Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Standard view
Full view
of .
Save to My Library
Look up keyword
Like this
0 of .
Results for:
No results containing your search query
P. 1

Ratings: (0)|Views: 13 |Likes:
Published by Celis González

More info:

Published by: Celis González on Apr 27, 2010
Copyright:Attribution Non-commercial


Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less





 D. Estrin, L. Girod 
UCLADepartment of Computer Scienceemail:
@cs.ucla.eduG. Pottie, M. Srivastava
UCLADepartment of Electrical Engineeringemail:
Pervasive micro-sensing and actuation may revolutionize the wayin which we understand and manage complex physical systems:from airplane wings to complex ecosystems. The capabilities fordetailed physical monitoring and manipulation offer enormous op-portunities for almost every scientific discipline, and it will alterthe feasible granularity of engineering.We identify opportunities and challenges for distributed sig-nal processing in networks of these sensing elements and investi-gate some of the architectural challenges posed by systems that aremassively distributed, physically-coupled, wirelessly networked,and energy limited.
The availability of low-power micro-sensors, actuators, embeddedprocessors, and radios is enabling the application of distributedwireless sensing to a wide range of applications, including en-vironmental monitoring, smart spaces, medical applications, andprecision agriculture [1][2]. Most deployed sensor networks in-volve relatively small numbers of sensors, wired to a central pro-cessing unit where all of the signal processing is performed [3]. Incontrast, this paper focuses on
distributed, wireless, sensor net-works
in which the signal processing is
along withthesensing.
Why distributed sensing?
When the precise location of a signal of interest is unknown in a monitored region, dis-tributed sensing allows one to place the sensors closer tothe phenomena being monitored than if only a single sen-sor were used. This yields higher SNR, and improved op-portunities for line of sight. While SNR can be addressedin many cases by deploying one very large sensitive sen-sor, line of sight, and more generally obstructions, cannotbe addressed by deploying one sensor regardless of its sen-sitivity. Thus, distributed sensing provides robustness toenvironmental obstacles.
Why wireless?
When wired networking of distributed sen-sors can be easily achieved, it is often the more advanta-geous approach. Moreover, when nodes can be wired to re-newable (relatively infinite) energy sources, this too greatlysimplifies the system design and operation. However, in
Supported by the DARPA SensIT program and NSF Special Projectsunder ANIR.
Supported by the DARPA SensIT program.
many envisioned applications, the environment being mon-itored does not have installed infrastructure for either com-munications orenergy, and therefore untethered nodes mustrely on local, finite, and relatively small energy sources, aswell as wireless communication channels.
Why distributed processing?
Finally, although sensorsare distributed to be close to the phenomena, one might stillconsider an architecture in which sensor outputs could becommunicated back to a central processing unit. However,in the context of untethered nodes, the finite energy budgetis aprimarydesign constraint. Communications is akey en-ergy consumer as the radio signal power in sensor networksdrops off with
[4] due to ground reflections from shortantenna heights. Therefore, one wants to process data asmuch as possible inside the network to reduce the numberof bits transmitted, particularly over longer distances.
The potential applications of wireless sensor networks are highlyvaried: e.g., Physiological monitoring; Environmental monitoring(air, water, soil chemistry); Condition based maintenance; Smartspaces; Military; Precision agriculture; Transportation; Factory in-strumentation and inventory trackingHabitatmonitoring[Cerpa-etal01, Hamilton, Steere-etal00]pro-vides a rich collection of sensing modalities and environmentalconditions and we use it tomotivate our technical discussion. Con-siderthegoalof supporting datacollectionandmodel developmentof complex ecosystems. Scientists andenvironmental impact mon-itoring authorities would like to monitor soil and air chemistry, aswell as plant and animal species populations and behavior. Forthe latter, the primary modalities are imaging and acoustics to lo-calize, identify and track species or phenomena based on implicitsignals (acoustic and seismic), or explicit signals (RF tags). Thesefacilities must be deployable in remote locations that lack installedenergy and communication infrastructures, motivating the need forlow-power wireless communication.The strategy for node cooperation strategy has significant con-sequences in terms of communication bandwidth and energy con-sumption. Forexample, considerthetaskofidentifying birdspeciesin view of several cameras. If it is to be accomplished throughimage analysis, we could stream all the video back to a humanoperator-a very costly approach. Alternatively, we could streamaudio to a central location, which then performs signal processingto identify and stream back only those streams that are most likelyto contain a target species. While this reduces communications
overhead greatly, it still suffers from communications latency andlacks scalability due to the need to stream audio thorough a centralprocessing point. Finally, we might distribute the problem further,hosting the audio signal processing software on the nodes, and de-veloping algorithms that require only local cooperation to make adecision to capture images. This approach is scalable in that nolong-range streaming of audio or video is necessary, resulting inmore efficient use of communications bandwidth and limited en-ergy resources.In the remainder of this paper we identify some of the tech-nical challenges associated with the design of wireless sensor net-works and discuss several algorithmic approaches.
Most envisioned sensornetworkapplications encounterone ormoreof the following challenges:
for energy and communication requiring max-imal focus on energy efficiency.
