Simplified Successive-Cancellation List Decoding of Polar Codes
Simplified Successive-Cancellation List Decoding of Polar Codes
Polar Codes
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 1 / 14
Motivation
In this talk:
We show how to speed up SCL without losing
error-correction performance!
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 2 / 14
Background
Polar Codes
Decoding schemes:
List-SD
Successive-Cancellation (SC)
Speed
SC List (SCL) SC
Sphere Decoding (SD) SCL
List-SD
SD
Error-Correction Performance
E. Arıkan, ”Channel polarization: A method for constructing capacity achieving codes for symmetric binary input
memoryless channels,” IEEE Transactions on Information Theory, vol. 55, no. 7, pp. 3051-3073, July 2009.
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 3 / 14
Background
Polar Codes
Decoding schemes: X
List-SD
Successive-Cancellation (SC)
Speed
SC List (SCL) SC
Sphere Decoding (SD) SCL
List-SD
SD
Error-Correction Performance
E. Arıkan, ”Channel polarization: A method for constructing capacity achieving codes for symmetric binary input
memoryless channels,” IEEE Transactions on Information Theory, vol. 55, no. 7, pp. 3051-3073, July 2009.
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 3 / 14
Background
= 0sz}|{
sz}|{ = 1 zs }|
= 2{ z =3
s }| { s=3
x0 α
0 β
s=2 l βr
0 x1 α lα r
0 x2 s=1 β
u3 x3
s=0
0 x4
u5 x5
u6 x6
u7 x7
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 4 / 14
Background
= 0sz}|{
sz}|{ = 1 zs }|
= 2{ z =3
s }| { s=3
x0 α
0 β
s=2 l βr
0 x1 α lα r
0 x2 s=1 β
u3 x3
s=0
0 x4
u5 x5
Exact formulation
u6 x6
1 + eαi +αi +2s−1
u7 x7 l
αi = ln
eαi + eαi +2s−1
αri =αi +2s−1 + (1 − 2βil )αi
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 4 / 14
Background
= 0sz}|{
sz}|{ = 1 zs }|
= 2{ =3
s }| s=3
z {
x0 α
0 β
s=2 l βr
0 x1 α lα r
0 x2 s=1 β
u3 x3
s=0
0 x4
u5 x5
Exact formulation
u6 x6
1 + eαi +αi +2s−1
u7 x7
αli = ln
eαi + eαi +2s−1
αri =αi +2s−1 + (1 − 2βil )αi
Hardware-friendly formulation
αli = sgn(αi ) sgn(αi +2s−1 ) min(|αi |, |αi +2s−1 |)
αri =αi +2s−1 + (1 − 2βil )αi
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 4 / 14
Background
SC
s=3
s=2
s=1
s=0
T = 14 time-steps
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 5 / 14
Background
Simplified SC (SSC)
s=3
s=2
s=1
Rate-0 Rate-1
s=0
T = 10 time-steps
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 5 / 14
Background
Fast-SSC
s=3
s=2
Rep SPC
s=1
s=0
T = 2 time-steps
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 5 / 14
Background
SCL Decoding
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 6 / 14
Background
SCL Decoding
Exact formulation i
X
−(1−2ûjl )αjl
PMil = ln 1 + e
j=0
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 6 / 14
Background
SCL Decoding
Exact formulation i
X
PMil = ln 1 + e−(1−2ûjl )αjl
j=0
Hardware-friendly formulation
(
PMi −1l +|αil |, if ûil 6= 21 (1 − sgn (αil )) ,
PMil =
PMi −1l , otherwise
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 6 / 14
Proposed Algorithm
SCL Issues
Simplified SCL:
Faster: simplified Rate-1, Rate-0 and Rep nodes
Guaranteed to preserve error-correction performance
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 7 / 14
Proposed Algorithm
Simplified SCL
Theorem
The path metric for a Rate-1 node of length 2s can be calculated as
s −1
2X
PM2s −1 = ln 1 + e−(1−2βi )αi ,
i =0
where αi and βi are relative to the top of the Rate-1 node tree.
s=1
s=0
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 8 / 14
Proposed Algorithm
Simplified SCL
Proof.
Induction (ηi = 1 − 2βi )
For Rate-1 node with N = 2:
α +α
l 1+e 0 1
α 0 = ln eα0 +eα1
r
α0 = α1 + η0l α0
SC
η0l = η0 η1
r
η0 = η1
l l
r r
PM1 = ln 1 + e−η0 α0 +ln 1 + e−η0 α0
Simplified SCL
Proof. Proof.
Induction (ηi = 1 − 2βi ) For Rate-1 node with N = 2s :
For Rate-1 node with N = 2:
α +α PM2s −1 = PMl2s−1 −1 + PMr2s−1 −1
l 1+e 0 1
α = ln s−1
= Σi2=0 −1 ln 1 + e−ηi αi
0
eα0 +eα1
r
α0 = α1 + η0l α0 + ln 1 + e−ηi +2s−1 αi +2s−1
SC
η0l = η0 η1
s
= Σ2i =0
r
−1
ln 1 + e−ηi αi ,
η0 = η1
l l
r r
The theorem holds for all N.
PM1 = ln 1 + e−η0 α0 +ln 1 + e−η0 α0
u0 β0 x0
l r l r
Substituting for α0 , α0 , η0 and η0 :
u1 β 1 x1
β
PM1 = ln 1 + e −η0 α0
+ ln 1 + e −η1 α1
u2 2 x2
u3 β3 x3
The theorem holds for N = 2.
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 9 / 14
Proposed Algorithm
Simplified SCL
Theorem
In the hardware-friendly formulation of SCL algorithm, the path metric for
a Rate-1 node of length 2s can be calculated as
2s −1
1X
PM2s −1 = sgn (αi ) αi − (1 − 2βi )αi ,
2
i =0
where αi and βi are relative to the top of the Rate-1 node tree.
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 10 / 14
Proposed Algorithm
Simplified SCL
Theorem
The path metric for a Rate-0 node of length 2s can be calculated as
s −1
2X
ln 1 + e−αi ,
PM2s −1 =
i =0
where αi is the LLR value at the top of the Rate-0 node tree.
[α0 , α1 , α2 , α3 ] ⇔ [0, 0, 0, 0]
s=2
s=1
s=0
Simplified Rate-0 nodes can be decoded in at most log2 N time-steps
The time complexity is reduced from O(N) to O(log2 N)
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 11 / 14
Proposed Algorithm
Simplified SCL
Theorem
The path metric for a Rep node of length 2s can be calculated as
s −1
2X
PM2s −1 = ln 1 + e−(1−2β2s −1 )αi ,
i =0
where αi is relative to the top of the Rep node tree and β2s −1 is relative to
the information bit in the Rep node tree.
[α0 , α1 , α2 , α3 ] ⇔ [β3 , β3 , β3 , β3 ]
s=2
s=1
s=0
Simplified Rep nodes can be decoded in at most log2 N time-steps
The time complexity is reduced from O(N) to O(log2 N)
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 12 / 14
Conclusion
Conclusion
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 13 / 14
Conclusion
Thank you!
Seyyed Ali Hashemi (McGill) SSCL Decoding of Polar Codes ISIT 2016 14 / 14