DSP Challenges in Future DSP Challenges in Future Wireless SystemsWireless Systems

http://infopad.eecs.berkeley.edu/

Acknowledgements: Profs. Randy Katz and Paul Gray as well as many UCB students

R.W. BrodersenDept. of EECS

University of CaliforniaBerkeley, CA

OutlineOutline

Past, Present and Future Wireless Applications

VLSI Technology Trends Wireless System Design

Wireless Technologies Wireless Technologies

1890’s - First radio transmission 1940’s - Pre-cellular - Analog (MTS & IMTS) 1970’s - First Generation cellular - Analog

Voice, Digital control (AMPS) 1985-1990’s Second Generation - Digital

Voice and Control (GSM, N-AMPS, CDMA) 2000 - Third Generation - Multimedia Data

(UMTS, Wideband CDMA)

Third Generation Telecommunication Third Generation Telecommunication ArchitecturesArchitectures

High-tier

Low-tier

Satellite

High Mobility Low MobilityWide Area

Regional Area

Local Area

UMTS» Universal multimedia information access with mobility spanning residences,

businesses, public/pedestrian, mobile/vehicular, and global regions» Data rates to 2 Mbits/sec per user indoor, 384kbit/sec outdoor - standards

being developed now

Cellular Subscriber Growth in Cellular Subscriber Growth in USUS

85 87 89 91 93 950

10000

20000

30000

40000

50000

44 million

Source: CTIA Web Page

Cellular telephonyremains the hottest andfastest growing sector of the telecomm market

Nordic Countries: 10 mobilephones being added for

each wireline phone!

Voice vs. Data?Voice vs. Data?

Previous telecommunication systems have been optimized for voice

however More than 50% of telecomm traffic in

Bay Area is now data, not voice» Internet FIND/SVP Survey

– 25% of Internet users make fewer long distance calls

– 32% watch less television

Exponential growth of Internet Exponential growth of Internet traffictraffic

(Source: GILDER Technology Report, Nov. 1996)

Tera

by t

e s P

er M

on

th400

300

200

100

0

1992

1993

1994

1995

1996

• Growing at the rate off 20% per month• Earlier this year increased 2x in 100 days!

Wireless Internet AccessWireless Internet Access

Voice communications » Has been the only driver for personal wireless

access » Will evolve to be just one of many services.

The ideal access device would allow multimedia internet access from any location (like a cellphone does for voice)

The access device should have the mobility and battery life of a PDA, with the multimedia capability of a PC

The progression towards a Wireless The progression towards a Wireless Personal Internet Access DevicePersonal Internet Access Device

Improving support for data But so far

» Low bandwidth - optimized for audio and outdoor mobile links

» Low user capacity» Wrong form factor and poor multimedia support

?

What is Needed (the home What is Needed (the home model)model)

Set-top box doubles as basestation Set-top box doubles as basestation and gateway from WANand gateway from WAN

Allows family Allows family andand personal use personal use

of set-top box accessof set-top box access

Internal Architecture Internal Architecture

DataFlow

Low Power Bus

RadioModem

Embedded Processor

AudioCodec

VideoDecomp

VideoAudio

Decomp Fifo

Graphics

Pen

Sched ECC Pact Interface

SRAM

DSPBlocks

OutlineOutline

Past, Present and Future Wireless Applications

VLSI Technology Trends Wireless System Design

Wireless Technology HistoryWireless Technology History

Precellular - Vacuum tubes, Crystals, cat whiskers, horse hair

First generation - Bipolar and GaAs Second generation - Analog bipolar radios

with digital CMOS control Third generation - Analog and Digital CMOS

for the baseband and bipolar for the RF Future - All CMOS

1GHz

75 77 79 81 83 85 87 89 91 93

3u

2u

1.5u

1u 0.8u

0.6u

GaAs

Bipolar

CMOS

ft

Year95 97 98

0.5u0.35u

3GHz

Hemts,HBTs

Why it will be all CMOS?Why it will be all CMOS?

