entropy

Article
Design of Low-Density Parity-Check Code Pair for Joint
Source-Channel Coding Systems Based on Graph Theory
Yijie Lv 1 , Jiguang He 2,3 , Weikai Xu 1, * and Lin Wang 1

1 Department of Information and Communication Engineering, Xiamen University, Xiamen 361005, China;
[email protected] (Y.L.); [email protected] (L.W.)
2 Technology Innovation Institute, Abu Dhabi P.O. Box 9639, United Arab Emirates; [email protected]
3 Centre for Wireless Communications, University of Oulu, 90014 Oulu, Finland
* Correspondence: [email protected]

Abstract: In this article, a graph-theoretic method (taking advantage of constraints among sets
associated with the corresponding parity-check matrices) is applied for the construction of a double
low-density parity-check (D-LDPC) code (also known as LDPC code pair) in a joint source-channel
coding (JSCC) system. Specifically, we pre-set the girth of the parity-check matrix for the LDPC
code pair when jointly designing the two LDPC codes, which are constructed by following the set
constraints. The constructed parity-check matrices for channel codes comprise an identity submatrix
and an additional submatrix, whose column weights can be pre-set to be any positive integer numbers.
Simulation results illustrate that the constructed D-LDPC codes exhibit significant performance
improvement and enhanced flexible frame length (i.e., adaptability under various channel conditions)
compared with the benchmark code pair.

Keywords: joint source-channel coding; low-density parity-check code; graph theory

1. Introduction
A joint source-channel coding (JSCC) system is capable of overcoming the short-
Citation: Lv, Y.; He, J.; Xu, W.; Wang, comings of a Shannon separation coding system and enhancing the performance of the
L. Design of Low-Density digital communication system via joint decoding [1]. In general, a JSCC system can be
Parity-Check Code Pair for Joint realized with different combinations of source and channel codes. For instance, examples
Source-Channel Coding Systems of these combinations are a variable-length code (VLC) cascaded with a convolutional code
Based on Graph Theory. Entropy 2023,
(CC) [2,3]; a VLC cascaded with a Turbo code [4–6]; a VLC cascaded with a low-density
25, 1189. https://doi.org/10.3390/
parity-check (LDPC) code [7,8]; and an LDPC code (falling into the category of fixed-length
e25081189
codes) cascaded with another LDPC code [9,10], termed as double LDPC (D-LDPC) codes.
Academic Editor: T. Aaron Gulliver Among these realizations, the D-LDPC code-oriented JSCC system has been inten-
sively investigated since its first appearance in [9,10], where a joint extrinsic information
Received: 29 June 2023
transfer (JEXIT) chart was employed to enhance the waterfall performance of the code pair
Revised: 26 July 2023
(one LDPC code for source code and another LDPC code for channel code). To reduce the
Accepted: 7 August 2023
hardware implementation complexity of the JSCC system and to improve its performance,
Published: 10 August 2023
double protograph LDPC (DP-LDPC) codes for JSCC were extensively studied in [11–16].
The source protograph LDPC codes were optimized to improve the error floor of the
JSCC system by maximizing the source decoding threshold [12]. Increasing the mutual
Copyright: © 2023 by the authors. information (MI) among the source code and the channel code by redesigning the channel
Licensee MDPI, Basel, Switzerland. protograph LDPC codes leads to performance improvements in the waterfall region [13].
This article is an open access article To reduce the transmission latency and decoding complexity, a concatenation of spatially
distributed under the terms and coupled LDPC codes with sliding window decoding for JSCC was proposed in [17]. How-
conditions of the Creative Commons ever, the aforementioned JSCC systems heavily rely on analyses of protograph LDPC codes
Attribution (CC BY) license (https:// and JEXIT. Moreover, the candidate frame length of the JSCC based on DP-LDPC code
creativecommons.org/licenses/by/ must be restricted to an integer fold of the column number of the base matrix.
4.0/).

