Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Comprehensive Study on CC-LDPC, BC-LDPC and

Polar Code
Kun Zhu, Zhanji Wu*
School of Information and Communication Engineering
Beijing University of Posts and Telecommunications
Beijing, China
*Corresponding author: Zhanji Wu, wuzhanji@bupt.edu.cn

Abstract— Channel coding scheme is very crucial for the Arvind Sridharan investigated the free distance and the lower
next sixth generation (6G) mobile communication. In this paper, bound of CC-LDPC in [3] [4]. According to the research,
we investigate the encoding and decoding methods of the state- CC-LDPC has been proved to have larger free distance than
of-the-art channel codes, which include the convolutional code
LDPC (CC-LDPC), block code LDPC (BC-LDPC) and polar BC-LDPC in regular ensemble with the same row and column
code to access the benefits from CC-LDPC. And then, we present weight. In [5] [6] [7], the regular terminated (TE) CC-LDPC
comprehensive evaluation to compare their error performance, is proved to have more excellent performance as compared
decoding complexity and latency. Our research shows that CC- with BC-LDPC under belief propagation (BP) decoding. In
LDPC has the advantages in terms of high reliability, low addition, the BP threshold of TE-CC-LDPC can reach the
complexity and low latency. Finally, we propose some open
research problems and solving ideas for CC-LDPC application maximum likelihood (ML) decoding threshold of BC-LDPC
in 6G communication. [2] [8]. This phenomenon is called ”threshold saturation
effect”. In other word, we can use linear complexity (CC
Index Terms—convolutional-code low-density parity-check structure) instead of exponential complexity (ML decoding)
(CC-LDPC), block code LDPC (BC-LDPC) , polar code, low to achieve the same performance. However, the TE also
latency, high reliability.
makes rate loss due to the slightly irregular structure in the
head and the tail of the codeword. For long code length, the
I. I NTRODUCTION
rate loss is very small, but for short code length, the rate loss
In the recent two decades, as turbo code, low-density is non-negligible. To solve this problem, M. B. S. Tavares etc.
parity-check (LDPC) code and polar code emerged, the proposed a Tail-Biting (TB) CC-LDPC in [9] [10]. However,
state-of-the-art channel coding schemes have approached the TB structure will worsen the codes performance. In [11],
the Shannon limit very closely. In the fifth generation authors compared the finite length TE-CC-LDPC, TB-CC-
(5G) mobile communication standard, the block code (BC) LDPC and BC-LDPC. Through the literatures, we can learn
LDPC and polar code are utilized for data channel and about the huge potential of CC-LDPC. Nevertheless, there
control channel, respectively. Although these schemes get are still many technical problems to be explored to reach
excellent performance, they still have obvious limitations. the potential. In this paper, we consider that the CC-LDPC
The BC-LDPC has some shortcomings in terms of slow has a semi-infinite structure, which means the code length is
decoding convergence, high complexity and long decoding infinite to avoid stopping problems.
latency. Meanwhile, it does not perform very well for the
short code length and low code rate. The polar code also The rest of this paper is structured as follows. We firstly
has unacceptable high decoding latency for long code length introduce the encoding and decoding methods of BC-LDPC,
due to the serial decoding calculation. For the future 6G polar code and CC-LDPC code in section II, III and IV,
communication scenario, higher reliability, lower latency and respectively. Secondly, we carry out comprehensive study to
higher throughput are required to meet the real-time high-rate compare their error performance, decoding complexity and
data transmission. The convolutional code (CC) LDPC has latency in section V, which turns out that CC-LDPC has
enormous potentials in terms of lower error floor, lower some obvious advantages in terms of high reliability, low
decoding latency and lower decoding complexity. complexity and low latency. In section VI, we sum up some
The CC-LDPC was first proposed by M. Lentmaier, open problems and solution ideas for CC-LDPC. Finally, we
A. Sridharan.et in [1], since its coding structure is like a reach some conclusions in section VII.
convolutional code. In [2], the term ”spatially coupled”
was firstly proposed. This term is a general concept to
denote the phenomenon of coupling several independent II. BC-LDPC
code as a convolutional-like structure. Many properties of BC-LDPC can be considered as the conventional LDPC
CC-LDPC has been proved to be better than BC-LDPC. code. The code is defined by a sparse N × K parity-check

‹,(((
Authorized licensed use limited to: University of Exeter. Downloaded on July 02,2020 at 14:38:51 UTC from IEEE Xplore. Restrictions apply.
