Document downloaded from:
This paper must be cited as:
The final publication is available at
Copyright
Additional Information
"(c) 2016 IEEE. Personal use of this material is permitted. Permission from IEEE must be
obtained for all other users, including reprinting/ republishing this material for advertising or
promotional purposes, creating new collective works for resale or redistribution to servers or
lists, or reuse of any copyrighted components of this work in other works.")
http://dx.doi.org/10.1109/TBC.2015.2505414
http://hdl.handle.net/10251/82084
Institute of Electrical and Electronics Engineers (IEEE)
Michael, L.; Gómez Barquero, D. (2016). Bit-Interleaved Coded Modulation (BICM) for ATSC
3.0. IEEE Transactions on Broadcasting. 62(1):181-188. doi:10.1109/TBC.2015.2505414.
Submitted to the IEEE Transactions on Broadcasting
1
Abstract—In this paper we summarize and expound upon the
choices made for the BICM (Bit-Interleaved Coded Modulation)
part of the next-generation terrestrial broadcast standard known
as ATSC 3.0. The structure of the ATSC 3.0 BICM consists of a
forward error correcting (FEC) code, bit interleaver and
constellation mapper. In order to achieve high efficiency over a
wide range of reception conditions and Carrier-to-Noise (C/N)
ratio values, several notable new elements have been
standardized. First, twenty four original Low-Density Parity
Check (LDPC) codes have been designed, with coding rates from
2/15 (0.13) up to 13/15 (0.87) for two code sizes: 16200 bits and
64800 bits. Two different LDPC structures have been adopted;
one structure more suited to medium and high coding rates and
another structure suited to very low coding rates. Second, in
addition to QPSK, Non-Uniform Constellations (NUCs) have been
chosen for constellation sizes from 16QAM to 4096QAM to bridge
the gap to the Shannon theoretical limit. Two different types of
NUCs have been proposed: one-dimensional NUCs (1D-NUC) for
1024- and 4096-point constellations, and two-dimensional NUCs
(2D-NUCs) for 16-, 64- and 256-point constellations. 2D-NUCs
achieve a better performance than 1D-NUCs but with a higher
complexity since they cannot be separated into two independent
I/Q components. NUCs have been optimized for each coding rate
for the 64800 bits LPDCs. The same constellations are used for
16200 bits LDPCs, although they have been limited up to
256QAM. Finally, a bit interleaver, optimized for each
NUC/coding rate combination, has been designed to maximize the
performance. The result is a BICM that provides the largest
operating range (more than 30 dB, with the most robust mode
operating below -5 dB C/N) and the highest spectral efficiency
compared to any digital terrestrial broadcast system today,
outperforming the current state-of-the-art DVB-T2 standard
BICM by as much as 1 dB in some cases. ATSC 3.0 will also
provide a considerable increase in the maximum transmission
capacity when using the high-order non-uniform constellations
such as 1024QAM and 4096QAM, which will represent a major
milestone for terrestrial broadcasting since the highest order
constellation currently available is uniform 256QAM. This paper
describes the coding, modulation and bit interleaving modules of
the BICM block of ATSC 3.0, and compares its performance with
other DTT standards such as ATSC A/53 and DVB-T2.
Index Terms—ATSC 3.0, BICM, Coding, LDPC, Modulation,
Non-Uniform Constellations.
Manuscript received July 31, 2015; reviewed September 13, 2015. Parts of
this paper have been published in the Proc. IEEE BMSB 2015, Ghent, Belgium.
L. Michael is with the Digital Analog System Development Department,
Research & Development Division, Device Solutions Business Group, Sony
Corporation, Atsugi-shi 243-0021 Japan. (e-mail:
lachlan.michael@jp.sony.com).
D. Gómez-Barquero is with the Institute of Telecommunications and
Multimedia Applications (iTEAM), Universitat Politecnica de Valencia,
Valencia 46022, Spain (e-mail: dagobar@iteam.upv.es).
I. I
NTRODUCTION
IT-Interleaved Coded Modulation (BICM) is the
state-of-the-art pragmatic approach for combining channel
coding with digital modulations in fading transmission
channels [1], where the modulation constellation can be chosen
independently of the coding rate. The structure of the BICM
block consists of the serial concatenation of a forward error
correction (FEC) code, a bit interleaver, and a constellation
mapper. Within the physical layer, the modulation and coding
constitute the main portions of the system that affect the overall
spectral efficiency, which is a key performance indicator to
make an efficient use of the scarce radio spectrum.
