[ieee 2007 4th annual ieee communications society conference on sensor, mesh and ad hoc...

7
SAFIRE: A Self-Organizing Architecture for Information Exchange between First Responders Nabeel Ahmed, Kamran Jamshaid, and Omar Zia Khan University of Waterloo, Waterloo, Ontario N2L 3G1 Email: {n3ahmed, kjamshai, ozkhan} @uwaterloo.ca Abstract- Disaster response requires quick and timely mobi- lization of relief efforts to save lives and property. Fundamental to these efforts is a reliable communications infrastructure that allows the disaster response teams to coordinate and exchange information in an efficient manner. Existing solutions for disaster response are inadequate as they suffer from interoperability problems and lack the appropriate amount of flexibility. In this paper, we propose SAFIRE, a novel multi-hop architecture for facilitating fast and reliable information exchange between first responders. The salient features of SAFIRE are (1) A decentralized cognitive radio-based approach for supporting direct communication between first responders, (2) A publish- subscribe mechanism for exchanging information among first responders, and (3) A flexible multi-layered policy framework for optimally configuring the system. We present the challenges in designing SAFIRE, and outline its basic components. We believe our exploration of such an architecture opens up a set of unique challenges related to the integration of different systems to realize SAFIRE, giving rise to new avenues for research in communication systems for disaster response. I. INTRODUCTION In a typical disaster scenario first responders, emergency service teams, and global relief agencies (hereafter collectively referred to as first responders) quickly mobilize to aid those affected by the disaster. The disaster zone can be large and geographically spread apart, as seen in the Indian Ocean Tsunami of 2004 and the South Asian Earthquake of 2005. Unfortunately, relief operations are often hampered due to communication and coordination issues between first respon- ders. These teams often have to act on partial information collected in some ad hoc and unreliable manner (such as through news agencies or word of mouth). Also, any local information collected by a team may not be available to other teams in adjoining areas. First responders often have to make crucial decisions based on the collected information. Failure to communicate accurate and timely information can cost lives - of both first responders as well as those they are trying to assist. Under these circum- stances, first responders can benefit from a reliable commu- nications infrastructure that provides the requisite information in a timely manner. Examples of data that may need to be shared include statistics related to human loss or property damage, outbreak of a disease, topographical information of the disaster site (e.g via images/video), rising water levels or other weather-related information. Existing communications systems fall short of providing the necessary support for these situations. Large disasters destroy Uninterrupted remote connectivity ~~A-~~ ~~ __ f ~ ~ ~ ~ ~ ~ ~ ~ ~ XP~ Loral C an nterbl H ,~~Mobile Hospitd I Main Commanld Center ,#- k <) I_ __ k & Intermittent connectivity between nodes on disaster site Fig. 1. Localized communication between teams on the ground requires fine-grained crisis information. Main command center, on the other hand, is interested in summarized information on relief operations. core terrestrial communications infrastructure, and backup networks are often unable to handle the necessary traffic volumes. Teams of first responders may need to communicate with either the central command center, or with other teams on the disaster site. In the first scenario, the long-distance com- munication link with the central command center is typically established through VSAT terminals that are mounted at se- lected mobile command centers. In our work, we focus on the second communication scenario. Specifically, we are interested in situations where first responders collaborate with each other to share information that is useful to other responders also operating on the ground or in the air (shown in Figure 1). This localized communication paradigm is inherently different from the wide-area communication between the disaster site and the main command center. While the main command center is mostly interested in overall statistics pertaining to relief operations, local networks must exchange more fine-grained information on shorter timescales for accurately coordinating relief efforts. Realizing a disaster response system that encapsulates the information sharing capabilities outlined above is non-trivial. Existing work has attempted to address this problem by a variety of techniques. However, a number of fundamental problems still need to be solved to realize a practical system for disaster response. The following are some of the key observations that motivate our work. Network Interoperability Limitations: Interoperability of network devices belonging to first responders is a key require- ment for coordinating relief efforts. It is a well-known fact that emergency response teams, even within the same municipality, have incompatible radio systems, prohibiting coordination of 1-4244-1268-4/07/$25.00 C)2007 IEEE 655

Upload: omar-zia

Post on 14-Mar-2017

212 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: [IEEE 2007 4th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks - (2007.06.18-2007.06.21)] 2007 4th Annual IEEE Communications Society

SAFIRE: A Self-Organizing Architecture for

Information Exchange between First RespondersNabeel Ahmed, Kamran Jamshaid, and Omar Zia KhanUniversity of Waterloo, Waterloo, Ontario N2L 3G1Email: {n3ahmed, kjamshai, ozkhan} @uwaterloo.ca