matrix H. K is the length of source sequence and N is the information sequence as u = [u0 , u1 , . . . , uK ] and codeword
codeword length, N < K. Each column of H represents a is v = [v0 , v1 , . . . , vN ]. The codeword satisfied as (2).
check node, and each row is denoted by a variable node.
Hv T = 0 (2)
Sparse means the number of zeros take very large proportion
of H. For an irregular BC-LDPC with a code rate of 1/2, The encoded sequence can be divided to two part
when its code length approaches infinity, the Eb/N0 threshold v = [v data , v check ]. The first part v data = u. The
distance required for reliable communication on the binary second part v check are the redundant check bits.
input AWGN channel is only 0.0045dB from Shannon limit
[12].
C. Decoding
With discovery of the Belief Propagation(BP) decoding
A. Construct method algorithm, LDPC reveals its great performance. The reliable
The BC-LDPC can be classified as regular and irregular soft messages are exchanged through adjacent nodes during
code. The regular code put forwarded by Gallager [13]. In iterations. On this basis, offset Min-Sum and scaled Min-Sum
2001 [14], irregular LDPC has been proposed by Luby et al. algorithms are derived.
For the result of introduce different constraints into the LDPC
code, the irregular LDPC code improve the performance
dramatically. In recent decade, there are many construct III. POLAR CODE
method to optimal the structure of code efficiently. The
Progressive Edge Growth (PEG) algorithm having the girth The core of the structure is processed by channel
as the target parameter. Approximate Cycle Extrinsic (ACE) polarization [17]. When the code tend to infinite long, part of
method adds more parameters to PEG for lower error floor. the channel will tend to be a perfect channel with a capacity
Degree Distributions (DD) is an important structure property
of the base matrix. The most common tool to determine close to 1 and the other part of channel will tend to be 0.
the DD is Density Evolution (DE) based analysis. After Using the reliable part to transmit the information bits to
determined DD, Protograph-based or Quasi Cyclic (QC) make few wrong bits as possible.
LDPC code construction have used the DD and lifting the
original basic matrix to obtain the optimized final matrix.
⎡ ⎤ A. Construct method
1 1 1 1 1 0
⎢ 1 0 1 1 1 1 ⎥ The basic idea of the polar coding structure is to find out
B=⎣
0 1 0 1 1 1 ⎦ the unreliable part of channel, which is usually called the
1⎡ 1 1 0 0 1 ⎤ Frozen Set. The mainstream of the method to construct a
D W1 D W4 D W7 DW10 DW13 −1 (1)
⎢ D W2 −1 D W8 DW11 DW14 D W16 ⎥ good polar code is based on DE and Gaussian Approximation
HQC =⎢
⎣ −1

D W5
−1 DW12 DW15 DW17 ⎦ (GA) [18]. The polar code which is used for comparison is
D W3 D W6 D W9 −1 −1 D W18 the systematic polar code [19].
We use the construct method, which is proposed by
Bocharova [15], to construct the QC-LDPC-BC. The detailed B. Encoding
introduction is summarized as follow two steps: Let u is the source code and the x is the codeword. The
1) As the example in equation (1), the B is the basic linear transformation process over a field F is expressed as:
matrix. To optimize B, there are five criteria to ensure
good circle characteristic in [16]. The main idea is based x = uG x, u ∈ FN , G ∈ FN ×N (3)
on ACE-like algorithm. N is the block size of generator matrix G, N = 2n . F is a
2) Then we replace the elements in the B by S ×S identity fixed binary field F2 . The N × N generator matrix is derived
diagonal Cyclic Permutation Matrixes (CPM) to obtain from the equation:
the final HQC . The 0 element is marked as all-zero  
matrix correspond to −1 in HQC . The permutation ⊗n 1 0
G N = F ΠN , F = (4)
factors DWi ∈ [0, S] is searched by optimizing Girth 1 1
Profile(GP). Where F ⊗n denotes the nth Kronecker power of F . The ΠN
In addition, we also use this method to construct the HQC represent an N × N permutation matrix.
