Dialog Codes for Secure Wireless Communications

1
Dialog Codes for Secure Wireless Communications

• Anish Arora Lifeng Sang

• IPSN 2009

2
Motivation

• Basic problem
• The feasibility of achieving perfect secrecy in
  wireless network communications without shared
  secrets
• Physical layer security
• The focus thus far has largely been on theory
• The development of practical codes in current
  wireless platforms is only just beginning

3
Our contributions

• Define a secure coding problem
• Present the general properties of any solution to
  the problem
• Design a class of dialog codes
• Channel model
• Receiver model
• Validate the jamming capability and implement the
  dialog codes at byte-level and packet-level
• CC1000
• CC2420

4
Outline

• Problem statement and system model
• System model
• The secure coding problem
• Properties of secure coding
• Dialog Codes
• Full duplex
• Half duplex
• Experimental evaluation
• Conclusion and future work

5
System model

• Goal protect s without shared secret

What if there is no Kjk ?
j
k
s
Kjk(s)
Receiver
Sender
e
Eavesdropper
6
Basic idea
x
y
k selectively jams with time sequence ?
j
k
z
e
?
7
The secure coding problem

• Problem
• Design coding functions f and f such that
• x f(s)
• f(y, ?) s
• Pr(s) Pr(sz)

• Assumption
• Cooperative jamming by the receiver is
  predictable
• The sender and receiver are synchronized so that
  bit level jamming is feasible
• The detection of jamming at bit level is hard
• (discuss later how they are relaxed)

8
Jamming model and receiver model

• Jamming model

• Receiver model
• Full duplex where k knows x completely
• Half duplex where k knows x partially

9
Outline

• Problem statement and system model
• System model
• The secure coding problem
• Properties of secure coding
• Dialog Codes
• Full duplex
• Half duplex
• Experimental evaluation
• Conclusion and future work

10
Properties of secure coding

• Proposition 1. To achieve perfect secrecy, it
  must be that
• Theorem 1. For the full-duplex model, the maximal
  coding rate, 100, is achievable if 0.5 p q
  1
• Theorem 2. For the half-duplex model, the optimal
  coding rate in any scheme that achieves perfect
  secrecy is 50
• Typical existing coding schemes not efficient
• LT codes, Raptor codes and secret sharing

11
Outline

• Problem statement and system model
• System model
• The secure coding problem
• Properties of secure coding
• Dialog Codes
• Full duplex
• Half duplex
• Experimental evaluation
• Conclusion and future work

12
Dialog codes for full duplex receiver
If 0.5pq1 xs the receiver jams every bit
with probability p such that pp0.5
Any other p and q add random preambles to lose e
We prove that after (t-1)-bit random preamble,
the probability of guessing next bit correctly
converges to 50. t depends on p and q
13
Dialog codes for half duplex receiver
Special case pq1
0 1
0 ? 1 ?
1
0
f
1
0
k jams either position with probability 50
However, p and q are typically less than 1 if
pqlt1, for instance, e knows it was 0 if she sees
0 0
14
Dialog codes for half duplex receiver
General solution add random preamble to lose e
f
k jams either position on each pair with
probability 50, and uses the remaining bit for
recovery
We proved that after (t-1)-bit random preamble,
the probability of guessing next bit correctly
converges to 50. t depends on p and q
15
Efficiency of dialog codes

• Complexity table lookup O(1), and coin tossing
• Coding rate dialog codes achieve perfect secrecy
  asymptotically with rate 1/t
• For example, t 4 when pq0.5 to have near
  perfect secrecy
• If perfect secrecy is not strictly required
• For example, a 29-byte message s without using
  preamble, the probability of guessing s correctly
  is lt when pq0.5,
• while 1024-bit public key scheme has strength
  around 280

16
Outline

• Problem statement and system model
• System model
• The secure coding problem
• Properties of secure coding
• Dialog Codes
• Full duplex
• Half duplex
• Experimental evaluation
• Conclusion and future work

17
Experimental setup

• Fixed location, varying jamming power
• Fixed power, varying es location
• Varying the bit value
• Middle-band links
• cc1000 and cc2420 platforms

18
Experimental results
19
Jamming observation

• p and q are non-trivial
• p and q change over time, sometimes even
  dramatically
• When jamming is effective, e does not necessarily
  benefit
• from sitting closer to the sender
• when k lowers its jamming power

20
Implementing dialog codes

• Implementation at byte level in cc1000 and packet
  level in cc2420
• Let XOR value of all bits in that byte (or
  packet) denote the intended bit

• Byte level implementation
• j encodes s using dialog codes
• j precedes each byte with two sync bytes
• k randomly injects a jamming byte to corrupt one
  of the two data bytes

21
Outline

• Problem statement and system model
• System model
• The secure coding problem
• Properties of secure coding
• Dialog Codes
• Full duplex
• Half duplex
• Experimental evaluation
• Conclusion and future work

22
Conclusion and future work

• This paper takes a first step in the development
  of practical codes for confidential
  communications at the physical layer
• Define the secure coding problem
• Present a class of dialog codes
• Demonstrate the feasibility of dialog codes on
  current wireless platforms
• Future work
• Explore symmetric secure coding if the assumption
  of the detection of jamming at bit level is hard
  will not hold
• Alternative codes that do not code at the level
  of bits
• Deal with failures (e.g. bit level
  synchronization)

23
Thank You!