Requirements, concept and design of downlink symbol structure for 802.16m

Document Number: IEEE S802.16m-08/120Date Submitted: 10-March-2008

Source:Yuval Lomnitz Yuval.Lomnitz@intel.comHuaning Niu Huaning.Niu@intel.comJong-kae (JK) Fwu Jong-kae.Fwu@intel.com Sassan Ahmadi Sassan.Ahmadi@intel.comHujun Yin Hujun.Yin@intel.comIntel Corp.

Venue:IEEE Session #54, Orlando FL.

Base Contributions:This presentation supports two contributions:IEEE C80216m-08/120: A high level description of a symbol structure concept for 802.16mIEEE C80216m-08/121: Symbol structure design for 802.16m - resource blocks and pilots

Purpose:For discussion as introduction to base contributions

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 >.

2

Scope and methodology• This contribution proposes an initial symbol structure concept and design

for downlink 802.16m• Symbol structure includes:

1. Subchannelization: – slot (logical “resource block”) definition– mapping slots to physical subcarriers (“permutation”)

2. PilotsThe two aspects are tightly related

• Relation between symbol structure and PHY functionality– Symbol structure is a tool to support PHY functionality (such as various

MIMO modes, interference mitigation, etc), so is dependent on this functionality

– However symbol structure affects feasibility of PHY features and a baseline symbol structure is needed to unify simulation assumptions and align thoughts

– To solve this tie we need to iterate between structure and functionality

Structure Function

3

Scope and methodology

• Process in 802.16e was– Symbol structure originally designed for SISO reuse 3

system (PUSC)

– Other features added as patches, creating additional zones, additional complexity, and performance/functionality limitations

• We suggest the following methodology:– Design initial symbol structure based on assumed PHY

feature set which is relatively broad

– Update symbol structure concepts as suitable when features are determined (added / removed) or additional considerations are made

4

Key points (1)

• General Concepts– Design symbol structure to support broadly potential PHY

features (FFR, MIMO modes, precoding, etc) with considerable flexibility

– Based on SRD explicit and derived requirements and trade-offs

– Allow flexibility to add future advanced features• Symbol Structure and Resource Blocks

– Hierarchical symbol structure for dynamic and effective resource allocation.

– Support 16m subframe or shortened subframe– Conditionally align with 16e AMC subchannels

5

Key points (2)

• Pilot Design– Maximize spectral efficiency by considering pilot overhead and

channel estimation loss– Support both dedicated (localized subchannel) and common pilots

(distributed group) with same structure– Support different MIMO modes, # of antennas, and # of streams– Pilots have different locations in different cells (“randomization”) to

reap the gain of boosting– Pilot patterns obtained by computer optimization with manual fine-

tuning• Specific SDD texts are proposed for

– DL Symbol Structure and general concepts– Various types of DL slot/”resource blocks”– Various DL Pilot patterns of different overheads and advantages

6

Contents of this contribution

• Requirements and tradeoffs

• General concept

• Resource block design

• Pilot design

• Usage scenarios and comparison with baseline system

• Review of text proposal

7

Requirements

8

Requirements from SRD

• Peak SE – To support high peak SE numbers in SRD overheads need

to be minimized

– We need low overhead to support the peak SE but can allow higher overhead in other cases

– To enable satisfying peak SE for 2x2 and 4x4 we limit the pilot overhead to:

• 11% for 2 streams

• 16% for 4 streams

Peak SE

9

Requirements from SRD• Mobility support (7.3)

– optimize for low mobility, graceful degradation up to 350Km/h

• Support of soft reuse (Annex A.2.1, informative) = FFR• Support of relay (Annex A.2.2, informative)• Support of 2,4 TX antennas in DL and 1,2 in UL

(baseline)– We would like the structure to be extensible to more (e.g. 8 DL)

• VoIP capacity requirements (7.2)

Mobility Performance

Stationary, Pedestrian 0 - 10 km/h Optimized

Vehicular 10 - 120 km/h Graceful degradation as a function of vehicular speed

High Speed Vehicular 120 - 350 km/h System should be able to maintain connection

10

PHY functions and implications

• Localized/diversity resources• Open-loop MIMO transmission• Precoded MIMO techniques

– Single user / Multi-user MIMO– Short term / Long term precoding– Fixed / Adaptive precoding

• FFR (reuse partitioning)• Relay• MBS• Power control

Localized/distributed resources

Open-loop MIMO transmission

Precoded MIMO techniques

FFR / soft reuse / reuse partitioning

MBS and Relay

11

Derived Requirements and tradeoffs

Requirements related to frame structure [7,8]• Fixed-duration 6 symbol subframe• Subframe shortening (e.g. to 4/5)

– For support of additional switching points in TDD (erase last symbol of subframe before the switch)

– For support of special transmissions (e.g. sounding, preambles)– Proposed mechanism is that first/last symbol of the subframe will be erased– To be considered in pilot and subcarrier mapping design

• Duplexing: similar structure for TDD and FDD• Slot (RB) size considerations

– Small slot => low pilot efficiency with dedicated pilots– Large slot => degrades small packet (VoIP, TCP ACK) efficiency– See [9] and slide: Small packet efficiency

12

Derived requirements and tradeoffs• Functional unification

– Situation in current 802.16 OFDMA spec:• Different features such as localized/distributed (AMC/PUSC),