for weakly coupled CC-LDPC. The construct method of CC-
C. Decoding
LDPC in this paper is based on the BC-LDPC construc-
t method above. We consider good Sliding-Window Girth
(SWG) for CC-LDPC in the step 2).

B. Encoding
LDPC code originate from linear code. The improvement
of this code is come from redundant check bits. We set Fig. 1. system of Polar code using CA-SCL

Authorized licensed use limited to: University of Exeter. Downloaded on July 02,2020 at 14:38:51 UTC from IEEE Xplore. Restrictions apply.
For polar code the standard decoding algorithm is CC-LDPC.
Successive Cancelation (SC). To further improve the In [26], the CC-LDPC be divided into weakly and strongly
performance A. Balatsoukas-Stimming propose the SC List coupled respectively. This idea is based on maximize
(SCL) decoder. This method splitting the decoding path into minimum distance. The lower bound of minimum distance is
some parallel path. The decoding complexity is in direct dmin ≤ (m + 1)(μm + 1). The m is denoted as the number of
proportion to the number of list. The CRC-aided SCL (CA- row in each sub-matrix. For saving computation complexity,
SCL) decoding algorithm is proposed in [20]. The system the syndrome former constraint length vs = m × (μ + 1)
model is shown as Fig.1. By the turbo decoding in the dotted should not be large [27]. Therefore, μ and m could not be
box, the performance of polar code get significant improve, large at the same time. In order to ensure the minimum
also the decoding complexity get reduced. In this paper, we distance is large enough, one of m and μ need to be large.
use the CA-SCL decoder to achieve the best performance. The paper defined the code which have large μ and micro
The polar code get limit performance in relative higher code m as strongly coupled. Conversely, the large m and small
rate, because of the less Frozen Set is required after channel μ code is defined as weakly coupled. For strongly coupled
polarization. code, author construct it by protograph-based method and
extend the non-zero elements to S × S CPMs. For the sake of
avoiding cycles of length 6, the strongly coupled code need to
IV. CC-LDPC
be time-varying. The weakly coupled code is easier to design
A. Construct method for it have relatively large sub-matrixes, so we construct it
As shown in (5) , the general check matrix of CC-LDPC as time-invariant. We construct the single basic matrix and
are consisted by μ + 1 sub-matrixes and arranged them as ensure the shift value for S × S CPMs by the BC-LDPC
a diagonal zigzag graph, μ is called the syndrome former construct method, which we have mentioned above. And
memory. It is necessary to comprehend that CC-LDPC is using edge spreading operation to split the matrix HQC into
divide to time-varying and time-invariant two kinds code prior μ + 1 sub-matrix. Experimental results show these two kinds
to construct CC-LDPC. These design methods reduce the of CC-LDPC get nearly similar performance.
problem of code construction to designing a one-period parity-
check matrix. So that the construction of giant check-parity
B. Encoding
matrix become practical. The (5) is the general expression
of CC-LDPC. The index in parentheses represents the time We let the m, n, S denote the rows, columns and circulating
parameter. Assuming the period is T, Hi (t) = Hi (t + T ), factor of each sub-matrix, respectively. The size of the sub-
when T = 1 the (5) is an time-invariant CC-LDPC. basic-matrices (sub-B) is m × n. The dim of Hi = mS × nS.
