IEEE COMMUNICATIONS LETTERS, VOL. 19, NO. 6, JUNE 2015 909
Reduced-Complexity Linear Programming Decoding
Based on ADMM for LDPC Codes
Haoyuan Wei, Xiaopeng Jiao, and Jianjun Mu
Abstract—The Euclidean projection onto check polytopes is the
most time-consuming operation in the linear programming (LP)
decoding based on alternating direction method of multipliers
(ADMM) for low-density parity-check (LDPC) codes. In this letter,
instead of reducing the complexity of Euclidean projection itself,
we propose a new method to reduce the decoding complexity of
ADMM-based LP decoder by decreasing the number of Euclidean
projections. In particular, if all absolute values of the element-wise
differences between the input vector of Euclidean projection in the
current iteration and that in the previous iteration are less than
a predefined value, then the Euclidean projection at the current
iteration will be no longer performed. Simulation results show that
the proposed decoder can still save roughly 20% decoding time
even if both the over-relaxation and early termination techniques
are used.
Index Terms—Linear programming decoding, alternating
direction method of multipliers (ADMM), reduce-complexity.
I. INTRODUCTION
L
INEAR programming (LP) decoding, proposed by
Feldman in [1], is an alternative decoding method for
low-density parity-check (LDPC) codes. Recently, based on
the optimization method of alternating direction method of
multipliers (ADMM) [3], Barman et al. proposed an efficient
LP decoding algorithm, called ADMM-based LP decoder [2],
that has complexity comparable to the belief propagation (BP)
decoder. The most time-consuming part of the ADMM-based
LP decoding is the Euclidean projection onto the check poly-
tope, which is a convex hull of all binary vectors with an even
number of 1s. The length of these binary vectors is equal to the
degree of check nodes.
To reduce the complexity of ADMM-based LP decoder
further, Zhang et al. proposed an efficient projection algo-
rithm [ 4] based on the cut search algorithm introduced in [5].
Later, in [6], the projection operation onto a check polytope is
transformed to a projection onto a simplex. The complexity of
projection operations in [2] is O(d log d), where d is the degree
of check node.The transformation makes the complexity of the
projection operation reduce to O(d) [6]. In addition to reducing
the complexity of the projection operation, the over-relaxation
and early termination are also two effective techniques [3] to
reduce the complexity of the ADMM-based LP decoder.
Some existing algorithms described above try to reduce
the complexity of ADMM-based LP decoder by reducing the
Manuscript received February 19, 2015; revised March 19, 2015; accepted
March 27, 2015. Date of publication March 31, 2015; date of current version
June 5, 2015. This work is supported by the National Nature Science Founda-
tion of China (No. 61471286 and No. 61271004) and the Fundamental Research
Funds for the Central Universities. The associate editor coordinating the review
of this paper and approving it for publication was M. Baldi.
The authors are with the School of Computer Science and Technol-
ogy, Xidian University, Xi’an 710071, China ( e-mail: hehewhy@gmail.com;
jiaozi1216@126.com; jjmu@xidian.edu.cn).
Digital Object Identifier 10.1109/LCOMM.2015.2418261
complexity of projection operation. In this letter, we propose an
efficient algorithm to decrease the number of projection oper-
ations for the ADMM-based LP decoding. Simulation results
show that the proposed algorithm can decrease the number of
projection operations significantly at high signal-to-noise ratio
(SNR) regions.
II. P
RELIMINARIES
Suppose an LDPC Code C is defined by a m × n parity
check matrix H.LetI = {1, 2,...,n} and J = {1, 2,...,m}
denote the set of variable nodes and check nodes of C, respec-
tively. Let N
i
denote the set of neighboring check nodes of
variable node i. Similarly, let N
j
denote the set of neighboring
variable nodes of check node j.Wealsoused
i
and d
j
to
denote the degree of variable node i ∈Iand check node j ∈J,
respectively. Suppose that a codeword x ∈Cis transmitted over
a memoryless binary-input output-symmetric channel, and a
vector y is received. In [1] the authors proposed an LP decoder
to solve the following optimization problem:
minimize γ
T
x
subject to P
j
x ∈ PP
d
j
, ∀j ∈J, (1)
where γ is the log-likelihood ratio (LLR) vector and the i-
th entry of γ is defined as γ
i
=log
Pr(y|x
i
=0)
Pr(y|x
i
=1)
, P
j
is the
operation of selecting the sub-vector of x corresponding to the
j-th check node, and PP
d
j
is defined as the convex hull of all
even-parity binary vectors of length d
j
, called check polytope.
The ADMM-based LP decoding of (1) is formulated as:
minimize γ
T
x
subject to P
j
x = z
j
,z
j
∈ PP
d
j
, ∀j ∈J, (2)
where z
j
s are auxiliary variables of length d
j
, aiming to
constrain x to satisfy the j-th check node.
The augmented Lagrangian for the above problem is
L
μ
(x, z, λ)=γ
T
x+
j
λ
j
(P
j
x−z
j
)+
μ
2
j
P
j
x−z
j
2
2
,
(3)
where λ
j
s are Lagrangian multipliers and μ is a constant that
needs to be optimized. The updates for each component of the
variables using ADMM to solve problem (2) are described as
follows [2]:
x
i
=
[0,1]
1
d
i
⎛
⎝
j∈N
v
(i)
z
(i)
j
−
1
μ
λ
(i)
j
−
1
μ
γ
i
⎞
⎠
, (4)
z
j
=
PP
d
j
(P
j
x + λ
j
/μ), (5)
λ
j
= λ
j
+ μ(P
j
x − z
j
), (6)
where
[0,1]
is the Euclidean projection to the interval [0, 1],
and
PP
d
j
is the Euclidean projection onto check polytopes.
1558-2558 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.