Entropy 2023, 25, 1189. https://doi.org/10.3390/e25081189 https://www.mdpi.com/journal/entropy

Entropy 2023, 25, 1189 2 of 10

Combinatorial mathematics is an important tool for constructing LDPC codes, which

includes the graph-theoretic method, combinatoric designs, finite geometries, and differ-
ence sets [18–21]. As an essential construction method for LDPC codes, the channel LDPC
codes constructed with the graph-theoretic method achieve promising performance [22].
In this paper, a construction method of the code pair is developed based on graph
theory, which transforms parity-check matrices of the source code and the channel code
into inter-set constraint problem and selects the sets which satisfy the constraint. In order to
match the parity-check matrix of the source code and the generation matrix of the channel
code in the encoding process, the channel code is first constructed during the construction
of the code pair via the graph-theoretic method. In contrast to the conventional method,
the proposed one can be obtained without constructing the base matrix and performing
‘copy and permute’ operations. Since the proposed channel code takes the systematic form,
the generator matrix is easily obtained by transposing the parity-check matrix. In addition,
the proposed code pair is constructed with set constraints from the graph-theoretic method,
which brings a more adaptive/flexible frame length than the conventional protograph
code pair.

2. Preliminaries of JSCC
2.1. Representations of JSCC
The D-LDPC system consists of two LDPC codes: one for source code and the other for
channel code. As a linear block code, a parity-check matrix can be used to express the LDPC
code. A systematic LDPC code can be completely characterized by a generator matrix
G ∈ Bk×n with B = {0, 1}, which corresponds to a parity-check matrix H ∈ B(n−k)×n of
the form in (1), where PT is the transpose of the matrix P and In−k is the identity matrix
with dimension (n − k) × (n − k).

1 0 0 ··· 0 p11 p21 ··· pk1

 

h i

 0 1 0 ··· 0 p12 p22 ··· pk2 

H = In−k | P T
=
 0 0 1 ··· 0 p13 p23 ··· pk3 
 (1)
 .. .. 
 . . 
0 0 0 ··· 1 p 1( n − k ) p 2( n − k ) ··· pk(n−k) (n−k)×n

The parity-check matrix of the JSCC system is represented as

 
HS HL
HJ = , (2)
0 HC

which includes three non-zero submatrices, i.e., HS with dimension ms × ns , HC with

dimension mc × nc , and H L with dimension ms × nc . Therefore, the overall dimension of
H J is (ms + mc ) × (ns + nc ). A simple example of a parity-check matrix H J is provided
in (3).
1 1 1 1 1 0 1 0 0 0 0 0
 
 0 1 0 0 1 0 0 1 0 0 0 0 
 
 1 0 1 1 0 1 0 0 1 0 0 0 
HJ = 
 0 0 0 0 0 0 1 1 0 1 1 0 .
 (3)
 
 0 0 0 0 0 0 1 0 1 0 0 1 
0 0 0 0 0 0 0 0 0 1 1 1
The Tanner graph associated with the parity-check matrix H J is depicted in Figure 1.
Six types of MI in Figure 1 are defined as follows:
Entropy 2023, 25, 1189 3 of 10

SC (iter )
* tv,c : MI from the variable node (VN) of the source code to the check node (CN) of
the source code.
SC (iter )
* tc,v : MI from the CN of the source code to the VN of the source code.
SC →CC (iter )
* tc : MI from the CN of the source code to the VN of the channel code.
CC →SC (iter )
* tv : MI from the VN of the channel code to the CN of the source code.
CC (iter )
* tv,c : MI from the VN of the channel code to the CN of the channel code.
CC (iter )
* tc,v : MI from the CN of the channel code to the VN of the channel code.

Figure 1. Tanner graphical representation of the JSCC system realized according to exemplary H J
in (3).

2.2. Encoding and Decoding of JSCC Systems

2.2.1. Encoder
The source sequence follows the binomial Bernoulli (p) distribution, whose entropy is
expressed as
H = − p log2 p − (1 − p) log2 (1 − p), (4)
with p (p 6= 1/2) being the probability of ‘1’.
The encoding process for the LDPC code pair is represented by

c = GTC o = GTC HS s, (5)

where s ∈ Bns denotes the source sequence, while o ∈ Bms stands for the compressed source
sequence. GC ∈ B(nc −mc )×nc is the generator matrix of the channel code. By following the
dimension of HS , the following constraint

nc − mc = ms (6)

needs to be satisfied.
The definition of the rate for the low-density parity-check code pair is given by

ns × (nc − mc )
RJ =
, (7)
ms × nc − n punc

where n punc is the number of punctured variable nodes in the channel code.
Entropy 2023, 25, 1189 4 of 10

2.2.2. Decoder
The joint decoding algorithm is applied at the receiver to recover the original source
sequence s. The initial source information J SC is denoted by ln((1 − p)/p), depending on
the source statistics. The initial channel information is denoted by J CC = 2yi /σ2 , where
yi = (1 − 2ci ) + ni is the additive white Gaussian noise (AWGN).