Ad hoc
deployment, requiring that the system identifiesand copes with the resulting distribution and connectivityof nodes.
environmental conditions requiring the system toadapt over time to changing connectivity and system stim-uli.
operation requiring configuration and recon-figuration be automatic (self-configuration)To address these challenging environments, several strategiesare likely to be key building blocks/techniques for wireless sensornetworks:
Collaborative signal processing
among nodes that haveexperienced a common stimulus will greatly enhance theefficiency (information per bit transmitted) of these sys-tems. Develop both
signal processing on smallclusters by a centralized entity within the cluster, and
processing with much less stringent synchroniza-tion requirements and applicable across larger numbers of more loosely coupled elements.
Exploiting redundancy
of hardware elements to compen-sate for ad hoc deployment of systems. If elements cannotbe carefully positioned relative to each other and the envi-ronment, then an alternate strategy to achieve “coverage”is to deploy a greater density of elements so that one canmake use of some subset that have the desired absolute andrelative position. In some contexts, even if elements canbe uniformly placed in 3-space, environmental conditionsmight be such that coverage is not uniform due to obstaclesand other sources of noise. Another application of redun-dancy is when the incremental cost of a node during initialdeployment is much smaller than the incremental cost of deploying new nodes or renewing node resources (e.g., en-ergy). In this case, one can exploit redundancy to extendsystem lifetime by adjusting duty cycle based on local den-sity and local demand.
is another strategy thatcan be exploited in sensor networks to make trade-offs be-tween energy, accuracy, and rapidity of results. Recogniz-ing that one is trying to detect non-deterministic phenom-ena in the presence of communication noise and sensor di-versity, thefidelityandtimelinessof thesignal processing atindividual sensor nodes can be adapted to energy resourcesand latency requirements.
A hierarchical, tiered architecture
can greatly contributeto overall system lifetime and capability. Whenever pos-sible, higher capacity system elements can be used to of-fload drain on small form factor elements, while the lattercan be exploited to obtain the desired physical proximity tostimuli. Moreover, even among elements with homogenouscapabilities, creating clusters and assigning special com-bining functions to cluster heads can contribute to overallsystem scalability. However, to avoid compromising ro-bustness, such clusters/hierarchy must be self-configuringand reconfiguring in the face of environmental or network changes.
We now describe three generic techniques that would enable dis-tributed signal processing tasks in wireless sensor networks.
4.1. Coherent processing algorithms
Coherent signal processing algorithms are distinguished from non-coherent methods in that information about the phase of the wave-front impinging on the nodes must be conveyed. Beamformingtechniques allow localization of signals that originate within theconvex hull of the participating nodes, higher SNR estimates of the signals compared to non-coherent methods, and determinationof bearing angles for signals that originate outside the convex hullof the participating nodes. The price is a higher level of synchro-nization (to within a small fraction of one oscillation), and com-munication of relatively high bit rate data streams consisting of sampled waveforms. Given its high resource cost, we should re-sort to coherent processing only when we cannot attain adequateaccuracy in the result with non-coherent methods such as combi-nation of likelihood functions.One way to organize the operations leading to coherent beam-forming isas follows. Nodes gothrough asequence of internal lev-els of signal processing before determining that neighbors shouldbe involved in a detection/localization decision. An ad hoc net-work is constructed for non-coherent decision-making using forexample the single winner election algorithm of [2]. The algo-rithm is optimized to minimize the overhead in finding a fusioncenter, since relatively little data must actually be communicated.However, if the decision has insufficient certainty or resolution,the same set of nodes become involved in a new network set-upthat seeks to minimize the energy consumption in conveying sam-pled waveforms to a common central processing point. To thisend significant overhead is acceptable since large amounts of datawill be conveyed in the local neighborhood. A multi-winner elec-tion algorithm to accomplish this is also described in [2]. Standardbeamforming techniques can now be applied using the data col-lected from the cluster of nodes.There is no requirement for uniform lay-down of nodes toachieve beamforming [5]. To track distant sources, two or moreclusters of nodes can be used, and with the intersection of thebearing lines used to establish location. Note that simply usingall nodes in the network to do one massive beamforming operation
could accomplish this end, but excessive communications and sig-nal processing complexity would be required. Rather, for a scal-able solution a signal processing step is required that recognizeswhether near or far objects are being tracked. A crude techniqueis to consider the SNR variations among nodes in a cluster and toneighboring clusters. If the SNR is similar, then the signal sourceis likely to be distant. Having made this determination, clustersmay decide to estimate lines of bearing or not, whether probabilis-tically or according to a predetermined schedule. The informationon the bearing lines is then conveyed to a central node designatedto perform the (noncoherent) fusion. Thus, there is never a case inwhich sampled waveforms must be conveyed over a large numberof hops.Achievingtherequiredlevelof synchronism forcoherent beam-forming is in principle relatively straightforward for systems inwhich every node possesses a radio. Since the propagation ve-locity of seismic and acoustic signals is six orders of magnitudeslower than that of radio waves, achieving data lock for RF com-munications would seem to already be much better synchroniza-tion than is required for beamforming. However, particular caremust be paid to the node architecture to take advantage of this tim-ing information. The typical interrupt cycles of general-purposeprocessors can be tens of milliseconds, an eternity with respect toeven acoustic signals. Thus, embedded real-time components arerequired in the nodes to deal with time-stamping of the data.
