A 4G System Proposal Based on Adaptive OFDM

Mikael Sternad

Part of SSF PCC, 2000-2002

A SSF funded project2002-2005

+Vinnova funding

www.signal.uu.se/Research/PCCwirelessIP.html

The Wireless IP Project

Visions and Goals

• A flexible, low-cost general packet data system allowing wide area coverage and high mobility (vehicular velocities)

– Perceived performance of 100 Mbit/s Ethernet– High spectral efficiency (10 fold increase vs.

3G)– Quality of service and fairness

Leads to an extreme system based on adaptive resource allocation

Design concepts

• Use short term properties of the channelinstead of averaging (predictive link adaptation)

• Interference control (smart antennas etc.)

• Scheduling among sectors and users (combined MAC and RRM)

• Cross-layer interaction(soft information)

Short-term Channel Properties• Typical time-frequency channel behavior (6.4 MHz, ~50 km/h)

• Data from Stockholm, Sweden @1900MHz (by Ericsson) Accurate channel prediction is needed

Coherence bandwidth 0.6 MHz

Coherence bandwidth 4.9 MHz

Channel Prediction

Adaptive Modulation and Prediction ErrorsModify thresholds to keep BER constant (single-user)

Smart Antennas: Simplest Case

MRC in mobile stations (MS)

Advantages BS: Efficient use of space (robust)

Low interference levels

MS: Improvement of SNR (robust)

Fixed lobes (sectors, cells) at base stations

Scheduling Among Users in a Sector

time

freq

user

4

53

2

1• Feedback info from each

mobile: Appropriate modulation level for each bin in a time slot.

• Perform scheduling based on predicted SNR in bins

• For each bin let the “best” user transmit; use adaptive modulation and ARQ scheme

• Modify to take QoS and fairness into account

Minimizing Interference Among Sectors• Exclusive allocation of time-frequency bins to users

within border zones between sectors of a base station.

• Frequency reuse 1 in inner parts of sectors

• Frequency reuse 3 in outer parts of sectors

• Multi-antenna terminals (IRC)

• (Power control)

• Slow resource reallocation

between sites and sectors,

based on traffic load

11

1 1112

22

222

f

time

Design Example: An Adaptive OFDM Downlink

• Maximize throughput. Ignore fairness and QoS

• Target speed 100 km/h +large cells Frequency-selective fading

• WCDMA frequency band (5 MHz bandwidth, 1900 MHz carrier)

• Adaptive modulation. Fixed within a bin (BPSK, 4-QAM, 8-QAM, 16-QAM, 32-QAM, 64-QAM, 128-QAM, 256-QAM)

• Simple ARQ

• No channel coding

Physical Layer

• OFDM system with cyclic prefix yielding low inter-channel interference– Symbol period is 111 s (100+11 cyclic prefix)– 10 kHz carrier spacing (500 subcarriers in 5

MHz)

• Time-frequency grid 0.667 ms x 200 kHz (120 symbols/bin; 5 are pilots)

– Channel ~ constant within each bin– Design target speed is 100 km/h

• Broadband channel predictor– Accurate over λ/4 - λ/2 2 - 4 slots @ 1900 MHz

and 100 km/h

Analysis of ThroughputSimplifying assumptions:

• Flat AWGN channel within each bin; Independent fading between bins

• MRC with L antennas at mobiles (one sector of BS)

• Average SNR = 16 dB / receiver antenna and info symbol (same for all users; slow power control)

• Adaptive modulation. Selection based on perfect channel prediction

• K users. Fairness between users, QoS requirements, and delay constraints are neglected

Analysis of Throughput (cont.)

Spectral efficiency (L antennas, K users):

Cyclic prefix:

Pilots:

11

,0

(1 ( )) ( )i

i

N

c p i FER ii

G G k P p d

1e,

LK

K

Kp L L

L

100 /111cG

108 /120pG

Thresholds

*,arg max (1 ( ))i i FER ii k P

Select the modulation level i as

Modulation

0 BPSK 1

1 4-QAM 2 8.70

2 8-QAM 3 13.53

3 16-QAM 4 16.89

4 32-QAM 5 20.46

5 64-QAM 6 23.59

6 128-QAM 7 26.86

7 256-QAM 8 29.94

ik (dB)ii

Spectral Efficiency and Throughput(one sector, 16 dB)

Thr

ough

put [

Mbi

t/s]

10

15

20

25

Observations• Scheduling gives multiuser selection

diversity (from both time and frequency selectivity of the channels)

• MRC leads to good initial SNR

• Good spectral efficiency improvement already at low to moderate load (#users)

• Not all bins can be used in every sector due to interference

• Uplink control information is required to signal modulation level

Work in Progress

• Evaluation of system level performance

– Intercell interference, QoS, and fairness– First indications give a reuse of 1.7, average SIR

16dB– Results in 1.25 bits/s/Hz/sector at K=1 user/sector (Reuse 1 combined with reuse 3, ”area-fair scheduling”, interference limited, full load, Rayleigh+path loss, L=1

ant.)

• Improved adaptive modulation systems

– TCM (See presentation by Sorour Falahati)– Prediction errors ( - ” -)– Feedback information– MIMO ( 2 x 2 MIMO quite reasonable)

• Development of a network simulator

– Study of TCP/IP interaction

• Design of uplink system

– Single carrier modulation or OFDM?

Work in Progress (cont.)

Network

Server

• Optimize scheduler

– QoS and fairness– Maximum Entropy Scheduler (using information about

buffer influx). Minimize average buffer contents.– Intercell scheduling

• Soft information

– Passing PHY soft information to application– JPEG 2000 application– Modifications to TCP and UDP– Format for soft information

Thank you!

Questions?