⎡ ⎤ If we terminate the code at time parameter t = LCC the
H0 (0) code length N = LCC nS. The information sequence and
⎢ H1 (0) H0 (1) ⎥
⎢ ⎥ code sequence can be recorded in chronological order as
⎢ .. .. ⎥
⎢ . H1 (1) . ⎥ u[0,t] = [u0 , u1 , · · · , ut ] and v[0,t] = [v0 , v1 , · · · , vt ]. In
⎢ ⎥
Hc = ⎢ ⎢ . .. ⎥ (5) each point of time j, uj = [u(j,0) , u(j,1) , . . . , u(j,m) ] vj =
H (0) .. . H (t) ⎥
⎢ μ 0 ⎥ [v(j,0) , v(j,1) , . . . , v(j,n) ]. For each vj = [vjdata , vjcheck ], the
⎢ . ⎥
⎢ Hμ (1) H1 (t) . . ⎥ front part vjdata = uj and the second part vjcheck is a check
⎣ ⎦
.. .. .. sequence of length mS. As the extension of the encoding
. . . algorithm of BC-LDPC, the encoding processes are show as
The first method to construct time-invariant code was follow equations:
proposed by Tanner [21]. Several years later, [1] proposed 0 = H0 v0T
the time-varying construct method. And they first used
= H1 v0T + H0 v1T
unwrapping procedure to obtain the parity-check matrix
of a periodically time-varying convolutional code from the = H2 v0T + H1 v1T + H0 v2T (6)
parity-check matrix of a block code. These two construct ..
methods can be defined as the primary construction method .
T T T
for CC-LDPC [22] [23] [24]. Since then, people bring forward = Hμ vt−μ + Hμ−1 vt−(μ−1) · · · H1 vt−1 + H0 vtT
many design ways aim to get better CC-LDPC. We construct a Sm × Sm unit matrix at the last part of H0 .
Protograph-based method is a very common way for According to the last equation in (6), we summarized the
CC-LDPC. Since Thorpe [22], recognized that protographs mathematic expression of the encoder as v(j,i) = u(j,i) , i ∈
provide an efficient way to constructing LDPC codes. The [1, (n − m)S]. For the check bits, the expression is shown as:
protograph-based LDPC inherent the properties of protograph.
(n−m)S
In [24] Mitchell use this method to obtain CC-LDPC. In (t−(n−m)S,k)
v(i,j) = H0 v(j,k) +
that work, the block protographs are coupling into a single k=1
μ nS (7)
chain by edge spreading operation. In [25] Mitchell, analysis (t−(n−m)S,L)
Hp v(j−p,k) , i ∈ ((n − m)S, nS]
the minimum distance and performance for protograph-based p=1 k=1

Authorized licensed use limited to: University of Exeter. Downloaded on July 02,2020 at 14:38:51 UTC from IEEE Xplore. Restrictions apply.
In the (7), the left part of equation known as the messages V. C OMPARISON AND A NALYSIS
which come from the other sub-matrices. This is the direct
A. Complexity and Decoding Latency
presentation of coupling part.
The N is defined as code length, L is the width of list
and d¯v is the average of variable nodes’ degree. In both
C. Decoding BC-LDPC and CC-LDPC, Ie is defined as the number of
iterations per code bit. The Ie in CC-LDPC is equivalent to
In this work, we consider BP algorithm as the basic the I in BC-LDPC. We compare the CC-LDPC, BC-LDPC
decoding method. Nevertheless, if we use the conventional and polar code in Table I. Assuming the d¯v is 3 in each
BP algorithm for long LCC code, the decoding latency will LDPC class code, N equal to 214 , Ie = 30 and L = 8. The
be unacceptable [28]. We use the Sliding Window Decoding bit decoding complexity of CC-LDPC, BC-LDPC and polar
(SWD) algorithm [15] and layered decoding algorithm [29] code are 690, 690 and 784, respectively.