MIMO/BF with different ranks, etc cannot be supported in same zone– Result: coarse fragmentation of resources and limitation on feature

support

• Each permutation and MIMO mode has different symbol parameters (useful subcarriers, pilots, etc)

– Result: complexity and inconsistency

– In 802.16m we are likely to have more features than current spec

We would like to be able to support all PHY features in the same subframe, with a unified permutation and pilot pattern

13

General concept

14

General concept (1)

• OFDMA parameters:– No change to sampling factor and CP length

• to allow smooth TDM (/FDM) between legacy and new structures

– Number of used subcarriers as in AMC (maximal Nused in 802.16) = 1+27/32NFFT (865 in FFT=1024)

• All modes supported within the same zone• One-dimensional resource block indexing over the

subframe– Motivation is simplicity, scheduling efficiency, sharing

between localized and distributed

15

General concept (2)• FFR and multi-mode support

– Hierarchy of resources from entire subframe to single slot

– A top of hierarchy, changes are infrequent (multi-cell coordination)

– At bottom, changes are per cell and per user

– Number of resources at each branch is flexible and controlled

Entire subframe

FFR group

LocalizedDistributed

Sc1Sc2Sc3Sc4Sc5Sc6

Multi cell

Cell/sector

User

FFR group

• Pilots– Pilots are precoded and organized in clusters (narrow-band)– Channel is estimated independently per cluster or two clusters– Pilots are dedicated to localized subchannels and common to a diversity

resource– Sounding symbols in DL and UL are used to estimate physical

(unprecoded) channelDL Slot Allocation Process

Pilot design

16

Rationale for precoded pilots

Wideband, not precoded

Precoding/ beamforming“open loop”

N streams < N antennasN streams = N antennas

FFR /boosting / power control

Flat power

Peak SERX

Interference cancellation

Narrowband, precoded

Since we assume use of precoding and FFR in our system, precoded narrowband pilots are preferred in most factors

Summary of differences precoded/non precoded – which is the best case for each type of pilots?

Details rationale for precoded pilots

17

Resource block design

Review of the text proposals

Usage scenarios and comparison with baseline system

18

Downlink Resource Block Design

• Flexible Structure– Support different modes simultaneously– Different frequency reuse groups (reuse factors, powers,

interference mitigation, etc)– Localized + distributed allocations– Different modulation modes (MIMO,BF, etc)

• Hierarchical symbol structure with– Semi static frequency reuse groups coordinated between

cells– dynamic resource groups and allocations– M x N as basic cluster

19

DL Slot/ Resource Block “RB” Definitions

• Cluster: – Physically contiguous group of M by N subcarriers – Default: 9 subcarriers by 6 OFDM symbols (TBD)

• miniBand – Nclusters physically adjacent clusters (default Nclusters =2, TBD)– Minimal unit for FFR group partition, I.e. FFR group is allocated several miniBands– Can be used for more efficient pilot design

• Slot / “Resource block (RB)”– Logical block of data subcarriers.– Physically contiguous or distributed– Default: 54 (TBD) subcarriers including pilots

• Localized RBs– A downlink localized RB can be used for frequency-selective scheduling and multi-user diversity.– Localized RBs comprising a subchannel are adjacent/contiguous in frequency– Organized in localized-groups

• Distributed RBs– A downlink distributed RB can be used for frequency-non-selective scheduling– distributed RBs comprising a subchannel are spreading across frequency within the frequency reuse

group– Organized in distributed-groups. Several groups in the same FFR group can be used (e.g. for long term

BF)

20

DL Slot Allocation ProcessP

hysi

cal f

requ

ency

Logi

cal m

inib

ands

FF

R g

rp 1

FF

R g

rp 2

FF

R

grp

2

Localized

Diversity group 1

(1)Distribute miniBands to

FFR groups

(2)Distribute miniBands to localized and diversity

groups

Localized

Diversity group 2

Diversity group 1

Localized

Subcarrier (Inner)

permutation

(3)Distribute subcarriers

to subchannels (logical resources)

Subcarrier permutation

Subcarrier permutation

010203040506070809

00

...(Outer) Permutation

of miniBands

to FFR groups

(Optional) Pseudo-random

(Opt

iona

l) P

seud

o- r

ando

m

dist

ribut

ion

Inter-cell (semi static) Intra-cell (potentially dynamic)

Resource groups Single resource

21

Slot allocation process illustrated

Subframe

6 symbols

MiniBandE.g. 24 sc.

•Divide spectrum to miniBands

•Distribute miniBands to FFR groups

•Collect minibands of each FFR group

•Distribute to localized and distributed groups

•For localized, distribute subchannels, add pilots

•For distributed, add pilots, distribute subchannels

MiniBand MiniBand

Localized

Distributed#1

MiniBandCluster=subchannel 1 slot

1 slot out of 6 in the group

22

DL Slot Allocation Process (2)1. Form M by N Clusters2. Form miniBands by grouping clusters3. Mapping miniBands into frequency reuse groups

– Mapping is to maximum spreading (Outer Permutation)– Based on number and size of FFR group defined– FFR groups are orthogonal and synchronized across cells

4. Split each frequency reuse group to localized and distributed resources– Based on number and size of localized & distributed group within each frequency

reuse group– Assign Localized resource based on CQI information– The distributed resources group is analogous to major group in DL-PUSC with

flexible size5. Form localized Slot “RB” and distributed Slot “RB”

