NEST ANSCD Self Organizing Wireless Sensor Network Middleware CleanPoint University of Virginia PI: John A. Stankovic December 2004 University of Virginia Outline NEST ANSCD • Operational Scenario • Goals • Overview and Status of Middleware • Middleware Services – Key Services • • • • Power Management/Sentry/Tripwire Service Group Management Service 3-Tier Classification Self-healing – Other Services • Lessons Learned • Remaining Work FY ‘05 University of Virginia Energy Efficient Surveillance System NEST ANSCD 1. An unmanned plane (UAV) deploys motes Diffusion Routing Neighbor Discovery Time Synchronization Parameterization Sentry Selection Coordinate Grid Zzz... Data Aggregation Data Streaming Group Management Leader Election Sentry Localization Network Monitor Tripwire Service Reconfiguration Reliable MAC 3.Sensor network detects vehicles and wakes up the sensor nodes Leader Migration 2. Motes establish an sensor network with power management Scheduling State Synchronization University of Virginia Goals NEST ANSCD • Develop an operational self-organizing sensor network of size 1000 • Cover an area of 1000m x 100m • Stealthy • Lifetime 3-6 months • Timely detection, track and classification – Large or small vehicle – Person, person with weapon • Wakeup other devices when necessary – Extend the lifetime of those devices as well • Exhibit self-healing capabilities University of Virginia Project Milestones FY04 March 3rd V1.0 May 28th V1.1 Aug. 6 V1.2 NEST ANSCD Oct.29 V1.3 Dec 6/13 Final University of Virginia Summary: Deliverables NEST ANSCD • ANSCD V1.3 middleware code delivered – About 40,000 lines of code and 600 files – About 30 Middleware services provided – Tested with a network with hundred(s) of nodes • ANSCD Data Packages V1.3 Delivered –System Architecture designed/documented –Mote-Relay Interface designed/documented –Relay requirements defined/documented –Requirements analysis/documented –Demo Test scenario design/documented –ANSCD & Mission GUI Manual documented –Wireless Download Manual documented • About 20 related papers published University of Virginia Summary: Objectives Achieved NEST ANSCD Metrics Objective Achieved Metrics Objective Achieved Coverage (T) 30m by 1000 m (O) 100m by 1000m Yet to be tested Sensor Modality Magnet and PIR (T), Acoustic and other (O) YES Scale 1000 motes Yet to be tested Self-Localization Real Coordinates (O) YES Deployment Manual (T); Airdrop (O) YES Reconfiguration True(T/O) YES Ad hoc Routing True(T/O) YES Robustness Backbone True(T/O) YES False Alarm <5% (T); <1% (O) YES Time Synchronization True(T/O) YES Detection True (T/O) YES Interface Control Doc True(T/O) YES Tracking True (T/O) YES Tracking Trace True (T/O) YES Classification True (T/O) YES Network Topology Report True(T/O) YES Accuracy 90%(T); 95%(O) YES Sentry Control True(T/O) YES Tracking Speed 30 mph (T); 50mph (O) YES Sentry Health Report True(T/O) YES Sentry Service True (T/O) YES Source Code True(T/O) YES Endurance 3 mo (T); 6 mo (O) YES Documents True(T/O) YES Energy Balance True(T/O) YES Technical Support True(T/O) YES Stealthness True (T/O) YES Multi-hop Reprogramming N/A YES Data Dissemination Relay/RSC (T) SISA(O) YES Golden Image N/A YES University of Virginia ANSCD Architecture V1.3 NEST ANSCD Application Layer EnviroTrack Velocity Regression Classification False Alarm Filtering Engine Relay Middleware Layer Time Sync Group Mgmt Sentry Service Dynamic Config Power Mgmt. Locali zation Tripwire Mngt Sensing Layer Network Layer Robust Diffusion Tree Report Engine Asymmetric Detection Radio-Base Wakeup Frequency-Filter Continuous Calibrator Data Link Layer MAC Interference avoidance MICA2 /XSM /XSM2 / MICA2DOT Motes Sensor Drivers Display at C&C University of Virginia Time-Driven System Operation Phase II Phase III Phase IV Time Sync Localization Asymmetri Detection NEST ANSCD Phase V Network Partition & Diffusion Tree Constrcution Dormant Section Phase I Start System Initialization Event Tracking Phase VI Sentry Selection Phase VIII Power Mgmt RESET Wakeup Service Phase VII Health Report Power Mgmt Phase VIII Event