Downlink link adaptation scheme for IEEE 802.16m
Document Number: C80216m-08/742
Date Submitted: 2008-07-07
Source:
Shanshan Zheng (shanshan.zheng@intel.com )
Intel Corporation
Yuval Lomnitz(yuval.lomnitz@intel.com )
Intel Corporation
Hongming Zheng(hongming.zheng@intel.com)
Intel Corporation
Yang-seok Choi (yang-seok.choi@intel.com)
Intel Corporation
Venue:
Call for contributions on project 802.16m SDD: Link Adaptation Scheme
Session #56: Denver, USA
Base Contribution:
IEEE C80216m-08/742
Purpose:
For discussion and adoption by IEEE 802.16m group
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 >.
1
Introduction
• Link Adaptation is the process of setting transmission parameters
according to link state information
• In this contribution, we deal with adaptive modulation and coding
selection (AMCS *) in the downlink which is based on channel
quality information (CQI)
• Adaptive Modulation and Coding Selection (AMCS)
–
–
–
–
Key points
Background
AMCS procedure
LA combined with HARQ
• Appendices
*: The term AMCS is used to distinguish this from 802.16 AMC permutation
2
Summary of key points
• We recommend BS controlled link adaptation
• CQI feedback : virtual MCS
– Represent achievable spectral efficiency of link
– Actual transmitted MCS is a subset of virtual MCS
– 802.16e has 11 MCS levels of actual transmission. The granularity of
virtual MCS can be higher for example 16 levels
• Separate mechanism required for submap adaptation
3
Background (1) - Concept
• What is MCS
– Code rate + modulation (Modulation and coding scheme)
– Tightly coupled with Rank & MIMO mode selection
• AMCS metric (CQI) - attempts to capture the quality of the channel
by a single number
• CQI usage
– MCS selection
– User selection
4
Background (2) – BS/MS controlled
Approaches to link adaptation
•
MS controlled: MS responsible for selecting MCS for future frames, BS
generally follows recommendation
– MS performs adaptation to mobility
– Pros: best knowledge of interference & channel information, mobility/dynamics,
and especially of specific modem performance
– Cons: MS cannot optimize H-ARQ, only optimizes for specific packet length &
PER. Recommendation is good only till certain time limit. MS is unaware of BS
(serving and interfering) behavior which affects the choice
•
BS controlled: MS reports a metric (e.g. CINR), BS selects MCS
– BS performs adaptation of the measurement to the CQI delay + mobility
– Pros: BS has system info (scheduling tradeoffs, QoS, delay) to optimize MCS for
each packet length, delay from CQI and H-ARQ parameters, BS knows system
interference behavior; BS adapts MCS to selected precoding/MIMO mode etc
– Cons: CINR metric may not accurately capture different receiver performance at
MS; Only coarse adaptation to specific modem performance
 Appendix A - Effects on link adaptation known to MS and BS
5
Background (3) - Outer loop
•
•
•
•
•
•
•
Preliminary knowledge, such as mobility, channel model, and BS specific
knowledge is used to generate table mapping from CQI to transmit MCS.
In fast link adaptation process, BS uses the table to map CQI feedback to
MCS decision
BS tracks the ACKs to measure actual packet error and updates the table.
This slow ‘learning’ process is the outer loop.
