NIMS networked info-mechanical system

NLOS non-line-of-sight

OFDM orthogonal frequency division multiplexing

PAM pulse amplitude modulation

PAR peak-to-average ratio

PCI peripheral component interconnect

PCMCIA Personal Computer Memory Card International Association

PDA personal digital assistant

pdf probability density function

psd power spectral density

QAM quadrature amplitude modulation

RBS reference broadcast system

RFID radio frequency identification

RREP route reply

RSA Rivest, Shamir, and Adleman

SDMA space division multiple access

SIR signal-to-interference ratio

SISO single-input single-output

SMACS self-organizing medium access control for sensor networks

SNR signal-to-noise ratio

SQL structured query language

SRAM static random access memory

SSL secure socket layer

SSS strict-sense stationary

STEM system topology and energy management

TCP transport control protocol

TDMA time division multiple access

TDOA time difference of arrival

TH-SS time-hopped spread spectrum

TNEA thermal noise equivalent acceleration

TNERR thermal noise equivalent rotation ratio

UART universal asynchronous receiver/transmitter

UDP universal data protocol

USB universal serial bus

UTC coordinated universal time

VCC voltage-controlled clock

VCO voltage-controlled oscillator

VOR VHF omni-directional ranging

WSS wide-sense stationary

Embedded network systems

1.1 Introduction

Continuing advances in integrated circuit technology have enabled the integration ofcomputation and communication capabilities into devices which monitor or controlphysical processes Digital controllers and sensors are found in automobiles, homeappliances, factories, aircraft, cellular telephones, video games, and environmentalmonitoring systems Indeed, the vast majority of processors now being manufacturedare used in embedded applications (i.e., having connection to physical processes)rather than in what would ordinarily be thought of as a computer Many are net-worked within the confines of a local control system, typically in master/slave config-urations However, advances in wireless technology and in the understanding ofdistributed systems are now making possible far more elaborate compositions ofembedded systems that may function as the connection of the Internet to the physicalworld Embedded network systems (ENS) are poised to become pervasive in theenvironment with the potential for far-reaching societal changes that have hithertobeen the subject of science fiction It is the purpose of this book to lay out the founda-tions of this technology, the emerging design principles and applications, and some

of the interesting societal questions raised by ENS This chapter provides some examples

of ENS, discusses relevant technological trends, explores some implications of scaling

to very large numbers of ENS nodes, and places the technology in a historical context.Figure1.1is the block diagram of an ENS node Essential hardware componentsare some type of energy supply (e.g., battery or external connector), a sensor and/oractuator (e.g., microphone or speaker, camera or display), a processing unit (signalprocessing and storage capability), a communications device (e.g., radio or Ethernetport), and packaging Some of these may be integrated together, while others may bediscrete components The components are connected by data buses (dashed) andpower lines, with various devices to provide interfaces Also essential is the softwarethat enables the management of the platform resources, signal processing, and externalcommunications The basic characteristic of such devices is that it is their aggregationinto a network that provides the required functionality

1

ENS networks provide distributed network and Internet access to sensors, controls,and processors embedded in equipment, facilities, and the environment Integratedcircuit technology now enables the construction of sensors, radios, and processors atlow cost and with low power consumption, enabling mass production of sophisticatedbut compact systems that link the physical world to networks Scales range from local

to global, with applications including medicine, security, factory automation, onmental monitoring, and condition-based maintenance Compact size and low costallow ENS to be embedded and distributed at a small fraction of the cost of conven-tional wired sensor and actuator systems This can enable there to be hundreds orthousands of sensors per user, resulting in many new system design challenges.Centralized methods for sensor networking present heavy demands on cable instal-lation and network bandwidth However, by processing at source and conveyingdecisions rather than raw data, the burden on communication system components,networks, and human resources is drastically reduced This same observation holdstrue whenever there are relatively thin communications pipes between a source and theend network, or when dealing with large numbers of devices The physical world cangenerate an unlimited quantity of data to be observed, monitored, and controlled, butthere are finite resources that can be put into wireless telecommunications infrastruc-ture Moreover, the end user needs to be presented with some redacted form of thisstream or suffers from information overload as the number of sensors increases Thus,from the perspectives of both network resources and the finite capacity of the end user,scaling to large numbers of devices implies that sensor nodes will become increasinglyautonomous, doing a large fraction of the data processing and decision-making in situ