3. Construction of LDPC Code Pairs

3.1. Definitions and Theorems
In Figure 1, the cycles of source and channel codes of the JSCC system in the Tanner
graph are underlined with bold lines. The minimum cycle length in a given bipartite
graph is the so-called girth of the graph. These cycles directly affect the performance of
waterfall and error-floor regions for the LDPC code pair. Particularly, short-length cycles
are more detrimental, since the information sent out by the message-passing decoder comes
back after a small number of hops to the same node that sent it. This results in biases in
the decoding algorithm. All in all, the short-length cycles are particularly deleterious to
decoding performance, and thus in the design of the code pair, one should intentionally
avoid them. Due to the correspondence between the non-zero elements in the parity-check
matrix and the edges in Figure 1, it is possible to avoid short cycles during the construction
of the LDPC code pair through structural design. Specifically, each column of the parity-
check matrix for both the source and channel codes can be represented by a set, with the
row indices of the non-zero elements forming this set. Consequently, short cycles are
effectively avoided by designing the relationships between these sets.
A graph consists of vertices (or nodes) and the edges (lines or branches) that connect
the vertices. In the sequel, we provide multiple definitions and theorems related to graphs.

Definition 1. A connected graph, G = (V
, E ), includes vertex-set V = {v1 , v2 , . . .} with elements

called vertices and edge-set E = vi , v j with pairs of vertices called edges. The end vertices of
the edge are the vertices vi and v j that are connected to an edge (vi , v j ).

Definition 2. Let T1 , T2 , . . . , Tn be n candidate sets, Ti contains W elements from {1, . . . , m},

where i ∈ {1, . . . , n}. Then, for any positive integer t, ∪t denotes the union of t sets selected from
the set of all pairwise intersections of the candidate sets, i.e., Ti ∩ T j for 1 ≤ i, j ≤ n.

Theorem 1. Let E (n) denote a set with n elements and Ti , T j be subsets of E (n). If ∀t ∈
 
{2, 3, . . . , s}, max | ∪t∀T ,T ⊆E (n),i6= j (Ti ∩ T j )| < t (the union is not calculated when t takes
i j
the value 2, i.e., calculating the intersection of Ti and T j only), then the cycle length of the Tanner
graph is larger than 2s, i.e., the girth of the Tanner is larger than or equal to 2(s + 1).


Proof. When s = 2, max |(Ti ∩ T j )| < 2, it means that the number of common elements
between any two subsets is less than 2. In this case, the cycle length of the Tanner graph
exceeds 4. When s > 2, an increase in its value means that new connection(s) between the
check node and the variable node is added to the Tanner graph. For each additional edge
in the Tanner graph, the cycle length and the girth will increase by 2. This is due to the
fact that the total of the degrees of the vertices for the graph is twice the number of edges,
which implies that the number of common elements between  the subsets will increase
 by 1.
That is, if there exists a length-2N cycle, then t = N, max | ∪t∀T ,T ⊆E (n),i6= j (Ti ∩ T j )| = N.
i j
For example, if there exists a length-6 cycle, ∃T1 , T2 , T3 satisfying |(T1 ∩ T2 ) ∪ (T2 ∩ T3 ) ∪
(T1 ∩ T3 )| = 3, which contradicts with Theorem 1 (since according to it, |(T1 ∩ T2 ) ∪ (T2 ∩
T3 ) ∪ (T1 ∩ T3 )| < 3).
Entropy 2023, 25, 1189 5 of 10

Theorem 2. According to Theorem 1, the LDPC code pair is constructed for optimizing the
performance of the JSCC system. Each column of the parity-check matrix can be represented by a
subset (a subset refers to a group of row indices representing the positions of non-zero elements in a
column of the parity-check matrix for the source or channel codes, excluding the identity matrix).
The maximum number of columns in the LDPC source code is determined via

mc (mc − 1)(mc 2 − mc − Wc 2 + Wc )
 
Ps ≤ , (8)
Ws Wc 2 (Ws − 1)(Wc − 1)2

where the column weights for the source and channel LDPC codes are denoted as Ws and Wc .

