SDD text proposal for IEEE 802.16m HARQ and FEC
Document Number:
IEEE C802.16m-08/606
Date Submitted:
2008-07-07
Source:
Tom Harel
tom.harel@intel.com
Yuval Lomnitz
yuval.lomnitz@intel.com
Olga Kapralova
olga@vu.spb.ru
Intel Corp.
Venue:
Call for Contributions on Project 802.16m System Description Document (SDD): Hybrid ARQ (PHY aspects)
Base Contribution:
Purpose:
Accept changes for SDD
Notice:
This document does not represent the agreed views of the IEEE 802.16 Working Group or any of its subgroups. It represents only the views
of the participants listed in the “Source(s)” field above. It is offered as a basis for discussion. It is not binding on the contributor(s), who
reserve(s) the right to add, amend or withdraw material contained herein.
Release:
The contributor grants a free, irrevocable license to the IEEE to incorporate material contained in this contribution, and any modifications
thereof, in the creation of an IEEE Standards publication; to copyright in the IEEE’s name any IEEE Standards publication even though it
may include portions of this contribution; and at the IEEE’s sole discretion to permit others to reproduce in whole or in part the resulting
IEEE Standards publication. The contributor also acknowledges and accepts that this contribution may be made public by IEEE 802.16.
Patent Policy:
The contributor is familiar with the IEEE-SA Patent Policy and Procedures:
<http://standards.ieee.org/guides/bylaws/sect6-7.html#6> and
<http://standards.ieee.org/guides/opman/sect6.html#6.3>.
Further information is located at <http://standards.ieee.org/board/pat/pat-material.html> and <http://standards.ieee.org/board/pat >.
Key points
• Select 802.16e CTC as the FEC (Forward Error Correction) scheme
• Extensions for performance improvement:
– Minimal code-rate ⅓, and optional extension to ¼
– Support additional block lengths
• HARQ-IR (Incremental redundancy)
–
–
–
–
Flexible mother code rate
Constellation re-arrangement with version per symbol
Variable re-transmission size (adaptive or static)
Exploit frequency selectivity in re-transmission
• How many MCSs we need, and how to describe them
– Rate matching to accommodate variable number of symbols and pilots
– Padding is a waste of resources => “Continuous” code-rate instead of
MCS specification
2
FEC selection
• We suggest to use only 802.16e CTC coding scheme (section
8.4.9.2.3.1) for all data traffic and maps
• An MS must implement 802.16e CTC for backward compatibility,
and designing a fresh new FEC scheme will double the HW
– CTC/LDPC decoder has considerable hardware complexity compared
to overall PHY
– Convolutional Code used for FCH backward compatibility only
• Good performance/complexity tradeoff for the relevant block lengths
(48-4800 bits)
3
Minimal code rate
• We considered a simple low-complexity extension of 16e code to
minimal code rate ¼ instead of ⅓
– Higher coding gain improves performance in low SNR range by ~0.20.4dB
– For QAM modulation (higher SNR range), re-transmission of bits using
constellation re-arrangements performs better than sending more parity
bits
• Since the overall system performance improvement of rate ¼ is
small, minimal code-rate = ⅓ is recommended
See Optional code-rate ¼ support (1)
4
FEC block length granularity
•
Block lengths in 802.16e, including IR HARQ (tables 524 and 525):
–
•
•
•
•
•
6, 9, 12, 18, 24, 27, 30, 36, 48, 54, 60, 120, 240, 360, 480, 600
Data block size (bytes)
Granularity: finer granularity of supported block length (2-3 bytes for small
packets), to reduce the padding waste
Maximal block size: the marginal gain from larger block size is negligible,
and implementation (memory) complexity is high, so we propose 600 bytes
(4800 bits) as maximal block size
For new lengths, new interleavers should be designed
However, because of tail-biting, the length cannot be a integer multiple of 7
We propose a method to generate new block lengths from larger CTC
block, called “known padding”:
1. Select a larger block size
2. Add pair of zeros to the message as padding (spread them among the message
bits) to get the larger block length
3. Diminish the systematic padding bits from the sequence of encoded bits, as well
as the parity bits from the same step in the trellis
4. Transmit only the remaining bits
•
Simulation results show that this method is approx. equivalent to design of
a smaller block length.