Tracking Tripwire Section University of Virginia Key Software Components (1) NEST ANSCD • 2-way software interface to RSCC and Avalanche (see ICD) • Flexible Tripwire based power management with sentry and wakeup services • Group-Based Entity Tracking (EnviroTrack) • Hierarchical Multi-tier Detection and Classification via heterogeneous sensors (4 PIRs (motion), acoustic, magnetometers) • Frequency-Filter and continuous threshold adaptations for robust sensing University of Virginia Key Software Components (2) NEST ANSCD • Flow control with Aggregate display/health/Tracking message • Localization (walking GPS) • Radio-based network wakeup • Asymmetry detection for robust routing establishment • Robust velocity calculation with least squares estimation • Wakeup service for relay to conserve energy University of Virginia Key Software Components (3) NEST ANSCD • Stripped-down version of Vanderbilt clock sync • Multi-hop Dynamic reconfiguration • Multi-hop wireless download (Berkeley’s Deluge) • Golden image support • Modified B-MAC to avoid communicationsensing interference University of Virginia System Scenario Supported (1) NEST ANSCD • Flexibility to define various system architectures • Independent deployment with Tripwires – ANSCD Middleware V1.3 – ANSCD GUI Laptop1 Base1 802.11g Router (optional) Laptop0 802.11g Base0 Laptop2 Base2 50 meters Road 300 Meters ( 3 tripwire section with 100 motes (5 x 20 ) in each section) University of Virginia System Scenario Supported (2) NEST ANSCD • ANSCD Middleware 1.3 • Single RSCC • Mission GUI RSCC Laptop Base0 50 meters Road 300 Meters ( 3 tripwire section with 100 motes (5 x 20 ) in each section) University of Virginia System Scenario Supported (3) • ANSCD Middleware V1.3 with Tripwires • RSCC • Relay • C2PC • SISA •… Additional RSCC and Sensor Networks Long Haul (LH) Comms Link NEST ANSCD C2PC Client RELAY Comms Antenna RF SENSOR FIELD SEIWG Antenna Ground Station Element C2PC Gateway (& Client) Long Haul Radio TACTICAL DISPLAY Mission GUI Socket Socket RS232 Interface IR/EO CAMERA(s) MOC Server Interface FCD MOTE FIELD MOTEFIELD (SENSOR NETWORK) Hardwired Sensors SOPHISTICATED SENSORS SENSOR (SS) (SSU) Mission GUI RSCC RSCC LH Socket Converter TCP/IP Portal CStat Socket LH Server Interface MOC/P MOC/P Courtesy of Northrop Grumman University of Virginia ANSCD GUI – Vehicle & Person NEST ANSCD University of Virginia ANSCD GUI – Person w/ Weapon NEST ANSCD University of Virginia Mission GUI NEST ANSCD University of Virginia Mote - Relay Interface V1.5 Address AM ID Group ID Byte Count Flags Record Type Source ID Message ID Data NEST ANSCD CRC Format of notification and command messages Notification Data Records • Tracking • Node status • Network configuration Command Data Records Request Reset Tracking record Cmd ID Node ID Event ID XCoord Event Type Event Type Leader ID YCoord Attribute Type Velocity Parent ID XCoord # Sentries Confidence Value YCoord # Nodes Accuracy Conf. Level Voltage Magnet Number Motion Number Acoustic Number Aggregate status record Periodicity Request record University of Virginia Power Management NEST ANSCD • Sentry Service • Tripwire • Rotation 3 Sentry 10mA@3v Base node 1 2 4 Non-Sentry University of Virginia Tripwire-based Surveillance NEST ANSCD • Partition sensor network into multiple sections. • Turn off all the nodes in dormant sections. • Apply sentry-based power management in tripwire sections • Periodically, sections rotate to balance energy. Road Dormant Active Dormant Active Dormant Dormant Active Dormant Active University of Virginia Estimation of Network Lifetime NEST ANSCD • Lifetime is determined by – Individual Mica 2 mote consumption • Energy plot for a sentry node • Energy plot for a sleep node University of Virginia Tripwire + Sentry NEST ANSCD Power Draw (Tripwire+SBPM vs SBPM) (Based on 10 events per day, 24/7 full Coverage ) Wakeup Surveillance Communication Tripwire+SBPM Event Process Sleep SBPM Initialization 0 0.5 1 1.5 Power Draw(mA@3v) One tripwire section out of every 4 sections with 10% sentry expected 142 days (20x) lifetime. University