Re-Configurable Multimode, Multiband Information Transfer Systems

8/18/2019 Re-Configurable Multimode, Multiband Information Transfer Systems

1/47

Re-Configurable

Multimode, Multiband

Information

Transfer Systems

(Low

Power,

Smal l Size,

Software Defined Radio,

Secure

Re-Configurable

Mobile

Communicat ions

Computing)

M . J .

Flynn,

M .

M o r f

Computer

Systems

Laboratory

Stanford

University

Stanford, CA 94305

Final

Report,

May 1999

Supported

by

th e

DARPA

ACS

Adaptive Computing

Systems

Program

and th e DARPA

GloMo

Global

Mobile

Information Systems

Program

A R M Y Contract

DABT63-96-C-0106-P00003

Contractor

identification:

00028088

DnC QU A LI TY INSP EC TED

2

2 0 0 0 0 5 2 2 0 3 7


2/47

Principal Investigator:

Michae l J .

Flynn

Address:

E.E.

Dept. , Gates

Computer Science

Building,

Room

334,

Stanford,

CA

94305 ,

e-mail: [email protected]

Phone: (650)

723-1450,

Fax:

(650)

725-6949

Investigator:

artin

M orf

Address: E.E. Dept. ,

Gates

Computer Science

Building,

Room

335,

Stanford,

CA

94305 ,

e-mail:

[email protected], Phone:

(650)

723-0140,

Fax:

(650) 725-6949

Subcontractor:

ark

Cummings,

enVia,

Address: 348

Camino al

Lago

Atherton,

CA.

94027,

e-mail : cummings@envia .com,

Phone/Fax:

(650)

54-4406

Fina l Report

Tab le of

Conten t

0.

roject Summar y

1 .

roblem

Statement

2.

roposed Work

3.

roject Organization

4 .

esearch Resul ts

5. esult ing

Publ icat ions

6.

emo s

7. eliverables

8.

echnology

Transfer

9. ppendices

A ,

B , C.


3/47

Project S u m m a r y

O ur

research

was

a imed

at

tw o

Darp a programs,

l gor i thms nd econfigurable rchitectures

as

art

f he

dap t ive

ompu t ing

ys t ems

A C S ) rogram, nd ppl ica t ions o ireless

networks ,

ncluding od ing nd

ecurity

s

part

f

th e

Globa l

M o b i l e

nformation

Systems

(GloMo)

rog ram .

.

Problem

Statement:

W e

focused

on

tw o

research

problem

doma ins : l gor i thms nd econfigurable rchitectures

as

part

of

Darpa 's

Adap t ive

ompu t ing

ys t ems

A C S )

rog ram , nd

ppl ica t ions o wireless

networks, ncluding od ing nd

ecurity

s

art

f

arpa 's

loba l obi l e

nformation

Sys t ems

GloMo) rog ram .

ireless

e tworks

re deal est omain s

or

econfigurable

architectures.

Tomorrow's

wireless

networks

can

be

charac ter ized

by n

ncreas ing

number

f

t ransmit ters

and receivers

and

l a rger and

larger

data

rates,

and

a

growing

profusion

of

m o d e s

f operation

waveforms . he mobile ommunications

se r

s

confronted ith a myr iad

f

ncompatible

c ommu n ic a t ions

serv ices .

ach has it s wn

technical ,

eographic , nd eature e t.

mobile

defense profess ional

moving through

a

miss ion moves n

and out

f

different overage

reas;

henc e heir commun ica t ion

eeds

change onstan t ly .

Users

need

a

ingle

communication

device

ha t c ommu n ic a t e s

with

nyone ,

nywhere ,

with

any

va i lable nfrastructure.

herefore ,

n e

f

th e oals f this

roject was o esearch

he

necessary

cos t

effect ive

t echnologies

that enab le such

devices .

Project

oals:

T he

roposed goal

of the

esearch as o

nderstand

n d

erify he ull x t en t f

he

opportunit ies

to

sat isfy ecure efense mob i l e ommunications

equirements or

mult i mode

mult i and

and

e ld

nits

with

econfigurable

omputing

rchitectures.

i lestones

were

ba sed

on th e

ana lys is ,

imu la t ion ,

n d

mplemen t a t ion n

reconfigurable

ardware

es t bed

of new archi tec tures nd l gor i thms urrounding he ddi t ion

f widely

eployed

mobile

infrastructure.

Research

Summary :

Stanford's

reconf igurab le

architecture

research in C S

was

ba sed

n

ts

P a m B l o x

pproach

to

P G A

rogramming,

n

bjec t

riented

es ign methodology

hich

as

hown

om e

s ign if ican t

dvantages .

arious

rithmetic nd

ecurity

l gor i thms ere

mplemented

n

eff ic ien t

l ow-power

F P G A

des igns ,

inc lud ing D ES

and

IDEA

encrypt ion

s t andards .

