Exact Split Information Function for SPC
Y. Min
⇤
, F. C. M. Lau
⇤†
and C. K. Tse
⇤
⇤
Department of Electronic and Information Engineering, Hong Kong Polytechnic University, Hong Kong
†
The Hong Kong Polytechnic University Shenzhen Research Institute
Email: encmlau@polyu.edu.hk, Fax: +852 2362 8439, Tel: +852 2766 6206
Abstract—Split information functions are used in deriving
closed-form Extrinsic Information Transfer (EXIT) curves of
super-variable nodes (SVNs) in doubly-generalized low-density
parity-check (DGLDPC) codes. In this letter, we derive an exact
split information function for single-parity-check (SPC) codes. The
function is very easy to compute and has been verified against
the results obtained using the traversal method.
Index Terms—Doubly generalized LDPC codes, EXIT curve,
split information function.
I. INTRODUCTION
Low-density parity-check (LDPC) codes make use of rep-
etition codes at the variable nodes and single-parity-check
(SPC) codes at the check nodes. Doubly-generalized LDPC
(DGLDPC) codes are formed when the repetition codes and
the SPC codes are replaced by more complex linear block
codes. Subsequently, the nodes are called “super-variable
nodes” (SVNs) and “super-check nodes” (SCNs). Researchers
have proposed using various types of constituent codes such
as Hamming codes and BCH codes in the generalized LDPC
and DGLDPC codes [1]–[7].
Similar to the standard LDPC decoders, the iterative decoder
of DGLDPC codes can be regarded as two concatenated
component codes, including the super-variable-node (SVN)
decoder and the super-check-node (SCN) decoder, as shown in
Figure 1. For an m
a
⇥n
a
adjacency matrix, the corresponding
DGLDPC code has m
a
SCN decoders and n
a
SVN decoders
which are connected by an edge-interleaver.
In each iteration, each SVN decoder takes the channel
information C and the a priori information A
svn
as the input,
and then outputs the extrinsic information E
svn
. E
svn
, after
passing through the edge-interleaver, becomes the a priori
information A
scn
of the neighboring SCN decoder. Based on
A
scn
, each SCN decoder generates the extrinsic information
E
scn
and passes it, via the edge-interleaver, back to the
SVN decoder as the a priori information A
svn
. Consequently,
two types of channels, namely communication channel and
extrinsic channel, exist in the decoder model.
In [8], ten Brick has proposed using Extrinsic Information
Transfer (EXIT) charts to analyze the convergence behavior
of turbo codes. Later, the principle of EXIT charts has been
successfully applied to study other iteratively-decoded codes
such as parallel concatenated codes (PCCs) [9], serially con-
catenated codes (SCC) [10], convolutional codes [11], LDPC
codes [9], [12], repeat-accumulate codes [13], generalized
LDPC codes and DGLDPC codes [7], [14], [15].
Fig. 1. An iterative decoder of DGLDPC codes.
In [16], it has been shown that codes with capacity-
approaching performance over the binary erasure channel
(BEC) can be designed by matching exactly the EXIT curves
of the component codes. Empirically, this approach can also
be applied to the more general cases, i.e., the binary-input
additive-white-Gaussian-noise (BI-AWGN) channel. While the
closed-form EXIT functions of LDPC codes over both the
BEC and the BI-AWGN channel have been derived already,
they cannot be applied to the analysis of DGLDPC directly.
One common method is to use Monte Carlo simulations
to estimate the EXIT charts of DGLDPC codes. However,
intensive simulations have to be performed and are very time-
consuming.
Another way is to derive the closed-form EXIT functions
for all types of SVNs and SCNs in the DGLDPC codes over
the BEC. Assume that the communication channel and the
extrinsic channel are BECs and the corresponding erasure
probabilities of these two channels are denoted by q and p,
respectively. For a DGLDPC code, the closed-form EXIT
functions of any SVN type and any SCN type over the
BEC have been derived in [16]. For a general ( n
svn
,k
svn
)
constituent code with code length n
svn
and k
svn
information
bits used at the SVN, the closed-form EXIT function of this
SVN over the BEC is expressed by [16]
I
BEC
e,svn
(p, q)
=1
1
n
svn
n
svn
1
X
t=0
k
svn
X
z=0
p
t
(1 p)
n
svn
t1
q
z
(1 q)
k
svn
z
⇥[(n
svn
t) ee
n
svn
t,k
svn
z
(t + 1)ee
n
svn
t1,k
svn
z
]
(1)
where ee
g,h
is named as the (g, h)-th split information function.
Moreover, ee
g,h
is defined as the summation of the ranks of
all the possible sub-matrices (denoted by S
g,h
) obtained by
choosing g columns in the corresponding generator matrix
with the size of k
svn
⇥n
svn
and h columns in the correspond-
ing k
svn
⇥k
svn
identity matrix. Figure 2 depicts the definition