Abstract- Disaster response requires quick and timely mobi-lization of relief efforts to save lives and property. Fundamentalto these efforts is a reliable communications infrastructure thatallows the disaster response teams to coordinate and exchangeinformation in an efficient manner. Existing solutions for disasterresponse are inadequate as they suffer from interoperabilityproblems and lack the appropriate amount of flexibility. Inthis paper, we propose SAFIRE, a novel multi-hop architecturefor facilitating fast and reliable information exchange betweenfirst responders. The salient features of SAFIRE are (1) Adecentralized cognitive radio-based approach for supportingdirect communication between first responders, (2) A publish-subscribe mechanism for exchanging information among firstresponders, and (3) A flexible multi-layered policy frameworkfor optimally configuring the system. We present the challengesin designing SAFIRE, and outline its basic components. Webelieve our exploration of such an architecture opens up a set ofunique challenges related to the integration of different systemsto realize SAFIRE, giving rise to new avenues for research incommunication systems for disaster response.

I. INTRODUCTION

In a typical disaster scenario first responders, emergencyservice teams, and global relief agencies (hereafter collectivelyreferred to as first responders) quickly mobilize to aid thoseaffected by the disaster. The disaster zone can be large andgeographically spread apart, as seen in the Indian OceanTsunami of 2004 and the South Asian Earthquake of 2005.Unfortunately, relief operations are often hampered due tocommunication and coordination issues between first respon-ders. These teams often have to act on partial informationcollected in some ad hoc and unreliable manner (such asthrough news agencies or word of mouth). Also, any localinformation collected by a team may not be available to otherteams in adjoining areas.

First responders often have to make crucial decisions basedon the collected information. Failure to communicate accurateand timely information can cost lives - of both first respondersas well as those they are trying to assist. Under these circum-stances, first responders can benefit from a reliable commu-nications infrastructure that provides the requisite informationin a timely manner. Examples of data that may need to beshared include statistics related to human loss or propertydamage, outbreak of a disease, topographical information ofthe disaster site (e.g via images/video), rising water levels orother weather-related information.

Existing communications systems fall short of providing thenecessary support for these situations. Large disasters destroy

Uninterrupted remote connectivity

~~A-~~~~ __f ~ ~~ ~ ~ ~ ~ ~ ~

XP~ Loral C an nterbl H,~~Mobile Hospitd

I

Main Commanld Center

,#-k <)I___k&

Intermittent connectivity between nodes on disaster site

Fig. 1. Localized communication between teams on the ground requires fine-grainedcrisis information. Main command center, on the other hand, is interested in summarizedinformation on relief operations.

core terrestrial communications infrastructure, and backupnetworks are often unable to handle the necessary trafficvolumes. Teams of first responders may need to communicatewith either the central command center, or with other teams onthe disaster site. In the first scenario, the long-distance com-munication link with the central command center is typicallyestablished through VSAT terminals that are mounted at se-lected mobile command centers. In our work, we focus on thesecond communication scenario. Specifically, we are interestedin situations where first responders collaborate with each otherto share information that is useful to other responders alsooperating on the ground or in the air (shown in Figure 1). Thislocalized communication paradigm is inherently different fromthe wide-area communication between the disaster site andthe main command center. While the main command centeris mostly interested in overall statistics pertaining to reliefoperations, local networks must exchange more fine-grainedinformation on shorter timescales for accurately coordinatingrelief efforts.

Realizing a disaster response system that encapsulates theinformation sharing capabilities outlined above is non-trivial.Existing work has attempted to address this problem by avariety of techniques. However, a number of fundamentalproblems still need to be solved to realize a practical systemfor disaster response. The following are some of the keyobservations that motivate our work.Network Interoperability Limitations: Interoperability of

network devices belonging to first responders is a key require-ment for coordinating relief efforts. It is a well-known fact thatemergency response teams, even within the same municipality,have incompatible radio systems, prohibiting coordination of

1-4244-1268-4/07/$25.00 C)2007 IEEE

655

Page 2: [IEEE 2007 4th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks - (2007.06.18-2007.06.21)] 2007 4th Annual IEEE Communications Society

relief work [19]. A poignant reminder of this fact is the humanloss that was associated with the 9-11 disaster. When policeofficials realized that the south tower of the World Trade Cen-ter was about to collapse, they immediately issued evacuationorders to their personnel. Unfortunately, this information wasnot available to fire officials at the scene, resulting in the tragicloss of many of their lives.