envir-In this overview, two scenarios are considered illustrating different aspects of thedesign tradeoffs In the first, an autonomous network of sensors is used to monitorevents in the physical world for the benefit of a remote user connected via the web Inthe second, a space is instrumented to enable interactions with people A generalarchitecture is depicted in Figure1.2, for now neglecting the details of how servicesare actually supported by Internet-connected devices To supply some service, twodifferent clusters of sensor nodes are connected through their respective gateways tothe Internet The nodes are assumed to be addressable either through an Internet

Energy

supply

Digital signal processor and memory

Interface

Sensor/

actuator suite

protocol (IP) address or some attribute (location, type, etc.), and are distinguishedfrom pure networking elements in that they contain some combination of sensors and/

or actuators That is, they interact with the physical world The gateway may itself be asensor node similar to other nodes in its cluster, or it may be entirely different,performing, e.g., extra signal processing and communications tasks and having nosensors In the cluster in the top left of Figure1.2, nodes are connected by a multihopnetwork, with redundant pathways to the gateway In the bottom cluster, nodes may

be connected to the gateway through multihop wireless networks or through othermeans such as a wired local area network (LAN) The nodes in the different clustersmay all be of one type, or they may be different within or across clusters In a remotemonitoring situation, there may be part of the target region with no infrastructure, andthus the multihop network must self-organize, while in other parts of the region theremay already be assets in place that are accessible through a pre-existing LAN There is

no requirement that these assets be either small or wired; the point is to make use of allavailable devices for providing the desired service

The remainder of this chapter describes some design heuristics, discusses a number

of different research efforts to deploy large networks in areas without infrastructuresupport, radio frequency identification (RFID) tag systems, some possibilities forenacted public spaces, and presents technological trends within a historical context

1.2 ENS design heuristics

A description of some of the fundamental physical constraints for sensing, detection,communication, and signal processing cost is provided in later chapters Implicit inany discussion of constraints is that an optimization problem emerges in which thequality of some basket of services is traded against the resource costs The basic designconstraints that emerge for ENS are:

(1) There are many situations in which reliable detection demands sensors in closeproximity to a physical event, causing numbers to scale (e.g., physical obstructions

to cameras) With large numbers of sensors, the type of information obtained isqualitatively different than that obtained with remote arrays.

(2) Sensors, radios, and signal processing will all ride the integrated circuit technologycurve down in cost, but batteries and other energy sources improve in cost onlyslowly with time

(3) Communications energy cost per bit is in many instances many orders of tude larger than the energy required for making decisions at source, and commu-nication is limited in its efficiency by fundamental limits, whereas the processingcost is to first order limited only by current technology

magni-(4) Human labor does not scale; networks must be self-organizing to be economical.(5) Scaling with physical responsiveness demands hierarchy, with distributed opera-tion at lower layers and increased centralized control at higher layers

Note that hierarchy does not necessarily imply heterogeneous devices If one considershuman organizations, the native processing abilities are roughly equal at all levels.What differs most in progressing up the chain is that different information is pro-cessed, i.e., information at different levels of abstraction/aggregation Moreover,commands progressing down the chain also differ in their level of abstraction, frompolicies down to work directives, with varying scope for interpretation This flexibilityenables lower levels to deal with local changes in the situation much faster than if acentral controller needed to be consulted for each action, while enabling global goals

to be pursued With machines, of course, there is the possibility of providing thedevices with highly differentiated abilities at different levels of the hierarchy Thesecan bring important advantages, e.g., a backbone long-range high-speed communica-tion pipe can greatly reduce latency compared to relying only on multi-hop links.Thus, while the logical rather than physical hierarchy is arguably much more impor-tant in enabling scalability, it behooves the designer of large-scale systems to considerboth Homogeneity is in any case impractical in long-lived systems composed ofintegrated circuit components; as for the Internet, the architecture must accommodatethe addition of successive generations of more powerful components