to reduce the decoding latency and complexity. As the Fig.2 For the decoding latency, the CC-LDPC and BC-LDPC
shows an example of SWD decoding, the size of decoding adopt parallel decoding, the decoding time complexity is
window is (W + μ)nS × W mS. In the example, W = 4 is O(N · Ie ), which is in direct proportion to N . Polar code
the length of decoding window, length of memory μ = 2. The employ serial decoding, the decoding time complexity is
blue area indicate the window being decoded and the pink O(LN log2 N ), which is in direct proportion to N log2 N .
area transmit the memorized information to current decoding Therefore, the decoding delay of CC-LDPC is an order of
window. With the time index changed from t to t + 1 the magnitude better than polar code. For long length code, the
decoding window slide to the area in dot dash line. We can CC-LDPC and BC-LDPC save many operations compare
also know the data in and data out position when time is t from to polar code. In a word, the CC-LDPC get same decoding
Fig.2, each sub-block of message need pass through whole complexity and latency with the BC-LDPC for Nw = N ,
window until it be decoded. We let the number of iterations Ie = I and same DD. The polar code get higher complexity
of the window as Iw . So that, each bit will get equivalent and latency compare to others, especially for long length
iterations Ie = Iw ×W times. We adopt the Layered Decoding code. We will concrete analysis decoding complexity and
Algorithm (LDA) into the CC-LDPC decoding scheme to latency according to the simulation in next subsection.
improve the converge speed [30].

B. Simulation and Analysis


We consider the performance comparison in Additive White
Gaussian Noise (AWGN) channel for Quadrature Phase Shift
Keying (QPSK) modulation. In spite of the irregular LDPC
codes can reach the better performance, there is randomness
in the search process. Thus, we construct the regular codes and
choose the best kind of codes into the comparison to ensure the
fairness of contrast. In these comparison, we let the window
length Nw = (W + μ)nS for weakly coupled CC-LDPC and
Nw = (W + Tμ )nT S for strongly coupled CC-LDPC. In
order to guarantee the constant code length comparison, Nw
need equal to code length N in BC-LDPC. The CC-LDPC are
semi-infinite construct, which means there is no truncation.
Firstly, we construct the regular weakly and strongly coupled
CC-LDPC compare with regular BC-LDPC and systematic
Fig. 2. Example of SWD polar code. In order to ensure the comparisons are between
the best performance, we construct the BC-LDPC and CC-
LDPC in different choice of degree, then we pick the best one
TABLE I in each kind of code. In the process of practice, we found the
C OMPLEXITY C OMPARISON P ER B IT regular BC-LDPC is good at dv = 3, the regular CC-LDPC is
good at dv = 4. Hence, we use this condition in the following
CC-LDPC BC-LDPC Polar-code
Operation experiments.
LDA-Log-BP Log-BP CA-SCL
Fig.3 illustrates the codes comparison at about 16000 code
addition/subtraction (4d¯v − 1)Ie (4d¯v − 1)Ie 3L · log2 N
length and 0.5 code rate in terms of bit error rate (BER) and
table look-up 4d¯v Ie 4d¯v Ie 3L · log2 N
frame error rate (FER). The P, B and C represent polar code,
comparison - - L · log2 N
BC-LDPC and CC-LDPC, respectively. The corresponding
total complexity (8d¯v − 1)Ie (8d¯v − 1)Ie 7L · log2 N
parameters are shown in Table II. In Fig.3, the polar and
BC-LDPC perform better than CC-LDPC in low SNR region.

Authorized licensed use limited to: University of Exeter. Downloaded on July 02,2020 at 14:38:51 UTC from IEEE Xplore. Restrictions apply.
