BFTPR

1
Performance Implication of Environmental Mobility
in Wireless Networks
Maneesh Varshney and Rajive Bagrodia Department
of Computer Science University of California, Los
Angeles
(Proc. of INFOCOM 2007)
2
Motivation Types of Mobility

• Nodal Mobility (NM)
• Movement of the transmitting and/or receiving
  nodes
• Environmental Mobility (EM)
• Ambient movement of entities such as people,
  vehicles etc. in the vicinity of wireless
  communication
• Communication nodes may or may not be static
• Examples
• customers move about in cafes and offices
• students enter and leave classrooms or
  conferences
• vehicles moving on a freeway
• animals moving in a forest sensor deployment

Nodal mobility has been widely researched
Environmental mobility has been relatively
unexplored
3
Focus
1) Effect of EM on wireless link behavior and
protocol performance 2) How to model EM? Is the
model efficient? Is it also simple? 3) Including
EM channel model in network simulation 4)
Helping network protocols become aware of the
underlying channel conditions
4
Talk Outline

• Four Parts
• 1) Measurement Results
• 2) Environmental Mobility (EM) Channel Model
• 3) Qualnet Simulations
• 4) Cross-layer Optimization Recovery from
  Earlier Good State (REGS)
• Conclusions
• Future Work

5
Measurement Results

• IBM Thinkpad T-42 laptops
• Orinoco Proxim WD-8470 b/g gold card operating
  in 802.11b mode and Linksys DWL-AG660 card
  operating in 802.11a mode
• The transmitter broadcasts traffic at the rate
  of 2000 packets/sec for 11a and 500 packets/sec
  for 11b mode
• The receiver captures the Received Signal
  Strength Indicator (RSSI) for each packet using
  ethereal software
• About 60 hours of channel traces at different
  locations on a university campus, during
  different times of the day and spread over
  multiple months
• Indoor scenario (the bottom figure)
• inside building corridor with classrooms on both
  sides
• Outdoor scenario (the top figure)
• an outdoor portion of an on-campus coffee shop

6
Measurement Results (contd)

• a) No mobility measurements done late at night
• b) Far mobility no mobility was allowed within
  about 10 meters of the radios
• c) Unrestricted mobility pedestrians were free
  to move without restriction

a) fairly static channel b) Multi-path fast
fading modeled well with Ricean distribution c)
The EM channel different from multi-path
fading - EM channel is a superimposition of low
frequency people shadowing (hundreds of ms.,
person(s) crossing a given line) on multipath
fast fading (hundreds of µs.)
7
Measurement Results (contd)
Average error burst duration for fast fading and
EM channels

• Avg. error burst duration for EM channels is
  significantly greater than that of fast fading
• Longer error bursts ? higher correlation in bit
  errors ? EM channel has higher memory

8
Measurement Results (contd)
Observation 1 Real life channels with pedestrian
mobility cannot be sufficiently captured by
multipath fast fading theory and models.
9
Measurement Results (contd)
Experiment 1 Shadowing loss due to motion of
multiple people

• Loss is maximum at the ends and is almost a
  constant value in the middle
• Fading value increases in the case of two people
• Fading value does not only depend on the number
  of people

10
Measurement Results (contd)
Observation 2 Shadowing loss in an EM channel
depends not only on relative location of people
with respect to transmitter and receiver but also
with respect to each other.
11
Measurement Results (contd)
Experiment 2 Effect of EM on prob. distribution
of signal strength

• People moving randomly close to Rx, and their
  number varies between 1 and 3
• Similar trends are observed when number of
  people is fixed but their speed varies

• No movement curve is conventional Ricean
  fading distribution
• For the other cases, the fading distribution
  exhibits a secondary peak
• Key observations
• Entire curve can be thought of as a sum of two
  distributions where mean value of second
  component is displaced and its relative weight is
  dependent on number of people
• Basis for the channel model EM Channel is
  modeled as alternating between two states
  corresponding to the two distributions

