Slides Arik An

Polarization Encoding Decoding Construction Performance
Polar Coding Tutorial
Erdal Arıkan
Electrical-Electronics Engineering Department

Bilkent University
Ankara, Turkey
Jan. 15, 2015

Simons Institute
UC Berkeley
The channel
Let W : X → Y be a binary-input discrete memoryless channel
X W Y
◮ input alphabet: X = {0, 1},

◮ output alphabet: Y,
◮ transition probabilities:
W (y |x), x ∈ X,y ∈ Y
The channel
X W Y

W (y |x), x ∈ X,y ∈ Y
The channel
X W Y

W (y |x), x ∈ X,y ∈ Y
The channel
X W Y

W (y |x), x ∈ X,y ∈ Y
Symmetry assumption
Assume that the channel has “input-output symmetry.”

Symmetry assumption
Examples:
BSC(ǫ)
1−ǫ
0 0
ǫ
ǫ
1 1
1−ǫ
Symmetry assumption
Examples:
BSC(ǫ) BEC(ǫ)
1−ǫ 1−ǫ
0 0 0 0
ǫ ǫ
?
ǫ ǫ
1 1 1 1
1−ǫ 1−ǫ
Capacity
For channels with input-output symmetry, the capacity is given by

∆
C (W ) = I (X ; Y ), with X ∼ unif. {0, 1}
Capacity
For channels with input-output symmetry, the capacity is given by

∆
C (W ) = I (X ; Y ), with X ∼ unif. {0, 1}
Use base-2 logarithms:
0 ≤ C (W ) ≤ 1
The main idea
◮ Channel coding problem trivial for two types of channels

◮ Perfect: C (W ) = 1
◮ Useless: C (W ) = 0
◮ Transform ordinary W into such extreme channels
The main idea

The main idea

The main idea

The method: aggregate and redistribute capacity
Original channels
(uniform)
W
W
Original channels
(uniform)
W
Vector
b
channel
b
Wvec
b
Combine
Original channels New channels

(uniform) (polarized)
W W1
Vector
b
channel b
b b
Wvec
b b
W WN−1
W WN
Combine Split
Combining
◮ Begin with N copies of W , X1 Y1

W
◮ use a 1-1 mapping
X2 Y2
GN : {0, 1}N → {0, 1}N W
◮ to create a vector channel
Wvec : U N → Y N XN YN
W
Combining
◮ Begin with N copies of W , U1 X1 Y1

W
U2 X2 Y2
GN : {0, 1}N → {0, 1}N W
GN
Wvec : U N → Y N UN XN YN
W
Combining
◮ Begin with N copies of W , U1 X1 Y1

W
U2 X2 Y2
GN : {0, 1}N → {0, 1}N W
GN
Wvec : U N → Y N UN XN YN
W
Wvec
Conservation of capacity
Combining operation is lossless:

◮ Take U1 , . . . , UN i.i.d. unif. {0, 1} U1 X1 Y1
◮ then, X1 , . . . , XN i.i.d. unif. {0, 1} W
◮ and U2 X2 Y2
W
C (Wvec ) = I (U N ; Y N ) GN
= I (X N ; Y N )
= NC (W ) UN XN YN
W
Wvec

◮ and U2 X2 Y2
W
= I (X N ; Y N )
= NC (W ) UN XN YN
W
Wvec

◮ and U2 X2 Y2
W
= I (X N ; Y N )
= NC (W ) UN XN YN
W
Wvec
Splitting
C (Wvec ) = I (U N ; Y N )
U1 Y1
Ui−1
Ui Wvec Yi
Ui+1
UN YN
Splitting
C (Wvec ) = I (U N ; Y N )
U1 Y1
N
X
N i−1
= I (Ui ; Y , U )
i=1
Ui−1
Ui Wvec Yi
Ui+1
UN YN
Splitting U1
Ui−1
C (Wvec ) = I (U N ; Y N )
U1 Y1
N
X
N i−1
= I (Ui ; Y , U )
i=1
Ui−1
Ui Wvec Yi
Ui+1
Define bit-channels
Wi : Ui → (Y N , U i−1 )
UN YN
Wi
Splitting U1
Ui−1
C (Wvec ) = I (U N ; Y N )
U1 Y1
N
X
N i−1
= I (Ui ; Y , U )
i=1
N
X Ui−1
= C (Wi )
i=1 Ui Wvec Yi
Ui+1
Define bit-channels
Wi : Ui → (Y N , U i−1 )
UN YN
Wi
Polarization is commonplace
◮ Polarization is the rule not the

