Reliable Deniable Communication: Hiding Messages from Noise

Reliable Deniable Communication:
Hiding Messages in Noise
Pak Hou Che
Mayank Bakshi
Sidharth Jaggi
The Chinese University
of Hong Kong
The Institute of
Network Coding
Alice
Bob
Reliability
Alice
Bob
Reliability
Deniability
Willie
(the Warden)
Alice’s Encoder
M
T
𝐼𝑓 𝐓 = 0, 𝐗 = 𝟎
𝐼𝑓 𝐓 = 1, 𝐗 = 𝐸𝑛𝑐(𝐌)
π‘šπ‘’π‘ π‘ π‘Žπ‘”π‘’ 𝐌 ∈ {1, … , 𝑁}
tπ‘Ÿπ‘Žπ‘›π‘ . π‘ π‘‘π‘Žπ‘‘π‘’π‘  𝐓 ∈ {0, 1}
𝑁 = 2πœƒ( 𝑛)
π‘‡β„Žπ‘Ÿπ‘œπ‘’π‘”β„Žπ‘π‘’π‘‘ 𝜏 = log 𝑁
log 𝑁
π‘…π‘’π‘™π‘Žπ‘‘π‘–π‘£π‘’ π‘‘β„Žπ‘Ÿπ‘œπ‘’π‘”β„Žπ‘π‘’π‘‘ π‘Ÿ =
𝑛
𝐗
Alice’s Encoder
M
T
𝐼𝑓 𝐓 = 0, 𝐗 = 𝟎
𝐼𝑓 𝐓 = 1, 𝐗 = 𝐸𝑛𝑐(𝐌)
Message 𝐌 ∈ {1, … , 𝑁}
Trans. Status 𝐓 ∈ {0, 1}
𝑁 = 2πœƒ(
𝑛)
Bob’s Decoder
𝐗
BSC(pb)
π˜π‘
𝐌 = 𝐷𝑒𝑐(π˜π‘ )
𝐌
1 βˆ’ πœ– π‘Ÿπ‘’π‘™π‘–π‘Žπ‘π‘™π‘’
Pr 𝐌 = 𝐌 > 1 βˆ’ πœ–
Alice’s Encoder
M
T
Bob’s Decoder
𝐼𝑓 𝐓 = 0, 𝐗 = 𝟎
𝐗
BSC(pb)
𝐼𝑓 𝐓 = 1, 𝐗 = 𝐸𝑛𝑐(𝐌)
𝐌 = 𝐷𝑒𝑐(π˜π‘ )
𝐌
1 βˆ’ πœ– π‘Ÿπ‘’π‘™π‘–π‘Žπ‘π‘™π‘’
Pr 𝐌 = 𝐌 > 1 βˆ’ πœ–
Message 𝐌 ∈ {1, … , 𝑁}
Trans. Status 𝐓 ∈ {0, 1}
𝑁 = 2πœƒ(
π˜π‘
𝑛)
BSC(pw)
π˜π‘€
𝐓 = 𝐷𝑒𝑐(π˜π‘€ )
𝐓
Willie’s (Best) Estimator
Bash, Goeckel & Towsley [1]
Shared secret
𝑂( 𝑛 log 𝑛) bits
AWGN channels
But capacity only 𝑂
𝑛 bits
[1] B. A. Bash, D. Goeckel and D. Towsley, β€œSquare root law for communication with low
probability of detection on AWGN channels,” in Proceedings of the IEEE International
Symposium on Information Theory (ISIT), 2012, pp. 448–452.