12
Measurement Results (contd)
Observation 3 The fading distribution, in
presence of EM, is distorted to exhibit a
secondary peak. The relative magnitude of the
second peak depends on the number of people and
their speed.
13
EM Channel Model
A two-state continuous-time Markov process

• The transition between the two states is
  governed by the dynamics of mobility of the
  people
• Transition rates between good and bad states
  were studied using Monte-Carlo simulations

(Unobstructed Link)
14
EM Channel Model (contd)
Average duration of good states

• A large terrain of size D x D
• ? density of the people
• ?D2 number of people, placed at random initial
  positions
• v the speed of movement, follows a random
  way-point model
• d distance between Tx and Rx
• a line segment of length d is placed in the
  middle of the terrain
• Consider the time instances of events where any
  person crosses this line segment
• Found that these events follow the Poisson
  distribution, implying that the durations when
  the channel in not shadowed is exponentially
  distributed
• Average duration of good states depend on ?, v,
  and d

15
EM Channel Model (contd)
Average duration of good states (contd)
16
EM Channel Model (contd)
Average duration of bad states

• Duration of shadowed channel state is also
  exponential
• Mean value determined experimentally to be 300
  ms.
• Prob. of more than one person crossing the link
  simultaneously
• Define ?i to be the difference in time instances
  between the ith and the (i1)th crossings
• If ?i lt (0.8 2W)/v (W is the width of the
  human body), assume that two people crossed the
  link simultaneously
• Procedure is recursively applied to derive the
  data for simultaneous crossings of more than two
  people

17
EM Channel Model (contd)
Additional loss due to blocking (Ldiffraction) in
the bad state

• Determined according to the Three Knife-edge
  diffraction model
• M. Varshney, Z. Ji, M. Takai and R. Bagrodia,
  Modeling Environmental Mobility and its Effect
  on Network Protocol Stack, Proc. of WCNC 2006.

18
EM Channel Model (contd)
Brief overview of the diffraction model Varshney
et al. 2006

• Radius of curvature at edges of human body are
  negligible when compared with communication range
  ? knife-edge
• Fading due to one person
• a three knife-edge model head, left arm and
  right arm of a person
• Fading due to multiple people
• Use the Epstein-Peterson model Parsons01 of
  multiple knife-edge diffraction
• E.g., for the case of two obstacles, the two
  losses are added and an extra correction term is
  added to the sum
• Model extension to account for various
  combinations of paths that the waves may take to
  reach the receiver
• E.g., in Figure (b), there are 5 paths from Tx
  to Rx, each encountering 2 diffraction edges.
  Components are added to produce the total loss

19
EM Channel Model Validation
Experiment 1 (Revisited)
20
EM Channel Model Validation (contd)
Experiment 2 (Revisited)
Note Fraction of time that channel was in good
state was taken as 0.75 and 0.65 respectively
21
Qualnet Simulation
Objectives - Effect of EM channel behavior on
classes of protocols that maintain state via
feedback from the channel - Is it possible that
the link can switch to a good state while the
protocol is still operating under the conditions
of the bad state recorded in the recent past? -
If yes, then protocols are NOT able to adequately
utilize the good state ? performance
degradation - How to remedy that? How effective
is the solution?
22
Qualnet Simulation (contd)
MAC layer data rate adaptation

• Interaction of SampleRate Bicket05 with the
  EM 2-state Markov channel model
• One transmitter sending a backlogged UDP traffic
  over 802.11a link to a single receiver
• Tx-Rx separation varies 50-300m. Pathloss Two
  Ray
• Expected performance (throughput)
• f fraction of time that the channel remains in
  the good state
• Good, Bad and EM respectively refer to exclusive
  good state, exclusive bad state and EM channel
  (alternation of previous two)
• Rationale Any deviation from this expected
  value implies that the protocol maintains some
  memory of past channel conditions which
  prevents it from utilizing current conditions
  efficiently