exception U1 X1 Y1
◮ A random permutation W
U2 X2 Y2
GN : {0, 1}N → {0, 1}N W
GN
is a good polarizer with high
probability
UN XN YN
◮ Equivalent to Shannon’s random W
coding approach

exception U1 X1 Y1
U2 X2 Y2
GN : {0, 1}N → {0, 1}N W
GN
probability
UN XN YN
coding approach

exception U1 X1 Y1
U2 X2 Y2
GN : {0, 1}N → {0, 1}N W
GN
probability
UN XN YN
coding approach
Random polarizers: stepwise, isotropic

1
0.9
0.8
0.7
0.6
Capacity
0.5
0.4
0.3
0.2
0.1
0
5 10 15 20 25 30
Bit channel index
Random polarizers: stepwise, isotropic

1
0.9
0.8
0.7
0.6
Capacity
0.5
0.4
0.3
0.2
0.1
0
5 10 15 20 25 30
Bit channel index
Isotropy: any redistribution order is as good as any other.

The complexity issue
◮ Random polarizers lack structure, too complex to implement

◮ Need a low-complexity polarizer
◮ May sacrifice stepwise, isotropic properties of random
polarizers in return for less complexity


Basic module for a low-complexity scheme
Combine two copies of W
X1 Y1
W
X2 Y2
W
U1 X1 Y1
+ W
U2 X2 Y2
W
G2
U1 X1 Y1
+ W
U2 X2 Y2
W
G2
and split to create two bit-channels
W1 : U1 → (Y1 , Y2 )
W2 : U2 → (Y1 , Y2 , U1 )
The first bit-channel W1
W1 : U1 → (Y1 , Y2 )
U1 Y1
+ W
random U2 Y2
W
The first bit-channel W1
W1 : U1 → (Y1 , Y2 )
U1 Y1
+ W
random U2 Y2
W
C (W1 ) = I (U1 ; Y1 , Y2 )
The second bit-channel W2
W2 : U2 → (Y1 , Y2 , U1 )
U1 Y1
+ W
U2 Y2
W
The second bit-channel W2
W2 : U2 → (Y1 , Y2 , U1 )
U1 Y1
+ W
U2 Y2
W
C (W2 ) = I (U2 ; Y1 , Y2 , U1 )
Capacity conserved but redistributed unevenly

U1 X1 Y1
+ W
U2 X2 Y2
W
◮ Conservation:
C (W1 ) + C (W2 ) = 2C (W )
◮ Extremization:
C (W1 ) ≤ C (W ) ≤ C (W2 )
with equality iff C (W ) equals 0 or 1.

Capacity conserved but redistributed unevenly

U1 X1 Y1
+ W
U2 X2 Y2
W
◮ Conservation:
C (W1 ) + C (W2 ) = 2C (W )
◮ Extremization:
C (W1 ) ≤ C (W ) ≤ C (W2 )
with equality iff C (W ) equals 0 or 1.

Notation
The two channels created by the basic transform
(W , W ) → (W1 , W2 )
will be denoted also as
W − = W1 and W + = W2
Notation
The two channels created by the basic transform
(W , W ) → (W1 , W2 )
will be denoted also as
W − = W1 and W + = W2
Likewise, we write W −− , W −+ for descendants of W − ; and W +− ,

W ++ for descendants of W + .
For the size-4 construction

+ W
W
... duplicate the basic transform

+ W
+ W
W
... obtain a pair of W − and W + each

W−
W+
W−
W+
... apply basic transform on each pair

+ W−
+ W+
W−
W+
... decode in the indicated order

U1 + W−
U3 + W+
U2 W−
U4 W+
... obtain the four new bit-channels

U1 W −−
U3 W +−
U2 W −+
U4 W ++
Overall size-4 construction

X1 Y1
U1 + + W
X3 Y3
U3 + W
X2 Y2
U2 + W
X4 Y4
U4 W
“Rewire” for standard-form size-4 construction