Global harmonization of standards and spectrum for publicsafety is difficult. While the U.S. and other developed countriesare planning to allocate chunks of the soon-to-be-vacant analogTV bands for public safety purposes [17], these bands willcontinue to be used in many developing countries in theforeseeable future.

Bootstrapping and Congestion Issues: While use of publicsafety bands may address interoperability problems for regionsunder the jurisdiction of a single regulatory body, they maystill suffer from network deployment and performance issues.Traditional radio systems require setting up and configuringthe system, which can incur substantial overhead. Moreover,we can also run into network congestion problems due tolarge network traffic volumes. In its report, the 9-11 Commis-sion blamed communication failures partly on simultaneousattempts to use a common tactical radio channel by thousandsof first responders operating at the disaster site [2].

Intermittently Connected Networks: In a large disaster,relief operations would likely occur in tiny pockets that aregeographically far. In such scenarios, information sharingbetween relief workers can be facilitated through the use ofinformation relays (e.g., rescue helicopters), that pick up anddeliver information between relief sites. Although this is akinto an ad hoc network, ad hoc networks typically assume end-to-end connectivity (direct or multi-hop) between communi-cating entities. This may not always be possible when thedisaster site is large and nodes are continuously mobile. Suchintermittent connectivity places additional constraints on thenetwork architecture, requiring disconnected nodes to bufferdata and opportunistically discover transmission opportunities,as and when they become available.More recently, there has been a surge of interest in de-

veloping software reconfigurable radios (also called software-defined radio, or SDR) that make a radio interface interoper-able across different standards and network types. Cognitiveradios [14] are SDRs that dynamically re-tune their radio pa-rameters based on feedback they obtain from their surroundingenvironment. This functionality can allow the cognitive radioto discover other transmitting nodes in its vicinity, and to self-organize into an information-sharing network, subject to thepolicy specifications outlined by its users.

Based on these observations, SAFIRE makes the followingcontributions:

A decentralized cognitive radio based approach thatsupports interoperability, improved performance, and en-hanced usability for first responders. It allows dynamicdiscovery and formation of networks, with communica-tion characteristics that may be altered on-the-fly to meetthe constraints of a given application. It also reduces setupand configuration times, thus allowing first responders toconcentrate on their core tasks.

. A publish-subscribe system that sits atop a disruption-tolerant networking substrate. This allows SAFIRE tosupport publish-subscribe functionality even in environ-ments where end-to-end connectivity is not always avail-able.

. A policy framework feeding multiple layers of the pro-tocol stack, to allow optimal configuration of the in-formation sharing overlay. We propose cross-layeringtechniques to allow a bottom-up approach to networkconfiguration.

It turns out that the complex inter-coupling between theseSAFIRE components brings forth a set of unique challengesrelated to building such a system for disaster response. Wepresent details of these components and list challenges as-sociated with them in this paper. The rest of the paper isorganized as follows. Section II discusses the requirementsand challenges that need to be addressed in realizing such adisaster response system. Sections III and IV provide an archi-tecture overview and some details on the policy specificationmechanism of SAFIRE. Section V discusses related work andSection VI ends with a conclusion and some directions forfuture work.

II. CHALLENGES

In this section, we discuss the requirements and challengesfaced in establishing an information sharing system for disasterresponse. Interoperability is one key problem that can beaddressed through the use of cognitive radios. We list furtherrequirements and challenges below.

. Rapid deployment The communications infrastructureneeds to be rapidly deployable. As such, it should be self-organizing, and discover other nodes and communicationopportunities without requiring manual intervention. Thisrequires the capability to correctly infer the current con-text, determine the best sequence of actions, and thenreact accordingly.

. Adaptable. Network and communication characteristicscan vary considerably over the course of the disasterrelief operation, and thus cannot be pre-determined. Thisincludes characteristics such as the network topology,type and size of exchanged traffic, as well as changingapplication service requirements, e.g., the type of datacollected can range from aggregate statistics on casualtynumbers, to images/video content of a specific site.This necessitates that the communication infrastructurebe flexible enough to accommodate these changing re-quirements without sacrificing system performance orhindering other operational aspects of the network.