The spectral efficiency of the first-generation Digital
Terrestrial Television (DTT) standard ATSC (Advanced
Television Systems Committee) A/53 [2], currently used in the
U.S., Canada, Mexico and South Korea, is far from the
theoretical capacity Shannon limit. ATSC A/53 employs
8-level Vestigial Side Band (8VSB) modulation, and a FEC
coding scheme based on the concatenation of a convolutional
inner code and an outer Reed-Solomon (RS) code. Receiver
operation point is approximately 15.0 dB Carrier-to-Noise
(C/N) ratio to ensure a good reception with Additive White
Gaussian Noise (AWGN), and its transmission capacity is
about 19.4 Mbps in a 6 MHz radio frequency (RF) channel,
which results in a spectral efficiency of 3.23 bps/Hz. Taking
into account these values, and using the Shannon’s capacity
formula:
log
2
1 (1)
where BW is the system bandwidth and Γ is the C/N in linear
units. It can be shown that the operating point of ATSC A/53 is
6.77 dB and 10.7 Mbps away from the Shannon limit in a 6
MHz RF channel.
Compared to the first-generation ATSC A/53 DTT standard,
the next-generation DTT ATSC standard, known as “ATSC
3.0” [3], is required to provide at least 30% capacity increase at
the same operating point, and to be significantly more robust
[4], in to cope with the broadcast spectrum shortage due to the
rapidly growing demand for wireless broadband services and
the upcoming broadcast TV spectrum incentive auction in the
U.S. [5]. But ATSC 3.0 aims to become the reference terrestrial
broadcasting technology worldwide, outperforming existing
terrestrial broadcast standards [6], [7], and leveraging recent
research into digital terrestrial broadcasting [8]-[10].
Current state-of-the-art terrestrial broadcasting standards are
Bit-Interleaved Coded Modulation (BICM)
for ATSC 3.0
Lachlan Michael and David Gómez-Barquero
B
Submitted to the IEEE Transactions on Broadcasting
2
very close to the Shannon theoretical limit using Low-Density
Parity Check (LDPC) codes and implementing uniform QAM
(Quadrature Amplitude Modulation) constellations up to
256QAM. However, for ATSC 3.0 several notable new
elements have been standardized for the BICM block in order
to bridge the remaining gap to the theoretical limit without the
need for additional transmission power or bandwidth, and
achieve extremely high efficiency over a wide range of
reception conditions and C/N values. In particular, ATSC 3.0
has adopted:
New LDPC codes with coding rates from 2/15 (0.13) up to
13/15 (0.87) for two code sizes: 16200 and 64800 bits [11].
Non-Uniform Constellations (NUCs) from 16-QAM to
4096-QAM, in addition to QPSK. Two different types of
NUCs have been adopted: two-dimensional NUCs
(2D-NUCs) for 16-, 64- and 256-point constellations, and
one-dimensional NUCs (1D-NUC) for 1024- and
4096-point constellations, respectively [12].
Bit Interleavers (BIL), optimized for each combination of
coding rate and constellation pattern.
The result is that the BICM block of ATSC 3.0 provides not
only the highest spectral efficiency compared to any DTT
system today, but it also provides the largest operating range
(more than 30 dB C/N), and a significant increase in the
maximum transmission robustness and transmission capacity,
with the most robust mode (QPSK 2/15) operating below -5 dB
C/N, and the highest capacity mode (4096QAM 1D-NUC
13/15) with a spectral efficiency of 10.4 bps/Hz. As a reference,
the most robust and highest capacity modes of DVB-T2 are
QPSK 1/2 and uniform 256QAM 5/6, respectively. Compared
to ATSC A/53, ATSC 3.0 is almost 4 dB and 7 Mbps closer to
the Shannon limit in a 6 MHz RF channel, and the gain
compared to DVB-T2 reaches up to 1 dB in some cases.
This paper describes the coding, modulation and bit
interleaving blocks of the BICM module of ATSC 3.0, and
compares its performance with the ATSC A/53 and DVB-T2
standards. A summary version of the paper was presented in
[13]. The rest of the paper is structured as follows. Section II
presents a brief overview of the ATSC 3.0 physical layer.