of Virginia Lifetime Analysis Network Life Time NEST ANSCD Number of Tripwires (10 regions, 30% sentry, 7 day life) 4 2 AA Batteries 50 days 4 AA Batteries 100 days 3 2 70 days 105 days 140 days 210 days 1 210 days 420 days University of Virginia Group Management NEST ANSCD IR Camera University of Virginia Group Management NEST ANSCD IR Camera University of Virginia Detection Delay NEST ANSCD DETECTION DELAY (S) CLASSIFICATI ON DELAY (S) VELOCITY DELAY (S) REPORTED VELOCITY (MPH) ACTUAL VELOCITY (MPH) 2.7 3.2 3.2 25.0/10.9 N/A 1.8 3.2 3.2 24.6 N/A 1.7 2.7 3.2 17.6 N/A 3.8 4.8 5.3 9.3 N/A 1.7 2.7 2.8 11.1 10 2.6 3.1 3.6 18.5 20 1.9 2.4 2.4 23.0 20 2.6 2.9 3.2 12.7 12 0.9 2.5 2.5 22.1 20 4.5 8.1 8.1 6.2 N/A University of Virginia 3-Tier Classification NEST ANSCD Base mote Group Report Performing base level classification Group Group leader, performing group level classification Group Normal mote, performing sensor (mote) level classification University of Virginia First Tier: Robust Sensing NEST ANSCD • PIR Sensing • Magnetic Sensing • Acoustic Sensing • Commonality: – Initial Threshold Calibration – Continuous Threshold Calibration with changing environment – Power & Frequency Filtering University of Virginia PIR Sensing Module (1) NEST ANSCD • The current PIR detection algorithm using XSM sensors can distinguish walking persons in a range of 12-20 ft in hot environments – About 19 ft/person running – About 12 ft/person walking • 30-40 ft in cool environments. • Almost all false alarms are reliably removed. • Radio interference has been also removed. University of Virginia PIR Sensing Module (2) NEST ANSCD • Environmental factors – Grass and Trees. – Temperature. – Wind and Sunshine. • Frequency Analysis – Uses high/low-pass filters to filter out noise, so that no false alarms are generated due to environmental effects. • Self-adaptive – Continuous filtering and calibration to adapt to environment. • Data sampling is turned off for 60 ms when there is radio transmission. University of Virginia PIR Sensing Module (3): Data This figure displays the raw data, the dynamic threshold, and the confidence of the detection. The detection report is based on frequency analysis of the signals and compared with an adaptively adjusted threshold. University NEST ANSCD of Virginia Magnetometer Sensing (1) NEST ANSCD • Requirement – Detect vehicles and persons with a weapon • Challenges – ADC reading may saturate – Response latency – Magnetic and electric noise from environment and mote circuitry – Thermal reading drift – Radio/Mag interference – Short range – XSM-2 has greater noise than XSM-1 University of Virginia Magnetometer Sensing (2) NEST ANSCD • Raw ADC reading can saturate • Translate the pair of POT/ADC values to a single scaled mag point • Moving average of recent scaled ADC readings. • Compare to difference between slow and fast moving average University of Virginia Magnetometer Sensing (3) NEST ANSCD Response time • Mag sensor chain needs about 40ms to settle. • ADC readings need about 50ms to settle after a potentiometer change. • The averaging algorithm needs at least 3 initial readings to perform computation. • A fast-detect logic speeds up detection of obvious signals University of Virginia Magnetometer Sensing (4) NEST ANSCD – Signal/noise ratio • Signals (Scaled ADC readings) are hard to distinguish for small targets or targets at far distances – Signals for iron bar moving at 5 ft. • Use a moving average of recent readings (Mag Points) to filter out noise. • Mag Points show signals whose amplitude is often lower than that of noise – Mag Points for iron bar moving at 5 ft. University of Virginia Acoustic Sensing (1) NEST ANSCD • Properties: – Power based approach. – Automatic and continuous calibration due to temperature fluctuations, noisy environments and individual sensor characteristics. – Differentiates between vehicles, humans, background noise and wind (collaboration with PIR sensors necessary). • Limitations: – No differentiation between small-big vehicles currently available. University of Virginia Acoustic Sensing (2) NEST ANSCD Three Cars Initial Calibration No Detection Detection when Energy Crosses Standard Deviation University of Virginia Acoustic Sensing (3) NEST ANSCD • Moving average curve plus 3 times the standard deviation curve = THR curve (called standard deviation on previous slide) • Count number of crossings of THR out of the last N readings and if percentage is greater than x% then this is a target – X is about 60% University of Virginia Second Tier: Group Aggregation NEST ANSCD DOA controls minimal aggregation degree to reduce false alarms Awareness Range Node Member Follower Leader Detection Range University of Virginia System Issues: False alarms NEST ANSCD Impact of DOA on False Alarms 0.7 • Probability of false positives reduces as DOA increases Probability of false alarms 0.6 0.5 • Probability of false negatives false positives increases as DOA increases false negatives 0.4 •With DOA = 3 we had zero false alarms 0.3 0.2 0.1 0 1 2 3 4 Degree of aggregation (DOA) 5 6 •The DOA parameter can be tuned based on sensing range and the density with which motes are deployed Spatial-temporal correlated data aggregation can effectively reduce false alarms University of Virginia Third Tier: Base Mote (1) NEST ANSCD • The base mote keeps received tracking messages in FLASH. • It then makes use of the spatio-temporal correlation to decide which target a tracking message belongs to. (e.g., 30 m and 5 sec) • When a specific target gets enough (according to a adjustable parameter) messages for one target, a “detection” report is sent from the base mote to the RSCC. University of Virginia Third Tier: Base Mote (2) NEST ANSCD • After the “detection” report is sent and enough information is gathered for classification, a “classification” report is sent from the base mote to RSCC. (2 additional reports beyond detection) • Afterwards, send reports according to an adjustable flow rate parameter. X Distance • The base mote also uses a least square calculation to calculate the velocity of the target. A “velocity” report is sent to RSCC. (5 additional reports beyond classification) Slope = X Velocity ( Least Square Estimation) University of Virginia Time Classification Scheme PIR Sensor Detection X Detection X[n] Detection X Detection X Freq. Analysis Group Size Magnetic Sensor X X X X Status Done Any Target Done Person Done X Person with Weapon Done X Vehicle Done Big/Small Vehicle Potenti al Big/Small Vehicle Potenti al Big/Small Vehicle Potenti al X X Target Type False Alarm by Wind X Detection Num Hits Acoustic Sensor NEST ANSCD X University of Virginia Detection/Classification/Velocity Delay NEST ANSCD DETECTION DELAY (S) CLASSIFICATIO N DELAY (S) VELOCITY DELAY (S) REPORTED VELOCITY (MPH) ACTUAL VELOCITY (MPH) 2.7 3.2 3.2 25.0/10.9 N/A 1.8 3.2 3.2 24.6 N/A 1.7 2.7 3.2 17.6 N/A 3.8 4.8 5.3 9.3 N/A 1.7 2.7 2.8 11.1 10 2.6 3.1 3.6 18.5 20 1.9 2.4 2.4 23.0 20 2.6 2.9 3.2 12.7 12 0.9 2.5 2.5 22.1 20 4.5 8.1 8.1 6.2 N/A University of Virginia Self-Healing (1) NEST ANSCD • Wide spectrum of capabilities • Not binary • In Routing – Multiple parents in backbone tree • • • • No cost for periodic probing Stealthiness is maintained Local decision on choosing alternative parent is fast Re-create n-parent tree on system rotation • In MAC – For unicast – retransmission of lost packet University of Virginia Self-Healing (2) NEST ANSCD • At Application Level – Critical messages are transmitted multiple times to better ensure delivery • In Sensing – Fail-stop – use of many sensors as targets move avoids problems here – Byzantine failure – detect node continuously reporting and shut it down • In Localization – If node fails to obtain location during walking GPS, it gets info from neighbors and uses tri-lateration University of Virginia Self-Healing (3) NEST ANSCD • In System Initialization – Each phase is coordinated and sequential – If a node is not in-step it becomes silent until next system rotation • In Tracking – If group leader fails, info is still with the members and is passed to next leader • In Wakeup – Decentralized and if some nodes fail to wake-up it