This work
No shared secret
BSC(pb)
pb < pw
BSC(pw)
Alice’s Encoder
M
T
Bob’s Decoder
𝐼𝑓 𝐓 = 0, 𝐗 = 𝟎
𝐗
BSC(pb)
𝐼𝑓 𝐓 = 1, 𝐗 = 𝐸𝑛𝑐(𝐌)
𝐌 = 𝐷𝑒𝑐(π˜π‘ )
𝐌
1 βˆ’ πœ– π‘Ÿπ‘’π‘™π‘–π‘Žπ‘π‘™π‘’
Pr 𝐌 = 𝐌 > 1 βˆ’ πœ–
Message 𝐌 ∈ {1, … , 𝑁}
Trans. Status 𝐓 ∈ {0, 1}
𝑁 = 2πœƒ(
π˜π‘
𝑛)
BSC(pw)
π˜π‘€
𝐓 = 𝐷𝑒𝑐(π˜π‘€ )
𝐓
Willie’s (Best) Estimator
Hypothesis Testing
Willie’s Estimate
Alice’s
Transmission
Status
β€’ 𝛼 = Pr 𝐓 = 1 𝐓 = 0 , 𝛽 = Pr 𝐓 = 0 𝐓 = 1
Hypothesis Testing
Willie’s Estimate
Alice’s
Transmission
Status
Hypothesis Testing
Willie’s Estimate
Alice’s
Transmission
Status
Hypothesis Testing
Willie’s Estimate
Alice’s
Transmission
Status
Intuition
β€’ 𝐓 = 0, 𝐲𝑀 = 𝐳𝑀 ~Binomial(𝑛, 𝑝𝑀 )
Intuition
β€’ 𝐓 = 0, 𝐲𝑀 = 𝐳𝑀 ~Binomial 𝑛, 𝑝𝑀
β€’ π‘Šβ„Žπ‘’π‘› 𝐓 = 1,
Theorem 1 (Wt(c.w.))
(high deniability => low weight codewords)
β€’ Too many codewords with weight β€œmuch” greater
than 𝑐 𝑛, then the system is β€œnot very” deniable
Theorems 2 & 3
(Converse & achievability for reliable & deniable comm.)
Theorems 2 & 3
𝑝𝑀
1/2
pb>pw
0
1/2
𝑝𝑏
Theorems 2 & 3
𝑝𝑀
1/2
𝑁=0
0
1/2
𝑝𝑏
Theorems 2 & 3
𝑝𝑀
pw=1/2
1/2
0
1/2
𝑝𝑏
Theorems 2 & 3
𝑝𝑀
1/2
N ο‚» 2(1ο€­ H ( p b ))n
(BSC(pb))
ο‚ ο€ 
0
1/2
𝑝𝑏
Theorems 2 & 3
𝑝𝑀
1/2
pb=0
0
1/2
𝑝𝑏
Theorems 2 & 3
𝑝𝑀
𝑁 = 2𝑂(
1/2
0
𝑛 log 𝑛)
,
𝑛
= 2𝑂(
𝑛
1/2
𝑛 log 𝑛)
𝑝𝑏
Theorems 2 & 3
𝑝𝑀
1/2
pw>pb
0
1/2
𝑝𝑏
Theorems 2 & 3
𝑝𝑀
𝑁 = 2𝑂(
𝑛)
1/2
β€œStandard” IT inequalities
+
Wt(β€œmost codewords”)<√n
(Thm 1)
0
1/2
𝑝𝑏
Theorems 2 & 3
𝑝𝑀
1/2
Main thm:
Achievable region
𝑁 = 2Ξ© ( 𝑛)
0
1/2
𝑝𝑏
log
𝑛
β‰ˆπ‘›
𝑛/2
logarithm of
# codewords
0
n
𝑀𝑑𝐻 (π’šπ‘€ )
log(# codewords)
𝑛𝐻(𝑝𝑀 )
𝐱=0
Pr(𝑀𝑑𝐻 𝐲𝑀 )
𝐙𝑀
𝑂(1
𝑛)
0
𝑝𝑀 𝑛 βˆ’ 𝑂( 𝑛)
𝑝𝑀 𝑛 𝑝𝑀 𝑛 + 𝑂( 𝑛)
n
𝑀𝑑𝐻 (𝐲𝑀 )
log(# codewords)
𝑛𝐻(𝑝𝑀 βˆ— 𝜌)
Pr (𝑀𝑑𝐻 𝐲𝑀 )
𝑐 𝑛
𝐌,𝐙𝑀
𝑂(1
𝑛)
n
0
(𝑝𝑀 βˆ— 𝜌)𝑛 βˆ’ 𝑂( 𝑛)
(𝑝𝑀 βˆ— 𝜌)𝑛
(𝑝𝑀 βˆ— 𝜌)𝑛 + 𝑂( 𝑛)
𝑀𝑑𝐻 (𝐲𝑀 )
Theorem 3 – Reliability proof sketch
Random code
Weight 𝑂( 𝑛)
1000001000000000100100000010000000100
0001000000100000010000000010000000001
0010000100000001010010000000100010011
2𝑂(
𝑛) codewords
.
