Sensor Network-Based Countersniper System

1
Sensor Network-Based Countersniper System

• Gyula S, Gyorgy B, Gabor P, Miklos M, Branislav
  K, Janos S, Akos L, Andras N, Ken F
• Presenter
• Yamuna Krishnamurthy
• 2/7/05

2
Talk Outline

• Problem/Solution
• PinPtr
• System Architecture
• Middleware Services
• Time Synchronization
• Message Routing
• Sensor Localization
• Signal Detection
• Sensor Fusion
• Consistency Function
• Search Algorithm
• Performance Results
• Future Work and Conclusion
• Limitations/Discussion

3
Problem/Solution

• Problem
• Locate snipers in urban environments
• Work with constraints of the urban environment
• Multipath effects
• Poor coverage due to shading effect of buildings
• Overcome limitations of existing systems
• Require direct line of sight
• Rely on muzzle flash that can be suppressed
• Centralized system not tolerant to sensor failure
• Cost effectiveness

• Solution
• Use an ad-hoc wireless sensor network-based
  system
• Utilize many cheap sensors for
• good coverage of direct signal
• tolerance to failures
• Detect via acoustic signals like muzzle blasts
  and shockwaves

4
PinPtr - System Architecture

• Ad-hoc wireless network of inexpensive sensors
• Sensors can
• detect muzzle blasts and acoustic shockwaves
• Measure their time of arrival (TOA)
• Message routing service delivers TOA to a base
  station
• User Interface through base stations or PDAs
• System field tested at the US Army McKenna MOUT
  (Military Operations in Urban Terrain) facility
  at Fort Nenning, GA

5
PinPtr System Architecture
6
Custom Sensor Board and Mica2 Mote
PinPtr Application
Middleware

• Time synchronization
• Message routing
• Data aggregation

Operating System

• Tiny OS (UC Berkeley)
• Task scheduling
• Radio communication
• Clocks and timers

• I/O
• Power management

Hardware

• Mica Mote
• Microcontroller
• Multichannel receiver
• 4KB RAM, 128KB Mem
• Extension Interface

• Acoustic Sensor Board
• 3 acouastic channels
• FPGA
• Signal Processing
• Measure TOA

7
Middleware Services (MS)
8
MS - Time Synchronization

• Flooding Time Synchronization Protocol (FTSP)
• Synchronize local clock to clock of selected root
  node
• Time stamping broadcasted radio message multiple
  times at sender and receiver nodes
• Time stamps made when sending/receiving
  individual bytes
• Reduce uncertainties of encoding/decoding and
  interrupt handling times
• Final error corrected value embedded into message
  before end of transmission
• Estimate global time by synchronizing with nodes
  a level above
• Less communication overhead

9
MS - Time Synchronization

• Alternate algorithm
• Power conservation
• Does not require continuous radio broadcast to
  synch time
• Use data packets for time synch
• Each node adds an age (prev age) (time pkt
  resent time pkt received)
• Time of event T(recv) - age

10
MS - Message Routing

• Gradient-Based best effort converge-cast protocol
• Assign a root node
• Route data from all nodes to the root node
• Each node rebroadcasts data packets upto 3 times
• Data packets reach the root through multiple
  paths
• Fast and robust
• Does not guarantee message delivery
• Has significant message overhead

11
PinPtr System Architecture
12
MS - Sensor Localization

• Self-Localization procedure using ranging
  procedure
• Ranging procedure
• Broadcast a radio message followed by multiple
  chirps
• Destination node samples each chirp by streaming
  microphone
• Adds the samples together to increase signal to
  noise ratio
• After recording, a digital band pass filter and
  peak detector estimate start of the first chirp
  