– Each M by N contiguous localized subcarriers is a localized slot– Interleave (inner Permute) the distributed subcarriers within the group to form

distributed slot

23

Resource Block Comparison –(1)Cluster & miniBand Sizes

12x6 cluster /

12x6 miniBand

12x6 cluster /

24x6 miniBand

9x6 cluster /

9x6 miniBand

9x6 cluster /

18x6 miniBand

18x6 cluster /

18x6 miniBand

DL Pilot Efficiency*

Less efficient, 12x6 pilots patterns

Efficient, 24x6 and 12x6 pilots patterns

*24x6 ~3-4% more efficient than 18x6

Less efficient, 9x6 pilots patterns

Efficient, 18x6 and 9x6 pilots patterns

Efficient, 18x6 pilots patterns

UL Pilot Efficiency

Efficient, 4xM, 6xM tiles

Efficient, 4xM, 6xM tiles

Less efficient, 3x3 or 3x6 tile

Less efficient, 3x3 or 3x6 tile

Efficient, 3xM, 6xM, 9xM tiles

UL RB design Can use 16e UL 4x3 tiles

Can use 16e UL 4x3 tiles

Can not reuse 16e 4x3 tiles

Can not reuse 16e 4x3 tiles

Can not reuse 16e 4x3 tiles

DL RB design Not directly aligned with 802.16e AMC subchannel

Not directly aligned with 802.16e AMC subchannel

Aligned with 802.16e AMC subchannel

Aligned with 802.16e AMC subchannel

Aligned with 802.16e AMC subchannel

•Efficiency is defined in slide 26•Alignment with AMC is beneficial for CQI reuse and FDM with 802.16 which is considered for UL.

24

Resource Block Comparison - (2)Cluster & miniBand Sizes

12x6 cluster /

12x6 miniBand

12x6 cluster /

24x6 miniBand

9x6 cluster /

9x6 miniBand

9x6 cluster /

18x6 miniBand

18x6 cluster /

18x6 miniBand

VoIP / small packet

Good Good Better Better Worse

Diversity Gain (< 1 dB difference)

Smaller RB has better DG for small diversity group

Slightly less DG due to larger RB and miniBand

Smaller RB has better DG for small diversity group

Slightly less DG due to larger RB and miniBand

Slightly less DG due to larger RB and miniBand

Scheduling Gain (< 1 dB difference)

Smaller RB has better SG for small local group

Slightly less SG due to larger RB and miniBand

Smaller RB has better SG for small local group

Slightly less SG due to larger RB and miniBand

Slightly less SG due to larger RB and miniBand

CQI overhead High Lowest High Medium Medium

Overall Not Favorable Fair Not Favorable Favorable Fair ? (due to concern related to VoIP efficiency)

Based on this our current recommendation is 9x6 cluster (=subchannel) with 18x6 miniBand

25

Pilot design

26

Pilot targetsPurposes of pilots

1. Channel estimation2. Measurements: Signal, Noise + interference

• CQI• interference mitigation/cancellation

3. Frequency offset estimation4. Time offset estimation (UL)

• Purposes are ordered in decreasing order of estimation difficulty (estimation error) and effect of estimation error on system performance

• Channel estimation has the most severe impact on system performance• Our current analysis focuses mainly on optimizing this effect

Non purposes of pilots• cell identification• broadcast/control information transmission

27

Pilot Optimization• Optimization target is maximum spectral efficiency

considering pilot overhead and channel estimation loss [1,2,3]– SE abstraction

Ideal SE is based on link level simulation with perfect channel knowledge

– Equivalent SNR at the decoder output with MMSE estimator is

– Maximize SE by optimizing overhead, pilot location and boosting

cluster one in tonesofnumber theisNt pilot, ofnumber theis Np

11

1

tN

tdecoder

tTX

t

p tSNRidealSEN

NN

NSE

1,

1

1,

2

2

2

2

2

2

2

21

1

MMSEHTX

MMSEH

eTX

e

eff

decoderSNRNSNR

SNR

hEN

h

hEhSNR

28

Main concepts of pilot design• Pilots are dedicated to a localized subchannel and common to distributed group

– Pilot streams: • In precoded case, equals to the number of data streams • Un-precoded case, may equal the number of Tx antennas (or less if CDD/similar used)

– Pilots are precoded/boosted the same way as data is precoded/boosted– “common” means same pilots are used for more than one burst (they can still be

precoded)• Pilots are boosted

– Desirable to have equal number of pilot per symbol to reduce power fluctuation (which reduces TX power and range)

– Scattered pilots (in time) are better than concentrated since they allow higher boosting values

• For MIMO channel estimation, pilot separation between streams by different locations combined with nulls (perfect separation)

• Pilot separation between BS by randomization (selection from fixed range of options)

• For MU-MIMO where interference is small, pilot streams are CDM to reduce overhead

• Pilot patterns we propose are a result of computer based optimization with manual fine-tuning

29

DL Pilot Patterns• Total of 3 pilot patterns to support of 1,2,and 4 pilot streams • Each pattern has several orthogonal versions. In next figures these versions are

shown overlaid– Pattern A:

• Designed for 18x6 (mini-band), for support of 1,2 streams• Used for distributed group and localized burst with two clusters based RB• 5.5% overhead per pilot stream (6 pilots)• 6 versions