.
.
0000100000010000000000010000000010000
Theorem 3 – Reliability proof sketch
Weight 𝑂( 𝑛)
β€’ E(Intersection of 2 codewords) = O(1)
β€’ β€œMost” codewords β€œwell-isolated”
1000001000010000100100000010000000100
0001000000100000010000000010000000001
0010000100000001010010000000100010011
.
.
.
0000100000010000000000010000000010000
Theorem 3 – dmin decoding
x
+
𝑂( 𝑛)
x’
β€’Pr(x decoded to x’) < 2βˆ’π‘‚(
𝑛)
Theorem 3 – Deniability proof sketch
β€’ Recall: want to show 𝑉 𝐏0 , 𝐏1 < πœ–
log(# codewords)
𝑛𝐻(𝑝𝑀 βˆ— 𝜌)
Pr (𝑀𝑑𝐻 𝐲𝑀 )
𝑐 𝑛
𝐌,𝐙𝑀
𝑂(1
𝑛)
n
0
(𝑝𝑀 βˆ— 𝜌)𝑛 βˆ’ 𝑂( 𝑛)
(𝑝𝑀 βˆ— 𝜌)𝑛
(𝑝𝑀 βˆ— 𝜌)𝑛 + 𝑂( 𝑛)
𝑀𝑑𝐻 (𝐲𝑀 )
Theorem 3 – Deniability proof sketch
β€’ Recall: want to show 𝑉 𝐏0 , 𝐏1 < πœ–
𝐏0
𝐏1
Theorem 3 – Deniability proof sketch
log(# codewords)
0
n
Theorem 3 – Deniability proof sketch
logarithm of
# codewords
0
n
𝑀𝑑𝐻 (π’šπ‘€ )
Theorem 3 – Deniability proof sketch
𝑬π‘ͺ (𝐏1 )!!!
𝐏0
𝐏1
Theorem 3 – Deniability proof sketch
β€’ 𝑉 𝐏0 , 𝐏1 ≀ 𝑉 𝐏0 , 𝑬π‘ͺ (𝐏1 ) + 𝑉 𝑬π‘ͺ (𝐏1 ), 𝐏1
𝑬π‘ͺ (𝐏1 )!!!
𝐏0
𝐏1
Theorem 3 – Deniability proof sketch
𝑬π‘ͺ (𝐏1 )
𝐏1
Theorem 3 – Deniability proof sketch
logarithm of
# codewords
0
𝑝𝑀 𝑛 βˆ’ 𝑂( 𝑛)
𝑝𝑀 𝑛 𝑝𝑀 𝑛 + 𝑂( 𝑛)
𝑀𝑑𝐻 (π’šπ‘€ )
n
Theorem 4
logarithm of
# codewords
0
n
𝑀𝑑𝐻 (π’šπ‘€ )
Theorem 4
Too few codewords
=> Not deniable
logarithm of
# codewords
0
n
𝑀𝑑𝐻 (π’šπ‘€ )
Summary
Summary