X1 Y1
U1 + + W
X2 Y2
U2 + W
X3 Y3
U3 + W
X4 Y4
U4 W
Size 8 construction
U1 X1 Y1
+ + + W
U2 X2 Y2
+ + W
U3 X3 Y3
+ + W
U4 X4 Y4
+ W
U5 X5 Y5
+ + W
U6 X6 Y6
+ W
U7 X7 Y7
+ W
U8 X8 Y8
W
Demonstration of polarization
Polarization is easy to analyze when W is a BEC.
W
If W is a BEC(ǫ), then so are W −
and W + , with erasure probabili- 1−ǫ
ties 0 0
∆ ǫ
ǫ− = 2ǫ − ǫ2
?
and ǫ
∆
ǫ+ = ǫ2 1 1
1−ǫ
respectively.
If W is a BEC(ǫ), then so are W − W−

and W + , with erasure probabili- 1 − ǫ−
ties 0 0
∆
ǫ− = 2ǫ − ǫ2 ǫ−
?
and ǫ−
∆
ǫ+ = ǫ2 1 1
1 − ǫ−
respectively.
If W is a BEC(ǫ), then so are W − W+

and W + , with erasure probabili- 1 − ǫ+
ties 0 0
∆
ǫ− = 2ǫ − ǫ2 ǫ+
?
and ǫ+
∆
ǫ+ = ǫ2 1 1
1 − ǫ+
respectively.
Polarization for BEC( 21 ): N = 16

Capacity of bit channels
1
0.9
0.8
0.7
0.6
Capacity
0.5
0.4
0.3
0.2
0.1
N=16
0
2 4 6 8 10 12 14 16
Bit channel index

1
0.9
0.8
0.7
0.6
Capacity
0.5
0.4
0.3
0.2
0.1
N=32
0
5 10 15 20 25 30
Bit channel index

1
0.9
0.8
0.7
0.6
Capacity
0.5
0.4
0.3
0.2
0.1
N=64
0
10 20 30 40 50 60
Bit channel index

1
0.9
0.8
0.7
0.6
Capacity
0.5
0.4
0.3
0.2
0.1
N=128
0
20 40 60 80 100 120
Bit channel index

1
0.9
0.8
0.7
0.6
Capacity
0.5
0.4
0.3
0.2
0.1
N=256
0
50 100 150 200 250
Bit channel index

1
0.9
0.8
0.7
0.6
Capacity
0.5
0.4
0.3
0.2
0.1
N=512
0
50 100 150 200 250 300 350 400 450 500
Bit channel index

1
0.9
0.8
0.7
0.6
Capacity
0.5
0.4
0.3
0.2
0.1
N=1024
0
100 200 300 400 500 600 700 800 900 1000
Bit channel index
Polarization martingale
C (W )
0
C (W2 )
C (W )
C (W1 )
0 1
1
C (W ++ )
C (W2 )
C (W +− )
C (W )
C (W −+ )
C (W1 )
C (W −− )
0 1 2
1
C (W ++ )
C (W2 )
C (W +− )
C (W )
C (W −+ )
C (W1 )
C (W −− )
0 1 2
1
C (W ++ )
C (W2 )
C (W +− )
C (W )
C (W −+ )
C (W1 )
C (W −− )
0 1 2 3
1
C (W ++ )
C (W2 )
C (W +− )
C (W )
C (W −+ )
C (W1 )
C (W −− )
0 1 2 3 4
1
C (W ++ )
C (W2 )
C (W +− )
C (W )
C (W −+ )
C (W1 )
C (W −− )
0 1 2 3 4 5
1
C (W ++ )
C (W2 )
C (W +− )
C (W )
C (W −+ )
C (W1 )
C (W −− )
0 1 2 3 4 5 6
1
C (W ++ )
C (W2 )
C (W +− )
C (W )
C (W −+ )
C (W1 )
C (W −− )
0 1 2 3 4 5 6 7
1
C (W ++ )
C (W2 )
C (W +− )
C (W )
C (W −+ )
C (W1 )
C (W −− )
0 1 2 3 4 5 6 7 8
1
Theorem (Polarization, A. 2007) 1−δ
The bit-channel capacities {C (Wi )} polarize: for any
δ ∈ (0, 1), as the construction size N grows

no. channels with C (Wi ) > 1 − δ
−→ C (W )
N
and

no. channels with C (Wi ) < δ
−→ 1 − C (W )
N
Theorem (Rate of polarization, √

A. and Telatar (2008)) δ
Above theorem holds with δ ≈ 2− N . 0
1
Theorem (Polarization, A. 2007) 1−δ
The bit-channel capacities {C (Wi )} polarize: for any
δ ∈ (0, 1), as the construction size N grows