Stanford

ls o

nvest iga ted ew pace- t ime

od e

iversi ty

ased l gor i thms

or

ireless

communication

n

he

on t ex t

f

oftware

efined

adios,

it h

he

oa l

f

roviding

r

extending

imul taneous

ecurity

pr ivacy ,

ntegrity,

eliabil i ty,

vailabil i ty, nd rotection

from

everse ngineering)

y

xploiting

aral lel ism nd

ca lable

ierarch ica l

multiplexing

s c hemes to

bu i ld

software

def ined

adio rch i tec tures .

tanford eve loped ew

Space-Time-

C o d e

Divers i ty

( S T C D )

based

ecurity

architectures

ha t

uppor t

ll

these

equirements

nd

m ap

well

onto

software

def ined

radio t echnologies .

Stanford

emons t r a t ed ,

s

roposed,

ul t i-path

pproach

o

ecure

ireless

a ta

t ransmiss ion n

9 9 8

s ing

hree aptop

ompu t e r s , ased

n wired,

m o d e m ,

R r Ethernet

l inks , and commerc ia l

wireless

Ricoche t or

Freewave

l inks.

Morphics'

goa l was to how

that

mul t ip le

ig i ta l

t andards

an e mplemen t ed n

ing le

ba seband

architecture

without

mult ip le ,

paral lel process ing

paths.

he

func t ional

target was

multi-standard

ardware ogic ect ion ~ he ew, reviously

ndeveloped

a rt f mul t i -


4/47

Standard

commun ica t ion s

a seband ,

o

nable

multi-standard

digital

modem. T he ervices

considered dur ing

th e projec t

focused on

T D M A

based

services.

Morp h ic s

resen ted

wo

demos :

G S M

F P G A

baseband ect ion,

nd

oftware

t ructural

des ign

fo r a second service.

he

G S M

s t andards

were

mp lemen t ed

nd he S-54 /136

was

in

progress.

tes t bed was as semb l ed based on Lucen t ' s G S M test platform (Sceptre) .

Morp h ic s

em o

howed

heir

ompl e t ed

es ign

nd

emonst ra ted

t

n

heir

ardware

Pla t form,

S M ogether it h

he

na lys is

f

S-136

North mer ican

D M A ) ,

nd S-95

( C D M A ) ,

ndica te

ha t

mul t i -mode

ommunication

evice s

poss ib le

s ing

econfigurable

archi tec ture

in

comb ina t ion

with

in s t ruc t ion-set

rocessor(s)

or aseband

ig i ta l

rocessing.

T he

Morph ic s

evelopment

t

t ime

f

conclus ion f

he ont rac t

uccessfu l ly

hows

ha t

complete

range

f

T D M A

t andards

an e

m p l e m e n t e d . he ex t t age would

be

o

dd

th e C D M A

capabil i t ies .

enVia's

work focused on th e R F

frontend

of a mul t ip le service rototype ys tem upporting

G S M

and

D M A

& M P S

pera t ing

t

cellu lar ,

00-900

Hz, and

C S ,

800 -1900 Hz

f requenc ies .

he

urren t mplemen t a t ion

nvo lved

S I

cale

iscrete

na log

omponents.

V L S I ingle

h ip

olut ions

re

eginning

o

xhibit cceptab le

erformance.

emaining

discrete

omponen t s re A W

ilters,

ielectr ic

esonator ilters,

rystals .

igi tal /RF

interfaces

e tween

he

inal e m o d

tage n

he

F

rontend

nd

he

ig h

peed

reconfigurable

S P

backend

nc lude :

na log ,

a seband ,

nd

& Q .

ontrol

igna ls

n to

h e

R F

frontend a re ana log .

This

range

of

services

requires three

differen t

interface bandwidths .

enVia

demonstrated an R F

frontend

suppor t ing

4

modes ,

2 bands , and 3

bandwid t h s ,

M P S ,

IS -136 ,

S 5 4 A ,

S 5 4 B ,

S 5 4 C ,

S - 9 5 ,

C S ,

C S 1 9 0 0 ,

S 1 3 6 + ,

n d

S - 9 5 +

n F es t

boards : Receiver , Transmit ter ,

#1 O, IF ,

C o m b i n e r Switchboard

and Auxi l ia ry L O.

enVia 's

R F

frontend

des ign

involved

a m ix

of

discreet

componen t s

to

m e e t

th e

roject

oa ls

fo r

performance ,

power eff ic iency , size, and cost . or

IS -136 ,

G S M ,

and IS -95 ,

hree

F

A W

filters

were

used

with

f requenc ies

in th e 10 & 211

M Hz

bands , fo r p ac kag e size

and

cost .