The target of BS is to refine table (CQIMCS) that optimize performance
(e.g. optimize throughput with given PER for given delay and packet size)
The outer loop adds robustness to unexpected/unmodeled behavior (e.g. of
an interfering BS)
The outer loop enables to implicitly adapt to CQI delay, estimation errors,
interference dynamics etc, etc without explicitly modeling these effects
Observation: With outer loop, specific metric has small effect on
performance
BS
MS
Slow table
Update
MCS Design
Channel
Measurement
receive
Appendix B - Effect of different metrics with outer loop
Appendix C - Outer loop abstraction
6
AMCS proposal
•
•
Target: get the gain from both MS controlled and BS controlled approaches
BS controlled
– MS feeds back CQI
– MS is not responsible for mobility/interference dynamics/delay
– BS decides actual transmitted modulation and coding scheme
•
MS-dependant metric, representing specific MS capabilities
– Aid user selection
– Faster convergence of outer loop
•
We propose to use Virtual MCS metric
7
Virtual MCS
•
•
•
•
•
Based on channel condition, MS reports its preferred MCS that would have been
used (in given conditions: distributed/localized, precoding, etc), in the measured
frame
This MCS is the MS estimate which MCS would achieve maximum throughput
with the maximum FEC block size
This metric is a practical embodiment of “mutual information” – how much
information this channel can carry
E.g. based on measuring SINR/MI and fitting to modem LLS performance
Advantages compared to PCINR/other metrics
–
–
–
–
Represents specific MS conditions for faster outer loop convergence & can be directly
used for user selection
Externally testable for conformance tests
Not dependent on receiver structure (As opposed to post processing CINR which is
based on receiver detection mode (MRC/MMSE/MLD). MLD receiver doesn't have post
processing SINR)
Unified metric for different MIMO modes (As opposed to post processing CINR which
has different effective range. See Appendix D Post processing CINR CDF of different
MIMO scheme)
8
MCS granularity
•
Currently EMD defines 11 types of MCS can be used as actual transmission
QPSK 1/2 Rep 2, 4, 6, QPSK (1/2; 3/4), 16QAM (1/2; 3/4), 64QAM (1/2; 2/3; 3/4; 5/6)
•
•
•
•
Coding scheme of 802.16m not defined yet
We assume 802.16m is likely to support rate matching and/or continuous
code rates therefore MCS scale doesn’t strictly exist as in 802.16e
Additional MCS granularity can be added in order to improve link adaptation
performance (Appendix E Different actual transmission MCS comparison)
Virtual MCS as CQI for feedback
– Granularity of virtual MCS can be higher than the granularity of actual transmitted
MCS
– For example, 16 types of MCS can be selected as virtual MCS
QPSK 1/2 Rep 2,4,6, QPSK(1/2; 3/5; 2/3; 3/4)
16QAM (1/2; 3/5; 2/3; 3/4; 4/5), 64QAM (3/5; 2/3;3/4; 5/6)
– The higher virtual MCS granularity is for differentiating user quality and achieving
more user scheduling gain
9
Link adaptation with HARQ
• Several H-ARQ modes can be considered
– Fixed MCS, fixed transmission size (regularly assumed)
• In this case LA problem is to choose initial MCS
• When MS reports the MCS BS should use higher MCS for H-ARQ (in order to
increase initial PER and gain from cases where channel is better than
expected)
– Fixed MCS, fixed set of transmission sizes
• E.g. second transmission is always ½ of first (see **)
• Similar to first option in terms of link adaptation
– Fully adaptive H-ARQ
• BS selects MCS and/or duration of each retransmission
• BS can utilize MS CQI reports in order to:
– Learn about future: channel changes during H-ARQ process
– Learn about past: information accumulated in MS receiver from past transmissions
• Requires further study
• We believe most of gain can be obtained using burst-duration adaptation and
SE/MCS related CQI
10
Miscellaneous issues
• CQI feedback overhead
– Localized permutation : CQI/sub-band/user or best-M CQI/user
– Distributed permutation : CQI/user
• Map adaptation
– Telescopic maps / submaps are already supported in 802.16 and allow
reduction of map overhead by adapting map of unicast IE-s MCS to user
conditions
– However 802.16 doesn’t have suitable link adaptation mechanism
– We propose for 16m to add a slow message-based link adaptation
(similar to subheader / REP-RSP) for the maps
– The MS will report preferred map-MCS based on long term PER for each
of the map portions
11
Proposed Text for SDD
11.x Link adaptation
11.x.1 Downlink Adaptive modulation and coding selection
All MS should estimate individual channel quality and feed back CQI to BS. Based on
CQI feedback, BS should decide actual transmitted MCS
11.x.1.1 Generic AMCS Architecture
The generic AMCS architecture is shown in figure xxx
MS
Slow table
Update
MCS Design
BS
Channel
Measurement
receive
11.x.1.2 BS controlled
Preliminary knowledge, such as mobility, channel model, and BS specific knowledge is
used to generate table of mapping from CQI to transmitted MCS. BS tracks the ACKs to
measure actual packet error and updates the table.
12
Proposed Text for SDD (Cont’d)
11.x.1.3 MS-dependant metric
Based on channel condition, MS reports its preferred MCS that would have been
used (in given conditions: distributed/localized, precoding, etc) in the measured
frame. This MCS as virtual MCS is the MS estimate which MCS would achieve
maximum throughput with the maximum FEC block size.