Fig. 3. Error Performance Comparison at 0.5 code rate Fig. 4. Error Performance Comparison at 0.8 code rate

TABLE II
CODE PARAMETERS OF F IG .3 error performance of polar code get worse as compared with
other LDPC codes. C3 (Ie = 10) and C4 (Ie = 15) perform
code name CPM size size of B(sub-B) μ W N (Nw ) 0.15dB better than B2 (I = 25) at 1e-6 BER and 0.4dB better
P1 - - - - 16384 than P2 (L = 8) at 1e-5 BER. The decoding complexity of
B1 683 × 683 12 × 24 - - 16392 C3 (Ie = 10) and C4 (Ie = 15) are as much as 54% and 80%
C1 68 × 68 20 × 40 1 5 16320 that of B2 (I = 25), respectively, while the decoding latency
C2 49 × 49 2×4 12 10 16464 (iteration number) of C3 (Ie = 10) and C4 (Ie = 15) are as
much as 40% and 60% that of B2 (I = 25), respectively. For
the same latency Ie = I = 25, C3 (Ie = 25) and C4 (Ie = 25)
However, in the high SNR area, the CC-LDPC codes have have 0.25dB coding gain as compared with B2 (I = 25) at
obvious advantage. As shown in Fig.3, the C1 (Ie = 10) and 1e-6 BER.
C2 (Ie = 20) have the same performance as B1 (I = 50)
and P1 (L = 8). Given the same performance, the decoding VI. O PEN R ESEARCH P ROBLEMS
complexity of C1 (Ie = 10) is as much as 27% of B1 (I = 50)
Although CC-LDPC exhibits excellent potential, there
or 39.5% of P1 (L = 8). The decoding complexity of C2
are still many problems to be further studied [31], and we
(Ie = 20) is as much as 54% and 79% for B1 (I = 50)
summarize them as follow:
and P1 (L = 8), respectively. Similar results are also for FER
performance, so CC-LDPCs have the best performance and
the fastest convergence. Increasing the Iw to Ie = 30, C1 and 1 The first problem is that the current CC-LDPC codes do
C2 codes get larger coding gain, while the complexity is only not perform very well at short-to-moderate code length.
80% of B1 (I = 50). We speculate that CC-LDPC codes are more sensitive
to the minimum distance than BC-LDPC. We need to
propose a special construction method to maximize the
TABLE III minimum distance of the SC-structured base matrices for
CODE PARAMETERS OF F IG .4 short-to-moderate code length, or we may slightly change
the coupling structure to reduce the sensitivity.
code name CPM size size of B(sub-B) μ W N (Nw )
2 As mentioned above, the TE-CC-LDPC codes get sig-
P2 - - - - 32768
nificant coding gain. However, the LCC leads to some
B2 1632 × 1632 4 × 20 - - 32640
rate-loss for the code. A proper termination scheme can
C3 182 × 182 6 × 30 1 5 32670 take the potential of CC-LDPC to mitigate the rate-loss
C4 82 × 82 1×5 30 10 32800 without performance cost. It still needs more study to
solve this issue .
In Fig.4, a high code rate 0.8 is considered, and the 3 A key problem for CC-LDPC is rate-compatible and
corresponding parameters are shown in Table III. In Fig.4, the hybrid ARQ (HARQ) design method. Some methods

Authorized licensed use limited to: University of Exeter. Downloaded on July 02,2020 at 14:38:51 UTC from IEEE Xplore. Restrictions apply.
were proposed to construct it [32] [33]. We still need [11] Leven, Andreas, and Laurent Schmalen. “Status and recent advances on
to find an efficient rate compatible scheme to make CC- forward error correction technologies for lightwave systems.” Journal of
Lightwave Technology 32.16 (2014): 2735-2750.
LDPC more flexible and reliable for various code rates [12] Richardson, Thomas J., Mohammad Amin Shokrollahi, and Rdiger L.
and HARQ applications. Urbanke. “Design of capacity-approaching irregular low-density parity-
check codes.” IEEE transactions on information theory 47.2 (2001): 619-
VII. C ONCLUSION R EMARKS 637.
[13] R. G. Gallager, “Low-Density Parity-Check Codes,” IRE Trans. Info.
The aim of this paper is to present the advantages of CC- Theory, vol. 8, no. 1, Jan. 1962, pp. 21C28.
