LDPC hard decision decoder for high-speed wireless data communications
View Patent ↗A method for low-density parity-check hard decision decoding includes computing, for every decoding iteration, a discrepancy of extrinsic messages responsive to channel inputs of a receiver, performing a flipping of the channel inputs responsive to a comparison of the discrepancy of extrinsic messages to a flipping threshold, the flipping threshold for each decoding iteration being determined based on a threshold computation responsive to a channel error probability estimation in a first iteration of a decoding of the channel inputs, and check node decoding responsive to the flipping of channel inputs.
1. A method for low-density parity-check hard decision decoding comprising the steps of:
computing, for every decoding iteration, a discrepancy of extrinsic messages responsive to channel inputs of a receiver;
performing a flipping of said channel inputs responsive to a comparison of said discrepancy of extrinsic messages to a flipping threshold, said flipping threshold for each said decoding iteration being determined based on a threshold computation responsive to a channel error probability estimation in a first iteration of a decoding of said channel inputs; and
check node decoding responsive to said flipping of channel inputs.
2. The method of claim 1 , wherein said bit flipping is responsive to v c→v,j (l) and u c→v,j (l) , which are input and output extrinsic messages, respectively, for said check node decoding.
3. The method of claim 2 , wherein v c→v,j (l) and u c→v,j (l) of said flipping of channel inputs are related according to
v
v
->
c
,
k
(
ℓ
+
1
)
=
{
-
u
n
if
m
k
(
ℓ
)
≥
d
(
ℓ
)
,
u
n
,
otherwise
,
where d (l) is said flipping threshold and m k (l) is a metric for said discrepancy at variable nodes.
4. The method of claim 1 , wherein flipping threshold comprises an optimal flipping threshold responsive to one of a K-exp-OSS and an L-add-OSS.
5. The method of claim 4 , wherein said K-exp-OSS comprises assuming said flipping threshold d (l) is given by d (l) =d* (|l/K|) and d* (l) to be the optimal flipping threshold obtained analytically according to
1
-
p
^
c
p
^
c
≤
(
1
+
ρ
(
1
-
p
(
l
)
)
1
-
ρ
(
1
-
p
(
l
)
)
)
d
,
where
ρ
(
x
)
=
∑
i
ρ
i
x
i
-
1
(
{
ρ
i
}
is a check node profile from an edge perspective), and p (l) is the error probability of traversing messages in the l-th decoding iteration, which can be analytically computed based on an estimated channel error probability {circumflex over (p)} c and degree profiles of LDPC codes, and p (l) being estimated analytically.
6. The method of claim 4 , wherein said L-add-OSS comprises supposing the first L 1 optimal flipping thresholds are d 1 , the following L 2 are d 2 , and the next L 3 are d 3 , . . . , then for the L-add-OSS, the first L 1 +L optimal flipping thresholds are d 1 , the following L 2 +L are d 2 and the next L 3 +L are d 3 , . . . .
7. The method of claim 1 , wherein encoded messages for said channel inputs comprise multilevel encoding responsive to said computing and performing steps to include partitioning n mapping of bits into m non-overlapping subsets, and employing different rate codes for the mapping bits in different subsets.
8. The method of claim 7 , wherein said different rate codes are responsive to one of parallel independent decoding PID and multistage decoding MSD.
9. The method of claim 7 , wherein said multilevel encoding comprises optimizing profiles of component codes given a channel signal to noise relationship and allocating rates and optimizing said profiles for component codes responsive to an overall code rate.
10. The method of claim 7 , wherein said multilevel encoding comprises given said channel input signal with a noise ratio of E s /N 0 , for 1≦i≦m, we evaluate error rates e (i) for signals S (i) , and perform a rate optimization for a code C (i) by the code rate optimization; letting R PID (E s /N 0 ) be an average rate of the component codes and given an overall code rate r, component codes C (i) for 1≦i≦m being selected using a bisection search.
11. The method of claim 10 , wherein said bisection search starting with a selection of said E S /N 0 selection from a low value (E s /N 0 ) l , and a high value (E s /N 0 ) h , then letting (E s /N 0 ) m =((E s /N 0 ) h +(E s /N 0 ) l )/2, and computing said R PID ((E s /N 0 ) m ); if said R PID ((E s /N 0 ) m )<r, then update (E s /N 0 ) l =(E s /N 0 ) m ; otherwise update (E s /N 0 ) m =(E s /N 0 ) m , until R PID ((E s /N 0 ) m ) is sufficiently close to r; and assuming (E s /N 0 ) m =(E s /N 0 )* when the bisection search terminates, output (E s /N 0 )* as the threshold and the optimized profiles of component codes for (E s /N 0 )* as the designed code profiles.