5/47

1. roblem Statement

W e

focused

on

tw o

research

problem

doma ins : lgorithms and

econfigurable architectures

as

part

of

Darpa 's

A dap t ive Compu t ing

S y s t e m s

( A C S )

program, and pplications o wireless

networks,

nc l ud ing oding nd ecurity

s

art

f arpa's

loba l obi le

nformation

Systems GloMo)

rogram.

ireless

e tworks

re dea l es t omain s or econfigurable

architectures.

Wireless

Network Appl ica t ions

Tomorrow's

wireless

networks

can be charac ter ized y

n

ncreas ing number f t ransmit ters

and

receivers

and

l a rger and

l a rger data

rates,

and

a

growing profus ion

of m o d e s f

operation

wave forms ( M o d / D e m o d ,

R F

IF

&

Baseband

bandwidths , coding t echn iques ,

security, etc .)

T he

obile

ommunications

ser s onfronted it h yr iad f ncompatible

commun ica t ion s serv ices . ach

has ts

ow n

t echn ica l frequency, modu l a t ion ,

rotocol, tc . ) ,

geograph ic

( coverage ,

etc.),

and

feature

set

(voice,

aging ,

ata , etc .) .

s

th e

mobi le efense

profess ional

m o v e s

through

a

miss ion ,

hey

m o v e n

nd

u t

f

different

overage reas

nd

their

commun ica t ion eeds hange onstan t ly .

In

th e

U S

today there are a t

leas t

four major cel lu lar s tandards ( A M P S , IS -54 , S-95 ,

D P D ) ,

and

s ix

P C S

s t andards .

ne

way

paging

is

being

expanded

with

tw o

way

pag ing .

here

re

growing

n u m b e r of

wireless

packe t data

serv ices .

n urope, here re

our

imi la r

t andards

( G S M , CT-2 ,

D E C T ,

and D C S 1 80 0 )

and

more

to

o m e .

n

apan,

here re

ersions

f

t hese

plus

Hand ip hone .

he

n u m b e r of

serv ices

s

clearly

ncreas ing . here

re t

leas t

four

ow

Earth

rbit ing

atellite ys t ems

L E O S )

nder

evelopment.

ireless

A N

roduct s

re

becoming

ore va i lable .

ireless

P BX' s have

rr ived.

h is am e

rocess

s

oing

n

n

Europe ,

As ia

and

ther

arts

f

th e

world.

or

geopolitical

eason s

here

re ikely

o e

t

least

hree

on-compatible

uper ets

f

ervices. ur ing h is eriod f nnova t ion ,

he

number

of incompat ib le

serv ices

is

l ikely

to

inc rease .

It

is

confusing ,

x p ens ive ,

nconven ien t

nd

angerous

o ea l with

this

jungle f

ervices,

with

ingle

evice or

ach

ervice.

Current

echnology

olu t ions

nvolve

xpens ive ,

ot

miss ion

dequate

ys t ems

hich

an

roduce

ogist ical

i gh tmares

uch s

equiring

per sonne l o

arry more

weigh t

n

at teries

han

n ood ~ n

ncent ive

or

oot

oldiers

o

l eave

quipment

eh ind .

e

ote

ha t

ur ing

eser t

torm

t

eas t

iv e

enera t ions

f

wireless

commun ica t ion

sy s t ems were

s imul taneously

ac t ive ,

and expec t ed to

cooperate

W h a t users want

is

a

single device

t ha t

will

al low

them o

ommunicate

with

w h o m e v e r they

need

to , wherever

they

re ,

with

what ever

infrastructure

appen s o

e

va i lable .

ne f

th e

major

oals

f h is

roject

as o

esearch

os t

ffect ive

echnologies

hat

nab le

uch

devices .

Cur ren t D S P A S I C

so lu t ions

suffer

from

problems with size an d ower consumption.

os t

sy s t ems

emp loy

an A S I C

chip

set fo r each commun ica t ion s

serv ice .

hese

h ip

e ts

typically

inc lude

umber

f

SICs

er

ervice.

he

oa l

s learly o es ign

econfigurable

computing

rch i tec tures

ha t

ave

he ecessary

lexibil i ty

o

dap t

o

ifferent

communication nv i ronmen t s .

uch rchitectures ave o e b le

o rovide

he ecessary

c omp u t e

power , while

sat isfying const ra in t s

such as

cos t

and

power .

P o w e r con sumpt ion rob l ems re he

esu l t

of

lock

peed

o

ch ieve

esired andwid th s

and requenc ies .

he rchitectural i l e m m a ha t confronts

es igners oday

s

ha t

rocessor

speeds

are

inc reas ing s t ead i ly ,

b u t

m e m o r y speeds

and

access

t imes

are

agg ing . he esu l t

s

that

a s imple

inc rease

in processor speed

even

of a

factor

of tenfold or more m ay

n ot

esult

n

a

ign if ican t

ncrease

n

ys tem erformance.

he

olut ion o h is

undamental

roblem

determines

tomorrow's

designs .