– Pattern B: • Design for 18x6 (mini-band) for support of 4 streams• 15% overhead • 1 version

– Pattern C:• Design for 9x6 (cluster), 1/2 streams• used for localized users with one cluster based RB • 7.5% overhead per pilot stream (4 pilots)• 6 versions

• To support of shortened sub-frame with 5 OFDM symbols, the last OFDM symbol together with pilots will be punctured

30

Patterns A and C

(a) Pattern A: 6 pattern overlay (b) Pattern C: 6 pattern overlay

(c) Pattern A 1+2 for 2 pilot stream (d) Pattern C 1+2 for 2 pilot stream

0 1 2 3 4 5 6 7-10

-9

-8

-7

-6

-5

-4

-3

-2

-1

0

symbol

sub

carr

ier

Pattern C 1+2 (9 x 6), OH 7.4 % per stream

0 1 2 3 4 5 6 7-10

-9

-8

-7

-6

-5

-4

-3

-2

-1

0

symbol

sub

carr

ier

Patterns C (9 x 6), OH 7.4 % per stream

0 1 2 3 4 5 6 7

-18

-16

-14

-12

-10

-8

-6

-4

-2

0

symbol

sub

carr

ier

Pattern A 1+2 (18 x 6), OH 5.6 % per stream

0 1 2 3 4 5 6 7

-18

-16

-14

-12

-10

-8

-6

-4

-2

0

symbol

sub

carr

ier

Patterns A (18 x 6), OH 5.6 % per stream

31

Combining Patterns

• The single antenna patterns can be used for single stream or combined to create 2 antenna patterns (orthogonal)

• Multiple patterns is to allow randomization, or planning– Total of 6 patterns: 2 per sector (MIMO 2x2) X 3

reuse

32

Randomization to enjoy boosting gain

• The gain from pilot boosting in white noise (when total TX power is constant) is substantial

• In current spec pilots collide with pilots and therefore boosting gain is lost in interference limited case

• To gain from boosting in interference scenario pilots should collide with data subcarriers

• This is always better than having pilots collide with pilots, since colored interference is better than white noise

PilotsDataNoise

Pilots boostedData deboostedSame avg power

Same done for the interferer

33

Pattern B – 4 Pilot stream

• Support 4 TX streams transmission

• Used also for full channel sounding for close loop beamforming with 4 TX antennas

• Not considered for x3 reuse since boosting leads to small/zero power fluctuations

0 1 2 3 4 5 6 7

-18

-16

-14

-12

-10

-8

-6

-4

-2

0

symbol

sub

carr

ier

Pattern B (18 x 6), OH 3.7 % per stream

34

Boosting values

– Optimal boosting value is based on white noise optimization, but is also good for interference

– The boosting value is the difference between pilot and data power

– The average power should be normalized to 1, and therefore different power for data tones may result in each case – this is not an issue since the pilots are dedicated

Pattern N streams

Boost value

A 1 5.6dB

2 7.4dB

B 4 8.2dB

C 1 5dB

2 5dB

35

Analysis for 2x2 MIMO

•These figures present spectral efficiency estimated using the aforementioned model, for V=120Km/h, without precoding.

•The left figure presents SE and the right shows SE normalized by the SE of the best scheme.

•The results are compared against the LTE design for 2 streams with boost=3dB [6]. LTE use wideband pilots which are optimal for non-FFR non-precoded case.

•The narrow band pilot design we propose performs well even in this scenario. We expect much better performance in FFR and beamforming cases.

-10 0 10 20 30 400.55

0.6

0.65

0.7

0.75

0.8

0.85

0.9

0.95

1Effective SE / best SE

SNR[dB]

ratio

Pattern C: 9x6LTE 2 streamPattern A: 18x6Wideband 18x6

-10 0 10 20 30 400

2

4

6

8

10

12Effective SE

SNR[dB]

SE

[B

/s/h

z]

Pattern C: 9x6LTE 2 streamPattern A: 18x6Wideband 18x6Ideal SE

36

Usage scenarios and comparison with baseline system

37

Support of precoded MIMO options

Mode Short term precoding (channel knowledge)

Long term precoding (correlation/DoA knowledge)

no precoding (OL-MIMO) •N/A •distributed group with e.g. 2 pilot streams, pilots are not precoded. •For 4 TX, modulation can be STBC/SM (per user) with CDD (common), or 4 streams for 4x2 configuration (less preferred due to pilot efficiency)

Fixed precoding:

(“OL-MU-MIMO” or “OL-SU-MIMO” with precoder)

•Localized resource with fixed precoding, MS measures CINR as basis to select the resource•Precoder doesn’t have to be conveyed to MS (and can change, infrequently)•Size of group changes slowly (since requires new measurements)

•distributed group with fixed precoding and fixed number of streams. •Number of pilots streams = number of data streams.•Number of different precoders is limited due to size of distributed groups•MS can measure CINR on groups to determine preferred group•Distributed groups inside FFR group can be randomized (cell-ID dependent permutation)

Adaptive precoding (“CL-MU-MIMO”, “CL-SU-MIMO”)

•Localized resource, adaptive precoding, MS reports according to sounding symbols

•distributed group allocated to a single user if traffic justifies

38

Parameters and comparison with baseline system

Item WiMAX 1.0 16m proposal Reasoning

OFDMA parameters Nused=varying

CP length = 1/8

Nused as in AMC

CP length = 1/8

Subcarrier spacing remains the same

Symbol alignment with baseline system, maximum used BW.