. Resilient and Robust. The communication infrastructureneeds to be resilient and robust. This is challengingas nodes can be highly mobile, resulting in frequentdisconnections and link failures. As previously described,an end-to-end connection may never exist between pairsof nodes wishing to share information. This imposeschallenges on both the communication protocols as wellas the communications infrastructure. The communica-tion protocols need to be designed such that they are

656

Page 3: [IEEE 2007 4th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks - (2007.06.18-2007.06.21)] 2007 4th Annual IEEE Communications Society

disruption-tolerant. Similarly, the communications infras-tructure needs to be modified to enable a store-and-forward approach, whenever disconnections are encoun-tered.Incrementally Deployable. Any feasible replacement forthe current generation of disaster response communica-tion systems must be incrementally deployable, so as notto require a complete overhauling of the existing infras-tructure. It must also be backwards compatible, wherenew users can benefit from using the information sharingnetwork, while also having the ability to communicatewith legacy radio systems.

* Power conservation. End user devices must be small andportable, limiting their battery size. Power conservationis an important issue as teams of first responders mayremain deployed at forward bases for days, or evenweeks.

. Multi-layered Policies: Existing policy specificationframeworks for cognitive radios only account for physicallayer information to tune the radio. Their goal is tooptimize use of the RF spectrum. For disaster response,we advocate that these radios create topologies that areconducive to optimizing the operation of the information-sharing overlay. This necessitates policies that make useof higher layer information in order to configure theunderlying radio.

. Security and privacy requirements. Depending on ap-plication requirements, network security and data privacyrequirements can vary significantly as well. e.g., whenthe network is used primarily for facilitating exchange ofstatistical information, data confidentiality (or anonymity)might not be particularly required, though maintainingdata integrity might still be important. However, if thenetwork is used to transmit medical files for patientsundergoing treatment, then user authentication and end-to-end data confidentiality may be needed, as per govern-mental regulations (e.g., HIPAA requirements for securityand privacy of health data in the U.S. [1]). Balancingthese security issues against the requirements for timelyand efficient exchange of information in life-and-deathsituations involves both policy and technical issues thatneed to be understood.

III. DESIGN AND ARCHITECTURE

We now provide an overview of the SAFIRE architecture.SAFIRE is completely decentralized, with all functionalityresiding at the end nodes. A component-based layout of aSAFIRE node is illustrated in Figure 2.

The four core components comprising this architectureare the Pub-sub module, the Routing/Forwarding engine, theRadio module, and the Policy module. We discuss each ofthem in turn.

A. Pub-Sub ModulePublish-subscribe systems are a natural fit for our data-

centric information sharing network. Data is published by pub-lishers without specifying any receivers. Subscribers register

Application Layer

Transport Layer

Network Layer

MAC

(a) Standard Network Stack

Ontology Engine

Pub-Sub Engine I l

Application Overlay

Pub-Sub Module

Data-centric Routing J

Topo Gen Storage

Routing/Forwarding Module

+~~ ............... U.

- -I-

Sen ing ~!RFTuning

RF Front End

Radio Module

(b) SAFIRE Architecture

Fig. 2. (a) Illustrates the standard network stack, while (b) illustrates how componentsof the SAFIRE architecture map to this stack

their interests in different types of content, without explicitlylisting the source of the data. Updates of published contentare propagated to the respective subscribers, thus decouplingthe publishing and information delivery process.

Given the lack of any single central entity in our system, weadvocate a decentralized implementation of publish-subscribefunctionality in SAFIRE. This design choice naturally leadsto the publish-subscribe system being implemented atop anapplication-based overlay, such as a DHT [5] or CAN [18].The overlay is responsible for establishing communicationlinks between participating nodes, which may even result inmulti-hop routes. Due to this overlay functionality, publishersand subscribers can operate without explicitly requiring anyknowledge of the underlying network topology.

It may be noted that publish-subscribe systems allow up-dates to be disseminated using either a push or pull approach[9]. These approaches have different merits based on thecurrent state of the network and subscribers. SAFIRE does notrestrict usage to any particular method of information access,and can be configured to support both push- and pull-basedmechanisms.

While the pub-sub provides an effective mechanism todisseminate information, it does not solve naming issues, inthe event that different teams use different naming schemes.We propose the use of ontologies to mitigate this problem.Ontologies provide a way to express different concepts thatare represented in a standard yet decentralized manner. Ifthe participating teams can agree on a basic ontology, fur-ther extensions can be made by individual teams to supportadditional domain-specific information. Moreover, translationsbetween ontologies can also be made to map between differentconcepts. The working of the pub-sub module is depicted inFigure 3. We reuse recent work done on combining ontologieswith pub-sub systems so that information is not only properlystructured, but also effectively distributed to interested entities[13].