LDPC code as compared with BC-LDPC and polar. Firstly, [14] M. G. Luby et al., “Improved Low-Density Parity-Check Codes using
Irregular Graphs,” IEEE Trans. Info. Theory, vol. 47, no. 2, Feb. 2001,
we briefly introduce the construct methods, encoding and pp. 585C98.
decoding algorithms for these codes. Secondly, we compare [15] Bocharova, Irina E., Boris D. Kudryashov, and Rolf Johannesson.
their error performance, decoding complexity and latency. “Searching for binary and nonbinary block and convolutional LDPC
codes.” IEEE Transactions on Information Theory 62.1 (2015): 163-183.
The results show that the CC-LDPC has obvious advantages [16] Wu, Zhanji, and Xin Huang. “A LDPC convolutional code optimization
in terms of lower error floor, faster decoding convergence and method for FTN systems.” 2017 3rd IEEE International Conference on
lower decoding complexity. We also discuss about several Computer and Communications (ICCC). IEEE, 2017.
[17] Arikan, Erdal. “Channel polarization: A method for constructing
open problems and solution ideas for CC-LDPC. In a word, capacity-achieving codes for symmetric binary-input memoryless chan-
CC-LDPC is a very promising channel coding scheme with nels.” IEEE Transactions on information Theory 55.7 (2009): 3051-3073.
high reliability and low latency for the next 6G mobile [18] Mori, Ryuhei, and Toshiyuki Tanaka. “Performance and construction
of polar codes on symmetric binary-input memoryless channels.” 2009
communication. IEEE International Symposium on Information Theory. IEEE, 2009.
[19] Arikan, Erdal. “Systematic polar coding.” IEEE communications letters
15.8 (2011): 860-862.
ACKNOWLEDGMENT [20] Niu, Kai, and Kai Chen. “CRC-aided decoding of polar codes.” IEEE
Communications Letters 16.10 (2012): 1668-1671.
This work is sponsored by the National Natural Science [21] Tanner R M.“ Convolutional codes from quasi-cyclic codes: A link
Fund (61171101) and the Fundamental Research Funds for between the theories of block and convolutional codes”[M]. University
the Central Universities of P.R. China. of California, Santa Cruz, Computer Research Laboratory, 1987.
[22] Thorpe, Jeremy. “Low-density parity-check (LDPC) codes constructed
from protographs.” IPN progress report 42.154 (2003): 42-154.
[23] D. G. M. Mitchell, A. E. Pusane, K. S. Zigangirov and D. J. Costello,
R EFERENCES “Asymptotically good LDPC convolutional codes based on protographs,”
[1] A. Jiminez Felstrom and K. S. Zigangirov, “Time-varying periodic 2008 IEEE International Symposium on Information Theory, Toronto,
convolutional codes with low-density parity-check matrix,” IEEE Trans. ON, 2008, pp. 1030-1034. doi: 10.1109/ISIT.2008.4595143
Inf. Theory, Vol. 45, No. 6 (1999). [24] Mitchell, David GM, Michael Lentmaier, and Daniel J. Costello. “S-
[2] S. Kudekar et al, “Threshold saturation via spatial coupling: Why patially coupled LDPC codes constructed from protographs.” IEEE
convolutional LDPC ensembles perform so well over the BEC,” IEEE Transactions on Information Theory 61.9 (2015): 4866-4889.
Trans. Inf. Theory, Vol. 57, No. 2, p. 803 (2011). [25] Mitchell, David GM, Ali E. Pusane, and Daniel J. Costello. “Minimum
[3] A. Sridharan, D. Truhachev, M. Lentmaier, D. J. Costello and K. S. distance and trapping set analysis of protograph-based LDPC convolu-
Zigangirov, “Distance Bounds for an Ensemble of LDPC Convolutional tional codes.” IEEE Transactions on Information Theory 59.1 (2012):
Codes,” in IEEE Transactions on Information Theory, vol. 53, no. 12, 254-281.
pp. 4537-4555, Dec. 2007. doi: 10.1109/TIT.2007.909113 [26] Schmalen, Laurent, et al. “Spatially coupled soft-decision error correc-
[4] A. Sridharan, D. V. Truhachev, M. Lentmaier, D. J. Costello and K. S. tion for future lightwave systems.” Journal of Lightwave Technology
Zigangirov, “On the free distance of LDPC convolutional codes,” Inter- 33.5 (2014): 1109-1116.