Diversity/Localized support

Open loop / Precoded MIMO support

Different zones, different permutations (PUSC, AMC)

Unified in same symbol structure

Reduce frame fragmentation and unify and simplify solution.

FFR support Partial & rigid (by PUSC major groups) in time domain (zones)

More flexible, multiple reuse factors in same symbol

Optimize FFR and make it more usable

Slot size 48 logical subcarriers 54 physical subcarriers, varying logical subcarriers

Remove constraint on fixed pilot overhead. Rate matching used to bridge the gap.

Size of WiMAX 1.0 slot w.r.t MAC packets is reasonable choice

Slot duration in time 1-3 symbols 6 symbols (subframe) Fixed duration in time & frequency-dimension scheduling simplify design

Pilot density (DL) 14% PUSC, 11% AMC. Pilots decimated between 1TX and 2TX

5.5% 1TX, 11% 2TX, 15% 4TX. More pilots added with NTX

Optimized overhead per number of antennas. In WIMAX1 too much overhead for 1TX and 4TX.

39

Item WiMAX 1.0 16m proposal Reasoning

Pilot modulation in precoding

Precoded & dedicated pilots (mainline)

Precoded pilots Precoded pilots more efficient when data is precoded/boosted

Pilot locations per BS Pilots hit with pilots Pilots are “randomized” (choose from set of locations)

Pilots hitting pilots eliminates boosting gain in interference limited case; #1 malady of WiMAX1 pilot design

Pilot boosting 2.5dB,5.5dB for 1,2 streams (DL)

5.6dB, 7.4dB, 8.2dB for 1,2,4 streams

Optimized boosting (higher)

Distributed permutation design

FUSC: wideband pilots, scattered subchannels

PUSC: narrowband pilots, major groups, clusters, subchannels spread over groups; two layer permutation.

Use PUSC concept but extend it to unified feature support

PUSC concept is good for multi-mode and FFR support and for distributed transmission with precoded pilots

Permutation concept Fixed and rigid structure, optimize hit ratio between subchannels

Flexible structure, less optimization of hit ratio possible

Hit ratio less important when interference is managed (FFR); tougher to maintain in flexible and dynamic structure.

Permutation unit PUSC cluster = 14 sc

AMC subchannel = 18 sc

Cluster = 9 sc

miniBand = 18 sc

See Resource Block Comparison –(1)

Parameters and comparison with baseline system

40

Review of text proposal

41

Review of the text proposals

• The text proposals are split as follows:– Generic/high level part:

• C80216m-08/120: Proposal for IEEE 802.16m Downlink Symbol Structure Concept

– Detailed part:• C80216m-08/121: Symbol structure design for 802.16m

- resource blocks and pilots

42

Specific Text Recommendations forSDD* Section 11 – PHY Layer – high level

Section 11.x: DL Symbol Structure Subsection 11.x.1 – motivation

– The downlink symbol structure will support different modes simultaneously in the same subframe, for example

(1) Localized and distributed resource allocations, (2) Different frequency reuse groups (different reuse factors, powers and other

properties related to interference mitigation and management)(3) Different MIMO modulation modes: Open loop / closed loop, single-user and

multi-user.– The symbol structure is optimized for precoded MIMO and fractional

frequency reuse (FFR) operation.Subsection 11.x.2 – high level concept descriptionSubsection 11.x.2.1 – general• The same OFDMA parameters (sampling factor and CP duration) of the

baseline system shall be used where the number of used subcarriers will be equal to the Nused of AMC permutation. The resource allocation is done in frequency domain only over the 6-symbol subframe, i.e. a slot equals one subchannel by 6 symbols.

* System description document [9]

43

Specific Text Recommendations forSDD Section 11 – PHY Layer – high level

Subsection 11.x.2.2 – subchannelization concept• FFR and multiple modulation modes will be supported by a hierarchical structure

as depicted in fig XX [slide General concept (2)], where the symbol is first split into FFR groups, then these groups are split into localized and distributed resource groups, and then these groups are split into individual subchannels. The number of resources at each branch will be flexible and controlled by the cell (the level of flexibility and required control channels are TBD).

• The motivation for the hierarchical structure is that at the top of this hierarchy changes are infrequent (multi-cell coordination), and at bottom, changes are more frequent (per cell and per user), so this structure yields most flexibility for the cell/user while reducing the amount of coordination/broadcast information required. By stating the structure explicitly we also allow the MS to identify resource groups required for various measurements.

Subsection 11.x.2.3 – pilot structure concept • The targets of pilots are: Channel estimation, measurements (such as Signal,

Noise + interference for CQI and interference mitigation/cancellation), Frequency offset estimation, Time offset estimation (UL)

• Pilots will be dedicated to localized subchannels and common to each group of diversity resources. The pilots will be precoded (using the same precoding and/or boosting of the information burst(s)). Sounding symbols in DL and UL may be used to estimate physical (unprecoded) channel.

44

Specific Text Recommendations forSDD Section 11 – PHY Layer (1)

Section 11.x: DL Symbol Structure – A cluster is physically contiguous group of M subcarriers by N

symbols, with default value of 9 subcarriers by 6 OFDM symbols, i.e. M=9, N=6.