no. channels with C (Wi ) > 1 − δ
−→ C (W )
N
and

no. channels with C (Wi ) < δ
−→ 1 − C (W )
N
Theorem (Rate of polarization, √

A. and Telatar (2008)) δ
Above theorem holds with δ ≈ 2− N . 0
Polar code example: W = BEC( 12 ), N = 8, rate 1/2

I (Wi )
Y1
0.0039 U1 + + + W
Y2
0.1211 U2 + + W
Y3
0.1914 U3 + + W
Y4
0.6836 U4 + W
Y5
0.3164 U5 + + W
Y6
0.8086 U6 + W
Y7
0.8789 U7 + W
Y8
0.9961 U8 W

I (Wi ) Rank
Y1
0.0039 8 U1 + + + W
Y2
0.1211 7 U2 + + W
Y3
0.1914 6 U3 + + W
Y4
0.6836 4 U4 + W
Y5
0.3164 5 U5 + + W
Y6
0.8086 3 U6 + W
Y7
0.8789 2 U7 + W
Y8
0.9961 1 U8 W

I (Wi ) Rank
Y1
0.0039 8 U1 + + + W
Y2
0.1211 7 U2 + + W
Y3
0.1914 6 U3 + + W
Y4
0.6836 4 U4 + W
Y5
0.3164 5 U5 + + W
Y6
0.8086 3 U6 + W
Y7
0.8789 2 U7 + W
Y8
0.9961 1 data U8 W

I (Wi ) Rank
Y1
0.0039 8 U1 + + + W
Y2
0.1211 7 U2 + + W
Y3
0.1914 6 U3 + + W
Y4
0.6836 4 U4 + W
Y5
0.3164 5 U5 + + W
Y6
0.8086 3 U6 + W
Y7
0.8789 2 data U7 + W
Y8
0.9961 1 data U8 W

I (Wi ) Rank
Y1
0.0039 8 U1 + + + W
Y2
0.1211 7 U2 + + W
Y3
0.1914 6 U3 + + W
Y4
0.6836 4 U4 + W
Y5
0.3164 5 U5 + + W
Y6
0.8086 3 data U6 + W
Y7
0.8789 2 data U7 + W
Y8
0.9961 1 data U8 W

I (Wi ) Rank
Y1
0.0039 8 U1 + + + W
Y2
0.1211 7 U2 + + W
Y3
0.1914 6 U3 + + W
Y4
0.6836 4 data U4 + W
Y5
0.3164 5 U5 + + W
Y6
0.8086 3 data U6 + W
Y7
0.8789 2 data U7 + W
Y8
0.9961 1 data U8 W

I (Wi ) Rank
Y1
0.0039 8 frozen U1 + + + W
Y2
0.1211 7 frozen U2 + + W
Y3
0.1914 6 frozen U3 + + W
Y4
0.6836 4 data U4 + W
Y5
0.3164 5 frozen U5 + + W
Y6
0.8086 3 data U6 + W
Y7
0.8789 2 data U7 + W
Y8
0.9961 1 data U8 W

