Sensing Location

1
Sensing Location
2
References

• P. Bahl, V. Padmanabhan, "RADAR An In-Building
  RF-based User Location and Tracking System" IEEE
  INFOCOM 2000, vol. 2, pp. 775-784.
• Nissanka B. Priyantha, Anit Chakraborty and Hari
  Balakrishnan, " The Cricket Location-Support
  System " Proc. 6th ACM MOBICOM, A ugust 2000, pp.
  32-43.
• Andy Hopper, Pete Steggles, Andy Ward, Paul
  Webster, " The Anatomy of a Context-Aware Applica
  tion " Proceedings of the 5th Annual ACM/IEEE
  International Conference on Mobile Computing and
  Networking (Mobicom '99), Seattle, Washington,
  USA, August 1999.
• Special Notes Special thanks goes to MIT for a
  presentation that has great pictures.

3
Introduction

• The proliferation of mobile computing devices and
  LANs has fostered a growing interesting in
  location-aware systems and services.
• In these systems the application information
  and/or interface presented to the user is a
  function of their physical location.

4
Introduction

• Granularity of location information needed varies
  from one application to another
• The nearest printer has a coarser granularity
  then finding the location of a book in the
  library.
• Possible Applications
• Navigation
• Lost child
• Resource discovery

5
Introduction

• A context-aware system uses in addition to
  location information, information about the user,
  state of the physical environment (e.g.,
  temperature), state of the computing system,
  history human-computer interaction, etc

6
Introduction

• Location is the most heavily used context.
• What about other context?
• Not used that heavily.
• Why? Possible reasons
• Not useful
• Not easy to sense
• For now, we will focus on location.

7
Sensing Location

• For outdoor use, we have the Global Positioning
  System (GPS).
• GPS basics
• GPS determines the distance by measuring the time
  it takes a signal to propagate from satellite to
  receiver
• Need to have very good synchronization of clocks
• Receive signal from three satellites to determine
  location
• Need a fourth satellite to estimate elevation
• Satellite GPS accuracy is getting reasonable
  (10-20 meters)

8
Sensing Location

• GPS doesnt work indoors because the satellite
  signal is weak or reflected which means lowers
  accuracy.
• Indoor location systems is an active research
  area.
• Ideal location sensor in indoor environments have
  the following properties
• Provide fine-grain spatial information at a high
  update rate.
• Unobtrusive, cheap, scalable and robust.

9
RADAR

• RADAR attempts to use common off-the-shelf
  components. For example, 802.11b base stations.
  Basically, RADAR makes use of WLAN technology.
• RADAR assumes that the access points (AP)s
  provide overlapping coverage over area of
  interest.
• The user carries a mobile device which helps in
  determining location e.g. laptop, palmtop, badge.

10
RADAR Method for Location Sensing

• RADAR records information about the radio signal
  as a function of the users location.
• Off-line Analysis Use signal information to
  construct and validate models for signal
  propagation during off-line analysis.
• Real-time Analysis Infer the location of a user
  in real time.

11
RADAR Method for Location Sensing

• Need to be able to measure signal strength (SS)
  and the signal-to-noise ratio (SNR) (later
  experiments showed that SNR didnt seem to matter
  much).
• A base station (bs) records the signal strength
  (ss) measurement with a time stamp. Basically,
  it is recording (t,bs,ss).
• A driver on the mobile host extracts the signal
  strength and signal-to-noise information from the
  network interface card. This can be then be made
  available to an application.

12
RADAR Method for Location Sensing

• During the off-line phase, the user indicates
  his/her current location by clicking on a map of
  the floor.
• The users coordinates (x,y) and timestamp t are
  recorded.
• Users orientation is also important
• There is strong signal strength if there is
  direct line-of-sight to a base stations antenna.
• In the opposite orientation, a persons body may
  form an obstruction.
• This implies that the direction, d
  (north,south,east,west) should also be recorded.
• Information collected by the mobile host is
  denoted by (t,x,y,d).
• Clocks on the mobile host and the base stations
  must be synchronized.