national Symposium onInformation Theory, 2004. ISIT 2004. Proceed- [27] Battaglioni, Massimo, et al. “Design and analysis of time-invariant
ings., Chicago, IL, 2004, pp. 312-312.doi: 10.1109/ISIT.2004.1365349 SC-LDPC convolutional codes with small constraint length.” IEEE
[5] A. Sridharan, M. Lentmaier, D. J. Costello, Jr., and K. S. Zigangirov, Transactions on Communications 66.3 (2017): 918-931.
“Convergence analysis of a class of LDPC convolutional codes for the [28] Iyengar, Aravind R., et al. “Windowed decoding of protograph-based
erasure channel,” in Proceedings of the 42nd Allerton Conference on LDPC convolutional codes over erasure channels.” IEEE Transactions
Communication, Control, and Computing, (Monticello, IL, USA), 2004. on Information Theory 58.4 (2011): 2303-2320.
[6] M. Lentmaier, A. Sridharan, K. S. Zigangirov and D. J. Costel- [29] Hocevar, Dale E. “A reduced complexity decoder architecture via
lo, “Terminated LDPC convolutional codes with thresholds close layered decoding of LDPC codes.” IEEE Workshop onSignal Processing
to capacity,” Proceedings. International Symposium on Information Systems, 2004. SIPS 2004.. IEEE, 2004.
Theory, 2005. ISIT 2005., Adelaide, SA, 2005, pp. 1372-1376.doi: [30] D. Chang et al., “LDPC convolutional codes using layered decoding
10.1109/ISIT.2005.1523567 algorithm for high speed coherent optical transmission,” OFC/NFOEC,
[7] D. J. Costello, L. Dolecek, T. E. Fuja, J. Kliewer, D. G. M. Mitchell Los Angeles, CA, 2012, pp. 1-3.
and R. Smarandache, “Spatially coupled sparse codes on graphs: theory [31] Costello, Daniel J., et al. “Spatially coupled sparse codes on graphs:
and practice,” in IEEE Communications Magazine, vol. 52, no. 7, pp. Theory and practice.” IEEE Communications Magazine 52.7 (2014):
168-176, July 2014.doi: 10.1109/MCOM.2014.6852099 168-176.
[8] S. Kudekar, C. Measson, T. J. Richardson, and R. L. Urbanke, “Thresh- [32] Nitzold, Walter, Gerhard P. Fettweis, and Michael Lentmaier. “Spatially-
old saturation on BMS channels via spatial coupling,” in Proc. Int. Symp. coupled nearly-regular LDPC code ensembles for rate-flexible code
Turbo Codes Iterative Inf. Process. (ISTC), Brest, France, Sep. 2010, pp. design.” 2014 IEEE International Conference on Communications (ICC).
309C313. IEEE, 2014.
[9] M. B. S. Tavares, K. S. Zigangirov and G. P. Fettweis, “Tail- [33] Mitchell, David GM, et al. “Randomly punctured spatially coupled
Biting LDPC Convolutional Codes,” 2007 IEEE International Sym- LDPC codes.” 2014 8th International Symposium on Turbo Codes and
posium on Information Theory, Nice, 2007, pp. 2341-2345. doi: Iterative Information Processing (ISTC). IEEE, 2014.
10.1109/ISIT.2007.4557569
[10] M. B. S. Tavares, K. S. Zigangirov and G. P. Fettweis, “Tail-Biting
LDPC Convolutional Codes Based on Protographs,” 2007 IEEE 66th
Vehicular Technology Conference, Baltimore, MD, 2007, pp. 1047-
1051.doi: 10.1109/VETECF.2007.227

Authorized licensed use limited to: University of Exeter. Downloaded on July 02,2020 at 14:38:51 UTC from IEEE Xplore. Restrictions apply.

You might also like