Ten Unsolved Hot Problems in Information and ...lampe/Fettweis/GF UBC Green Visiting Prof... · Ten...

Vodafone Chair Mobile Communications Systems, Prof. Dr.-Ing. G. Fettweis

Ten Unsolved Hot Problems

in Information and Communications Technology

Gerhard Fettweis – Green Visiting Professor – UBC

– Vodafone Chair Professor – TU Dresden

– IEEE SSCS Distinguished Lecturer

Trickle / Pipeline of Technology

Vodafone Chair Funding

• e.g. fundamental limits of cellular

DFG funding

• e.g. fundamentals of interference cancellation

BMBF

• e.g. EASY-C

• e.g. Cool Silicon

EU

• e.g. Artist4G

• e.g. EARTH

TU Dresden Gerhard Fettweis Slide 2


Broadband


Coverage: Cellular

1995 2000 2005 2010 2015

Short links (1m)

Cellular (100m)

GSM GPRS

HSPA

HSDPA

LTE

WiMAX

WLAN (10m)

3G R99 / EDGE

LTE Advanced

100Gb/s

10Gb/s

1Gb/s

100Mb/s

10Mb/s

1Mb/s

100Kb/s

10Kb/s

802.11ac/ad

802.11n 802.11ag

802.11

802.11b

USB 1.0

USB 2.0

USB 3.0

UWB intention

802.15.3c


The Wireless Roadmap

1995 2000 2005 2010 2015

Short links (1m)

Cellular (100m)

WLAN (10m)

100x

10x

ITRS Roadmap: Continues until 2020

100Gb/s

10Gb/s

1Gb/s

100Mb/s

10Mb/s

1Mb/s

100Kb/s

10Kb/s


Fairness

High SINR

High data rate

eNodeB eNodeB

Low SINR

Low data rate

High SINR

High data rate

Unfair !!!


Fairness & Data Rate

0.5

0.4

0.3

0.2

0.1

0.0

-10 -5 0 5 10 15 20 25 30

SIR in dB

Reuse 1

Power Control

E[SIR] = -0.2dB

Reuse 1

Power Control

10 Interferers cancelled

Macro/Distributed MIMO

E[SIR] = 8.1dB

pdf of SIR

fairness

Interference Cancellation: Fairness & High Data Rate

data rate

Talk at RAEng 2009-09-13 Gerhard Fettweis Slide 8

Realtime 3D Multimedia Rendering

© FHG HHI Berlin

© MPI Saarbrücken


CoMP: Coordinated Multi-Point

We thus believe that next generation systems will include multi-cell cooperative

signal processing (“network MIMO” or CoMP):

Backhaul

infrastructure

between sites

Cell phone jointly detected by

3 base stations

Shaded area: One site containing

three base stations (i.e. cells)

with four antennas each


Potential Gains of CoMP

Uplink Downlink

Okumura-Hata pathloss model, ITU pedestrian A

Link-to-system mapping (MIESM), 8 MCS schemes

Spectral eff. losses through guard bands / intervals

Assuming perfect channel est., 2 rx ant. per eNB

Linear joint transmission,

assuming perfect channel

knowledge at the eNBs

2 tx ant. per eNB

World’s Largest Operational

LTE-Advanced Algorithm Testbed

11

© Google Earth

Hbf-Süd

Karstadt

Postplatz

Lennéplatz

Mitte

Kongresszentrum

Fritz-Förster-Pl.

Strassburger Pl.

WTC

Hbf

ICC 2009 Dresden

April-16 2010 Dresden

Recent Uplink Field Trial Results Observed CoMP Gains

• Moderate average gains

Scheme Avg. gain

Conv. -

CoMP C = 2 19.0 %

CoMP C = 3 22.6 %

0 1 2 3 40

0.2

0.4

0.6

0.8

1

Rate [bpcu]

Cum

ula

tive D

ensity

UE1 (conv.)