Proof. The parity-check matrix of the systematic channel LDPC code consists of an identity
matrix I and a matrix Pc . The parity-check matrix of the source code is formed with matrix
Ps . The construction of matrix Ps and matrix Pc are based on Theorem 1. If there does
not exist any length-4 cycles, ∀Ti , T j ⊆ E (n), max(|(Ti ∩ T j )|) < 2, the number of columns
m ( m −1)
of the matrix P is less than or equal to W (W − 1 )
(the upper bound of row weight of the
m −1
additional submatrix is W −1 ), the number of columns of the systematic channel LDPC code
m·(W 2 −W +m−1)
is less than or equal to W ·(W −1)
. (Here, we remove the subscripts for m and W to
make it more general, since this statement can be applied to both source code and channel
code.) The condition that a larger-length (>4) cycle does not exist is pre-conditioned on
the fact that there does not exist any length-4 circles. Therefore, when a larger-length cycle
does not exist, the number of columns of the constructed parity-check matrix has to be
reduced to satisfy the stricter set constraint in Theorem 1.

3.2. Algorithm Description

According to Theorem 1, the corresponding pseudo-code for the proposed construction
algorithm of the LDPC code pair is given in Algorithm 1, described as follows:
Step 1: Set up the column weight Ws of the parity-check matrix for the source code
and the column weight Wc of the parity-check matrix for the channel code, and calculate
their dimensions according to Theorem 2 and (6) (i.e., first calculate the number of columns
ns needed in the parity-check matrix of the source code HS based on the corresponding
source parameters (e.g., message length and desired code rate). Then, use the desired
column weights Ws and Wc , along with ns , to calculate the number of rows mc required for
the parity-check matrix of the channel code HC .
Step 2: List all the subsets (of set {1, 2, · · · , mc }) containing Wc different elements.
mc
The number of such subsets is (W c
).
Step 3: Select the desired subsets (from the previous step) satisfying Theorem 1 for
constructing the column vector of the parity-check matrix for the channel code HC (the
elements in each subset indicate the row indices of ‘1’ in the associated parity-check matrix),
which specifies the constraints that the parity-check matrix must satisfy to ensure good
error-correction performance.
Step 4: Calculate ms according to the relationship in (6) and the constructed channel
code from the previous step, and list all the subsets (with Ws elements) of set {1, 2, · · · ms }
intended for the construction of the parity-check matrix for the source code HS .
Step 5: According to Theorem 1, select the desired subsets for constructing the column
vector of the parity-check matrix for source code HS . This matrix must satisfy certain
constraints to ensure effective error-correction performance.
Entropy 2023, 25, 1189 6 of 10

Algorithm 1 Construction of LDPC code pair

Require:
mc : row number of source LDPC code
Ws : column weight of source LDPC code
Wc : column weight of channel LDPC code
s: girth of the constructed source code and channel code.
Ensure:
Initialize: parity-check matrix for channel code HC = [Imc ], nc = mc .
1: Begin
2: T = {T1 , T2 , · · · , Ti }; . All elements in T are a subset of {1, · · · , mc } containing Wc
elements, i.e., Ti = {ei,1 , · · · , ei,Wc }.
3: T1 = { e1,1 , · · · , e1,Wc };
4: HC = [ HC | hT1 ]; . The elements in T1 indicate the row indices of ‘1’ in the associated
column vector hT1 .
5: nc = nc + 1;
m
6: for i = 2 : (Wc ) do
c
7: ts = s; . If the number of columns in the parity-check matrix is less than mc + s,
the inter-set constraint conditions will change.
8: if nc − mc + 1 < s then
9: . The variable nc represents the number of columns in the parity check matrix HC .
10: ts = nc − mc + 1;
11: end if
12: l = 0;
13: for t = 2 : ts do
14: if t = 2 then