11.x.2 Link adaptation with HARQ
Several H-ARQ modes can be considered, such as fixed MCS and transmission
size (regularly assumed), fixed MCS and set of transmission sizes, fully adaptive HARQ.
13
Appendices and backup
14
Appendix A - Effects on link adaptation known to
MS and BS
•
The following factors affect MCS selection:
Factor
Best
known by
Comments
Current channel and interference
condition
MS
Channel dynamics
MS
BS can estimate from CQI dynamics
Interference dynamics
BS
Related to other BS
scheduling/precoding algorithms
Precoding/boosting
BS
Resource / power / interference
budget
BS
E.g. in light load BS decides to use
more resource with lower power
MS receiver performance in various
scenarios
MS
Difference between different MS
depends on scenario and specific
channel instance (interference /
MIMO etc)
15
Appendix B - Effect of different metrics with outer
loop (1)
These plots show two metrics ability to forecast PER
–
–
Each point is PER in one channel instance (over random data & noise instances)
We can see in this comparison post-processing performs much better
[2.08] PER per channel v.s. Post Processing Physical SNR
[2.09] PER per channel v.s. Sum of SNR (sum(S/N)) over antennas
- matched receiver
- mismatched receiver
- matched receiver
- mismatched receiver
-1
-1
10
10
PER
•
•
Pre-processing CINR
Post processing CINR
PER
•
-2
-2
10
-3
10
10
-3
10
-4
-2
0
2
4
Sum of SNR (sum(S/N)) over antennas [dB]
CINR(15a ) 
Nant
I
n 1
S
(n)
(n)
N
(n)
0
 n
-2
-1
0
1
2
3
4
5
Post Processing Physical SNR (mismatched, receiver independent) [dB]
6
 h
 Ev
k
k
2
k ,n
2
k ,n
SNRPP 

k
hk
k E h k h k
*
2
1
2
 v k 

16
Appendix B - Effect of different metrics with outer
loop (2)
•
Following plot shows
performance (w/o H-ARQ)
in interference scenario
SIMO, ITU-Ped-B 3Km/h
Perfect outer loop
operation
Single strong interferer
Post-CINR and Pre-CINR
(eq 15 and 15a from CINR
RPD)
Although post and preCINR have very different
accuracy in forecasting
PER (see prev slide), there
is a very small difference
in link adaptation
performance
With H-ARQ the difference
is further decreased
•
•
•
•
•
•
•
Conclusion: with outer loop, link adaptation dependence on specific metric is
small (as long as there is sufficient correlation with decoding quality)
17
Appendix C - Outer loop abstraction
•
•
•
•
It’s possible to abstract the outer loop by considering very long convergence
times that can be used for optimal adjustment
In simulation, the optimal table values for given metric, delay, transmission
mode, etc can be computed
There is a statistical relation between the CQI and the PER of the packet
which is function of the MCS, channel model, velocity, CQI delay, retransmit
delay, measurement error, etc.
Possible criteria for optimization:
– Maximize throughput under PER constraint
– “Throughput” and “PER” can be average or in 99% confidence
– We optimize average equal-bandwidth PER under average PER constraint
(closed-form solution)
Channel
Measurement
Metric v.s.
MCS
Table
instantaneous
Packet loss &
Throughput
Offline
table
optimization
18
Appendix D - Post processing CINR CDF of different
MIMO scheme
Different MIMO Mode Post-SINR CDF at 4 Lamda
1
0.9
0.8
0.7
CDF
0.6
0.5
0.4
SVD based CL R2 MCW 2nd stream
SVD based CL R2 MCW 1st stream
0.3
SVD based CL R2 SCW
SVD based CL R1
0.2
SM SCW AMC
Long termPUSC R1(144x22)
0.1
STBC AMC
STBC PUSC(144x22)
0
-20
-15
-10
-5
0
5
10
15
Average post-SINR per RB(dB)
20
25
30
19
Appendix E – Different actual transmission MCS
comparison
•
•
SLS is followed EMD
Ideal channel estimation
5ms CQI feedback delay
Improvement SE Gain
11 MCS levels
16 MCS levels
Distributed SU STBC/SM
0
3.98%
Localized SU STBC/SM
0
2.83%
Localized Predefined MU
0
3.06%
Localized Channel Aware MU
0
3.29%
Without CQI feedback delay
Improvement SE Gain
11 MCS levels
Improvement Gain
Localized SU STBC/SM
0
7.41%
Localized Predefined MU
0
4.62%
20