1.3 Remote monitoring

To make the discussion more concrete, consider an application requiring identification

of particular classes of signal sources passing through a remote region These sourcesmay be vehicles, species of animals, pollutants, seismic events on Mars, or, on a smallerscale, enzyme levels in the bloodstream or algal blooms in the ocean In any case, it isassumed that there is no local power grid or wired communications infrastructure, butlong-range communication means exist for getting information to and from a remoteuser Then, in laying out a network such as the one depicted in Figure1.2, both energyand communications bandwidth may be critical constraints As noted above, when thenetwork must scale in the number of elements, this effectively means that much of thesignal processing must be performed locally For example, in studying the behavior ofanimals in the wild, a dense network of acoustic sensors may be employed The nodescontain templates for the identification of the species emitting calls Nodes that make atentative identification can then alert their immediate neighbors so that the location of

the animal can be roughly determined by triangulation Infrared and seismic sensorsmay also be used in these initial identification and location processes Then finallyother nodes may be activated to take a picture of the source location so that a positiveidentification can be made This hierarchy of signal processing and communicationscan be orders of magnitude more efficient in terms of energy and bandwidth thansending images of the entire region to the gateway Further, the interaction of diversetypes of nodes can more simply lead to automation of most of the monitoring work,with humans only brought into the loop for the difficult final visual pattern recogni-tion on preselected images On positive identification, the audio and infrared filescorresponding to the image can be added to a database, which may subsequently bemined to produce better identification templates Note that the long-range commu-nication link (via the gateway) potentially enables the full use of web-accessibleutilities, so that the end-user need not be present in the remote location, and databases,computing resources, and the like may all be brought to bear on interpreting the(processed) data.

There is tension between experimental apparatus for initially exploring an tion domain and what will actually be needed for large-scale deployment Becausenetworked sensors have hitherto been very expensive, relatively few array data sets areavailable for most identification purposes, and sensors have typically been placedmuch further from potential targets than is possible with ENS This means paradox-ically that initially fairly powerful nodes need to be constructed to conduct large-scaleexperiments to collect raw data, so that suitable identification algorithms can bedeveloped using the resulting data set Likewise, in experimenting with differentnetworking algorithms, it is desirable from the point of view of software developmentinitially to provide a platform with considerable flexibility

applica-Example 1.1 Evolution of a habitat monitoring system

An overview of one set of sensor deployments in the James Reserve in the San Jacintomountains near Palm Springs CA is provided in Figure1.3 There are very large elevationchanges and consequently a broad set of species, some endangered, represented in a smallarea Early sensor deployments included cameras in bird nest boxes, and a moss camera thatrecords its growth and change in appearance with moisture A first phase of wireless sensornetwork deployment focused on measuring microclimates, with the results correlated withplant growth

However, it became apparent very quickly that logistical tasks such as battery replacementthat are easy with ten nodes become extremely tedious as numbers scale, given that sustaineddeployments are required Further, studies of plant growth and animal behavior couldbenefit immensely from having a broad set of sensors that span from roots to tree tops.Many of the interesting locations are heavily shaded, rendering solar power an unattractivepower source Having some automated infrastructure would be very valuable in such anapplication, both for moving sensors and supplying them with energy and communications

A system of robotic nodes that move along wires at treetop level was devised (see alsoChapters12and13) The nodes can lower a secondary platform to the forest floor to allowsensing in 3-space, and the repositioning or replacement of nodes, as depicted in the sche-matic in Figure1.4 Since these nodes have access to reliable energy supplies, higher endprocessors can be used The energy and communications resources also allow high-qualityimagers to be used, with elevation above the underbrush providing a large viewing volume

Figure 1.3 Early sensor deployments (Courtesy of Dr Michael Hamilton).

Figure 1.4 Interaction of robotic elements with fixed sensor nodes

Several such transects are planned to enable sustained study of factors thought to beimportant to the health of the ecosystem.