4.2. Localization
Node location is employed by routing protocols that use spatial ad-dresses, and by signal processing algorithms (e.g. beamforming)that are used for tasks such as target tracking. The underlying al-gorithm problem is that of localization whereby the nodes in thenetwork discover their spatial coordinates upon network boot-up.When the sensor nodes are deployed in an unplanned topology,there is no a priori knowledge of location. The use of GPS insensor nodes is ruled out in many scenarios because of power con-sumption, antenna size, and overhead obstructions such as densefoliage. The ad hoc nature of deployment rules out infrastructurefor many scenarios of localization. It is critical that sensor net-work nodes be able to estimate their relative positions without as-sistance, using means that can be built-in.Thelocalizationprobleminitselfisagood exampleof asignal-processing task that the sensor network needs to solve. The ba-sic approach would be for sensor nodes to gather sufficient num-ber of pair-wise distance estimates via some suitable mechanism,and then use multilateration algorithms to estimate positions of thenodes. To begin with, a few nodes might know their position viaother means (beacon nodes), but at the end of the localization pro-cess every node would hopefully know its position.A key problem however is that in conventional formulationsof multilateration [6][7] one needs to estimate the location of anentity given estimates of its distance to 3 or more beacons withknown positions. In sensor networks a very high density of bea-cons nodes would be needed. To keep the required beacon densityand energies low, a preferred method would be to jointly estimatepositions of all the non-beacon nodes via a collaborative multilat-eration formulation based on criterion such as least-square errorminimization. Besides being computationally hard for large num-ber of nodes, doing this would require a centralized node where allthe distance estimates would be collected at significant commu-nication and associated energy cost. A more scalable solution islocally distributed iterative multilateration [8] whereby a node cal-culates its position and is promoted to a beacon as soon as enoughof its 1-hop neighbors are beacons. Starting with a critical densityof beacons, a percolation-like phenomenon would result in grad-ually all the nodes discovering their position. With a sufficientbeacon density, a small number of successive multilateration stepslead to rapid convergence of location estimates. The communi-cation overhead is much lower than in centralized approach as allmessage exchange is strictly local and is easily piggybacked onrouting messages.Another challenge in localization is estimation of distance be-tween a pair of nodes. Using time-of-flight of radio signals (as inGPS) is ruled out when the distances are too tiny and radio fre-quencies not very high. A readily available method would be touse the received signal strength indication (RSSI) provided be theradio. The RSSI data can be cheaply piggybacked on regular rout-ing and data. The accuracy of this approach can be improved byusing a parameterized channel, path loss model whose parametersare also estimated together with position [8]. However, in practice,the RSSI based approach works only in the absence of significantmultipath effects. In most environments other than open spacesmultipath is an issue. A promising alternative technology is to es-timate distance by time of flight of acoustic or ultrasound signals,and using the much faster radio signal to establish time reference[9][10][11].
4.3. Distributed power management
Dynamic power management techniques such as shutdown anddynamic voltage scaling have emerged as powerful methods forpower-aware computing. Power-aware operation is even more im-portant for wireless sensor networks, and requires distributed ver-sions of power management techniques.As an example, consider shutdown, which is widely used inportable computing systems such as laptops. In sensor nets onecould exploit redundant nodes by turning on only a time- varyingsubset of nodes, where the subset is selected for desired sensor andradio coverage. The remaining nodes can be shutdown, only to bewoken up to provide additional sensor readings or communicationroutes when something interesting happens [12].A key problem in such a distributed shutdown scheme is thestrategy to select which node to shutdown and which to turn onat any given instant. A good way to model this problem is to op-timally divide the sensor nodes into several subsets such that anygiven subset provides a baseline level of sensing and communica-tion coverage. The different subsets can then be turned on and off according to a duty cycle determined by a repetitive schedule. Asnodes die by depleting their batteries, the subsets are changed.Unfortunately, modeling the problem in this fashion requiresone to gather global information to find the subsets. Since com-munication is expensive in energy, the cost of the power manage-ment algorithm would swamp the savings from power manage-ment!
This illustrates the dilemma that so often arises in prob-lems in sensor nets: the seemingly optimal way of solving aproblem often results in algorithms whose communication en-ergy costs exceed their benefits.
Therefore, a better strategy isto use algorithms that only shoot for good though sub optimal re-sults but require only locally distributed processing with minimalcommunication costs.This suggests that the decision regarding when to shutdownand wakeup a node should be made using information in the local

You're Reading a Free Preview

/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->