23
Qualnet Simulation (contd)
MAC layer data rate adaptation (contd)
as high as 50 performance degradation ? due to
the memory effect
24
Qualnet Simulation (contd)
Brief overview of SampleRate Bicket 2005

• State maintained for each data rate
• whether four consecutive losses have occurred
• average packet transmission time (using the
  packet length, bit-rate, and the number of
  retries (including 802.11 back-off)).
• An averaging window (10 sec.) is used to avoid
  stale info.
• After a packet has been transmitted
• based on information from the device driver,
  update the average transmission time of the
  current data rate and record if 4 consecutive
  losses have occurred
• After every 10th packet
• switch randomly to one of the remaining data
  rates that have lower lossless transmission time
  than the average transmission time of the current
  data rate

25
Qualnet Simulation
bad states
MAC layer data rate adaptation (contd)

• After 4 consecutive losses, the 36Mbps will not
  be used for the next 10 seconds!
• Avg. Tx time for the 24Mbps increases (due to
  large number of retries), but it is not able to
  reset fast enough
• Protocol ends up using the lower 18Mbps rate ?

26
Qualnet Simulation (contd)
Observation 4 Protocols that maintain state can
acquire negative information in the bad state
which is discarded only very slowly or not at all
even when the channel conditions improve. The
effect of this phenomenon is that the protocols
are not able to utilize adequately the durations
when the channel conditions are good.
27
Cross Layer Optimization
Recovery from Earlier Good State (REGS)

• Basic Idea
• Forgetting the bad memories of the bad state
• Operation in the good state is not influenced by
  decisions made or state variables changes in the
  bad state ? performance is not degraded in the
  good state
• Cross-layer Operation
• LL monitors incoming packets and infers current
  channel state from RSS
• maintains smoothed trace of signal strength
• utilizes Hidden Markov Model training and is
  able to predict state transitions without false
  positives or negatives and with a latency of less
  than 5 ms.
• LL announces this information. Higher layers
  subscribe to these announcements
• Switch from a good state to a bad state
• A REGS-enabled protocol checkpoints the current
  protocol state
• Switch from a bad state to a good state
• A REGS-enabled protocol discards all the
  protocol memory acquired since the previous
  announcement and starts to operate from the
  earlier checkpointed state

28
Cross Layer Optimization (contd)
Application of REGS to SampleRate

• Switch from a good state to a bad state
• REGS-enabled SampleRate marks a checkpoint at
  the current index in the history
• Switch from a bad state to a good state
• REGS-enabled SampleRate goes back in the history
  and erases everything until it encounters the
  marked checkpoint
• Statistics such as average transmission are then
  recalculated

29
Cross Layer Optimization (contd)
Qualnet Simulation of SampleRate (Revisited)
Performance of REGS-enabled SampleRate is close
to expected performance ? Memory acquired during
the bad state was the cause of performance
degradation.
30
Conclusions

• A systematic study to analyze the effect of EM
  on wireless link behavior and protocol
  performance using
• measurements channel traces
• analysis model two-state Markov Process, simple
  but efficient model
• simulations thanks to its simplicity and
  efficiency, the model was incorporated into
  the Qualnet network simulator
• Several case studies at different layers of the
  protocol stack
• MAC layer data rate adaptation in a wireless
  network
• Routing (AODV and DSR) in ad-hoc networks
• TCP flow in wired-in-wireless path of a mixed
  network
• See http//pcl.cs.ucla.edu/papers/ for more
  details
• A cross-layer optimization scheme, called
  Recovery from Earlier Good State (REGS)
• REGS allows protocols to become aware of the
  underlying channel conditions

31
Future Work

• Role of nodal mobility in relation with
  environmental mobility
• nodes themselves are moving and can shadow other
  communications
• resulting impact on protocols at the different
  layers of the protocol stack
• Impact of shadowing by objects with dimensions
  larger than humans
• vehicles in Vehicular Ad Hoc Networks (VANETs)