15: I = max |(T j ∩ Tk )|, ∀T j , Tk ⊆ E (n), j 6= k , j, k ∈ [1, nc − mc + 1]; . Calculate
the maximum number of common elements between any two subsets when
s = 2.
16: else  
17: I = max | ∪t∀T ,T ⊆E (n),j6=k (T j ∩ Tk )| , j, k ∈ [1, nc − mc + 1]; . Calculate the
j k
maximum number of common elements between subsets when s > 2.
18: end if
19: if I < t then
20: l = l + 1;
21: else
22: break;
23: end if
24: end for
25: if l = ts − 1 then
26: . Determine whether the subset Ti satisfies Theorem 1.
27: H C = [ H C | h Ti ];
28: nc = nc + 1;
29: else
30: HC = HC ;
31: end if
32: end for
33: Construct the parity-check matrix of the source code HS = [HTi ] similarly according to
lines 2 to 32.
34: End

4. Simulation Results
We evaluate the performance of the proposed JSCC system over AWGN channels
in terms of bit error rate (BER) in this section. The simulation results are obtained by
considering binary phase-shift keying modulation and employing joint belief propagation
iterative decoding. The benchmark code pairs are generated using the progressive edge-
growth algorithm with ‘copy-and-permute’ [23]. In contrast to the benchmark code pair,
Entropy 2023, 25, 1189 7 of 10

the channel code of the proposed code pair does not include punctured VNs. The complex-
ity of proposed code pairs is primarily determined via the column weight of the proposed
code, resulting in a complexity of O(nW ), where W is the larger value between the source
code column weight Ws and the channel code column weight Wc . The maximum number
of iterations is set as 50, and that of erroneously decoded frames is set as 100 for all the
signal-to-noise ratios (SNRs), denoted as Eb /N0 in Figures 2 and 3.
Figure 2 presents the simulated BER values for the benchmark code pair (repeat-by-4-
jagged-accumulate (R4JA) [24], Accumulate-repeat-by-4-jagged-accumulate (AR4JA) [25]),
and the proposed code pair according to Algorithm 1. The source and channel codes of the
benchmark code pair are rate-1/5 R4JA and rate-2/3 AR4JA. The girth of the proposed code
pair is 8, i.e., s = 3. The column weights of the source and channel codes of the proposed
code pair are Ws = 7 and Wc = 8, respectively. The frame length of the proposed code pair
is 1693, and the corresponding frame length of the benchmark code pair (R4JA, AR4JA)
is 1700. The code rates of the source and channel codes of the proposed code pair are 0.2
and 0.66. The code rate of the proposed code is slightly lower than that of (R4JA, AR4JA).
Nevertheless, for p = 0.01, the proposed code pair brings a 0.5 dB improvement over the
benchmark code pair (R4JA, AR4JA) when the BER is at the level of 10−5 .

10-4
BER

-5
10
0.5 dB

10-6

2.5 3 3.5 4 4.5 5

Eb/N0 (dB)
Figure 2. Comparison of BER performance between the proposed code pair and the code pair (R4JA,
AR4JA) for p = 0.01.

Figure 3 shows the simulated BER performance of the code pair (R4JA, AR4JA) and the
proposed code pair. The source and channel codes of the benchmark code pair are rate-1/4
R4JA and rate-2/3 AR4JA. The proposed code pair successfully avoids both length-4 and
length-6 cycles. The column weights of the source and channel codes are Ws = 7 and
Wc = 7. The code rates of the source and channel codes are 0.25 and 0.67. The girth of
the proposed code pair is 8. In this study, the frame length of the proposed code pair is
1088, and the corresponding frame length of the benchmark code pair is 1088. Nevertheless,
for p = 0.015, the proposed code pair brings 0.7 dB gain over the benchmark code pair
when the BER is at the level of 10−5 .
Figure 4 presents the simulated BER performance of the proposed code pair along-
side [15,16]. The code rates of the source and channel codes in [15,16] are 0.25 and 0.5,
respectively. The proposed code pair effectively eliminates cycle of length-4 and length-6.
The column weights of the source and channel codes are Ws = 8 and Wc = 12, respectively.
The code rates of the source and channel codes are 0.24 and 0.5, respectively. The frame
Entropy 2023, 25, 1189 8 of 10

length of the proposed code pair is 1404, whereas the code pairs in [15,16] correspond
to a frame length of 1400. When the probability p is set to 0.025, the proposed code pair
outperforms the code pair in [15] by yielding a gain of 1 dB at the bit error rate of 10−4 .
Similarly, compared with the code pair in [16], the proposed code pair achieves a gain of
0.6 dB at the same bit error rate.

