Transcript Document
Sensor Networks – Trend, Applications, & Design Polly Huang iSpace Labs EE & GINM, NTU http://cc.ee.ntu.edu.tw/ ~phuang Road Map Overview What a sensor network is? Applications How can the sensor network be useful? Problems & Designs How should the sensor network work? Sensor Network Sensors? camera pressure mic accelerometer thermal gyro GPS Also the biomedical sensors: EMG, EKG, pulses, emotions, etc Sensor Nodes Today MICA, 2001-2002 5.7cm X 3.18cm Spec, March 2003 2mm X 2.5mm 4 MHz CPU 128K ROM 512K RAM 40kbps Radio range x00 feet Sensors, battery not included CPU, memory, RF transceiver Sensors, battery, antenna not included < 1 dollar if mass-produced Embedded Sensor Node Intel® Xscale CPU Analog and digital radio Flash and SRAM memory Sensors Embedded Sensor Network Applications Home, Office, Healthcare, Science & Nature, Agriculture Intel Digital Home UCSB Bren School of Environmental Science and Management Day-lighting controls Operable windows control taking advantage of on-site ocean breezes Airflow controls based on CO2 level Long-Term Nursing Home • Fall detection • Vital sign monitoring • Dietary/exercise monitoring Camera Orientation sensor Pressure sensor Accelerometers Muscle activity The Channel Islands Fox Fancy Californian Winemaking Temperature Soil moisture Pest/disease Taiwan Value! Home: Security (Crime Rate) Industry: IC (Energy crises) Healthcare: Elder Care (健保 Burden) Agriculture: Orchid Growing (Big Export Business) Our Applications Emergency Alarm (chronic disease) Location Tracking (cognitive declined) Background Music Player (just lonely) The Elder’s Companion Sensor Network WiFi MicaZ Sensor Node Xscale-based Embedded System Full-Fledged Server GPRS Idea of Emergency Detection Activity Level Normal Now 0 hr 6 hr 12 hr 18 hr 24 hr Prototype Front Courtesy: Steven Chiu Back Neck 1. 2. 3. The value level does not imply the level of activity Going for the difference to distinguish active vs. inactive periods Microphone is useful detecting conversation Accelerometer - x Microphone Accelerometer - y Converting to Activity Level Wavelet decomposition to extract the difference f(x) Activity_level(xk) = avg(f(xi), i=k-4,…k) x Sensor Network Design Major Difference to other Networks: 1. Wireless Technology 2. Communication Model Wireless Technologies Mote PicoNet IEEE 802.15.4 bluetooth WLAN Range xm xm 10 m 100 m Max BW 40 kbps 224 kbps 723 kbps 11Mbps Band 310, 433, 868-928 868/916M MHz, Hz 2.4GHz 2.4GHz 2.4GHz Power (data mode) (1mA) 30mA 250mA (10mA) Communication Models Address centric (for example IP) Name the nodes Data disseminated by the destination address Data centric Nodes desiring data expressing ‘interest of content’ Data disseminated based on the content IP Communication (Address Centric) Organize system based on named nodes Per-node forwarding state Senders need to push data to the node address of sink To Bob To Alice: To Bob To Alice: To Bob To Alice: My name is Alice. My name is Alice. My name is Alice. Like to meet girls II am II am II am am aBob 19-yr old am aBob 19-yr old am aBob 19-yr old I I am am Bob Alice I am Bob I I am am Bob Alice Tell me about you Tell me about you Web search Tell meAlice about you II am am Bob Alice girl… girl… girl… -> Alice Bob there Chris Alice there Bob there Bob Alice there Data-Centric Communication Organize system based on named data Per-data diffusion state Sinks need to be specific about what data they’d pull Here’s a 19-yr old Tell me Here’s a 19-yr old Here’s a 19-yr old Tell meTell me Tell me girl… about girls about girls about girls girl… girl… about girls Girl info goes there Girl info goes there The Need to Go Data Centric Node addressing not scalable Difficulty of configuring/tracking small nodes scattering around Cost of re-configuration of nodes moving around Server infrastructure not efficient Difficulty of deployment DNS and search engines in sensor networks? The sum of maintenance traffic (energy) Additional delay Information might be outdated by the time of communication Research Topics Data dissemination (Routing) Load balancing Service differentiation Naming and forwarding Data aggregation Time synchronization Energy efficient MAC Data Dissemination The problem Setting up ‘content-based’ states of interest to direct data to the rightful destination Naive dissemination may result in unreliable or long-delay in delivery Challenge Medical data are mission critical The approach Shortest multiple paths Magnets and nails Our Design Magnetic Diffusion Medical Applications Mission-critical data Timeliness Reachability Average user Wireless sensor nodes a must Energy efficiency wireless communication The Idea – Magnetic Diffusion Establishing Magnetic Field 3 4 4 3 2 2 3 1 Data Attracted by the Magnet 3 4 4 3 2 2 3 1 Reachability Overhead Latency-Base Latency-Mobility Latency-On Off Major Findings Going for timeliness Magnetic diffusion, flooding, opp Going for reachability Flooding, magnetic diffusion, opp Going for energy efficiency Opp, magnetic diffusion, flooding Best for Medicare Applications Healthcare applications Mission-critical