10GHz

30GHz 0.25u

100GHz

Characteristics of State-of-the-art Characteristics of State-of-the-art CMOSCMOS

Dimensions» .25 Micron gate length (significantly shorter than

the wavelength of visible light)» oxide thickness of 25 atomic layers» 10’s to 100’s of millions of transistors

Delay through a logic gate» ~20 picoseconds without loading (time it takes for

light to go less than a cm)» Near gigahertz clocks

The problem is not area or performance but is energy consumption for wireless systems

Reduce number of operations» Shutdown modules that aren’t being used» Perform processing in network servers - an

I/O device not a computer Reduce energy/operation (CV2)

» Advanced fabrication technology» Reduce C through logic and circuit style and

transistor sizing

How to Minimize Energy How to Minimize Energy ConsumptionConsumption

More on Energy ReductionMore on Energy Reduction Clock rate reduction doesn’t help

» Number of operations = Nops

» Energy/operation = CV2

» Total Energy = CV2 * Nops

» Energy consumption is independent of clock rate!

Reducing the clock rate only degrades speed, but no savings in battery life - unless the voltage is changed

Architectural optimization

Dependence of Energy Consumption Dependence of Energy Consumption on Architectureon Architecture

Conventional General Purpose processors » Clock rate is everything … somehow we’ll get the

power into it and back out..» 10-100 Watts, 100-1000 Mips = 100 mW/Mip

Energy Optimized but General Purpose» Keep the generality, but reduce the energy as

much as possible - e.g. StrongArm» .5 Watts, 160 Mips = 3 mw/Mip

Energy Optimized and Dedicated» < .01 mW/Mop

Energy improvements due to Energy improvements due to technologytechnology

10X each 2.5 years

(Ref. Walt Davis. Motorola)

mW

MIP

Dedicated Arch.

< .01mW/MIP

Energy Reduction for Signal Energy Reduction for Signal Processing ComputationProcessing Computation

Requires a fixed number of operations per sample period (real-time constraint)

Reduce the energy/operation (CV2) by reducing the voltage, but the delays increase

Processing speed can be retained by parallelism

2

init

final

initV

VE

initEfclk

In

Out

fclk fclk fclk

Out

In

3 3 3

CDMA Digital Baseband CDMA Digital Baseband ArchitectureArchitecture

Phase Control

Com

para

tor

CorrelatorDelay

WalshDecode

Correlator

ClockMux

LoopGain

P/N

De

scra

mbl

er

256 MHzClk

Timing Recovery

Data Recovery

RAKE Combiner

Correlator(x3)

Ana

log

RF

Se

ctio

n

Dat

aM

ux

Digital Clocks

Correlator(x3)

Channel Estimator

CorrelatorAdjacent Cell Scan

(Bits Out)

(Delay Locked Loop)

128 MHz 64 MHz

1 MHz

Power = 27 mW @1.5 V. (.8 process)

Potential to do much more complex algorithms

(S. Sheng et. al., 1996 ISSCC)

OutlineOutline

Past, Present and Future Wireless Applications

VLSI Technology Trends Wireless System Design

The GoalThe Goal

Conventional cellular phone solution

Single-chip multi-modal smart radio using scaled CMOS technology

DECT receiver, ISSCC 1997

DECT baseband

GSM baseband

CDMA Wideband

CMOS density now allows complete CMOS density now allows complete System-on-a-chip SolutionsSystem-on-a-chip Solutions

ViterbiEqual.

Demodandsync

phone

bookkeypad

intfc

protocolcontrol

de-intl&

decoder

RPE-LTPspeechdecoder

speechquality

enhancement

voicerecognition

phonebookDMA

S/P

DSP core

P core

RAM & ROM

Dedicated logic

A

D

digitaldownconv

Analog

P core» protocol» user interface

Dedicated logic» DSP

acceleration» bit level

manipulation Programmable

DSP core Analog

» A/D» RF, baseband

The Challenge of Exploiting The Challenge of Exploiting Technology ScalingTechnology Scaling

Effectively have arbitrary amounts of DSP available - how to use it?