5
HARQ-IR (1)
• We propose to support only HARQ-IR (incremental redundancy) for
traffic channel
– Chase-combining is a special case
– Non-HARQ is a special case
• Each transmission can start in various positions in the sequence of
mother code bits
– This sequence’s length is defined by burst size and Mother-Code-Rate
(MCR)
• Support flexible MCR for memory limited, high throughput users
– Compared to chase combining, HARQ-IR requires more buffering in
receiver, i.e. supports lower throughput for same buffer size
– By changing MCR we can allow high throughput when required with
smaller robustness and robust transmission at low throughput
– E.g. support MCR = ¾, ½, ⅓
– Without flexible MCR the throughput achieved with given buffer size
would reduce when changing to code rate 1/3
6
HARQ-IR (2) - CoRe
•
Constellation re-arrangement (CoRe) – change the bits mapping in retransmission for 16-QAM, 64-QAM
– The gain is from improving the LSB quality and equalizing the LLR quality of the
constellation
•
•
Only 2 versions (switch LSB and MSB) achieve most of the gain with minor
overhead
Version is selected for each QAM symbol (rather than same version over
the packet) and can be changed within a single (re) transmission as
depicted below
– For MCR=⅓, CCR=½
7
CoRe and Mother-Code-Rate (1)
SISO Ped-B 3 km/h
QPSK, R=1/3,1/2,5/12,5/8; 16-QAM, R=5/12,1/2,5/8 ; 64-QAM, R=5/9,5/6
3
Gray, MCR=1/2, LLR combining
CoRe, MCR=1/2, LLR combining
CoRe, MCR=1/3, LLR combining
CoRe, MCR=1/4, LLR combining
Spectral efficiency [bits/sec/Hz]
2.5
2
1.5
1
0.5
0
0
2
4
6
8
SNR [dB]
10
12
14
16
18
8
CoRe and Mother-Code-Rate (2)
SISO Ped-B 3 km/h
QPSK, R=1/3,1/2,5/12,5/8; 16-QAM, R=5/12,1/2,5/8 ; 64-QAM, R=5/9,5/6
18
CoRe, MCR=1/2, LLR combining
CoRe, MCR=1/3, LLR combining
CoRe, MCR=1/4, LLR combining
Spectral efficiency improvement [%]
Compared to: Gray, MCR=1/2, LLR combining
16
14
12
10
8
6
4
2
0
-2
0
2
4
6
8
SNR [dB]
10
12
14
16
18
9
Variable re-trans size HARQ-IR
•
In .16e the size of the retransmission is the same as the
original transmission (“same MCS”).
For high SE we see better
performance with smaller retransmission size (static)
Two types:
– Static: constantly set smaller size of
re-transmission
• Simple to design, define and
simulate, ~8% SE improvement
– Adaptive: set the re-transmission
size according to channel state or
another feedback
• More complicated design. Up to
~20% SE improvement by
information theoretic analysis
• Yet to be explored
•
We expect higher gain from
opportunistic scheduling with
smaller re-transmission
SIMO Ped-B
Slow link adaptation
5
HARQ-IR re-transmission size equals original
HARQ-IR with half sized re-tranmissions
4.5
Spectral efficiency (bit/channel use)
•
4
3.5
3
2.5
2
1.5
1
0.5
0
-2
0
2
4
6
8
10
SNR [dB]
12
14
16
18
10
20
Frequency diversity in re-trans.
• For slowly-changing frequency-selective channel, it is important to
change the mapping of symbols to subcarriers in re-transmissions to
improve frequency diversity
• In case of “chase-combining” (I.e. high MCR), the symbols position
shall be changed in re-transmission (E.g. by starting point selected
by SPID)
11
How many MCSs we need?
On one hand
–
–
•
With HARQ the spectral efficiency is not very
sensitive to MCS selection
It shows that we have enough MCSs in 16e
On the other hand
–
We have a rough granularity of resource
allocation
•
–
–
SISO Ped-B 3 km/h
16-QAM, R=5/12,1/2,5/8 ; 64-QAM, R=5/9,5/6
Slot size is selected according to considerations
like: channel estimation, diversity etc.
With small number of MCSs a lot of padding is
needed
Zero-padding is expensive
•
•
It waste system resources (energy, BW)
It enlarges the error probability (detection error
on unimportant parity bits)
3.5
3
Spectral efficiency [bits/sec/Hz]
•
2.5
2
1.5
1
0.5
0
8
10
12
14
16
SNR [dB]
18
20
22
24
12
An alternative – almost continuous CR
• We suggest to support finer resolution of burst size (to avoid
padding) while keeping the same slot size
– Need more MCSs.