I (Wi ) Rank
Y1
0.0039 8 frozen 0 + + + W
Y2
0.1211 7 frozen 0 + + W
Y3
0.1914 6 frozen 0 + + W
Y4
0.6836 4 data U4 + W
Y5
0.3164 5 frozen 0 + + W
Y6
0.8086 3 data U6 + W
Y7
0.8789 2 data U7 + W
Y8
0.9961 1 data U8 W
Encoding complexity
Theorem
Encoding complexity for polar coding is O(N log N).
Proof:
◮ Polar coding transform can be represented as a graph with
N[1 + log(N)] variables.
◮ The graph has (1 + log(N)) levels with N variables at each
level.
◮ Computation begins at the source level and can be carried out
level by level.
◮ Space complexity O(N), time complexity O(N log N).
Encoding complexity
Theorem
Proof:
level.
level by level.
Encoding complexity
Theorem
Proof:
level.
level by level.
Encoding complexity
Theorem
Proof:
level.
level by level.
Encoding: an example
0 Y1
frozen + + + W
0 Y2
frozen + + W
0 Y3
frozen + + W
1 Y4
free + W
0 Y5
frozen + + W
1 Y6
free + W
0 Y7
free + W
1 Y8
free W
0 Y1
frozen +0 + + W
0 0 Y2
frozen + + W
0 Y3
frozen +1 + W
1 1 Y4
free + W
0 Y5
frozen +1 + W
1 1 Y6
free + W
0 Y7
free +1 W
1 1 Y8
free W
0 Y1
frozen +0 + 1
+ W
0 0 1 Y2
frozen + + W
0 Y3
frozen +1 1
+ W
1 1 1 Y4
free + W
0 Y5
frozen +1 + 0
W
1 1 0 Y6
free + W
0 Y7
free +1 1
W
1 1 1 Y8
free W
0 Y1
frozen +0 + 1
+ 1
W
0 0 1 1 Y2
frozen + + W
0 Y3
frozen +1 1
+ 0
W
1 1 1 0 Y4
free + W
0 Y5
frozen +1 + 0 0
W
1 1 0 0 Y6
free + W
0 Y7
free +1 1 1
W
1 1 1 1 Y8
free W
Successive Cancellation Decoding (SCD)
Theorem
The complexity of successive cancellation decoding for polar codes
is O(N log N).
Proof: Given below.

SCD: Exploit the x = |a|a + b| structure

u1 x1 y1
+ + b1 + W
u2 b2 x2 y2
+ + W
u3 b3 x3 y3
+ + W
u4 b4 + x4 y4
W
u5 a1 x5 y5
+ + W
u6 a2 x6 y6
+ W
u7 a3 x7 y7
+ W
u8 a4 x8 y8
W
First phase: treat a as noise, decode (u1 , u2 , u3 , u4 )

u1 x1 y1
+ + b1 + W
u2 b2 x2 y2
+ + W
u3 b3 x3 y3
+ + W
u4 b4 + x4 y4
W
noise a1 x5 y5
W
noise a2 x6 y6
W
noise a3 x7 y7
W
noise a4 x8 y8
W
End of first phase

û1 x1 y1
+ + b̂1 + W
û2 b̂2 x2 y2
+ + W
û3 b̂3 x3 y3
+ + W
û4 b̂4 + x4 y4
W
u5 a1 x5 y5
+ + W
u6 a2 x6 y6
+ W
u7 a3 x7 y7
+ W
u8 a4 x8 y8
W
Second phase: Treat b̂ as known, decode (u5 , u6 , u7 , u8 )

known b̂1 y1
+ W
known b̂2 y2
+ W
known b̂3 y3
+ W
known b̂4 + y4
W
u5 a1 y5
+ + W
u6 a2 y6
+ W
u7 a3 y7
+ W
u8 a4 y8
W
First phase in detail

u1 x1 y1
+ + b1 + W
u2 b2 x2 y2
+ + W
u3 b3 x3 y3
+ + W
u4 b4 + x4 y4
W
noise a1 x5 y5
W
noise a2 x6 y6
W
noise a3 x7 y7
W
noise a4 x8 y8
W
Equivalent channel model

