Wireless Integrated Network Sensors Barbara Theodorides April 15
Wireless Integrated Network Sensors Barbara Theodorides April 15, 2003
Paper • G. J. Pottie and W. J. Kaiser, Wireless Integrated Network Sensors, Communications of ACM, 43(5), May 2000.
WINS • Initiated in 1993 at the UCLA, 1 G fielded in 1996 • Sponsored by DARPA LWIM program began in 1995 • In 1998, WINS NG • Distributed network • Internet access to sensors, controls and processors • Low-power signal processing, computation, and low-cost wireless networking • RF communication over short distances ( < 30 m ) • Applications: Industries, transportation, manufacture, health care, environmental oversight, and safety & security.
A general picture worldwide user local area low power networking sensing Internet wireless communication n signal processing / event recognition nt atio e ev orm inf
Concerned about… • The Physical principles dense sensor network • Energy & bandwidth constraints distributed & layered signal processing architecture • WINS network architecture • WINS nodes architecture
Physical Principles • When are distributed sensors better? A. Propagation laws for sensing All signals decay with distance e. g. electromagnetic waves in free space (~ 1/d 2) in other media (absorption, scattering, dispersion) distant sensor requires costly operations • In free space favor large array • However, almost every scenario of interest distributed array Ø regardless the array size objects behind walls If the system is to detect objects reliably, it has to be distributed, whatever the networking cost
Physical Principles (cont) • What are the fundamental limits driving the design of a network of distributed sensors? B. Detection & Estimation Detector: given a set of observables {xj} determines which of the hypotheses {hi} are true Target presence/absence: based on estimates parameters {fk} of {xj} Ø Selected Fourier, wavelet transform coefficients Ø Marginal improvement Formally: Decide on hi if Reliability: Complexity: p(hi | {fk}) > p(hj | {fk}) ∀j≠i #independent observations, SNR dimension of feature space, #hypotheses Either a longer set of independent observations or high SNR Decrease the #features and the #hypotheses
Physical Principles (cont) Use of practical Algorithms: Ø Apply deconvolution and target-separation machinery to exploit a distributed array (deal with only 1 target and no propagation dispersal effects) - reduces feature space & #hypotheses cons: complexity Ø Deploy a dense sensor network - homogeneous environment within the detection range - reduces #environmental features size of decision space attractive method
Physical Principles (cont) C. Communication Constraints • Spatial separation (e. g. low lying antennas) Ø Surface roughness, reflecting & obstructing objects However spatial isolation, reuse of frequencies • Multipath propagation (reflections off multiple objects) Recover ~ space, frequency, and time “diversity” But for static nodes, time diversity is not an option spatial diversity is difficult to obtain Diversity in frequency domain • “Shadowing”: dealt with by employing a multihop network The greater the density, the closer the nodes, and the greater the likelihood of having a link with sufficiently small distance and shadowing losses.
Physical Principles (cont) D. Energy Consumption • Limits to the energy efficiency of CMOS communications and signalprocessing circuits • Limits on the power required to transmit reliably over a given distance Networks should be designed so that radio is off as much of the time as possible and otherwise transmits only at the minimum required level • ASICs can clock at much lower speeds consume less energy ASICs maintain a cost advantage
Signal-Processing Architecture • We want: low false-alarm & high detection probability Processing Hierarchy Human Sophisticated Methods Collaboration of WINS nodes Precision Cost Higher-energy processing & sensing Energy thresholding If application & infrastructure permit: process data locally / multihop routing Play the probability game only to the extent we have to
Signal-Processing Architecture (cont) • Application Specific e. g. Remote security application • WINS node: 2 sensors (seismic & imaging capability) Ø Ø Seismic senor requires little power constantly vigilant Simple energy detection triggers the camera’s operation Collaborative WINS nodes (e. g. target location) Send image & seismic record to a remote observer WINS node: simple processing at low power Radio: does not need to support continuous transmission of images
WINS Network Architecture Characteristics • • • Support large numbers of sensor Low average bit rate communication ( < 1 -100 Kbps ) Dense sensor distributions Exploit the short-distance separation multihop communication Protocols: designed so radios are off MAC address should include some variant of time-division access Time-division protocol Ø Exchange small messages: performance information, synchronization, bandwidth reservation requests Ø Abundant bandwidth few conflicts, simple mechanisms At least one low-power protocol suite has been developed feasible to achieve distributed low-power operation in a flat multihop network
WINS Network Architecture (cont) Link Sensor Network to the Internet • Layering of the protocols (and devices) is needed WINS Gateways: Support for the WINS network and access between conventional network physical layers and their protocols and between the WINS physical layer and its low-power protocols System Architect – Responsibilities Ø Application’s requirements (reduced operation power, improved bit rate, improved bit error rate, reduced cost) Ø How can Internet protocols (TCP, IPv 6) be employed? - need to conserve energy, unreliability of physical channels Ø Where should the processing and the storage take place? - at the source / reducing the amount of data to transmit
WINS Node Architecture 1993: Initiated at the UCLA 1 G of field-ready WINS devices and software was fielded (1996) 1995 : DARPA sponsored - the LWIM project multihop, self-assembled, wireless network algorithms for operating at micropower levels - the joint, UCLA and Rockwell Science Center of Thousand Oaks, program platform for more sophisticated networking and signal processing algorithms (many types of sensors, less emphasis on power conservation) Lesson: Separate real-time from higher-level functions
WINS Node Architecture (cont) 1998: WINS NG developed by the authors contiguous sensing, signal sensor actuator interface processing for event detection, local control of actuators, event classification, communication at low power Ø Event detection is contiguous micropower levels Ø Event detected => alert process to identify the event Ø Further processing? Alert remote user / neighboring node? Ø Communication between WINS nodes signal processing for event detection control continuously vigilant operation Processing event classification & identification wireless internet interface low-duty cycle operation
WINS Node Architecture (cont) Further Generations (Future work): Ø Support plug-in Linux devices Ø Small, limited sensing devices interact with WINS NG nodes in heterogeneous networks Ø Scavenge energy from the environment photocells
Why WINS ? • Low power consumption ( 100 μW average ) Ø Separation of real-time from higher level functions Ø Hierarchical signal-processing architecture • Application specific • Communication facility ( WINS gateways ) Ø Remote user • Scalable Ø Reduce amount of data to be send scalability to thousands of nodes per gateway
Conclusion • • • Densely distributed sensor networks (physical constraints) Layered and heterogeneous processing Application specific networking architectures Close intertwining of network processing Development platforms are now available
- Slides: 19