10-3

10-4
BER

10-5
0.7 dB

10-6

1.5 2 2.5 3 3.5 4 4.5 5

Eb/N0 (dB)
Figure 3. Comparison of BER performance between the proposed code pair and the code pair (R4JA,
AR4JA) for p = 0.015.

-3
10
BER

10-4 0.6 dB 0.4 dB 0.5 dB

1 1.5 2 2.5 3 3.5

Eb/N0 (dB)
Figure 4. Comparison of BER performance involves the proposed code pair, the code pairs
from [15,16], and the code pair (R4JA, AR4JA) for p = 0.025.
Entropy 2023, 25, 1189 9 of 10

5. Conclusions
In this article, an algebraic construction method for LDPC code pairs of JSCC systems
has been proposed based on the graph-theoretic approach. The constructed code pairs could
avoid short-length cycles by following set constraints. The simulation results has shown
that the proposed code pairs achieve significant performance improvement compared with
the benchmark code pair (R4JA, AR4JA), [15,16]. The performance enhancement has been
observed in both the waterfall region and the error-floor region. In addition, the proposed
code pairs customized for the JSCC system have flexible frame lengths (i.e., enhanced
adaptability over various channel conditions) unlike the protograph code pairs.

Author Contributions: Conceptualization, Y.L., J.H., W.X. and L.W.; methodology, Y.L. and J.H.;
software, Y.L.; validation, Y.L.; data curation, Y.L.; writing—original draft preparation, Y.L.; writing—
review and editing, J.H., W.X. and L.W.; supervision, J.H., W.X. and L.W.; funding acquisition, W.X.
All authors have read and agreed to the published version of the manuscript.
Funding: This work was supported in part by the National Natural Science Foundation of China
under Grant 61871337 and 61671395. This work was supported in part by the Key Laboratory of
Southeast Coast Marine Information Intelligent Perception and Application, Ministry of Natural
Resources (MNR), No. KFJJ20220201.
Institutional Review Board Statement: Not applicable.
Data Availability Statement: The data that support the findings of this study are available from the
corresponding author upon reasonable request.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Shannon, C.E. A mathematical theory of communication. Bell Syst. Tech. J. 1948, 27, 379–423. [CrossRef]
2. Guyader, A.; Fabre, E.; Guillemot, C.; Robert, M. Joint source-channel turbo decoding of entropy-coded sources. IEEE J. Sel. Areas
Commun. 2001, 19, 1680–1696. [CrossRef]
3. Hedayat, A.; Nosratinia, A. Performance analysis and design criteria for finite-alphabet source-channel codes. IEEE Trans.
Commun. 2004, 52, 1872–1879. [CrossRef]
4. Lakovic, K.; Villasenor, J. Combining variable length codes and turbo codes. In Proceedings of the IEEE 55th Vehicular Technology
Conference, Birmingham, AL, USA, 6–9 May 2002; pp. 1719–1723.
5. Lakovic, K.; Villasenor, J. On reversible variable length codes with turbo codes, and iterative source-channel decoding. In
Proceedings of the IEEE International Symposium on Information Theory, Lausanne, Switzerland, 30 Jun–5 July 2002; p. 170.
6. Jaspar, X.; Vandendorpe, L. Design and performance analysis of joint source-channel turbo schemes with variable length codes.
In Proceedings of the IEEE International Conference on Communications, Seoul, Republic of Korea, 16–20 May 2005; pp. 526–530.
7. Poulliat, C.; Declercq, D.; Lamy-Bergot, C.; Fijalkow, I. Analysis and optimization of irregular LDPC codes for joint source-channel
decoding. IEEE Commun. Lett. 2005, 9, 1064–1066. [CrossRef]
8. Zribi, A.; Pyndiah, R.; Zaibi, S. Low-complexity soft decoding of Huffman codes and iterative joint source channel decoding.
IEEE Trans. Commun. 2012, 60, 1669–1679. [CrossRef]
9. Fresia, M.; Pérez-Cruz, F.; Poor, H.V. Optimized concatenated LDPC codes for joint source-channel coding. In Proceedings of the
IEEE International Symposium on Information Theory, Seoul, Republic of Korea, 28 June–3 July 2009; pp. 2131–2135.
10. Fresia, M.; Perez-Cruz, F.; Poor, H.V.; Verdu, S. Joint source and channel coding. IEEE Signal Process. Mag. 2010, 27, 104–113.
[CrossRef]
11. He, J.; Wang, L.; Chen, P. A joint source and channel coding scheme base on simple protograph structured codes. In Proceedings
of the International Symposium on Communications and Information Technologies, Gold Coast, Australia, 2–5 October 2012;
pp. 65–69.
12. Chen, C.; Wang, L.; Liu, S. The design of protograph LDPC codes as source codes in a JSCC system. IEEE Commun. Lett. 2018, 22,
672–675. [CrossRef]
13. Chen, Q.; Wang, L.; Hong, S.; Xiong, Z. Performance improvement of JSCC scheme through redesigning channel code. IEEE
Commun. Lett. 2016, 20, 1088–1091. [CrossRef]
14. Xu, Z.; Wang, L.; Hong, S.; Chen, G. Design of Code Pair for Protograph LDPC Codes-Based JSCC System with Joint Shuffled
Scheduling Decoding Algorithm. IEEE Commun. Lett. 2021, 25, 3770–3774. [CrossRef]
15. Chen, Q.; Wang, L.; Hong, S.; Chen, Y. Integrated design of JSCC scheme based on double protograph LDPC codes system. IEEE
Commun. Lett. 2019, 23, 218–221. [CrossRef]
Entropy 2023, 25, 1189 10 of 10