data Require timeliness, reachability Average user Energy efficiency Magnetic diffusion The solution to offer all the QoS required Load Balancing The problem Setting up ‘content-based’ states of interest to direct data to the rightful destination Naive dissemination may result in short network lifetime Uniqueness Co-existence of static and mobile sensor nodes The approach Mobility and power aware data diffusion Mobile, power limited nodes avoid propagating interests Service Differentiation The problem The network delay could be long when the demand is high The reliability might not be perfect Unique problem to health care Some data could be life-death critical The approach 2 class differentiation Urgent data going flooding and high priority in forwarding Non-urgent data going magnetic diffusion and regular forwarding System Architecture Services Core Urgent Data Dissemination Service Broadcast Adaptive Diffusion Routing Non-urgent Data Dissemination Service Priority Regular Forwarding Naming and Forwarding The problem Addressing data by the content Looking up all possible ‘states of interest’ to find the matching entry to direct data A real problem to sensor network for Different data types If not handled well, could be a serious performance problem The approach Efficient string matching algorithms for limited device Two Algorithms Preprocessing Time Space Suffix tree O(P) O(S) O(S) TST Nlog(N)+O(S) O(S) Searching Time O(log(N)+P) Note: N means the number of words S means the total length of words P means the length of input testing string Research Topics Data dissemination (Routing) Load balancing Service differentiation Naming and forwarding Data aggregation Time synchronization Energy efficient MAC Data Aggregation Concept Intermediate nodes have the computation power to process and aggregate the data passing through Achieve data reduction Why do we need to aggregate? Energy is a critical problem in sensor networks Without In-Network Processing Data are simply passed on Here’s a 19-yr old Tell me girl… about girls Girl info goes there Here’s a 20-yr old girl… Here’s a 20-yr old girl… Here’s a 19-yr old Tell me Here’s a 19-yr old Tell meTell me girl… about girls about girls about girls girl… Here’s a 20-yr old Tell me girl… about girls Girl info goes there Girl info goes there With In-Network Processing Data are aggregated and then passed on Here’s a 19-yr old girl… Girl info goes there Here’s a 20-yr old girl… Girl info goes there Here’re two 19+ yr old girls… Here’re two 19+ yr old girls… Girl info goes there Application-specific Aggregation Here! Average Dissipated Energy (Joules/Node/Received Event) Potential of In-network Processing 0.025 Diffusion Without Suppression 0.02 0.015 0.01 Diffusion With Suppression 0.005 0 0 50 100 150 200 250 300 Network Size Application-level suppression allows diffusion to reduce traffic and to surpass omniscient multicast. Our Design Data Aggregation for Resource Inventory Applications Resource Inventory Ecologist tracking the amount of animals over a certain area Retailers tracking the amount of in-stock merchandizes Factories tracking the amount of parts The Old vs. New Way Sampling & inferencing Sampling Inferencing the population statistically Issues: Bias Never sure what the real population is Sensor network monitoring Each sensor reporting location of objects Count the number of distinct locations Our Strategy Minimizing the energy consumption Send object counts within each sensor’s sensing range Aggregate location data right at the beginning of transmission One data per sensor, it scales well with increasing object population Energy Dissipation When the population increases Population Estimation A problem still exists how to get the population from the object count Cannot just sum up counts We could find the tight lower and upper bounds Given the location of the sensor Given the radius of the sensing range A Lower Bound Computation Find sets of disjoint regions Sum the counts The population is at least this much Get the maximal of these total counts Equivalent to the maximal independent set problem in graph algorithm A Upper Bound Computation Similar, But finding the set of regions that cover the entire area Sum the counts The population is no more than this Get the minimal of these total counts Equivalent to the Minimal coverage set problem in graph algorithm Estimated Population Estimation Range Upper bound remains on 70~80%, and lower bound remains on 30% Discussions The lower and upper bounds include all possible population, not a statistical result Because the estimation range is steady, possibly we could infer the exact population by lower and upper bound. The estimation range would strongly depend on the deployment of sensors Summary What more convenience can technologies bring? Context-aware and intelligent services anywhere and anytime We need small devices with processing, sensing, and communication capabilities (SOCs) See ‘communication’? We need to design all over again how to network these SOCs Questions? Polly Huang iSpace Labs EE & GINM, NTU http://cc.ee.ntu.edu.tw/~phuang