• Or: define MCS in an alternative manner
– Allocation fixes the number of resources (slots/LRU)
– Describe in the map the message size
• An exponential scale of possible sizes
• Only a few bits for the length of short messages. More bits for the larger
ones.
– => almost continuous code rate selection is supported
13
16m proposal versus 16e and WiMAX 1.0
Item
WiMAX 1.0
16e IR
16m proposed
Comment / Motivation
H-ARQ scheme
Chase
IR
IR
Higher SE
Mother code rate
½
1/3
¾ .. ⅓
Lower code rate
Variable M.C.R
Coding scheme
CTC
CTC
CTC
Good performance + backward
compatibility
Symbol mapping
Gray
Gray
Co-Re
Better Performance
Concatenation rule
Yes,
modulation
dependent
None
Yes, modulation
independent
FEC block granularity
~6B
Coarse
~2-3B for small
blocks
Variable re-trans size
No
No
Yes
Max FEC block size
(bits)
480
4800
4800
Padding overhead versus Map
overhead tradeoff
Performance v.s complexity tradeoff
14
Text proposal for SDD (1)
Insert the following text into Physical Layer clause (Chapter xx in [IEEE 802.16m-08/003r1])
------------------------------- Text Start -------------------------------
11.x Channel coding and Hybrid-ARQ
11.x.1 Block diagram
Concatenation rules
Burst
partition to
FEC blocks
n
Information
bits
(Burst)
MCR
HARQ
parameters
CoRe
version
FEC
encoder
Bit selection
Modulation
nk
Information bits
in the kth FEC
block
Collection
And repetition?
nk/RMC
Mother Code
bits for the kth
block
mk*M
Coded bits
mk
Tones for the kth
FEC block
m
Tones allocated
for the burst
Channel Code
Rate is
RCC=nk/(mk*M)
15
Text proposal for SDD (2)
11.x.2 Modulation
In IEEE 802.16m each data tone is modulated with QPSK, 16-QAM or 64-QAM constellation, and carries
2, 4 or 6 coded bits respectively.
For 16-QAM and 64-QAM the mapping of bits to constellation changes in HARQ re-transmission, and is
selected by CoRe-version. There are 2 possible mappings, between which the LSB and MSB bits are
switched.
11.x.3 Encoding
11.x.3.1 Coding scheme
IEEE 802.16m uses the CTC code rate ⅓ defined in 802.16e for encoding all traffic and control channels,
except UL control channels and FCH in backward compatible mode.
The coding scheme is extended to support additional FEC block sizes. The maximal block size is 4800
information bits. The resolution of block sizes for small blocks will be 16-24 bits. The definition of the code
rate in the map is TBD and will support granularity of 16-24 bits for small PDUs.
11.x.3.2 Rate matching
In IEEE 802.16m the number of subcarriers in a logical allocation (LLRU) varies due to e.g. varying
number of pilots and shortened subframes.
The amount of coded bits is adapted to the allocation size by using a flexible channel code-rate. The bitselection takes an arbitrary number of mother coded bits for transmission according to the allocation size.
16
Text proposal for SDD (3)
11.x.3.3 Repetition
Repetition is done when the number of transmitted bits is larger than the number of mothercoded bits (total number of information and parity bits generated by FEC encoder). The
coded bits selection is done cyclically over the buffer of encoded message.
11.x.3.4 Concatenation rules and interface to resource allocation
Concatenation rules are based on number of information bits and do not depend on the
structure of resource allocation (number of LLRUs and their size).
Concatenation rules include 2 parts: partition of the message bits, and partition of the
allocated tones.
11.x.4 HARQ
IEEE 802.16m implements Incremental Redundancy HARQ. The transmitted coded bits are
selected from the mother-code bits by the bit-selection that depends on the re-transmission
number. The SPID determines the starting points and the bits are taken sequentially and
cyclically. The mother code rate (MCR, between 1 and 1/3) may be controlled by the BS in
order to trade-off buffer requirements, throughput and robustness. When the mother code
rate is higher than 1/3, only the first nk/MCR encoder outputs are considered for HARQ.