13
RADAR Method For Location Sensing

• A radio map of building is created
• A radio map is a set of signal strength tuples
  collected at various points in the building
• An entry will look like
• (x, y, d, ss1.n)
• For a basic system of three APs an entry would
  look like this
• (x, y, d, ss1, ss2, ss3)

14
RADAR Method for Location Sensing

• The map of the building is used for on-line
  analysis.
• Given a set of signal strength measurements at
  each of the base stations, the location that best
  matches the observed signal strength data (from
  the off-line analysis phase) is determined.
• This is an example of a multi-dimensional search.
  There is a good deal of database research that
  describes data structures and algorithms for such
  searches for exact as well as closest matches.
• The RADAR prototype used a linear-time search
  algorithm.

15
RADAR Method for Location Sensing

• One technique for searching is the Nearest
  Neighbor in signal space (NNSS).
• Compute distance between the observed set of SS
  measurements ( ss1, ss2, ss3) and the recorded
  set ( ss1, ss2, ss3) at a fixed set of
  locations.
• Pick location that minimizes the distance.
• RADAR uses the Euclidean distance measure
• sqrt((ss1- ss1)2 (ss2- ss2)2 (ss3- ss3)2 )
• Another distance measure is the sum of the
  absolute differences for each base station
  (Manhatten).

16
RADAR Method for Location Sensing

• Instead of searching for the nearest neighbor it
  may be preferable to take the average of physical
  locations the N nearest neighbors. This will be
  referred to as AVG-NNSS
• The approach just described is better than the
  following approach
• Users location is determined to be the same as
  the location of the strongest signal.

17
RADAR Testbeds used in Experiments

• Two test beds were designed and deployed
• Both used different wireless hardware
• First Test Bed
• Second floor of a three story building
• Three APs cover entire floor
• During the off-line phase, signal strength
  information was collected in each of the 4
  directions at 70 distinct physical locations on
  the floor.
• For each combination of (x,y,d), 20 signal
  strength samples were collected.

18
(No Transcript)
19
RADAR Testbeds used in Experiments

• Second Test Bed
• Second floor of a four story building
• Five wall-mounted APs provide wireless coverage.
• The paper focuses on the first test bed.

20
(No Transcript)
21
RADAR Observations

• AVG-NNSS shows some benefit over using NNSS but
  was not considered very significant.
• Averaging over a large number of nearest
  neighbors accuracy degrades rapidly because
  points far removed from the true location also
  are included in the averaging procedure.
• Why arent the benefits high Often the nearest
  neighbors in signal space are not physically
  distinct as the result of same (x,y) coordinate
  but different direction.

22
RADAR Observations

• The more data points the more accurate the
  location determination.
• It is interesting to note though that the
  accuracy of the location determination is not
  that different between 40 and 70 points.
• Only a small number of samples is needed.
• Users orientation has a significant impact on
  the signal strength measured at the base
  stations.
• Tracking is possible if the user walks at a
  uniform pace.

23
RADAR Comments

• The approach just described seems to estimate
  user location with a good degree of accuracy (up
  to 2 to 3 meters).
• Problem Lots of effort is needed to collect
  samples.

24
RADAR Alternative Approach

• Use radio propagation
• Develop a mathematical model of indoor signal
  propagation to generate a set of
  theoretically-computed signal strength data which
  is akin to the empirically generated data.
• Apply the NNSS algorithm
• The empirical approach works better when smaller
  error distances are needed.

25
RADAR Comments

• RF is hostile
• Signal propagation dominated by reflection,
  diffractions and scattering of radio waves
• Multi path fading phenomenon occurs
• Number of people affects signal strength
• Suggestion Make use of multiple maps
  corresponding to different environmental
  conditions.

26
RADAR Comments

• Mobile node hears all APs
• APs operate at different frequencies.
• Mobile node need to scan all frequencies
• Potentially could cause quite a bit of overhead.

27
Cricket

• The goal of Cricket is to allow applications
  running on user devices and service nodes (both
  mobile and non-mobile) to learn their physical
  location.
• It is the decision of applications to determine
  who should receive the location information.
• Goals include preserving user privacy (by not
  having user tracking) and decentralization.
• Cricket uses a combination of RF and ultrasound.