 Active tag: the tag has the ability to send RF signals

 Active tag with sensors: the tag is a sensor node with a unique identity

Active tags have been used for a long time for such purposes as tracking wild animals

to infer behaviors Passive tags are now used extensively in animal husbandry, and willincreasingly be applied to the industrial supply chain for tracking of items at variousstages of production and distribution Very small tags will be applied to individualconsumer items for such purposes as inventory control and automated check-out.Their small size demands that readers be in close proximity to retrieve data Active tagsmay be attached to shipping pallets or containers to enable longer-range reading, andmay include sensors to determine the conditions under which the items were shipped(temperature, vibration, tampering) In the limit, these may form, for example, amultihop sensor network among shipping containers

The complete system consists of a set of tags, one or more interrogators, and abackend data management system Even with purely passive tags this system is asensor network, with the interrogators exciting responses and receiving as informationthe identity of the tags in range Unlike a system of bar codes and readers, RFIDsystems provide the capability of identifying not only the category of an item but alsoits unique identity Verification of individual identity may enable financial transac-tions based upon that identity Further, a decreased sensitivity to orientation and theability to read tags at a distance and through obstructions enables automation of manyfunctions

Since the tag readers have a limited range, the act of reading also provides positioninformation on the tagged item A sequence of readings can therefore yield a history ofthe motion of a tagged item (or person) Thus while the tag might be designedprincipally to provide identification, there are secondary inferences that can be madefrom the data retrieved by the interrogator network The locations that a taggedindividual visits can, for example, be used to infer shopping behavior (e.g., whatdisplays within a store most captured attention) Moreover, the combination ofdetermining identity and position enables binding of information to a tagged object

if that object is observed by multiple sensors Thus, while it is a difficult signalprocessing task to determine whether some particular individual or object is withinthe field of view of a camera, once tagged the uncertainty can be eliminated (in effect,identity is broadcast) and the image added to some database concerning that indivi-dual or object

Clearly there are large privacy concerns associated with RFID tags and consumerproducts If these become embedded in shoes and clothing, and interrogators arewidely placed throughout urban environments, there could be a huge intrusion intothe privacy of the individual To deal with such concerns, a kill switch has beenproposed for such tags, so that at the point of sale the tags are deactivated Lawshave also been proposed to govern the uses that can be made of information collected

by RFID readers However, the debate on privacy versus economic efficiency is not yetsettled

Consider a robotic gaming example, in which the space physically reconfigures inresponse to the activities of robotic teams controlled by human players Instead ofseeing a purely virtual environment, players would ‘‘see’’ and ‘‘hear’’ through camerasand microphones on the robots within the game, with some robots controlled directlythrough joy sticks and others run by remote agents Players might bring up differentwindows on their control panels to get views from different robots as they attempt tocoordinate diverse elements Some game elements might be virtual (e.g., explosions)but would nonetheless produce physical response (e.g., loss of particular functions,crumbling of buildings each of whose structural elements have some active compo-nents) Robots could manipulate physical elements to construct barriers, structures, orminiature cities in which the game would be carried out Flying elements could beapproximated through a network of wires allowing nearly complete mobility in3-space

A larger-scale system could have human players in an instrumented space thatreacts to player actions The costumes and props carried by the players may includesensors, readers, and signal sources that interact with the displays and activatedsurfaces in the game zone, as well as with the devices on other players or roboticelements Many of the necessary elements have already been developed in othercontexts (e.g., laser tags, virtual reality games, museum displays that change based

on sound and images) Others, such as weaving arrays of electronic elements intofabrics, are still at the stage of interesting research ideas Issues of cost and how toauthor a system with so many sophisticated components in an intuitive fashion alsoremain challenging However, there is interest in creating artificial realms of this typefor training for hazardous jobs as well as for entertainment

a brief historical overview of the technology.

The western technological revolution has its roots in Classical Greek rationalismdivorced from its social pessimism, fused instead with a belief in progress in humanaffairs That is, Nature is governed by laws, these laws are best discovered throughexperimentation and the use of Reason, and understanding of these laws leads toprogress towards a great destiny When combined with the competitive short-termambitions of city-states and later nation-states this potent mix has helped to launch ascientific age that began in the Italian Renaissance and has since become the mostpowerful force for social and economic change in the world The direction of scienceand technology is shaped by the philosophical, economic, and political imperatives ofthe day, and in turn profoundly affects society Thus in embarking on the creation ofinformation technologies with a large potential for societal impact it is well to considerdesired and undesired outcomes sooner rather than later