is not a problem because many others will be awake University of Virginia Self-Healing (4) NEST ANSCD • Limited Effect – Clock sync, neighbor discovery, etc. are highly decentralized and local. Single node failures only affect that node and does not propagate to the rest of the network. • System Rotation – Can correct many issues – Currently, only executed based on time – Could be extended to re-run when many failures are detected BUT this means extra messages which affects lifetime and stealthiness! University of Virginia Other Middleware Services NEST ANSCD • System Initialization – List of system parameters • MAC • Routing • Asymmetric Detection • Localization – Walking GPS • Clock Sync • Velocity Calculation University of Virginia System Initialization NEST ANSCD • Place motes in field – turn on mote; get location via walking GPS • Turn on relay and base mote • Turn on RSCC • RSCC requests system parameters to relay • Relay asks base mote for parameters (from flash) • Base mote sends to relay and relay sends to RSCC • RSCC then asks each other base mote the same thing in turn University of Virginia System Initialization NEST ANSCD • RSCC then sends out Origin of Reference – broadcast to all relays • Each relay adds its location to location of RSCC and sends to base mote • RSCC broadcasts master clock – essentially a start message • Relay sends start signal to base mote • Base mote sends out parameters and then begin mote field initialization, e.g., clock sync, localization, etc. University of Virginia System Parameters NEST ANSCD • Multi-hop reconfiguration with tunable parameters Parameter Name Units Description of the Parameter Value GRID_X meter Controls the topology of the network under static localization scheme Sentry Range meter Controls disperse/density of the sentries. Power Mode N/A Controls the power consumption of the non-sentries SD Threshold 1% Threshold to decide whether a link is symmetry or not Pm TimeOut second The duration a non-sentry should remain awake after it is waken by sentry nodes FlowRate second Specifies the minimum periodicity with which the tracking updates PIR Threshold N/A Used to tune the sensitivity of the PIR sensors DetectionThreshold N/A The minimum number of reports accumulated before a basemote declares the detection Magnetic Threshold N/A Used to tune the sensitivity of the magnetic sensors Acoustic Threshold N/A Used to tune the sensitivity of acoustic sensor sensors shutDownThreshold 1% Used to shutdown chaos motes second Controls the duration of each phase to accommodate Delay The duration of the tracking phase = TrackingPhaseCount * Phase Delay Settings N/A Defines various kind of binary control Schedule N/A Defines tripwire sleep/awake schedule Phase Delay TrackingPhaseCount University of Virginia MAC: B-MAC NEST ANSCD • A derivative version of CSMA – Listen before send. Linear back off if channel is busy. • Support dynamic noise floor during carrier sense • Support MAC Layer reliability through 1 byte ACK • Support flexible back off scheme to meet requirement of application • Support lower power listening to trade off fast response University of Virginia Routing (1) NEST ANSCD Reliability in routing infrastructure – Asymmetric link detection – MAC level delivery failure detection – Routing layer retransmission – Multi-Parent diffusion tree – Local parent switch in case of failure 2 – Robust to base failure 1 5 A 3 4 B Local Switch Symmetric Link Detection 6 7 University of Virginia Routing (2) NEST ANSCD • Robust diffusion tree with asymmetry detection – – – – It requires no location information. It requires small portion of nodes awake. Small cost to maintain (1 byte ACK detection). It matches to multiple relay scenario. • Robust diffusion tree with local switch – Robust to failure of parent nodes – Stealthiness (no need to maintain route periodically) – It requires small portion of nodes to be awake. University of Virginia Asymmetric Detection NEST ANSCD • Neighbors perform discovery via beacons • Neighbors then also exchange neighbor tables • Node must hear from a neighbor node