b1 x1 y1
+ W
b2 x2 y2
+ W
b3 x3 y3
+ W
b4 + x4 y4
W
noise a1 x5 y5
W
noise a2 x6 y6
W
noise a3 x7 y7
W
noise a4 x8 y8
W
First copy of W −
b1 x1 y1
+ W
b2 x2 y2
+ W
b3 x3 y3
+ W
b4 + x4 y4
W
noise a1 x5 y5
W
noise a2 x6 y6
W
noise a3 x7 y7
W
noise a4 x8 y8
W
Second copy of W −
b1 x1 y1
+ W
b2 x2 y2
+ W
b3 x3 y3
+ W
b4 + x4 y4
W
noise a1 x5 y5
W
noise a2 x6 y6
W
noise a3 x7 y7
W
noise a4 x8 y8
W
Third copy of W −
b1 x1 y1
+ W
b2 x2 y2
+ W
b3 x3 y3
+ W
b4 + x4 y4
W
noise a1 x5 y5
W
noise a2 x6 y6
W
noise a3 x7 y7
W
noise a4 x8 y8
W
Fourth copy of W −
b1 x1 y1
+ W
b2 x2 y2
+ W
b3 x3 y3
+ W
b4 + x4 y4
W
noise a1 x5 y5
W
noise a2 x6 y6
W
noise a3 x7 y7
W
noise a4 x8 y8
W
Decoding on W −
u1 (y1 , y5 )
+ + b1 W−
u2 b2 (y2 , y6 )
+ W−
u3 b3 (y3 , y7 )
+ W−
u4 b4 (y4 , y8 )
W−
b = |t|t + w|
u1 (y1 , y5 )
+ w1 + b1 W−
u2 w2 + b2 (y2 , y6 )
W−
u3 b3 (y3 , y7 )
+ t1 W−
u4 t2 b4 (y4 , y8 )
W−
Decoding on W −−
u1 (y1 , y3 , y5 , y7 )
+ w1 W −−
u2 w2 (y2 , y4 , y6 , y8 )
W −−
Decoding on W −−−
u1 (y1 , y2 , . . . , y8 )
W −−−
u1 (y1 , y2 , . . . , y8 )
W −−−
Compute
∆ W −−− (y1 , . . . , y8 | u1 = 0)
L−−− = .
W −−− (y1 , . . . , y8 | u1 = 1)
u1 (y1 , y2 , . . . , y8 )
W −−−
Compute
∆ W −−− (y1 , . . . , y8 | u1 = 0)
L−−− = .
W −−− (y1 , . . . , y8 | u1 = 1)
Set 
u1
 if u1 is frozen
û1 = 0 else if L−−− > 0

1 else

u1 (y1 , y2 , . . . , y8 )
W −−−
Compute
∆ W −−− (y1 , . . . , y8 | u1 = 0)
L−−− = .
W −−− (y1 , . . . , y8 | u1 = 1)
Set 
u1
 if u1 is frozen
û1 = 0 else if L−−− > 0

1 else

u1 (y1 , y2 , . . . , y8 )
W −−−
Compute
∆ W −−− (y1 , . . . , y8 | u1 = 0)
L−−− = .
W −−− (y1 , . . . , y8 | u1 = 1)
Set 
u1
 if u1 is frozen
û1 = 0 else if L−−− > 0

1 else

Decoding on W −−+
known û1 + (y1 , y3 , y5 , y7 )

W −−
u2 (y2 , y4 , y6 , y8 )
W −−
u2 (y1 , . . . , y8 , û1 )
W −−+
u2 (y1 , . . . , y8 , û1 )
W −−+
Compute
∆ W −−+ (y1 , . . . , y8 , û1 | u2 = 0)

L−−+ = .
W −−+ (y1 , . . . , y8 , û1 | u2 = 1)
u2 (y1 , . . . , y8 , û1 )
W −−+
Compute
∆ W −−+ (y1 , . . . , y8 , û1 | u2 = 0)

L−−+ = .
W −−+ (y1 , . . . , y8 , û1 | u2 = 1)
Set 
u2
 if u2 is frozen
û2 = 0 else if L−−+ > 0

1 else

u2 (y1 , . . . , y8 , û1 )
W −−+
Compute
∆ W −−+ (y1 , . . . , y8 , û1 | u2 = 0)

L−−+ = .
W −−+ (y1 , . . . , y8 , û1 | u2 = 1)
Set 
u2
 if u2 is frozen
û2 = 0 else if L−−+ > 0

1 else

u2 (y1 , . . . , y8 , û1 )
W −−+
Compute
∆ W −−+ (y1 , . . . , y8 , û1 | u2 = 0)

L−−+ = .
W −−+ (y1 , . . . , y8 , û1 | u2 = 1)
Set 
u2
 if u2 is frozen
û2 = 0 else if L−−+ > 0

1 else

Complexity for successive cancelation decoding
◮ Let CN be the complexity of decoding a code of length N

◮ Decoding problem of size N for W reduced to two decoding
problems of size N/2 for W − and W +
◮ So
CN = 2CN/2 + kN
for some constant k
◮ This gives CN = O(N log N)

◮ So
CN = 2CN/2 + kN
for some constant k

◮ So
CN = 2CN/2 + kN
for some constant k

◮ So
CN = 2CN/2 + kN
for some constant k
Performance of polar codes
Theorem
For any rate R < I (W ) and block-length N, the probability of
frame error for polar codes under successive cancelation decoding is
bounded as √ √
Pe (N, R) = o 2− N+o( N)
Proof: Given in the next presentation.