Section III describes the BICM methodology followed within
the standardization process. Section IV, Section V and Section
VI present the FEC coding, modulation and bit interleaving,
respectively. Section VII presents illustrative results of the
BICM performance of ATSC 3.0, and compares it with ATSC
A/53 and DVB-T2. Finally, Section VIII concludes the paper.
II. ATSC
3.0 PHYSICAL LAYER OVERVIEW
Fig. 1 shows the different block diagrams of the ATSC 3.0
physical layer architecture [3]. As can be seen in the figure,
ATSC 3.0 allows for the optional use of superposition
modulation LDM (Layer-Division Multiplexing) [15] in
addition to FDM (Frequency-Division Multiplexing) and TDM
(Time-Division Multiplexing) for SISO (Single-Input
Single-Output), MIMO (Multiple-Input Multiple-Output)
technology with cross/dual-polarized transmission [16], and
channel bonding [17], which consists in combining two RF
channels.
The BICM module is one of the most important modules as it
provides the error correction capability for the system, allowing
the broadcaster multiple choices to trade off robustness for
capacity. The improvements in this block are the major reasons
for improvements in efficiency compared to both the previous
generation ATSC standard A/53 and improvements compared
to DVB-T2. An outline of the system architecture follows.
The input formatting module takes input IP and other data
packet types and forms it into physical layer containers, known
as baseband frames. The SFN (Single Frequency Network)
STL (Studio-Transmitter Link) distribution interface
guarantees that the output of the physical layer is deterministic
for a given input, enabling SFNs. The framing and interleaving
module performs the time and frequency interleaving and
constructs the physical layer frame. The waveform module
consists of the pilot insertion, FFT (Fast Fourier Transform)
and guard interval, preamble, and allows for the optional use of
Multiple-Input Single Output (MISO) [14] and peak-to-average
power ratio (PAPR) reduction techniques.
Fig. 1. Block diagrams of ATSC 3.0 physical layer architecture for a single RF channel for SISO F/TDM (a), SISO LDM (b), MIMO (c), and for two RF channel
with channel bonding (d). The use of MIMO is not available with LDM and channel bonding.
FEC
BIL
MAP
Time Int.
Framing
Freq. Int.
Pilots
MISO
IFFT
PAPR
GI
Preamble
Bootstrap
FEC
BIL
MAP
FEC
BIL
MAP
LDM Injection Level
FEC
BIL
MAP
MIMO Precoder
MIMO
DEMUX
MAP
FEC
BIL
MAP
FEC
BIL
MAP
Stream Partitioner
Cell Exchange
Submitted to the IEEE Transactions on Broadcasting
3
Fig. 2 depicts the block diagram of the BICM module for a
single Physical Layer Pipe (PLP) for SISO, which, as noted
previously, consists of the FEC, bit interleaver and
constellation mapping. For MIMO there is also an antenna
stream demultiplexer, which distributes the output bits from the
bit interleaver into two constellation mappers, one for each
transmit antenna, see Fig. 1(c). The input to the BICM module
is a sequence of baseband frames carrying randomized data.
The FEC block is divided into an outer encoder and an inner
encoder. The inner code is an LDPC code and its use is
mandatory. For the outer code, there are several choices
possible. First, the use of BCH (Bose-Chaudhuri-
Hocquengham) code; second the use of a 32 bit CRC (Cyclic
Redundancy Check) and third not using any outer coding.
Regarding the support of multiple PLPs in ATSC 3.0, on the
transmitter side a maximum of 64 PLPs can be simultaneously
transmitted in a single radio frequency channel, however, a
single service consists of up to 4 PLPs. This means that
receivers must be able to receive and decode at least a minimum
of 4 PLPs (i.e., one service). PLPs of the same service share a
common time interleaver memory of 2
19
cells (constellation
symbols). In ATSC 3.0 an additional feature has been added,
similar to NGH, that allows for doubling of the time interleaver
depth, we call it “extended interleaving” here. In extended
interleaving in ATSC 3.0, double the number of cells is
allowed, up to 2
20
cells for single PLP for QPSK only. The
feature of extended interleaving is that the actual physical
memory need not be doubled, but by reducing the bits assigned
to each cell, two QPSK cells can fit into the same physical
memory as one cell for the higher modulations. This achieves
higher time interleaver depth without increasing physical
memory.