28
Cricket System Architecture

• Beacons disseminate information about a
  geographic space to listeners.
• A beacon is a small device attached to some
  location within the geographic space it
  advertises.
• Typically, it is obtained by the owner of the
  location
• To obtain information about the space, every
  mobile and static node has a listener attached to
  it.
• Listener infers its its current location from the
  set of beacons it hears and informs the device
  software about this via the API.

29
Cricket Determining the Location

• Use a combination of RF and ultrasound hardware.
• Speed of ultrasound signal is much smaller than
  RF.
• On each transmission, a beacon concurrently sends
  information about the space over RF, together
  with an ultrasound pulse.
• When the listener hears the RF signal, it uses
  the first few bits as training information and
  then turns on its ultrasonic receiver to listen
  for the ultrasonic pulse.
• The listener uses the time difference between the
  receipt of the first bit of RF information and
  the ultrasonic signal to determine the distance
  to the beacon.
• A time gap of x roughly corresponds to a
  distance of x feet from beacon

30
Machinery
B
Beacons on ceiling
SPACENE43-510 ID34 COORD146 272 MOREINFO
http//cricket.lcs.mit.edu/
Cricket listener
Mobile device
Mobile device
31
Cricket Reducing Interference

• RF transmissions from different beacons may
  collide.
• May cause a listener to wrongly correlate the RF
  data of one beacon with the ultrasonic signal of
  another, yielding false results.
• Ultrasonic reception suffers from severe
  multipath effects caused by reflections from
  walls and other objects.

32
Cricket Reducing Interference
Beacon A
Beacon B
Incorrect distance
t
Listener
RF B
RF A
US B
US A
33
Cricket Reducing Interference

• Cricket does not implement a full-fledged
  carrier-sense-style channel-access protocol to
  avoid collisions.
• Does not use use a fixed or deterministic
  transmission schedule.
• It uses randomization
• Transmission times are chosen randomly with a
  uniform distribution with an interval R1,R2 ms.
• Choice of random interval is governed by the
  number of beacons we typically expect will be
  within range of each other and the time it takes
  for the transmitted information to reach the
  listeners.
• Lower transmission frequency implies a longer
  amount of time to determine location.
• Higher transmission frequency implies collisions.

34
Cricket Reducing Interference

• S size of space advertisement
• b RF bit rate
• maximum propagation time for an ultrasonic
• signal in air between beacon and listener

S
?
b
Implies that any potentially correlated
ultrasound pulse bust arrive while an RF message
is being received.
35
Cricket Reducing Interference

• Envelop ultrasound by RF
• Interfering ultrasound causes RF signals to
  collide
• Listener does a block parity error check
• The reading is discarded.
• The randomized beacon transmission is used to
  prevent repeated occurrences of interference.
• Listeners do not simply use the first sample pair
  they get to infer their best location they
  collect multiple samples and then use an
  inference algorithm.

36
Cricket Beacon Position Inference

• Three algorithms can be used to determine closest
  beacon.
• Majority Picks the beacon with the highest
  frequency of occurrence in the data set.
• MinMean Calculates the mean distance from each
  unique beacon for the set of data points within
  the data set Select beacon with minimum mean
  distance.
• MinMode Compute the per-beacon statistical modes
  over the past n samples select beacon with
  minimum mode (found to be the best).

37
Cricket Beacon Position Inference
The listener can calculate distance from Beacon A
to Beacon B The listener knows coordinates of
Beacon A and Beacon B The listener can now
calculate its coordinates
38
Cricket Beacon Position Inference

• Its actually a bit more complicated since the
  user may not be standing still.
• Same principles apply but the geometry gets more
  complicated.

39
Cricket Beacon Positioning and Configuration

• Positioning of a beacon in a room is important
• Consider the positioning shown in the figure on
  the next page.
• Although receiver is in Room A, the listener
  finds the beacon in Room B to be closer.
• Solution Beacons should be placed at a fixed
  distance away from the boundary marking the two
  spaces.

40
Cricket Beacon Positioning and Configuration
Room A
Room B
I am at B
41
Cricket v1 Prototype
RF module (rcv)
RF module (xmit)
Ultrasonic sensor
Ultrasonic sensor
RF antenna
Listener
Beacon
Atmel processor
RS232 i/f
Host software libraries in Java Linux daemon
(in C) for Oxygen BackPaq handhelds Several apps
42
Deployment
43
Experimental Results

• Cricket units were able to correctly identify the
  room in which they were located in over 95 of
  cases when stationary.
• Achieve a location granularity of 4x4 feet.