Even though digital processing improves with technology scaling - the analog processing becomes more difficult.

The design of a system-on-a-chip involves a complete integration with software, hardware, analog and digital components

Lets see what this really means

Conventional Radio Conventional Radio ArchitectureArchitecture

90

DAC

DAC

VCOVCO

I

Q

I

Q

ADC

ADC

Transmitter

VCO

RF LNA/Mixer IF AGC

IF Mixer

Modulator

90

Receiver60 mW65 mW

40 mW

50 mW

45 mW

250 mW per ADC

165 mW40 mW

20 mW per DAC

The Future?The Future?

Direct conversion - mix to baseband without intermediate frequency filtering - minimize the analog

More appropriate for wideband (spread spectrum) systems

LNA and AGC

Digital Signal

Processing

A/D

X A/D

X

frf

90

Direct conversion problems...Direct conversion problems...

The rf oscillator is picked up by the antenna and reappears as a DC signal in the baseband processing

Can we use baseband DSP to reject this DC signal?

LNA and AGC

Digital Signal

Processing

A/D

X A/D

X

frf

90

Direct Sequence Spread Spectrum Direct Sequence Spread Spectrum modulation (e.g. UMTS)modulation (e.g. UMTS)

Xtbit

tchip

Spreading code

Data Input Spread output data

Transmit signal is expanded by tbit/tchip

Receiver (in gaussian noise) is a simple correlator - multiply with code and sum the products

+1

-1 -1

Design of the DS Spread Spectrum Design of the DS Spread Spectrum ReceiverReceiver

Design requires joint optimization of» Communication algorithms

» Analog circuit performance

» DSP architecture (e.g. dedicated vs. programmable)

Reducing A/D requirements are critical for implementation at low power levels

But the real problem is the user interference - can we do something about that???

10 30 50

Number of Users

1e-06

.0001

.01

1

Bit

Err

or R

ate

2 bit

3 bit

Unquantized

4 bit

Communications theory to the Communications theory to the rescue - Multiuser Detectionrescue - Multiuser Detection

0 5 10 15

Number of Active Users

10-6

1B

it E

rror

Rat

e (B

ER

)

Single User Correlator

MMSE Multiuser Detector

Multiple order of magnitude improvements in BER

.01

.0001

10-8

10-10

Multiuser Detection Multiuser Detection ImplementationImplementation

Baseband Signal

Despreading

X

Adaptive

ErrorSignal

Optimal MUD algorithm is exponential in the number of users, but …..

Adaptive algorithms using common error metrics and standard algorithms (LMS, RLS) approach optimal results

Dedicated accelerators make low power implementation possible (estim. 25 mW, 10 mm2)

Receiver adaption removes effect Receiver adaption removes effect of interference from other usersof interference from other users

0 100 200 300 400 500Symbol Periods

0

5

10

15

Sig

nal t

o In

terf

eren

ce R

atio

(S

IR)

12 users; equal transmit powers; (Eb/N0)=15 dB

Can adaption be used to reduce or compensate for performance of analog circuits and A/D ???

What happens to the A/D What happens to the A/D

requirements?requirements?

Higher resolution A/D converters are required to capturethe gains made possible by multiuser detection

Implies increased energy consumption

A/D converter technology ultimately limits performance

3 4

5 6 78 Unquantized

Number of bits in A/D

-10

0

20

40

10

30

100 200 300 400 5000 600

Symbol Periods

Ave

rag

e S

IR (

dB

)

To summarize - multiple disciplines To summarize - multiple disciplines are required for wireless systemsare required for wireless systems

Analog and RFCircuit Design Protocols and

Standards

Communications and DSP Theory Low Energy

Architectures

Multimedia Applications

ConclusionsConclusions

The capabilities of CMOS technology have now reached a threshold which allow completely new types of wireless systems

The future levels of CMOS integration will force an integration of multiple disciplines

Multiple areas of SPS are working on various aspects of wireless - we should work together!

Wireless system design is an exciting new opportunity