It should be noted that all other components in SAFIRE areconfigured to optimize the operation of the pub-sub module.

657

-0

}'l<EZ

Page 4: [IEEE 2007 4th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks - (2007.06.18-2007.06.21)] 2007 4th Annual IEEE Communications Society

Responderl

OntologyA

PublshI

ubcibe \

l Overlay

Responder2

OntologyA

Publish a--

__Su .bscribe

Overlay

Responder3

OntologyB:Map(B-A)

\ \* ~~~~~~Publish

\ Eu~~~~~~~~bscribe

~~~~~Overlay

Fig. 3. Responderl and Responder2 subscribe to data from Responder3. Responder3subscribes to data published by Responderl. Responderl and Responder2 share the sameontology (i.e., OntologyA), whereas Responder3 uses a translation from OntologyB toOntologyA and vice versa. Each pub-sub module sits atop the forwarding/routing, andradio modules as shown in Figure 2.

This necessitates the need for cross-layer interactions wherethe pub-sub module exposes statistics (such as publishingfrequency and importance of published information) to lowerlayers of the protocol stack. These layers then configurethemselves based on the provided information and periodicallypropagate their state up the stack to the pub-sub module, foroverlay optimization.

B. Routing/Forwarding Engine

There are three main functions of the routing/forwardingengine. First, it attempts to learn the topology of the SAFIREnetwork. Second, it uses the learned topology to decide thebest way to route packets within the network. Third, if an end-to-end connection does not exist, it stores packets so that theymay be forwarded later when a connection opportunisticallybecomes available. We describe each of these functions below.

Topology information changes over time due to node mo-bility, failure, or reconfiguration of radio parameters. Nodesemploy a link-state routing protocol to determine the currentnetwork topology, as discussed in [12]. This information isthen made available to the routing/forwarding engine (forrouting purposes), as well as the application overlay (to reor-ganize the overlay based on link availabilities in the underlyingnetwork).

The routing/forwarding engine considers topology informa-tion as well as packet priorities in the routing process. Packetprioritization is useful because situations may occur whereonly brief windows of opportunistic connections exist betweennetworks. In disaster situations, certain types of content maybe more crucial than others. e.g., an emergency alert versussimply reporting casualty statistics. Defining priority classesallows the system to push out higher priority data first. Duringrouting, nodes obtain data prioritization primitives as well asthe routing protocol to use from the policy module.Due to packet prioritization, the routing/forwarding engine

would buffer packets with lower priority that could not betransmitted earlier. As soon as another opportunistic connec-tion becomes available, these buffered packets would then besent out. DTNs [10] support such functionality so we employ

them for our purpose. It may be noted that the pub-sub modulefunctions without any knowledge of this routing behavior.

C. Radio ModuleThe radio module is responsible for managing the configura-

tion of the cognitive radio. There are two essential componentsof this module, a sensing component and a tuning component.The sensing component performs sensing of the RF spec-

trum to extract usage information on different channels. Spec-trum sensing is challenging for two reasons. First, the delayinvolved in periodically scanning different channels needs tobe minimized. Second, sensed information might itself beinaccurate, due to short-term variations of the wireless channel(e.g. channels undergoing a deep fade). Prior work suggeststhat coordination among radios can improve sensing accuracyby up to an order of magnitude [4].

Sensed information is then passed to the tuning componentthat subsequently tunes the radio. Traditional radio tuningpolicies only capture channel specific information such aschannel utilization and channel quality. We propose extendingthe policy framework to include policies that specify higher-layer constraints (through the policy module). Specifyingpolicies in this way makes radio tuning a multi-layered, multi-objective optimization problem. In Section IV, we illustrate anexample radio configuration algorithm that uses a given set ofconstraints described by higher layers of the protocol stack.

D. Policy ModuleThe policy module establishes the necessary ground rules

that impact the operating characteristics of all other modules.These policies are a comprehensive set of scenario-based rules,e.g., policies related to publishing and subscribing to differenttypes of data determine how the pub-sub overlay handlesrequests for shared information (Section III-A). Similarly,policies for the routing/forwarding engine correlate acceptabledelay bounds for different packet priorities with the routingprotocols to be used (Section III-B). Finally, the policy moduledescribes how the radio module should tune radio parametersbased on input from the pub-sub system. Next, we discuss anexample instantiation of policies for tuning the radio.