44
Cricket Comments

• Provides privacy
• Higher power consumption due to localized
  computations

45
Active Bat System Bat Unit

• Radio transceiver, controlling logic and an
  ultrasonic transducer.
• Each bat has a globally unique identifier.

46
Active Bat System Ultrasound Receiver Units

• Placed at known points on the ceiling of the
  rooms to be instrumented.
• Receivers are connected by a wired daisy-chain
  network.

47
Active Bat System Base Station

• Periodically transmits a radio message containing
  a single identifier (corresponds to a Bat unit).
• This causes the corresponding Bat to emit a short
  unencoded pulse of ultrasound.
• Receivers monitor the incoming ultrasound and
  record the time of arrival for any bat signal.

48
Active Bat System Base Station (continued)

• Base station transmits at the beginning of a
  timeslot. Timeslots are long enough so that
  receivers do not get confused.
• It takes 20 ms between bat readings 50
  timeslots per base station per second
• Location can be used to measure orientation
• Attach many bats to the same object. Use the
  measurements to infer the orientation
• Base station can provide Location
  Quality-of-Service(LQoS) to allocate time slots
  to bats based on the expected update frequency
• Bats carried by people few times a second
• Bats attached to workstation once every few
  minutes

49
Active Bat System Calculating Location

• Using the speed of sound in air, the times of
  flight of the ultrasound pulse from the Bat to
  receivers can be converted into corresponding
  Bat-Receiver distances.
• If distances from the Bat to three or more
  non-collinear receivers can be found, its
  position may determined.

50
Active Bat System Scalability

• LQoS allows for a more efficient distribution of
  timeslots for a set of Bats
• Scheduling is dynamic
• A person is monitored a few times a second
• A workstation may be monitored once every few
  minutes
• A workstation may be monitored more frequently if
  a person walks up to it.
• Scheduling can be used for power saving
• If a base station knows that an object will not
  be located for some time, it can command that
  Bats associated with that object to temporarily
  enter a low power sleep state.

51
Active Bat System Scalability

• Set of Bats to be tracked will change over time.
• If a base station sees no indication from
  receivers that a Bat has responded in its
  timeslots, then it is assumed that the Bat has
  left the operating space.
• When a Bat enters a space, it senses the base
  station when the base station is broadcasting. It
  sends a registration message (we are assuming an
  Aloha protocol).

52
Active Bat System Scalability

• Bats perform handover when moving from one base
  station to another (similar to the cellular
  networks)
• Hand off decisions can also be made based on the
  Bat location
• Battery consumption is low, power consumed
  depends on the update frequency and power state

53
Active Bat System Experiments

• Test Environment
• Two rooms and corridor
• Two base stations and 100 receivers to cover
  approximately 280m3.

54
Active Bat System Experiments

• In 100,000 measurements, 95 of readings had
  errors of less than 9cm.
• 15 degree error in 90 of measurements with a 22
  cm separation between the Bats.

55
General Discussion

• Which one of these approaches is better?
• Difficult to compare error rate.
• RF is not robust ultrasound systems are better
  but only if ceiling mounted.
• Lots of start-up cost with Active Bats same with
  Cricket but the beacons are independent in
  Cricket.
• RADAR is relatively inexpensive in terms of
  hardware but extremely time-consuming to do
  calibiration.
• RADAR needs network cards.

56
General Discussion

• All of the techniques discussed are based on a
  cellular approach. This does not have to be the
  case.
• Biometric approaches possible
• SMART FLOOR project at Georgia tech tries to
  identify persons by their footstep force
  profiles.
• Claim 90 accurate
• Unobtrusive
• Works only for people and not things.

57
General Discussion

• Cameras can also be used to track user location.
• These systems have line of sight problems such as
  IR and so far only work well with a small number
  of persons in a room.

58
General Discussion

• Cricket is decentralized Active Bats and RADAR
  are not (although RADAR could be made more
  decentralized).
• Ability of location systems to scale
  geographically is dominated by installation costs.