III. BICM
STANDARDIZATION METHODOLOGY
For each BICM technology item, a work plan was
established and over the course of approximately one year the
technology making up the BICM part was chosen by consensus
from the different proposals. A piecewise methodology was
followed to ensure the most suitable technology was chosen for
each part of the specification, starting with the FEC codes,
following with the constellations and finishing with the bit
interleavers. The main indicator in the selection of technologies
was performance, leaving complexity and other factors as a
secondary consideration. This made the task of technology
choice to be primarily a matter of determining the performance
of each technology proposal compared to other proposals. The
performance was confirmed not only by the proponent but by
multiple participants. To this end, the first step in the choice of
technology was to confirm full disclosure about each part, such
that each participant could freely implement other proposals
and perform cross-check simulations. This approach ensured
transparency and gave a high degree of confidence in the final
results.
Regarding the initial selection of the FEC codes, first the
constellation was fixed at QPSK and no bit interleavers were
used. Other evaluation conditions included setting the number
of decoder iterations to 50, and the target for bit error rate
(BER) and frame error rate (FER) curves down to 10
-8
and 10
-6
,
respectively were decided by consensus. Two channel models
were used in the comparisons: AWGN and i.i.d. Rayleigh, to
ensure that the chosen codes would be suitable for a wide range
of applications from static to portable reception.
All 136 FEC code proposals were then compared together in
a monolithic manner, classifying the FEC codes in two groups
consisting of shorter code lengths (less than 20,000 bits) and
longer code lengths (larger than 50,000 bits). For each group,
similar code rates were grouped together and compared. From
the best performing code in each code rate group, those codes
within 0.1 dB were allowed to remain for a second round of
comparisons, which included detailed cross-checks and
analysis of other factors including complexity. Finally, 24
LDPC codes were chosen, with coding rates from 2/15 (0.13)
up to 13/15 (0.87) for two code sizes of 16200 and 64800 bits.
Two different code structures were chosen to cover both the
medium/high and low code rates.
The next step consisted of choosing the constellations. Both
uniform and non-uniform QAM constellations were proposed,
although it became clear during the examination that some
NUCs were clearly superior in performance [18], [19]. In order
to limit the complexity at the receiver side, two-dimensional
non-uniform constellations (2D-NUCs) were adopted for 16-,
64- and 256-point constellations, and one-dimensional
non-uniform constellations (1D-NUCs) for 1024- and
4096-point constellations. QPSK was adopted as the most
robust constellation. A different NUC was designed for each
coding rate for LDPCs of 64800 bit length. The same
constellations were adopted for 16200 bit length LDPCs,
although the constellation size was limited up to 256QAM,
since the main use case proposed for the shorter length codes
was reduced implementation complexity.
The constellations were jointly evaluated with the bit
interleavers using the previously agreed FEC codes. Each
constellation and bit interleaver is potentially different for each
coding rate, although a common bit interleaver structure was
agreed before the evaluations to reduce unneeded
implementation complexity.
The final step consisted of examining the 72 MODCOD
(modulation and coding) combinations for long codes and the
48 MODCOD combinations for short codes, in order to reduce
the number of combinations to a manageable set without losing
flexibility. MODCODs were compared in terms of AWGN and
Rayleigh robustness (C/N threshold) and spectral efficiency,
and finally 46 MODCODs for long codes and 29 MODCODs
for short codes were chosen to be required combinations that
must be mandatorily implemented by all transmitters and
receivers. Nevertheless, to allow for complete flexibility for
unforeseen future situations, all the MODCOD combinations
Fig. 2. Block diagram of the ATSC 3.0 BICM module for SISO.
BICM
FEC
Inner
Encoder
(LDPC)
Mapper
Bit
Interleaver
PLPn
PLPn
Outer
Encoder
(BCH,
CRC)
Submitted to the IEEE Transactions on Broadcasting
4
will remain in the standard as options.
IV. ATSC
3.0 CODING
The FEC of the BICM is formed by concatenation of an outer
code and an inner code with the information part. The outer
code is either a BCH code, a CRC or none, whereas the inner
code is a LDPC code. BCH, CRC and LDPC codes are
systematic codes, such that the information part is contained
within the codeword. The resulting codeword is thus a
concatenation of information or payload part, BCH or CRC
parities and LDPC parities. Fig. 3 shows the structure of the
FEC frames at the output of the FEC sub-block when BCH or
CRC are used and when no outer code is used.