UE2 (conv)

UE1 (CoMP C = 2)

UE2 (CoMP C = 2)

UE1 (CoMP C = 3)

UE2 (CoMP C = 3)

• Peak CoMP gains up to 150%

Slide 12 Patrick Marsch, Michael Grieger, Jörg Holfeld 4GSM-Meeting

Ines Riedel Slide 13

From EASY-C to Artist4G

Achievements Technology Evolution

Project

Meeting

14

MU-MIMO

CoSCH CoMP

EASY-C identified major challenges concerning

Synchronization requirements,

Multi channel estimation,

Feedback compression,

Backhaul requirements,

UE Complexity

and proposed efficient solutions

09.1

0.20

10

Key Learnings Key CoMP Challenges Identified

The EASY-C consortium has gained vast experience in the implemen-

tation and challenges connected to coordinated multi-point (CoMP):

System Partitioning Reducing Backhaul /

Infrastructure Aspects

Scheduling

Synchronization in

time / frequency channel estimation &

obtaining transmitter

side CSI at eNBs

Impact of network MIMO

on higher protocol layers


Network Architecture

Cooperative Multi-Point (CoMP):

Power Efficient or Waste of Power?


Does the increase in spectral efficiency make up

for the additional energy for signal processing?

Metrics

Data transported per unity energy

Bit/J

Data delivered per unit area per unit energy

Bit/J/km2

Given site setup: site distance

Bit/J with site distance as paramenter


Bit per Joule Efficiency of

Cooperating Base Stations

TU Dresden Albrecht Fehske Slide 19

Extended linear power

model

Consideration of effective

rates taking into account

additional pilots and

feedback

Backhauling according to

centralized processing

per cluster

A. Fehske, J. Malmodin, G. Biczok, and G. Fettweis,

„Bit per Joule Efficiency of Cooperating Base Stations in Cellular Networks“

3rd Workshop on Green Communications, December 2010, Miami Florida, to appear

0 500 1000 1500 200010

20

30

40

50

60

70

Inter site distance in m

Bit p

er

Jo

ule

Effic

ien

cy in

kb

it/J

N

c = 1

Nc = 2

Nc = 3

Nc = 4

Nc = 5

Nc = 7

Base line processing: 128W

MIMO processing: 10%

UL Channel est.: 10%

Clustersize: 1 to 7

Transmit power 100 mW…40W

Bit per Joule Efficiency of

Cooperating Base Stations

500 1000 1500 200010

20

30

40

50

60

70

Inter site distance in m

Bit p

er

Jo

ule

Effic

ien

cy in

kb

it/J

Nc = 1

Nc = 2

Nc = 3

Nc = 4

Nc = 5

Nc = 7

TU Dresden Albrecht Fehske Slide 20

Base line processing: 128W

MIMO processing: 1%

UL Channel est.: 10%

Clustersize: 1 to 7

Transmit power 100 mW…40W

Extended linear power

model

Consideration of effective

rates taking into account

additional pilots and

feedback

Backhauling according to

centralized processing

per cluster

A. Fehske, J. Malmodin, G. Biczok, and G. Fettweis,

„Bit per Joule Efficiency of Cooperating Base Stations in Cellular Networks“

3rd Workshop on Green Communications, December 2010, Miami Florida, to appear

Processing determines whether cooperation

increases or decreases Energy Efficiency!

Network Optimization:

Micro/Macro Setup


The link budget is the part to watch out for

in terms of Energy Efficiency!

Example (cont„d)

Problem:

Minimize (weighted) number of BSs s.t. coverage constraints

Macro BS

Micro BS

Ptx,macro = 46 dBm

Ptx,micro = 33 dBm

Rx-Sensitivity = -97 dBm

Empirical path loss model

(WINNER II)

Based on LTE link budget

Slide 22 Ines Riedel

Optimal Area Power Consumption

2010-05-18, Taipei, Taiwan Fred Richter Slide 23

0 2 4 6 8 10 12 14 16 400

600

800

1000

1200

1400

1600

1800

2000

Target 10%-ile area spectral efficiency (bit/s/Hz/km2)

Op

tim

al a

rea

po

we

r co

nsu

mption

(W

/km

2)

Homogeneous macro

Heterogeneous, 1 micro site

Heterogeneous, 2 micro sites



Almost linear increase in area power consumption

Each deployment (hom., het.) optimal for a certain spectral efficiency range

Heterogeneous deployment beneficial for higher spectral efficiencies

High load

Access Via Alternatives

Overlapping Coverage = 39.0%

Macro BS

Micro BS

Multiple coverage:

• 1: 60.4%

• 2: 16.1%

• 3: 16.5%

• 4: 6.1%

Slide 24 Ines Riedel

TU Dresden, Peter Rost and Gerhard Fettweis Slide 25

Motivation

Base

Station

Relay

Node

Relay

Node

Relay

Node

BS coverage area

RN

coverage

areasSource: WINNER II

Network Optimization:

Micro/Macro Setup


How to manage and optimize the radio access over so many alternatives?