17
Text proposal for SDD (4)
For each transmitted bit the CoRe-version is selected by the number of
transmission of this bit. A possible mechanism is to indicate the version
of the first bit transmitted in each (re)transmission, and flip the version
when returning from the last encoded bit to the first.
The number of subcarriers used in each retransmission may vary in a
fixed or adaptive way (adaptive H-ARQ is for further study). IEEE
802.16m will support changing the allocation of coded bits to
subcarriers between retransmissions (the specific mechanism is TBD)
------------------------------- Text End -------------------------------
18
Backup slide
19
CoRe and mother-code-rate (1)
SISO Ped-B 3 km/h
16-QAM, R=5/12,1/2,5/8 ; 64-QAM, R=5/9,5/6
3
Gray, MCR=1/2, LLR combining
Gray, MCR=1/3, LLR combining
Gray, MCR=1/4, LLR combining
CoRe, MCR=1/2, LLR combining
CoRe, MCR=1/3, LLR combining
CoRe, MCR=1/4, LLR combining
2.5
Spectral efficiency [bits/sec/Hz]
2
1.5
1
0.5
20
0
0
2
4
6
8
SNR [dB]
10
12
14
16
18
CoRe and mother-code-rate (2)
SISO Ped-B 3 km/h
QPSK, R=1/3,1/2,5/12,5/8; 16-QAM, R=5/12,1/2,5/8 ; 64-QAM, R=5/9,5/6
18
Gray, MCR=1/3, LLR combining
Gray, MCR=1/4, LLR combining
CoRe, MCR=1/2, LLR combining
CoRe, MCR=1/3, LLR combining
CoRe, MCR=1/4, LLR combining
CoRe+1/4
is ~7%
better than
separately
16
Spectral efficiency improvement [%]
Compared to: Gray, MCR=1/2, LLR combining
14
12
CoRe
equals Gray
for QPSK
10
MotherCodeRate
1/3 performs
better than 1/4
for 16/64-QAM
with CoRe
8
6
4
2
0
-2
21
0
2
4
6
8
SNR [dB]
10
12
14
16
18
CoRe and mother-code-rate (3)
Average number of (re-)transmissions (for MCS achieving highest SE)
SISO Ped-B 3 km/h
QPSK, R=1/3,1/2,5/12,5/8; 16-QAM, R=5/12,1/2,5/8 ; 64-QAM, R=5/9,5/6
2.6
Gray, MCR=1/2, LLR combining
CoRe, MCR=1/2, LLR combining
CoRe, MCR=1/3, LLR combining
CoRe, MCR=1/4, LLR combining
2.4
2.2
2
1.8
1.6
1.4
1.2
1
0
2
4
6
8
SNR [dB]
10
12
14
16
18
22
Optional code-rate ¼ support (1)
•
•
•
Proposal: decrease minimal code rate to ¼ to improve coding gain
N=2400, BER R=1/3
vs R=1/4 the previous ones
Design
method: Optimal third polynomial
given
10
Pros:
0
Rate = 1/3
Rate = 1/4
–10 Simple low-complexity change (same trellis, modify only LLR calculation) – same
decoder
–10 Fully backward compatible with 16e CTC (as a feature subset)
– Improvement of maps and uplink implies better coverage
-2
-4
•
-6
10
Cons:
– Performance improvement is relatively small (0.2-0.3 dB) compared to code-rate ⅓
10
-0.5 [some simulations
0
0.5 gain]
1
1.5
2
showed higher
EbN0 [dB]
– Negligible improvement for high spectral-efficiency cases (16/64-QAM), even with
HARQ-IR
N=2400, FER R=1/3 vs R=1/4
-8
0
10
-1
10
-2
10
-3
10
-4
10
23
-5
10
-0.5
0
0.5
1
EbN0 [dB]
1.5
2
Optional code-rate ¼ support (2)
• Green- code rate ¼, Red – code rate 1/3, Blue – code rate ½.
• A gain of 0.4dB is obtained when the code rate is decreased from
480bits. QPSK
1/3 to ¼
0
10
-1
PER
10
-2
10
-3
10
-4
10
0
2
4
6
Eb/No [dB]
8
10
12
24
Optional code-rate ¼ support (3)
---
For the X parity bit: 0x3, equivalently 1+D
X1
X2
25
+
Optional code-rate ¼ support (4)
Sub-packet generation
X1
X2
subblock
subblock
26