For maximum flexibility and to achieve higher throughput
when the error correction capability of the inner code is deemed
sufficient, the outer code can be either a BCH with the ability to
correct up to 12 bit errors in a LDPC codewords, a CRC check,
or none at all. Nevertheless, the use of the BCH is expected to
be the most common use case, as it provides additional error
correction as well as error detection. CRC is provided to give
the choice of improved efficiency, however no additional error
correction is available, only error detection. Table I outlines
examples of the efficiency gain that can be achieved using the
CRC or without any outer code compared to the expected
reference case using BCH.
Regarding the inner LDPC, two different structures have
been chosen [11]. One structure is similar to that used in
DVB-T2, which provides excellent performance at medium and
high code rates, and another slightly different LDPC structure
which showed excellent performance at very low code rates (in
general, less than or equal to 5/15). The use of two different
inner code structures in the same standard is quite unusual, but
the wide range of code rates chosen, from 3/15 up to 13/15,
justifies this approach. Two different lengths of LDPC code
have been defined: 64800 and 16200 bits, as in DVB-T2 [6],
[8]. In general, 64800 bit codes are expected to be employed as
the performance is better, although for applications where
latency is critical, or a simpler encoder and decoder structure is
preferred, 16200 bit codes may also be used. The 16200 bit
codes adopted in ATSC 3.0 have lower latency but worse
performance than the 64800 bit codes.
V. ATSC
3.0 BIT INTERLEAVING
The role of the bit interleaver in the BICM chain is to match
the output of the LDPC codewords to the constellations. The bit
interleaver affects the performance and also the HW
implementation. ATSC 3.0 has adopted a 3-stage bit interleaver
structure introduced in the MIMO profile of DVB-NGH [7],
[9]. Fig. 4 shows the block diagram of the bit interleaver, which
consists of a parity interleaver followed by a group-wise
interleaver followed by a block interleaver. This structure
allows for parallel LDPC decoding while optimizing the
performance of the FEC codes to any constellation.
The role of the parity interleaver is to convert the staircase
structure of the parity-part of the LDPC parity-check matrix
into a quasi-cyclic structure similar to the information-part of
the matrix enabling parallel decoding. The group-wise
interleaving allows optimizing the combination between the
FEC code and the constellation, and hence it is optimized for
each combination of modulation and LDPC coding rate. Finally
the block interleaver provides the final allocation from bits to
constellation symbols.
VI. ATSC
3.0 MODULATION
ATSC 3.0 has adopted complex-valued quadrature
amplitude modulation (QAM) constellations. For the highest
robustness level, quaternary phase shift keying (QPSK) is used.
For higher order constellations with a higher spectral efficiency
but a lower robustness level, non-uniform constellations from
16QAM up to 4096QAM are defined with customized
constellations for each LDPC code rate [12].
Uniform QAM constellations, characterized with uniform
spacing between constellations points and square shape of the
constellations, have been traditionally used in many
communication standards because of their simplicity for
encoding and decoding. However, there is a significant gap
between the BICM capacity of uniform QAM constellations
TABLE I
E
FFICIENCY GAIN OF CRC AND NO OUTER CODE COMPARED TO BCH
LDPC
Codeword
LDPC
Code Rate
Outer
CRC
No Outer
Code
64800
bits
13/15 0.29% 0.34%
10/15 0.37%
0.45%
7/15 0.53% 0.64%
2/15 1.86% 2.27%
16200
bits
13/15 0.97% 1.21%
10/15 1.26% 1.58%
7/15 1.81% 2.27%
2/15 6.39% 8.43%
Fig. 3. Format of ATSC 3.0 FEC frame when BCH or CRC is used (top) and
format when only the inner LPDC code is used (bottom).
M
outer
M
inner
K
payload
N
inner
N
outer
InformationorPayload
InnerCode
Parity
OuterCode
Parity
M
inner
K
payload
N
inner
N
outer
InformationorPayload
InnerCode
Parity
Fig. 4. ATSC 3.0 three stage bit interleaver block diagram.
Bit Interleaver
PLPn
Parity
Interleaver
Group‐Wise
Interleaver
Block
Interleaver
PLPn