IV. RADIO TUNING ALGORITHM

We now discuss an example radio tuning algorithm for theSAFIRE architecture. Instead of espousing an optimal tuningalgorithm, the goal is to illustrate how multi-objective policiescan be incorporated into the radio parameter tuning process.The algorithm we describe incorporates policy information

from multiple layers of the protocol stack. The policies dic-tated by the application overlay optimize radio configurationbased on the requirements of the pub-sub system (e.g., datapriority that may bound the end-to-end delay experiencedon different communication paths). The policies dictated bythe network layer incorporate information pertaining to theunderlying multi-hop communication network (e.g., requiredper-node throughput). Channel policies take wireless channelcharacteristics into account in the radio parameter tuningprocess (e.g. channel utilization, channel quality, etc).

658

Page 5: [IEEE 2007 4th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks - (2007.06.18-2007.06.21)] 2007 4th Annual IEEE Communications Society

A threshold

2 3

Fig. 4. Example of spectral utilization in the 0-4 Ghz range

In the sample pseudocode shown in Algorithm 1, the goalis to ensure that node throughput requirements are met withminimal expended energy. Additional constraints specifiedare the delay requirements, which are in part determined bythe maximum hop-count between source/destination nodes,as well as acceptable channel noise levels. The tunable pa-rameters we consider are the radio's transmit power and itsfrequency/channel of use. The algorithm assumes that bothtransmit power and radio channels are discretized.

The sensing unit in the node first generates a Power SpectralDensity (PSD) map of the radio spectrum, an example ofwhich is shown in Figure 4. The node only considers fre-quencies/channels having an average noise level less than aspecified threshold A (lines 10-14). These acceptable channelsconstitute elements of set A. The node then finds the channelwith the lowest frequency in set A, as this is the channelwhich provides the best RF propagation characteristics, andcan therefore give the node the lowest hop count distanceamong all available channels in set A (Min function on line15). Starting with the minimum defined power level, the nodethen attempts to find the first power level that meets thehop count (or delay) requirements (ComputeHopDistancefunction on line 18). If such a power level exists (determinedin line 20), the node determines the available capacity ofthe channel (ComputeBW function on line 21). Note, thecapacity of a channel, as determined by Shannon's CapacityLimit theorem, is a function of the transmit power used andthe width of the channel employed during transmission. Ifchannel capacity requirements are met (based on the node'sthroughput requirement), the node assigns itself the giventransmit power level and the algorithm terminates. Otherwise,the node increases its transmit power by a step (line 17), andrepeats the process.

This example algorithm illustrates one way of realizing theconfiguration of a cognitive radio in the SAFIRE architecture.More complex instantiations that take into account additionalpolicy constraints can also be specified and are an interestingsubject for future work.

V. RELATED WORK

The need for integrated and interoperable systems for co-ordination and information sharing between first responders

Algorithm 1 Multi-Objective Policy Implementation1: ( = ordered array of power levels2: rj = set of PSD values across RF spectrum3: A = noise threshold on a channel4: 3 = throughput requirement for node being configured5: a = hop threshold6: A= candidate channels7: C = channel allocated to node8: p = transmit power of node9: Populate rj with collected PSD valuesio: for i= 1... Irj do11: if rji < A then12: A = A U Tli /* Channels with acceptable noise levels */13: end if14: end for15: C = min(A) /* Find the lowest frequency in set A *1

16: /* Start at lowest power */

17: for i= 1... 1 do18: T= ComputeHopDistance(C, (i)19: /* Find channel meeting hop threshold requirement *1

20: if T <= a then21: if ComputeBW(C, (i) >= 3 then22: p = (i /I Set transmit power of radio *1

23: Terminate Algorithm.24: end if25: end if26: end for

is obvious. Various governments are taking measures to thiseffect. In the U.S., the Office of Interoperability and Com-patibility is overseeing such efforts across local, state, andfederal agencies. The European Union expects to have similarsystems in place by the year 2010 [8]. While this may solveinteroperability issues, scalability concerns still need to beaddressed. Public safety bands are still susceptible to networkcongestion collapse due to communication overload in theaftermath of a disaster.Most of the current research literature for disaster man-