• Range is computed using time of flight of the
  chirp
• Assign unique time slots to adjacent nodes within
  two hops radius
• In each time slot acoustic ranging procedure is
  initiated between the node and its neighbors
• Measurements are propagated to the base
• Base estimates the relative position of node to
  known anchor points through optimization
  procedures
• Advantage
• Reconfigure dynamically when sensors fail and new
  sensors are added
• Disadvantages
• Requirement of all nodes having 4 neighbors
  within 10mts range is not practical
• Sounder makes sensors larger and consumes more
  power
• Audible frequency of sounder makes detection by
  adversary easy
• Ultrasonic sounders have lesser range

13
MS - Sensor Localization Cont.

• Passive acoustic sensor localization
• Use external acoustic sources
• In sniper scenario estimate sensor location
  through shots rather than sniper location with
  sensors
• Produce 6 shots at known locations at unknown
  times and locate 4 sensors using linearization
• Requires sensor to be in direct line of sight
• Sensitive to small individual measurement errors
• Non- analytic approach too slow
• Produce shots at known locations and known times
• Same as the active acoustic method
• PinPtr Methodology
• Due to shortcomings of above mentioned
  localization procedures PinPtr currently uses
  sensors at known locations.

14
Signal Detection

• Incoming acoustic signal samples at 1MHz
• Compressed using Zero Crossing (ZC) coding
• Interval between ZCs is coded by storing
• Start time of the interval (T)
• Length of the interval (L)
• Minimum or maximum signal value (Mm)
• Previous signal avg amplitude (P)
• Rise time (G)

• Detect muzzle blast patters in the coded stream
• Modeled with state machine states IDLE,
  POSSIBLE_START, DETECTED
• When DETECTED start time at POSSIBLE_START is
  returned as TOA
• Mote transmits TOA to base stations

15
Sensor Fusion

• Converge on the actual shooter position from
  positions calculated with TOA
• Consistency Function using 3D space and time
• Known values
• Location (xi,yi,zi) of sensor making i th
  measurement
• ti the time of arrival of detected muzzle blast
• CG(x,y,z,t) count (ti(x,y,z,t)-ti lt G)
• i1,K,N
• Search algorithm for convergence
• Generalized bisection based on interval
  arithmetic
• Create 4-dimensional spaces with the intervals
  xmin,xmaxx ymin,ymaxx zmin,zmaxx
  tmin,tmax
• Determine CG for each of the spaces
• Select the area with CGmax
• Bisect the area along the longest dimension and
  repeat from 1-4 till maximum region is less than
  vG/2 for space and G/2 for time
• Shooter position falls in this max area

16
Performance Results

• Experimental setup
• 56 motes
• 20 different known shooter positions were used
• 171 shots were fired

17
Performance Results

• Shooter localization error
• Elevation info eliminated for 2D
• 3D errors are more as sensors were mostly
  positioned on the ground

• Error Sources
• Time Synchronization errors
• Sensor localization errors

18
Performance Results

• Sensor Density
• Effects signal detection
• Increases shooter localization error

19
Performance Results

• Sensor Fusion
• Compare analytic solution to the fusion algorithm
• Analytic solution compares to fusion algorithm
  when no error readings are included
• With error readings fusion algorithm provides a
  much better solution

20
Future Work and Conclusion

• Future Work
• Provide power management
• Use shockwave signal
• Support multiple shots with sensor fusion
  algorithm
• Use post-facto time synchronization to conserve
  power
• Use system in other Concepts of Operations
• Reconnaissance missions
• Protect convoy routes
• Work on sensor self-localization techniques
• Conclusion
• PinPtr provides a counter-sniper system
• Provides efficient algorithms for time
  synchronization and shooter/sensor localization
• Good experiment to reassure actual deployment

21
Limitations/Discussion

• Since PinPtr does not employ shockwave signals
  and relies on muzzle blast it may not work when
  silencers are used
• Deployment of sensors in actual urban environment
  like NY is not trivial
• Does not provide for power conservation hence
  battery life can be an issue since these systems
  typically need to be available always
• Cannot deal with multiple shots fired by multiple
  snipers
• No self-localization performed hence cannot
  dynamically configure with change in number of
  sensors

22
Thank You