The electronic digital age had its beginnings with telegraphy in the 1840s, withhuman beings acting as actuators, sensors, and processors The story of communica-tions has been one of gradual replacement of human functions with computationalelements and switches, accompanied, as usual with machine technology, by higherspeeds and lower costs A prime driving force is described by Moore’s Law: theprocessing capability of computing devices doubles roughly every 18 months This is

a consequence of economic forces rather than silicon technology, as it has held in someform from mechanical relay computers, through vacuum tubes, and finally to inte-grated circuits in silicon There is an apparently unquenchable market for comput-ation and an expectation that it will become less expensive with time, but investmentmust be recovered in one generation of devices before the new generation can belaunched Resources are marshaled to achieve the expectation, and thus Moore’sLaw continues This evolutionary series of generations produces revolutionary results:with 66 generations in a century, processing power increases by a factor of 266(nearly

1020)! A startling consequence is that more artificial computational elements will beproduced in the current year than in all previous human history Projecting forward,there appear to be no insurmountable technological barriers to continue this growthfor the next human generation, with many alternative technologies being exploredshould the limits of silicon be reached Thus, it will be possible to support comput-ations and systems of vastly increased complexity over time

Integrated circuit fabrication technology may be regarded as a means to veryreliably make small things in large numbers at low cost This enables the construction

of arrays of sensors and actuators in a technology known as MEMS: mechanical systems MEMS is completely compatible with computation and commu-nication in the same small module ENS on a chip is thus a present rather than a distantpossibility, with the question being not whether it can be done but what kinds of ENSchips make sense The feature sizes in integrated circuits are approaching those of cells

microelectro-so that it is not far-fetched to suppose that the next great technological revolution will

be the merging of biological systems and information technology; many universitieshave established bioengineering departments with exactly this expectation If thisnanotechnology can then ride the Moore’s Law curve, the consequences will be far-reaching in all aspects of human endeavor

Yet there exist fundamental limits in activities that impinge upon the physicalworld Energy technology is mature, with storage densities improving only slowlyover time The energy required to affect physical objects does not change at all over time,and there are other limits in the realms of communications and the ability to observe theenvironment accurately Of course, even technology that is evolving at a rapid rate willalso be limited at any given point in time Thus engineers and computer scientists mustmake design tradeoffs The remainder of this book addresses the underlying physicalfactors that affect the design of ENS, performance criteria and some algorithms whichattempt to meet them, applications, and social implications of the technology

It is one of several articles on this topic in a special issue

A comprehensive study on ENS commissioned by the National Research Council is

D Estrin, Embedded Everywhere: A Research Agenda for Networked Systems ofEmbedded Computers National Academy Press, 2001

It contains a survey of technology, potential applications, and future directions One

of the recommendations is a study of the ethical, legal, and social implications of ENS.Some of the sensor deployments at the James San Jacinto Mountains Reserve can

be monitored on-line at:

While some comments are now dated, it is a good counterpoint to time spent inmathematics, physics, and engineering courses, and a stepping-off point for deeperinquiries into particular topics or periods It is striking how many seemingly modernproblems have historical analogs that have been dealt with in very sophisticatedways.

Representation of signals

Source detection, localization, and identification begin with the realization that thesources are by their nature probabilistic The source, the coupling of source tomedium, the medium itself, and the noise processes are each variable to some degree.Indeed, it is this very randomness that presents the fundamental barrier to rapid andaccurate estimation of signals Consequently, applied probability is essential to thestudy of communications and other detection and estimation problems This chapter isconcerned with the description of signals as random processes, how signals aretransformed from continuous (real-world) signals into discrete representations, andthe fundamental limits on the loss of information that result from such transforma-tions Three broad topics will be touched upon: basic probability theory, representa-tion of stochastic processes, and information theory

2.1 Probability

Discrete random variables

A discrete random variable (RV) X takes on values in a finite set X¼ {x1, x2, xm}.The probability of any instance x being xi, written P(x¼ xi), is pi Any probabilitydistribution must satisfy the following axioms:

(1) P(xi)
0

(2) The probability of an event which is certain is 1

(3) If xiand xjare mutually exclusive events, then P(xiþ xj)¼ P(xi)þ P(xj).That is, probabilities are always positive, the maximum probability is 1, and prob-abilities are additive if one event excludes the occurrence of another Throughout thebook, a random variable is denoted by a capital letter, while any given sample of thedistribution is denoted by a lower-case letter

Example 2.1 Binomial distribution

An experiment is conducted in which an outcome of 1 has probability p and the outcome of a zerohas probability (1 p) In Bernoulli trials, each outcome is independent, i.e., it does not depend

on the results of the prior trials What is the probability that exactly k 1s are observed in n trials?

