Group Seminar on 01272006

1
Group Seminar on 01/27/2006

• Presented by Jennifer C. Hou

2
Problem Considered

• How to detect the misbehavior of nodes in terms
  of violating back-off timers associated with IEEE
  802.11 MAC.
• Misbehaving nodes may choose a shorter back-off
  time or follow a different back-off/retransmission
  mechanism.
• Nice features to have
• Without a trusted centralized AP in ad hoc
  environments
• Minimal impact to the existing IEEE 802.11
  standard (minor change to 802.11
  hardware/firmware)
• As little information exchange as possible
• Taking into account of interference/sensing range
  of a node in analyzing and estimating the system
  state

3
A Glimpse at Proposed Solution

• Each node R monitors its neighbor(s) with respect
  to the sequence of back-off timer values they
  will have to follow, based on the system state
  of neighbor nodes.
• Procedure
• S1 Each node announces its pseudo-random
  sequence generator.
• S2 Every (RTS/data) packet sent by the sender S
  includes
• Sequence-offset number (SeqOff)
• Attempt number (Attempt)
• A message digest of the corresponding DATA packet
  (MD)

4
A Glimpse at Proposed Solution

• Procedure
• S3 From the information provided in S2, node R
  can infer the dictated sequence of back-off timer
  values.
• S4 Node R also keeps track of the system state
  (the number of busy (B) and idle (I) slots in a
  period of N observed slots), and infers the
  estimated sequence of back-off timer values. ?
  Need further discussion

5
A Glimpse at Proposed Solution

• Procedure
• S5 Because of hidden terminal problems, node R
  may not be able to deterministically determine if
  node S uses a legitimate back-off countdown
  process.
• The sequence of observed back-off timer values
  differs from the sequence of estimated ones ?
  Need further discussion
• S6 Node R uses hypothesis testing (Wilcoxon rank
  sum test) to determine if node S cheats. ? Need
  further discussion

6
Estimating Back-off Timers

• Node R continuously keeps track of the number of
  busy (B) and idle (I) slots in a chosen period of
  N observed slots (the sample interval).
• Node R deals with the hidden node problem with
  probabilistic analysis

• PI/I is the probability that sender S senses an
  idle channel, given that node R senses an idle
  channel as well.

7
Estimating Back-off Timers

• Under assumptions
• At any one time, at most one node is involved in
  transmission, if the channel is sensed busy.
• Nodes are uniformly distributed in the area of
  interest

m
n
j
k

• where r is the traffic intensity and is

0 if the node senses the channel idle in the
ith slot 1 if the node sense the channel to be
busy in the ith slot

• r is updated with the ARMA filter

Sample size
8
Estimating Back-off Timers

• The number of nodes in each area is calculated as
• where nT is the number of active nodes in the
  neighborhood, and is calculated using Bianchis
  method.

m
j
n
k
9
Estimating Back-off Timers

• Problems
• 1. How to determine Ai is not mentioned.
• Ai depends on the relative positions of R, S, T,
  and its calculation is expensive.

10
Statistical Test for Determining Uncertainty

• Recall
• The sequence of observed back-off timer values
  differs from the sequence of estimated ones Node
  R uses hypothesis testing (Wilcoxon rank sum
  test) to determine if node S cheats.
• Formulation of statistical test Given
• x the sequential population of the dictated
  back-off timers of node S
• Y the sequential population of estimated
  back-off timer
• Determine
• H0 S is well behaved.
• H1 S is malicious.

11
Statistical Test for Determining Uncertainty

• All the data from both populations are ranked,
  with the smallest value assigned rank 1, the
  second smallest value assigned rank 2, and so on.
• The ranks for each of the groups are added
  together, and the rank sums are then compared by
  means of available tables.
• Depending on the proximity of the rank sums, the
  tables yield a significance probability p. The
  smaller the value of p, the more unlikely H0 is
  true.

12
Validation of Derivation
13
Performance Evaluation
A malicious node has an associated percentage of
misbehavior m, if it transmits a packet after
counting down to (100-m) of the dictated backoff
value.
14
Critique

• Calculation of PB/I and PI/B is approximate and
  relies on several assumptions.
• How many neighbors does each node have to
  monitor?
• Globel states B and I.
• Per-neighbor states are Ai, pseudo-random
  sequence, Attempt, SeqOff.
• As the state to be kept and the calculation to be
  performed is expensive for each neighbor, can we
  devise a scheme so that each node is responsible
  for a small number of neighbors, and each node is
  at least monitored by a pre-determined number of
  nodes.
• essentially
  takes the mean path. Can we build a probability
  tree, so that the probability distribution of the
  estimated sequence can be derived and hypothesis
  testing can be used?