agement focuses on establishing some form of multi-hop adhoc communication network. Interoperability is not explicitlyaddressed, though the use of standards-based IEEE 802.11radios tries to obviate this problem. Each system only ad-dresses a point-problem for disaster management (e.g., [6]),and it is unclear if all the proposed modifications for aparticular application will work equally well in all scenarios.The architecture we propose allows cognitive radios to adaptdynamically as per the application's requirements.

There is active research in studying the individual com-ponents proposed in SAFIRE. Cognitive radio has been pro-posed for disaster scenarios. In [16], the authors proposea modification of the IEEE 802.11 MAC for cognitive ra-dios, while Rondeau et al. [19] describe their experiences inbuilding and testing a cognitive engine for a point-to-pointwireless link. More recently, the SDR Forum has announceda Smart Radio Challenge competition, where they proposethe idea of using cognitive radios for sharing informationbetween first responders [3]. The SAFIRE architecture not

659

l l~~~~~~~~~~~~~

(n

,LA

Page 6: [IEEE 2007 4th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks - (2007.06.18-2007.06.21)] 2007 4th Annual IEEE Communications Society

only complements such efforts but also extends them byadvocating a generalized multi-layered multi-objective policybased framework for tuning a cognitive radio's parameters.

Delay tolerant networks have only recently been studied[10]. Their applications include Internet connectivity for ruralareas [20], information collection in disaster scenarios [24],and even interplanetary communication [7]. In these applica-tions, lower layers of the protocol stack are unaware of thedelay-tolerant nature of the network. In our work, we employthe use of policies at the radio layer, that take into accountrequirements at higher layers (such as DTN). This aspect ofradio configuration has not been previously addressed in theresearch literature.

Publish-subscribe systems have been studied both on fixednetworks as well as mobile networks [11], [21], [22]. Systemsfor mobile networks provide inadequate support for discon-nections, requiring end-nodes to deal with network timeouts(e.g., by re-sending publications/requests). There is also somework in supporting publish-subscribe like functionality overchallenged networks such as DTNs [15], [23], [25]. However,the proposed approaches do not comprehensively exploreall aspects of the publish-subscribe architecture we outlinedearlier. As well, they do not explicitly explore techniques foroptimizing the underlying network, as per the requirements ofthe pub-sub overlay.

VI. CONCLUSION AND FUTURE WORK

Access to accurate and timely information in disaster sit-uations is critical for first responders. To facilitate this, aninformation sharing system between first responders can proveextremely useful during the course of the relief operations.However, the hostile nature of disaster environments imposes aset of crucial challenges in realizing such an information shar-ing system. We propose SAFIRE, a self-organizing architec-ture that supports information sharing among first responders.Its salient features are (1) A decentralized, disruption-tolerantcognitive radio based infrastructure for communication be-tween first responders, (2) A publish-subscribe mechanism forexchanging data, and (3) A flexible multi-layered policy-basedframework for optimally configuring the system.We believe we have only scratched the surface in terms of

fully realizing a disaster-response communication system. Weoutline two important directions for future research.

Spectral Sensing: The unique challenges for spectrumsensing outlined earlier need to be addressed. For coordination,in [4], the authors discuss an approach that makes use of aspatial decay function for establishing clusters of collaboratingnodes. We are exploring a similar approach for the SAFIREarchitecture.

Conflicting Policies: We have discussed how to specifymulti-layered policies for tuning a cognitive radio. However,situations may arise where conflicting policy constraints needto be simultaneously captured in the configuration process. Insuch scenarios, prioritizing policies can allow us to obviateproblems of conflicting interests. We are currently evaluatingthe use of utility-theory in solving this problem.

ACKNOWLEDGMENTS

This research was supported thanks to grants from the Nat-ural Sciences and Engineering Research Council of Canada,the Canada Research Chair Program, Nortel Networks, IntelCorporation, and Sprint Corporation. We would also like tothank our colleague, Saba Gul, and our anonymous reviewersfor their valuable feedback and suggestions.