and be in that node’s table => symmetric link • If link is asymmetric – drop neighbor from neighbor table University of Virginia Walking GPS NEST ANSCD • GPS Mote assembly: – Garmin eTrex Legend GPS device (WAAS enabled) – MICA2 mote – helmet, RS232 cable, board, wristband – Memory size: 17 Kbytes (code), 600 Bytes (data) • Sensor Node: – Mica2, XSM – Memory: 1 Kbytes (code), data: 120 bytes University of Virginia Walking GPS NEST ANSCD • The sensor node deployer (soldier or vehicle) has a GPS Mote assembly attached to it. • The GPS Mote periodically beacons its location. • Sensor Motes that receive this beacon infer their location based on the information present in this beacon. • From the localization perspective, two distinct software components exist. GPS Mote Sensor Mote GPS Localization University of Virginia Walking GPS: Sensor Mote NEST ANSCD • Two deployment types: – mote powered on at deployment • first INIT_LOCALIZATION packet gives the location – mote powered on all the time • INIT_LOCALIZATION stored in circular buffer, if RSSI > Threshold • Choose best value • Two stages for Localization: – at deployment time: Walking GPS – during system initialization: HELP_REQUEST/REPLY, if no location information present (for robustness) University of Virginia Walking GPS Evaluation NEST ANSCD • First deployment type: sensor motes turned on at the place of deployment, right before being deployed • Localization error: 0.8 meters • Standard deviation: 0.5 meters • Second deployment type: sensor motes turned on all the time. • Localization error: 1.5 meters • Standard deviation: 0.8 meters University of Virginia Walking GPS Evaluation NEST ANSCD • Second deployment type using two GPS devices • Each line along the length of the grid deployed with a different GPS device • Localization error: 1.6 meters • Standard deviation: 0.9 meters University of Virginia Clock Sync (1) NEST ANSCD • A strip-down version of Vanderbilt TimeSyn to meet the requirement of ANSCD system – Normal crystal accuracy 10~50 PPM. Worst case drift 0.03~0.142 second/per day. Average drift is even less. – Enough for ANSCD requirement • Used in ANSCD for: – Velocity calculation – Phase transition – Timestamp events University of Virginia Clock Sync (2) 1. NEST ANSCD Root node accepts time from RSCC through MASTER CLOCK command. 2. Disseminate time through flooding. 3. Time stamping performed right before Timestamp is sent out to avoid un-predictability in MAC access delay 4. Abandon continuous clock drift calibration to achieve stealthiness in operation 5. Rotation to compensate for clock drift University of Virginia Velocity Calculation NEST ANSCD • Performed at base mote attached to relay • Messages are ordered via the timestamps • Wait for “n” messages before calculating velocity • Calculate x-comp and y-comp of velocity separately using least squares curve fitting University of Virginia Lessons Learned (1) NEST ANSCD • System-wide energy solution is needed – Include system init; communication; sensing; use one flooding for multiple purposes • Many links are asymmetric – use conservative communication range and an explicit asymmetric detection module • Timely Delivery of hardware is crucial – Unstable hardware version costs us significant effort on continuous tuning the sensing & classification algorithms University of Virginia Lessons Learned (2) NEST ANSCD • Higher bandwidths and more data memory • Re-send lost messages based on semantics of messages (at application level) – too expensive to re-send every lost packet at MAC layer • System would be better with higher densities • Sensing ranges need to be increased University of Virginia Remaining Work FY ‘05 NEST ANSCD • Robustness testing and performance evaluation – Large scale testing (1000 motes) • Aggressive Power Management • Further reduce false alarms • Classify accurately – Classify small-large vehicles • Air Drop Localization • Increased self-healing properties • Supporting field tests/demo – Full integration and testing with sophisticated sensors University of Virginia NEST ANSCD End University of Virginia