12

Consider the number of ways there can be k 1s in a vector of length n, and the probability ofeach instance Each 1 has probability p, and each 0 has probability (1 p), so that each vectorwith k 1s has probability pk(1 p)nk The number of possible combinations of k 1s isn

to the tails (low-probability regions) due to an insufficient density of samples The trials thatoccur within the standard deviation represent roughly 2/3 of the possible outcomes

Note that if the trials are not independent, then little credibility can be attached to thisuncertainty measure

Consider now two RVs X and Y The joint probability distribution P(X, Y) is theprobability that both x and y occur In the event that outcomes x do not depend in anyway on outcomes y (and vice versa), then X and Y are said to be statistically indepen-dent and thus P(X, Y)¼ P(X)P(Y) For example, let x be the outcome of a fair coin tossand y the outcome of a second fair coin toss The probability of a head or tail on eithertoss is 1/2, independently of what will happen in the future or what has happened in thepast Thus the probability of each of the four possible outcomes of (head, head), (head,tail), (tail, head), (tail, tail) is 1/4

The conditional distribution P(X|Y) is the probability distribution for X given that

Y¼ y has occurred When X and Y are statistically independent, this is simply P(X).More generally,

Example 2.2 Campaign contributions and votes in Dystopia

In the corrupt political system of the town of Dystopia, campaign contributions Y influencevotes on legislation X Let the two outcomes of X be {yea,nay} and let the sum of campaigncontributions by persons favorable to the legislation less the sum of contributions by personsunfavorable to the legislation be drawn from the set {100, 100} These outcomes are equallylikely as are the voting outcomes P(yea| 100)¼ 1, P(nay|  100) ¼ 1 Determine P(X, Y) andP(Y|X)

Solution

Clearly P(yea| 100) ¼ P(nay|100) ¼ 0 Thus, P(yea, 100) ¼ P(yea|100)P(100) ¼ ½, andsimilarly, P(nay| 100) ¼ ½, P(yea|  100) ¼ P(nay|100) ¼ 0 Checking, the sum of all thepossibilities is 1 Working from this, P(Y|X)¼ P(X, Y)/P(X), i.e., P(100|yea) ¼ 0.5/0.5 ¼ 1 ¼P(100|nay); P(100|nay) ¼ P(100|yea) ¼ 0

Notice that while the causal relationship is that campaign contributions lead to a cular voting outcome, P(Y|X) still has meaning as an inference from the observations It isoften beneficial to use a form of this relationship known as Bayes’ Rule to rearrange problems

parti-in terms of probability distributions that are known or easily observed:

PðX j YÞ ¼ PðY j XÞPðXÞ=PðYÞ:

In particular this relation is invoked in the derivation of the optimal detectors for signals innoise

The expected value or mean of a random variable X is denoted by
¼ E[X ], given by

a Gaussian distribution The ubiquity of the Gaussian distribution is a consequence ofthe Central Limit Theorem:

For n independent random variables Xi, the RV X¼ 1=nXn

i¼1Xi tends to the Gaussiandistribution for n sufficiently large That is,

fðxÞ  1

 ffiffiffiffiffiffi

2p

p eðx
Þ2=22; (2:5)where
is the mean of the distribution and  the standard deviation

Associated with the Gaussian distribution is a function which relates to the areaunder its tails The Gaussian Q function is defined as

Details of the Q function are given in AppendixA

Example 2.3 Gaussian tails

The tail of a distribution is the region of low probability For a Gaussian, the tails arebounded by exponentials Using the approximation to the Q function, compute the prob-ability that y is between 2 and 2.1, where f(y) is a N(0, 0.25) distribution, i.e., a normaldistribution with zero mean and a variance of 0.25 (standard deviation=0.5)