– A miniBand consists of Ncluster (TBD, default 2) of these adjacent basic clusters. The miniBand can be used as basic units for CQI reporting, frequency reuse groups partitions and efficient pilot design.

– A slot (“Resource block/RB”) is a logical block of subcarriers which maps to an effective physical unit of M subcarriers by N symbols, with default value of 9 subcarriers by 6 OFDM symbols, i.e. M=9, N=6. The total 54 (MN) subcarriers in RB include both pilots and data tones.

• number of data subcarriers in slot varies with different pilot densities• for slots in distributed allocation the 54 subcarriers are pseudo-randomly

distributed across all available distributed subcarriers with the reuse group

45

Specific Text Recommendations forSDD Section 11 – PHY Layer (2)

• Section 11.x: DL Subcarrier to Resource Block Mapping– Add “DL RB Allocation Process” figure here.– A renumbering sequence (TBD), i.e. “Outer Permutation”, is applied

on miniBands to maximum spreading and map miniBands into a given size of frequency reuse group

– The frequency reuse group is partition into localized and distributed groups. The size of each group is flexible and the mapping is uniquely determined by the size of each group.

– The localized and distributed groups are further mapped into localized (by direct mapping) and distributed RBs (by “inner permutation”).

– BS scheduler can semi-dynamically change the Ratio between localized and distributed RBs for scheduling and diversity gain based on CQI feedbacks

46

Specific Text Recommendations forSDD Section 11 – PHY Layer (3)

• Section 11.x: DL Localized resource block format– Localized subchannels contain subcarriers which are contiguous in

frequency. – Pilots used for localized subchannels are dedicated and should be

precoded/boosted the same way as data.

• Section 11.x: DL Distributed resource block format– distributed subchannels contain subcarriers which are spread across

frequency within the frequency reuse group– The subcarriers within the distributed groups are permuted (“inner

permutation”) across distributed group to maximize frequency diversity.

– Common pilots are used in distributed groups and are shared within the group.

– The distributed group may be precoded in which case the pilots are precoded together with data.

47

Specific Text Recommendations forSDD Section 11 – PHY Layer (4)

• Section 11.x: Pilot patterns – Pattern A and C should be used for 1 and 2 pilot streams. Pattern B

should be used for 4 pilot streams. – Pattern A should be used as the common pilot in distributed resource

blocks. – Pattern C should be used when the localized burst is carried on 1

cluster. Pattern A should be used when the entire miniBand (2 clusters) is used by the same burst.

• Section 11.x: Combining pilot patterns – Pattern A and C each has 6 single patterns– The single antenna pattern should be used for single stream or

combined to create 2 antenna patterns (orthogonal)– Different base station should coordinate to choose different

combinations of the pilot patterns for randomization

48

References[1] Cosovic, I.; Auer, G., "Capacity Achieving Pilot Design for MIMO-OFDM over Time-Varying

Frequency-Selective Channels," Communications, 2007. ICC '07. IEEE International Conference on , vol., no., pp.779-784, 24-28 June 2007

[2] Hassibi, B.; Hochwald, B.M., "How much training is needed in multiple-antenna wireless links?," Information Theory, IEEE Transactions on , vol.49, no.4, pp. 951-963, April 2003

[3] Lang Tong; Sadler, B.M.; Min Dong, "Pilot-assisted wireless transmissions: general model, design criteria, and signal processing," Signal Processing Magazine, IEEE , vol.21, no.6, pp. 12-25, Nov. 2004

[4] IEEE 802.16m-07/002r4, “802.16m System Requirements”[5] IEEE 802.16m-008/004, “802.16m Evaluation Methodology”[6] 3GPP TS 36.211 V8.1.0 (2007-11) Technical Specification, 3rd Generation Partnership

Project; Technical Specification Group Radio Access Network; Evolved Universal Terrestrial Radio Access (E-UTRA); Physical channels and modulation (Release 8)

[7] IEEE C80216m-08/082, “Proposal for IEEE 802.16m Frame Structure”[8] IEEE C80216m-08/118r1, “Initial Document output of the IEEE 802.16 TGm Frame Structure

Rapporteur Group”[9] IEEE 80216m-08/003, “The Draft IEEE 802.16m System Description Document”

49

Backup Slides

50

Peak SE

• Calculation of peak SE– Symbol duration incl CP (1/8) = 102.85

– Useful subcarriers = 864 @10Mhz (AMC)

– MCS: 64QAM 7/8 (5.25 b/s/hz/stream)

– Pilot overhead: 11%, 16%

– Preamble overhead: not considered yet

– Peak thput = (Bits/Subcarrier)#Subcarriers (1-OH) / SymbolTime

– Conclusion: with same CP, we barely meet the requirement from below, but the difference doesn’t seem to justify new CP

Requirement Type Baseline Baseline Target Target

Link direction Downlink Uplink Downlink Uplink

MIMO Configuration 2x2 1x2 4x4 2x4

Required Normalized peak rate (bps/Hz) 8 2.8 15 5.6

Assumed theoretical spectral efficiency 10.5 21

Peak throughput @10Mhz 78.5 148.2

SE over entire BW (e.g. 10Mhz) 7.85 14.81

SE over occupied BW (e.g. 9.46Mhz) 8.29 15.66

51

Localized/distributed resources

• Distributed resource (“PUSC”) – spread across spectrum, yields frequency diversity, mainly used for high mobility / low feedback