SON: self optimizing networks

Mathematical framework (honey comb, reality, stochastic geometry)


Dirty RF


Dirty RF

Picking up “dirt” Nonlinear LNA

Feedthrough

Coupling

Phase noise

Aperture Jitter Ambiguity I/Q Imbalance RRC mismatch Flicker Noise Digital noise Nonlinear PA

DSP:

Living

With

Dirty

RF

easy


LOUL

LODL

AD

LNA

PA

Du

ple

xe

r

baseband

processing

Matched

Filter

Impulse

shaping

filter

Channel

select filter

AD

sBB[k]

sUL[k]

Power

Sens.

VGA

Limited Tx-Rx Isolation due to miniaturization and frequency agility

Tx Leakage (TxL)

Strong out-of-band interference in Rx branch

Transmit Leakage in FDD and Direct Conversion

Approaches for Solution:

Problem: Tx Leakage in FDD Terminals

Additional bandpass filter

Adaptive filter in parallel to duplexer

Dirty RF: Compensation of analog impairments by digital signal processing

+ Reconfigurable / standard independent / relaxed RF requirements

- Limited reconfigurability - Increased analog complexity

DL signal

-

+

AF

Tx Leakage

TxL

Estimation

-

+

sCMP[k]


Du

ple

xe

r AD

Power

Sens.

VGA

LOUL

LODL

Tx LeakageChannel

select

Filter

LNA

PA

sBB[k]

sUL[k]

AD

Tx Leakage in Zero-IF Receivers

Nonlinearity of I/Q down converter

Intermodulation product 2nd order of Tx Leakage (TxL-IM2)

Low freq. TxL-IM2 interfering with down converted DL signal of interest

Tx Leakage intermodulation products 2nd order (TxL-IM2)

DL signal

0

TxL

IM2Desired

signal

f

TxL

Desired

signal

fDLfUL

TxL-IM2

TxL channel

DC-


Antenna nearfield distorted by

metal interrupter moved within

distance [0.036, 0.303] m

to the antenna

Tx-Rx Isolation

Measurement with UMTS SAW Duplexer B7632 (EPCOS AG)

EPCOS

UMTS

Duplexer

B7632

Monopol -

Antenna

Metal

InterrupterNetwork -

Analyser

RX

TX

ANT

d

UL-Band DL-Band

B

Frequenz [GHz]

Tx-R

x I

sola

tionsdäm

pfu

ng [

dB

]

1.9 1.95 2 2.05 2.1 2.15 2.2

-70

-65

-60

-55

-50

mean

min / max

Frequency [GHz]

Tx-R

x isola

tio

n [d

B]

Tx-Rx isolation affected by nearfield

distortions

Only small variations in UL-Band

(ca. 2.2 dB)

TxL Estimation must be adaptive


SNR loss due to TxL

-5 0 5 10 15 20 25 30

0

2

4

6

8

10

12

14

16

18

20

~ ° TxL [ d B ]

¢ ~ ° w

. r . t w

/ o T

x L

[ d B

]

w/o TxL Cmp

Genie TxL Cmp

ML, t =3.5

LMS

t =3.5

meas.

t =1.5

dig. TxL Cmp required

TxL

negligible analog.

TxL

Cmp req.

OFDM susceptible to TxL for TxL < 30dB

digital TxL compensation

suitable for TxL > 0dB

TxL can be mitigated digitally up to 30 dB

Simulation Parameters:

DL / UL: 802.11a, 16QAM

DL channel: HiperLAN A

CSF EQ in TxL Est.

analog DC-1

ADC: 8-bit lin. mid-rise, =2

TxL: - TxL ch: 6-tap exp PDP

and meas. IR

- TxL SIR TxL = 10dB

- no TxL I/Q mismatch

TxL Est: - NB=4000,

ML reaches almost Genie Cmp. limit,

but suffers from high min. SNR loss

LMS performance strongly

depending on TxL channel property


Phase Noise and Clippping

-1.5 -1 -0.5 0 0.5 1 1.5-1.5

-1

-0.5

0

0.5

1

1.5

In

Qu

a

Phase Noise:

• Kalman Tracking

• Complete Sync

-1.5 -1 -0.5 0 0.5 1 1.5-1.5

-1

-0.5

0

0.5

1

1.5

In

Qu

a

Clipping:

• Analogue PAPR

• Saleh Model

• AM/PM distortion

-1.5 -1 -0.5 0 0.5 1 1.5-1.5

-1

-0.5

0

0.5

1

1.5

In

Qu

a

Clipping & PN:

• Rate

TU Dresden Wolfgang Rave Slide 34

Phase Noise in OFDM

TU Dresden Steffen Bittner Slide 35

BER Prediction

BER prediction

• Rayleigh fading with

exponential PDP

• 16-64-QAM

• Transmitter nonlinearities

and PN

• performance behavior

can be accurately

predicted

∆𝑓3𝑑𝐵 / 𝑓𝑠𝑢𝑏 = 1%

0 5 10 15 20 25 30 35 40

10-2

10-1

Average Subcarrier SNR in dB

BE

R

Simulation

Numerical

Reference

IBO = 0dB

16-QAM IBO = 3dB

64-QAM


Setup:

IEEE 802.11a SISO

Exponential PDP, 8 taps

64-QAM f = 4GHz, IBO = 0dB

∆𝑓3𝑑𝐵 / 𝑓𝑠𝑢𝑏 = 1%

Rate under Phase Noise & Clipping

-10 0 10 20 30 40 0

1

2

3

4

5

6

SNR [dB]

Rate

[bits]

no distortion

no compensation

I a

= 0.2, ICI 1

I a

= 0.4, ICI 2

I a

= 0.6, ICI 3

I a

= 0.8, ICI 4


Many Further Issues

Dirty RF: “Estimation and Detection” algorithm design

Nonlinearities

Phase noise

I/Q imbalance

Sampling jitter

Transmit leakage

Need understanding that analog chain is channel


M2M Sensors

WWRF Vision

“7 trillion wireless devices for 7 billion people by 2017”


Current Paradigm of Cellular


sensor

Slave 1000x Energy Problem ! Master

Required Paradigm of Cellular


sensor

slave master

Changes in GFDM

43 Ines Riedel

20 25 30 35 40 45 -50

-40

-30

-20

-10

0

10

PSD

RRC

RECT

ma

gn

itu

de

normalized frequency

RECT

(OFDM)

RRC

(GFDM)

time domain frequency domain

Possible matched filter pulse shapes:

GFDM System ...

10 June 2010 Rohit Datta 44

...

fading

channel

up-

conversion

tail biting

transmit filter

symbol

mapping

...

binary

data

cyclic prefixtail biting

receive filter

down-

conversiondetection

remove

cyclic prefixequalization

binary

data1

HCP -CP ..

.

...

ma

gn

itu

de

[d

B]

0 10 20 30 40 50 60 -80

-70

-60

-50

-40

-30

-20

-10

0

10

20


PSD

OFDM

primary

OFDM

secondary

0 10 20 30 40 50 60 -80

-70

-60

-50

-40

-30

-20

-10

0

10

20


ma

gn

itu

de

[d

B]

PSD

OFDM

primary

GFDM

secondary

GFDM system can adjust out of band interferences, with suitable pulse shapping.


Multi-Processors on Chip

Data Locality

1,0 Ni ikik xay

input output

Data Locality

memory

×

+

ia k ix

1,0 Ni ikik xay

Wireless Communications:

Processing Under Energy Constraint

Parallel Processing example: FIR filter

oddi iki

eveni ikik

xa

xay

,

,

Task parallelism within equation

4 memory reads per cycle

Parallel vector processing of task

2 memory reads per cycle

i ikik

i ikik

xay

xay

11

1z

Vector processing reduces memory I/O-bandwidth: Low-Power

1,0 Ni ikik xay

Data Locality

memory

×

+

×

+

memory

×

+

×

+

Data Locality

TU Dresden, 10/9/2010 Gerhard Fettweis Slide 50

Demonstrating Low Power & Die-Size

Source: T. Noll, RWTH Aachen

SAMIRA Core Measurement

8-SIMD, 32-bit float

130nm UMC

Tomahawk Cores 4-SIMD, 16-bit fixed-pt, 130nm

GP-CPU

Conventional DSP

FPGA

ASIC

Physically optimized

NXP OnDSP Measurement

8-SIMD, 16-bit fixed

90nm

Sandblaster Cores 4-SIMD, 16-bit fixed-pt, 90nm

full-custom

Data Locality


Heterogeneous MP-SoC: Task Scheduling

DSP

LMem

DSP

LMem

DSP

LMem

DSP

LMem

FLB

LMem

FLB

LMem

GPP

Shared

Memory

FLB

LMem

Task Task

Task Task

Task

Task

Task

Task

Task

Task

Task

Task

Task

Task Task

Task

Software

AD/DA

LMem

Scheduler

Router

Scheduler

Router

DSP

LMem

DSP

LMem

DSP

LMem

DSP

LMem

Scheduler

Router

GPP

GPP

GPP

Data Locality


Task:

Composed of Atomic Tasks

Data Flow Program

Contains control flow and “atomic” task calls

Task

Consumes and produces chunks of data

Contains a terminating program that operates on input data

Data locality exploited by operating on local copy of data

Resulting data committed to global memory after execution of task

Task A

Task B

Task C

Task D

Glo

bal m

em

ory

Task execution

Data transfer to core (Fetch)

Data transfer from core (put)

LMEM

TU Dresden, 10/9/2010 Slide 53

Software Scheduling

Design Time versus Runtime

Operating System

SoC

Program

Thread Thread

t1 t2

t3

t4 t5

t6

t1 t2

t3 t4

t6

t5

CP

PE1 PE2 PE3

PE4 PE5 PE6

PE7 PE8 PE9

Operating System

SoC

Program

Thread Thread

t1 t2

t3

t4 t5

t6

t1 t2

t3 t4

t6

t5

CP

PE1 PE2 PE3

PE4 PE5 PE6

PE7 PE8 PE9

Co

reM

an

ag

er

Data Locality


LTE FDD/ TDD/ WiMAX Single-Chip SDR:

Tomahawk - Die Photo

STA

LDPC Decoder/

Deblocker

ASIP

STA Vector DSP

2x STA SIOUX

Core

Manager

3x DDR Controller

10 mm

10 mm

2x Xtensa

DC212GP

Scratchpad

PLL

PLL

PLL

Silicon on

Jan 18 2008

In 45 nm CMOS:

Complete LTE BB

< 20mm2

< 200mW

130nm UMC

57M transistors


Heterogeneous MP-SoC: Task Scheduling

DSP

LMem

DSP

LMem

DSP

LMem

DSP

LMem

FLB

LMem

FLB

LMem

GPP

Shared

Memory

FLB

LMem

Hardware

Task Task

Task Task

Task

Task

Task

Task

Task

Task

Task

Task

Task

Task Task

Task

Software

AD/DA

LMem

Scheduler

Router

Scheduler

Router

DSP

LMem

DSP

LMem

DSP

LMem

DSP

LMem

Scheduler

Router

GPP

GPP

GPP

Data Locality

Embedded parallel computing:

how to architect hardware and software?

Silicon chip design cost:

how to architect the chips of the future as deep

sub-micron has driven design costs above the

$100M boundary?



Conclusions

Sensors

WAN for sensors

Energy autonomy

RF

Dirty RF

Flexible RF

Broadband

Sheer data rate

3D rendered

apps

Computing

HW/SW architect.

Templates beyond today‟s

Networks of hierarchy

Energy metrics & coverage

Self-X


No modulation buys us

orthogonality anymore !!!

Time to rethink for 5G !?!

A Piece of The Bigger Picture 3dim PHY / 3dim Traffic / 3dim design space


space

time

frequency

capacity

energy

fair coverage

today’s

traffic mix

m2m

rendered

multimedia


Thanks to UBC for inviting me as Green Visiting Professor !

Thanks to Vodafone for 16 years of continued support !

www.vodafone-chair.com

Ten Unsolved Hot Problems in Information and ...lampe/Fettweis/GF UBC Green Visiting Prof... · Ten...

Documents

Transcript of Ten Unsolved Hot Problems in Information and ...lampe/Fettweis/GF UBC Green Visiting Prof... · Ten...