Solution

Since the Q function is defined for a unit standard deviation normal distribution, onemust perform a change of variables, scaling by the ratio of the standard deviations (andadjusting for the non-zero mean as might be required in some cases) Let x¼ y/ ¼ 2y ThenP(2 < y < 2.1)¼ Q(4)  Q(4.2), which using the approximation is 3:35 105 1:44 105¼ 1:91  105

The Gaussian distribution may be used to construct a number of other importantdistributions for signal detection and communication Consider two Gaussian RVs X1

and X2 Suppose they represent the distributions for the real and imaginary parts of acomplex random variable Y (e.g., a signal transmitted over some channel) That is,

Y¼ X1þ jX2 Let the envelope of Y be denoted by Z¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

X2

1þ X2 2

The lognormal distribution results from the application of the Central Limit Theorem

to a multiplicative process, i.e., to a product of independent RVs Let f(x) be N(0, 2),where all quantities are expressed in decibels (dB) Then the lognormal distribution iswhat results after converting back to absolute units, i.e., Y¼ 10X/10

.Example 2.4 The Poisson distribution

Suppose the probability of an event happening in a short time interval
t is l
t, with theprobability of two or more events occurring in this short interval being very low This eventmight be the beginning of a telephone call or the reception of a photon in a detector Suchevents are usually termed arrivals The probability of no arrivals in this interval is then

1 l
t Suppose further that the probability of an arrival in any given interval is dent of the probability of an arrival in any prior interval Determine the probability of karrivals in a longer interval T

2.2 Stochastic processes

Frequency domain

Signals of interest in communication or detection problems are frequently ized in multiple domains, according to a representation that makes a particularproblem easier to solve Each of these domains provides a complete representation

character-of the signal, unless the representation is deliberately truncated for reasons character-of reducedcomplexity implementation of a system These two domains are related by orthogonaltransformations The classic example is the representation of signals passing throughlinear systems in both the time and frequency domains The Fourier transform X(f) ofsignal x(t) is the frequency domain representation, given by

(1) Linearity: if x1(t)$ X1(f ) and x2(t)$ X2(f ) then c1x1(t)þ c2x2(t)$ c1X1(f )þ c2X2(f ).(2) Time scaling:

xðatÞ $ 1

jajX

fa

 :

Example 2.5 Linear system

A signal x(t) passes through a linear filter with impulse response h(t) to produce the outputy(t) Then y(t) is given by the convolution integralR1

1xðtÞhðt  Þd Suppose h(t) ents the low-pass filter Asinc(2Wt) and x(t) is the sinusoid cos(2pfct), where the carrierfrequency is less than W Compute y(t)

repres-Solution

This integral is rather difficult to evaluate directly Instead, one may use the Fourier form pairs Asincð2WtÞ $ A=2Wð Þrect f=2Wð Þ and cosð2pfctÞ $1½ðf  fcÞ þ ðf þ fcÞtogether with property (10) above to obtain Y(f)¼ A/4W[(f  fc)þ (f þ fc)] Taking theinverse transform, y(t)¼ A/4Wcos(2pfct) Avoidance of difficult integrals in dealing withlinear systems is a major reason for use of the frequency domain Another is that many ofthese problems are more easily visualized in the frequency domain

trans-A question sometimes asked is what is the meaning of the frequency domain It isperhaps best visualized from the point of view of a Fourier series, whereby functionsare composed of a weighted sum of harmonic components The time and frequencydomains are merely abstract models of the behavior of systems, and since these canpotentially each be a complete representation they have equal value The frequencydomain is convenient in dealing with some aspects of linear systems (e.g., filters) whilethe time domain is useful for other aspects (e.g., description of pulses) The problemsolver needs facility in multiple abstract models, and indeed many orthogonal trans-formations have been developed for various signal processing purposes

As an example, the energy of a pulse can be computed in either time or frequencyusing the Parseval relation:

The chances are that the integration is much less difficult in one domain than in the other

Representation of stochastic processes

Thus equipped, the representation of random processes can now be addressed Let

X be a continuous RV with pdf f(x) A random or stochastic process consists of