• Localized resource (“AMC”) – localized in spectrum, enables frequency-selective scheduling, associated with low mobility / high feedback rate

• Support of both diversity and localized resource is required in order to guarantee high throughput for both low and high mobility users

52

Open-loop MIMO transmission• Open-loop MIMO transmission intended for user with high mobility or low rate of

CSI• Will typically be used in conjunction with distributed resource• We name a few options as reference• For 2TX: STBC (Matrix A), SM (Matrix B)• For 4TX:

– Techniques using 4 channel streams:• SM, rate 4• Stacked Alamouti (STBC+SMx2), rate 2• Large delay CDD + STBC, rate 1• May consider STBC design for 4 streams with rate 1

– Techniques using 2 channel streams• small-delay CDD + STBC, rate 1• small-delay CDD + SM, rate 2• Antenna-Frequency interleaving can be used instead/in addition to CDD• CDD information can be embedded into pilots; the advantage is in lower pilot overhead and

channel estimation loss

• For >4TX: combine CDD with 4TX/2TX scheme

53

Precoded MIMO techniques• Transmitted signal is modulated by a spatial linear precoder

x=VsV [NTX Nstreams] is the precoding matrix

• Precoder may be used to direct energy toward MS / reduce interference / etc.

• Single user / Multi-user MIMO (SU/MU)– Single user – all streams directed to same user. Achieves high peak SE.– Multi user – different streams directed to different users (SDMA).

Achieves high avg SE.

54

Precoded MIMO techniques

• Short term / Long term precoding– Short term – depends on approximate instantaneous channel knowledge.

Requires frequent feedback, yields best performance.– Long term – depends on long term channel statistics (e.g. correlation /

geometric direction), requires less feedback, can operate in higher mobility.• Fixed/Adaptive precoding

– Fixed precoding – BS determines precoder per subchannel, MS/BS selects subchannel (and/or stream) to best fit conditions. Guarantees predictable interference, low overhead, multi user diversity. Good solution when number of users is high.

– Adaptive precoding – precoder is a function of user conditions (by sounding / report / codebook selection). Good when number of users is small or some users require higher throughput.

• All combinations are useful in some scenarios{SU,MU} {Short,Long} {Fixed, Adpative}

• We assume most of these combinations will be used in 802.16m

55

FFR / soft reuse / reuse partitioning• FFR = fractional frequency reuse

– a way to increase system performance by using different reuse factors for different subchannel/user groups

– An example (supported in 802.16) is combination of reuse 1 and reuse 3

• Use model 1: move cell edge users suffering high interference to reuse 3 to improve cell edge throughput

• Use model 2: although on average over cell locations reuse 1 provides better SE than reuse 3, for a specific location/geometry, reuse 3 may be better. Choosing for each user the best reuse factor improves average and user throughput

– A softer version applies lower/higher power to some subchannel groups:

PHigh

PHigh

PHigh

PHigh

PHigh

PHigh

PHigh

PHigh

PHigh Preuse1

Preuse1

Preuse1

PLow PLow

PLow PLow

PLow PLow PLow

PLow

PLow

Reuse 3 Reuse 3/2 Reuse 1

Power

Power

Power

Frequency

Sector 3

Sector 2

Sector 1

w1 w2 w3 w12 w23 w13 w123

w

56

FFR / soft reuse / reuse partitioning• How many FFR groups?

– For 2 power levels and 3 segments (“colors”) we have 7 meaningful combinations (23-1 since the combination [Plow, Plow,Plow] is not useful)

– Too many groups cause resource fragmentation– We assume 1-7, typical 4 groups

• It is possible to use other attributes in addition to power. – For example, group using a single-stream or STBC transmission

improves ability for interference cancellation and may be more suitable for cell edge

• Dynamics of FFR: FFR can be static / semi-static or dynamic in two senses:– The allocation of users to groups – internal in the cell– The FFR groups (subchannels, attributes) - requires multi cell

coordination; we assume this process is slower than per-cell scheduling

57

FFR / soft reuse / reuse partitioning

• Symbol structure should support FFR in a generic way with minimum limitation to the scope of the solution

• For the purpose of symbol structure we generalize FFR as follows:– We divide the resource (spectrum x time) into orthogonal FFR groups– Each FFR group may have different attributes such as power, restriction

on type of modulation / number of streams, etc– We assume the structure changes on a slower basis than per-cell

scheduling and can be shared on slow basis (DCD/UCD)– FFR groups should be spread in frequency to maximize frequency

diversity and scheduling gains– As working assumption (for simulation/analysis) FFR groups are equal

(same subcarriers) across cells; in practice they may not be (due to traffic / deployment asymmetry)

58

Control channels

• Currently we do not discuss the design to control channels (DL maps, UL CQI, ranging, etc)

• We believe the design should first be optimized for traffic; then design for control channel should be a derivative

• General considerations for control channels:– In FFR, DL broadcast maps will use a single FFR group

(choose the FFR group that maximizes SE to reach target outage)

– Maps can be per-subframe or per-frame– UL control channels will use UL diversity resource or bits

of it

59

MBS and Relay

• Relay support may require – Concurrent DL transmission from several relays– Relay DL transmission with BS– This can be supported in a similar way to FFR, by

groups of clusters with associated pilots

• MBS– Requires pilot design for large delay spread

60

Small packet efficiency• Small packet efficiency (e.g. VoIP, TCP ACKs)