16. Hong, S.; Ke, J.; Wang, L. Global Design of Double Protograph LDPC Codes for Joint Source-Channel Coding. IEEE Commun. Lett.
2023, 14, 424–427. [CrossRef]
17. Golmohammadi, A.; Mitchell, D.G. Concatenated Spatially Coupled LDPC Codes with Sliding Window Decoding for Joint
Source-Channel Coding. IEEE Trans. Commun. 2021, 70, 851–864. [CrossRef]
18. Djurdjevic, I.; Lin, S.; Abdel-Ghaffar, K. Graph-theoretic construction of low-density parity-check codes. IEEE Commun. Lett. 2003,
7, 171–173. [CrossRef]
19. Ammar, B.; Honary, B.; Kou, Y.; Xu, J.; Lin, S. Construction of low-density parity-check codes based on balanced incomplete block
designs. IEEE Trans. Inf. Theory 2004, 50, 1257–1269. [CrossRef]
20. Kou, Y.; Lin, S.; Fossorier, M.P. Low-density parity-check codes based on finite geometries: A rediscovery and new results. IEEE
Trans. Inf. Theory 2001, 47, 2711–2736. [CrossRef]
21. Tang, H.; Xu, J.; Kou, Y.; Lin, S.; Abdel-Ghaffar, K. On algebraic construction of Gallager and circulant low density parity-check
codes. IEEE Trans. Inf. Theory 2004, 50, 1269–1279. [CrossRef]
22. Wu, G.; Lv, Y.; He, J. Design of high-rate LDPC codes based on matroid theory. IEEE Commun. Lett. 2019, 23, 2146–2149. [CrossRef]
23. Hu, X.Y.; Eleftheriou, E.; Arnold, D.M. Regular and irregular progressive edge-growth tanner graphs. IEEE Trans. Inf. Theory
2005, 51, 386–398. [CrossRef]
24. Divsalar, D.; Dolinar, S.; Jones, C.R.; Andrews, K. Capacity-approaching protograph codes. IEEE J. Sel. Areas Commun. 2009, 2009,
876–888. [CrossRef]
25. Divsalar, D.; Dolinar, S.; Jones, C. Construction of protograph LDPC codes with linear minimum distance. In Proceedings of the
IEEE International Symposium on Information Theory, Seattle, DC, USA, 9–14 July 2006; pp. 664–668.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.