Construction complexity
Theorem
Given W and a rate R < I (W ), a polar code can be constructed in
O(Npoly(log (N))) time that achieves under SCD the performance
√ √
Pe = o 2− N+o( N)
Proof: Given in the next presentation.

Polar coding summary
Summary
Given W , N = 2n , and R < I (W ), a polar code can be constructed
such that it has
◮ construction complexity O(Npoly(log (N))),
◮ encoding complexity ≈ N log N,
◮ successive-cancellation decoding complexity ≈ N log N,
√ √
◮ frame error probability Pe (N, R) = o 2− N+o( N) .
Summary
such that it has
√ √
Summary
such that it has
√ √
Summary
such that it has
√ √
List decoder for polar codes
Developed by Tal and Vardy (2011); similar to Dumer’s list

decoder for Reed-Muller codes.
◮ First produce L candidate decisions
◮ Pick the most likely word from the list
◮ Complexity O(LN log N)


Tal-Vardy list decoder performance

Length n = 2048, rate R = 0.5, BPSK-AWGN channel, list-size L.






List-of-L performance quickly approaches ML performance!

List decoder with CRC
◮ Same decoder as before but data contains a built-in CRC

◮ Selection made by CRC and relative likelihood
List decoder with CRC
◮ Same decoder as before but data contains a built-in CRC

◮ Selection made by CRC and relative likelihood
Tal-Vardy list decoder with CRC



Polar codes (+CRC) achieve state-of-the-art performance!

Summary
◮ Polarization is a commonplace phenomenon – almost

unavoidable
◮ Polar codes are low-complexity methods designed to exploit
polarization for achieving Shannon limits
◮ Polar codes with some help from other methods perform
competitively with the state-of-the-art codes in terms of
complexity and performance
Summary

unavoidable
Summary

unavoidable

Slides Arik An

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Slides Arik An

Uploaded by

Copyright:

Available Formats

Polarization Encoding Decoding Construction Performance

Polar Coding Tutorial

Electrical-Electronics Engineering Department

Jan. 15, 2015

Let W : X → Y be a binary-input discrete memoryless channel

◮ input alphabet: X = {0, 1},

Let W : X → Y be a binary-input discrete memoryless channel

◮ input alphabet: X = {0, 1},

Let W : X → Y be a binary-input discrete memoryless channel

◮ input alphabet: X = {0, 1},

Let W : X → Y be a binary-input discrete memoryless channel

◮ input alphabet: X = {0, 1},

Assume that the channel has “input-output symmetry.”

Assume that the channel has “input-output symmetry.”

Assume that the channel has “input-output symmetry.”

For channels with input-output symmetry, the capacity is given by

For channels with input-output symmetry, the capacity is given by

Use base-2 logarithms:

The main idea

◮ Channel coding problem trivial for two types of channels

The main idea

◮ Channel coding problem trivial for two types of channels

The main idea

◮ Channel coding problem trivial for two types of channels

The main idea

◮ Channel coding problem trivial for two types of channels

The method: aggregate and redistribute capacity

The method: aggregate and redistribute capacity

The method: aggregate and redistribute capacity

Original channels New channels

◮ Begin with N copies of W , X1 Y1

◮ to create a vector channel

◮ Begin with N copies of W , U1 X1 Y1

◮ Begin with N copies of W , U1 X1 Y1

Combining operation is lossless:

Combining operation is lossless:

Combining operation is lossless:

◮ Polarization is the rule not the

◮ Polarization is the rule not the

◮ Polarization is the rule not the

Random polarizers: stepwise, isotropic

Random polarizers: stepwise, isotropic

Isotropy: any redistribution order is as good as any other.

The complexity issue

◮ Random polarizers lack structure, too complex to implement

The complexity issue

◮ Random polarizers lack structure, too complex to implement

The complexity issue

◮ Random polarizers lack structure, too complex to implement

Basic module for a low-complexity scheme

Combine two copies of W

Basic module for a low-complexity scheme

Combine two copies of W

Basic module for a low-complexity scheme

Combine two copies of W

The first bit-channel W1

The first bit-channel W1

The second bit-channel W2

The second bit-channel W2

Capacity conserved but redistributed unevenly

with equality iff C (W ) equals 0 or 1.

Capacity conserved but redistributed unevenly

with equality iff C (W ) equals 0 or 1.

The two channels created by the basic transform