– The slot size (=granularity of burst size) affects the system efficiency for small packets

• The traffic size is not always known (20-40B), burst size varies because of MCS• With fixed MCS set, k slots map to kM bytes padding of 0..M-1 bytes• With continuous MCS scale, the number of slots determines coding rate, penalty

payed for mismatch between channel condition and coding rate.• Smallest slot size is preferred for small packet efficiency

– Too small slot size incurs PHY complexity, map overhead and pilot inefficiency (for dedicated pilots)

– We would like to set slot size to the maximum value that has small impact on VoIP capacity

This slide is t

o be updated

61

Rationale for precoded pilots

• We assume main mode of operation will be – Precoded (low mobility optimized) with varying rank– FFR will be used

• Consider two options:– Precoded & narrowband pilots:

• The channel estimation is narrow-band (as in PUSC, AMC). • The MS doesn’t have to know the precoder. • The number of pilot streams is the number of transmitted streams.

– Non-precoded wideband pilots• Channel estimation can be wide-band (as in FUSC). • BS has to provide MS with precoder matrix. • MS estimates physical channel H and uses precoder matrix to calculate

effective channel HW. • The number of pilot streams is the number of TX antennas.

62

The disadvantages of precoded pilots

• Start with the down side:– Narrowband pilots/channel estimation is less efficient

locally• However the difference can be made small (3-5% loss in SE), and

is more than compensated by the other advantages

– Non-precoded pilots allow estimation of the physical channel for feedback purposes

• With precoded pilots we need to add DL sounding symbols (“midamble”)

• However, channel estimation quality required for feedback is not high – therefore small overhead will be required to support the requirement.

63

The advantages of precoded pilots

• Precoded pilots have the following advantages for a precoded system:– Non-precoded pilots limit number of TX antennas, while precoded

pilots determine only number of streams– Non-precoded pilots require precoding codebook to be predefined +

signaling of the chosen codeword, whereas precoded pilots allow more flexibility in determining the solution and room for future optimization

– Precoded pilots enjoy the beamforming gain (the pilot energy is focused toward the MS direction), thus more than compensating for smaller effective number of pilots (compared to wideband pilots)

– Smaller overhead when number of streams is smaller than num of antennas.

• This occurs in most scenarios, if typical NTX is 2..4 while typical number of streams is 1-2

• Assuming 5.5% per stream the difference is significant 5-10%

Non precoded pilot

precoded

pilot

64

The advantages of precoded pilots

• Pilot adaptivity: by using local/dedicated pilots we can design pilots that fit the scenario

• Two examples– For multi-user MIMO

• Pilots can be optimized for low-cross talk. • For example, CDM/randomization instead of orthogonal streams can be

used to separate pilots of the users, thus reducing overhead• This is in addition to the advantages mentioned

– For open-loop MIMO• At low SNR, considering unknown channel (or channel estimation loss) it

is preferred to transmit on less degrees of freedom (=less antennas or streams) [Hassibi*]. Dedicated/adaptive pilots enables this matching.

• At high SNR open loop MIMO, narrowband pilots are indeed less efficient.

* Hassibi, B.; Hochwald, B.M., "How much training is needed in multiple-antenna wireless links?," Information Theory, IEEE Transactions on , vol.49, no.4, pp. 951-963, April 2003

65

The advantages of precoded pilots

• For FFR, we compare a case where pilots are dedicated to the group (top) to a case the pilots are common to all groups (bottom)

PilotsFFR Group 1FFR Group 2

PilotsFFR Group 1FFR Group 2

• Narrowband pilots belonging to FFR groups have the following advantages in FFR– Pilots which boost/deboost with data are more efficient than pilots

which have fixed boosting– Pilot quality matches data quality; pilots have the same reuse as data– Allow exact estimation of CINR of each FFR group (compared to

indirect deduction of CINR with pilots common to all groups), thus enabling robust link adaptation and dynamic choice of FFR group.

66

The advantages of precoded pilots

• RX interference mitigation– RX interference mitigation such as spatial LMMSE (IRC) yields link

level gains of 1-3dB and system SE gains of 10-15% (2RX)– Evaluation methodology document requires that usage of LMMSE will

be justified– Precoded pilots enable instantaneous interference estimation for RX

interference cancellation. Such estimation is not accurate with non-precoded pilots if interference is precoded / boosted

67

DL RB Allocation ExampleAssumptions:

– FFT_Size=1024, Nused=864, M =12, N =6, Nclusters=2 – FFR ratio =1/3, Localized Ratio=1/2

Steps:1. Form 12 x 6 Clusters 72 clusters (864/12=72)2. Form 36 miniBands (assume Nclusters=2) 72/2=36 miniBands3. Mapping miniBands into FFR groups (36 x 1/3=12 miniBands)

– Perform Outer permutation to maximum spreading– Assign 12 miniBands to the FFR groups

4. Assign localized and distributed resources (12 * ½ = 6 miniBands)– Assign 6 miniBand to Localized resource and 6 to distributed resource

5. Form localized RB and distributed RB– Form 12 localized RBs from the 6 localized miniBands– Form 12 distributed RBs after “Inner Permute” the distributed subcarriers

within the distributed 6 miniBands