LDPC Decoding

Parity check matrix & Tanner graph example

Notations:

Left side :variable nodes Vi , i ∈ {0, 1, 2 ⋯ , 7}

Right side: check nodes Cj , j ∈ {0, 1, 2, 3}
Γi denote the set of all checked nodes connected to variable node Vi
Υj denote the set of all variable nodes that connected to check node Cj
L(ci ) denote the LLR of encoded bit
Vi→j message from variable node i to check node j
Cj→i message from check node j to variable node i
Message(Belief) passing from variable node to check node in the nth round

∑ Ck→i
(n) (n−1)
Vi→j = L(ci ) +
k∈Γ i \{j}

LDPC Decoding 1
 Message passing from check node to variable node in the nth round

⎛ (n−1)
⎞
∏ tanh ( )
(n) (−1)
Vk→j
= 2 tanh
⎝k∈Υj \{i} ⎠
Cj→i
2
∣ (n−1) ∣
∏ sgn (Vk→j ) ⋅ min ( Vk→j )
(n−1)
≈
k∈Υ j \{i} ∣ ∣
k∈Υ j \{i}

Message Passing Decoding algorithm(belief propagation decoding)[2]

For transmitted encoded codeword c = [c0 , c1 , ...] with channel output y =

[y0 , y1 , ⋯ ], the input to the LDPC decoder is the log-likelihood ratio (LLR) value:

P r(ci = 0∣yi )
L(ci ) = log { }
P r(ci = 1∣yi
(0) (0)
Initialization : Vi→j = 0, Cj→i = 0, n=0
for each iteration n

for each variable node i uses received messages from previous round
and its accumulated LLR from all previous round, calculate and send
messages to its connected check nodes.

∑ Ck→i
(n) (n−1)
Vi→j = L(ci )(0) +
k∈Γ i \{j}

each check node j receives message from its connected variable

nodes, calculate and send message back.

⎛ (n−1)
⎞
tanh ( )
Vk→j
∏
(n) (−1)
= 2 tanh
⎝k∈Υj \{i} ⎠
Cj→i
2

update LLR

( )

LDPC Decoding 2
 L(ci )(n) = L(ci )(n−1) + ∑ Ck→i
(n)

k∈Γ i

Make decision based on final L(ci )

Layered decoding algorithm (layered BP) [3]

same performance but converge faster

(0) (0)
Initialization : Vi→j = 0, Cj→i = 0, n=0
for each iteration n

for each layer j (each layer corresponds to a check node or a row of

exponent matrix)

for each variable node i uses received messages from previous round
and its accumulated LLR from all previous round, calculate and send
messages to its connected check nodes.

(n−1)
L(yi ) = L(ci ) − Cj→i
(n)
Vi→j = L(ci )

the check node j receives message from its connected variable nodes,
calculate and send message back.

⎛ (n−1)
⎞
∏ tanh ( )
(n) (−1)
Vk→j
= 2 tanh
⎝k∈Υj \{i} ⎠
Cj→i
2
(n)
L(ci ) = L(ci ) + Cj→i

Make decision based on final L(yi )

min-sum decoding algorithm [4]

(n)
replace above Cj→i in layered decoding algorithms with its min-sum approximation

∣ (n−1) ∣
∏ sgn (V (n−1) k → j ) ⋅ min k ∈ Υj \{i} ( Vk→j )
(n)
Cj→i =
∣ ∣
k∈Υ j \{i}

LDPC Decoding 3
 Normalised min-sum decoding algorithm [4]
scale min-sum approximation with a positive factor less than 1

(n) (n)
Cj→i = αCj→i , 0 < α < 1

offset min-sum decoding algorithm [4]

if absolute of min-sum approximation less than offset, make it 0

(n) (n)
Cj→i = max(Cj→i − β, 0), 0 < α < 1

Performance
min-sum < normalized min-sum < offset min-sum < layered BP = BP

[2] Gallager, Robert G. Low-Density Parity-Check Codes, Cambridge, MA, MIT

Press, 1963.

[3] Hocevar, D.E. "A reduced complexity decoder architecture via layered decoding
of LDPC codes." In IEEE Workshop on Signal Processing Systems, 2004. SIPS
2004. doi: 10.1109/SIPS.2004.1363033
[4] Chen, Jinghu, R.M. Tanner, C. Jones, and Yan Li. "Improved min-sum decoding
algorithms for irregular LDPC codes." In Proceedings. International Symposium on
Information Theory, 2005. ISIT 2005. doi: 10.1109/ISIT.2005.1523374

LDPC Decoding 4