REFERENCES

[1] Standards for Privacy of Individually Identifiable Health Information.Office of the Secretary, Department of Health and Human Services,Federal Register, 67(157):53182-53273.

[2] The 9-11 Commission Report, July 2004.http://origin.www.gpoaccess.gov/91 1/.

[3] SDR Forum's Smart Radio Challenge, Sample Problems, June 2006.http://www.radiochallenge.org/SampleProblems.html.

[4] N. Ahmed, D. Hadaller, and S. Keshav. GUESS: Gossiping Updatesfor Efficient Spectrum Sensing. In ACM MobiShare - Ist InternationalWorkshop on Decentralized Resource Sharing in Mobile Computing andNetworking, September 2006.

[5] H. Balakrishnan, M. F. Kaashoek, D. Karger, R. Morris, and I. Stoica.Looking up data in p2p systems. Communications of the ACM, 46(2):43-48, 2003.

[6] S. Bouckaert, J. Bergs, D. Naudts, J. Baekelmans, E. D. Kegel, N. V.den Wijngaert, C. Blondia, I. Moerman, and P. Demeester. A MobileCrisis Management System for Emergency Services: from Concept toField Test. In First International Workshop on "Wireless mesh: movingtowards applications" (WiMeshNets), August 2006.

[7] V. Cerf. Interplanetary Internet, January 2004. Talk delivered atDARPA's Disruption Tolerant Networking Proposer's Day.

[8] G. R. Chaddock. Police, fire depts. still can't talk, September 2004.http://www.csmonitor.com/2004/0915/pOls03 -usgn.html.

[9] P. T. Eugster, P. A. Felber, R. Guerraoui, and A.-M. Kermarrec. Themany faces of publish/subscribe. ACM Computing Surveys, 35(2):114-131, June 2003.

[10] K. Fall. A delay-tolerant network architecture for challenged internets.In Proceedings of the 2003 conference on Applications, technologies,architectures, and protocols for computer communications (SIGCOMM),pages 27-34, 2003.

[11] Y Huang and H. Garcia-Molina. Publish/subscribe in a mobile enviro-ment. In Proceedings of the 2nd ACM international workshop on Dataengineering for wireless and mobile access (MobiDe), pages 27-34,2001.

[12] E. Jones. Practical routing in delay-tolerant networks, Master's Thesis,University of Waterloo (UW). 2006.

[13] 0. Z. Khan. Incremental deployment of context-aware applications,Master's Thesis, University of Waterloo (UW). 2006.

[14] J. Mitola and G. Maguire. Cognitive Radio: Making Software RadiosMore Personal. In IEEE Personal Communications, volume 6. IEEECommunications Society, August 1999.

[15] M. Musolesi and C. Mascolo. Spatio-Temporal Communication Primi-tives for Delay Tolerant Systems. In 3rd Minema Workshop, February2006.

[16] P. Pawelczak, R. Prasasd, L. Xia, and I. Niemegeers. Cognitive RadioEmergency Networks - Requirements and Design. In IEEE Symposiumon New Frontiers In Dynamic Spectrum Access Networks (DySPAN),pages 601 - 606, 2005.

[17] J. M. Peha. The digital TV transition: A chance to enhance public safetyand improve spectrum auctions. IEEE Communications, 44(6):22-23,2006.

[18] S. Ratnasamy, P. Francis, M. Handley, R. Karp, and S. Shenker. AScalable Content-Addressable Network. In Proceedings of the 2001conference on Applications, technologies, architectures, and protocolsfor computer communications (SIGCOMM), 2001.

[19] T. W. Rondeau, C. W. Bostian, D. Maldonado, A. Ferguson, S. Ball, S. F.Midkiff, and B. Le. Cognitive Radios in Public Safety and SpectrumManagement. In Proceedings of the 33rd Research Conference onCommunication, Information and Internet Policy, September 2005.

[20] A. Seth, D. Kroeker, M. Zaharia, S. Guo, and S. Keshav. Low-costCommunication for Rural Internet Kiosks Using Mechanical Backhaul.In Proceedings of the 12th annual international conference on Mobilecomputing and networking (MobiCom), September 2006.

660

Page 7: [IEEE 2007 4th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks - (2007.06.18-2007.06.21)] 2007 4th Annual IEEE Communications Society

[21] K. S. Skjelsvik, V. Goebel, and T. Plagemann. Distributed EventNotification for Mobile Ad Hoc Networks. In IEEE Distributed SystemsOnline, volume 5, August 2004.

[22] E. Yoneki and J. Bacon. Distributed Multicast Grouping for Pub-lish/Subscribe over Mobile Ad Hoc Networks. In IEEE WirelessCommunications and Networking Conference (WCNC'2005), March2005.

[23] Y Zhang, B. Hull, V. Bychkovsky, H. Balakrishnan, and S. Madden.IceDB: Continuous Query Processing in an Intermittently ConnectedWorld. In under submission, March 2006.

[24] W. Zhao, M. Ammar, and E. Zegura. A message ferrying approach fordata delivery in sparse mobile ad hoc networks. In Proceedings of the5th ACM international symposium on Mobile ad hoc networking andcomputing (MobiHoc), pages 187-198, 2004.

[25] W. Zhao, M. Ammar, and E. Zegura. Multicasting in delay tolerantnetworks: semantic models and routing algorithms. In Proceedingsof the 2005 ACM SIGCOMM workshop on Delay-tolerant networking(WDTN), pages 268-275, 2005.

661