6/47

T he

se f aches : r imary ,

ec onda ry ,

t reaming ,

tc .

s

he r imary ay hat

ndustry

addres sed such

p rob lems today .

uch caches

re

o m p o s e d

f

expen s ive

fast" memory nd

located

lose

o he rocessor .

u t

he bility f

hese

pproaches

o ffer

ontinued

performance

mp rovemen t s

s

severe ly

imi t ed .

W e ave

een

ooking

or

ther

eans

f reak ing hrough

his onstant ly

arrowing

bot t leneck .

This

was

th e focus

of

m u c h of

th e work

in architecture a t

Stanford .

Alterna t ive

o onven t ional

rocessor

pproaches

s he

reat ion f

igh ly

calar ,

eeply

pipel ined

mac h ine s

Super-Scalars)

nd

Very

ong

nstruct ion

Word

(VLIW) mac h ine s .

t

Stanford nd l sewhere both f these pproaches

re

ursued , ecen t ndus t r ia l mac h ine s y

Intel

(Pen t iu m)

and

T I

( T M S )

a re

of

th e

VLIW

t ype .

he

eal

ssues involve

th e

uest ion

f

how o m ix n d atch

hese

rch i tec tura l pproaches o pecif ic ets f

pplicat ions,

.g .

wireless

networks .

ntel l ready

nnounced

new

effort

o

evelop

rocessor

rchitectures

targeted

a t

(non-wireless)

networks .

T he

current

S P

processor

pproach aces narrowing m e m o r y ot t l eneck ,

esult ing

n

h igh

ower

onsumpt ion , imi t ing

he

ang e

f ppl ica t ions o

nes

it h

elat ively

ow

frequency igna ls .

mprovement

n ower consumption nd

rocessor

peed

m ay

p p ea r

s

par t of th e

orma l

volut ion

f

ch ip

echnology.

owever ,

sys tem

performance

nd

ystem

power con sumpt ion will

still be

l imi t ed

by th e memory bot t l eneck .

In

deep ly pipel ined

mach ines , process ing

elements

a re ar ranged

in

uch fashion ha t

er ial

process ing ccurs n

erially

rranged

rocess ing

omponen t s .

he

utput

f

ach

component

s irect ly

onnected

o

he

np u t

f

he

ex t

omponent.

hu s

here

s

o

m e m o r y a tency when

moving from on e process ing

te p to

nother .

ys t ems

d o ot av e o

ru n t

c lock ates

that

re

ig h

mul t ip les f

th e a ta

rate,

because

ach

l emen t n ly as

o

process data

at

th e

data

ate .

ach amp le

s

processed

ssent ial ly

n

ne amp le

ime .

h is

reduces ower consumption

nd aises

th e

eil ing

n

he

requency

f

ignals

ha t

an

e

processed.

Secur i ty

T he verr id ing

ifference e tween

O D

nd

on-DO D equirements

ie s

n he

ecurity

area.

n

n o n - D O D

areas

cost -benef i t

dr ives

he nves t men t way

from

esources

or ecurity.

In

OD

ppl ica t ions

se r

nd

iss ion

equirements

ictate

he

i n ima l

evel

f

ecurity .

Idea l

archi tec tures fo r

such

appl ica t ions require

a proper m ix

of s tandard

and

e-configurable

subsys tems , omponen t s ,

n d

uppor t ing oftware.

n

e-configurable

rchitectures,

nce

unneces sa ry conf igura t ion

data

is e m o v e d t

is

imposs ib le

o

etermine

what th e

ystem wa s

doing .

herefore

t

is very ifficult

o

everse engineer econfigurable ystem

where he

hardware

is

not

security

algorithm

specif ic .

Su c h ppl ica t ions

equire ew

pproaches

o ata

ecurity nd

rotec t ion . hese

approaches

re

part ly

ecess i t a ted

by ew t echnologies

ha t

uppor t wireless

networks .

a ta

security

nd protect ion

methods

common ly

used n oday 's wireless

networks

re

ikely

o

e

inadequate

n uture ireless

etworks.

or

ns tance , he

ncreas ing umber f ifferent

services nd

igher

a ta

ates

ak e

eliable

a ta

ransfer

rotocols

ha t

ely n

re t ransmiss ions

e c o m e

nusable

when

th e

m o u n t f a ta

ha t

m u s t

e

uffered

ecomes

excess ive . herefore

a ta protection

n

a ta

ompress ion

n wireless

e tworks

will

epend

largely

on

orward

rror

orrection

nd

oding

met hods .

hese

echniques

mu s t

dap t

o

variety

of

data loss

cenarios, anging

ro m

elat ively

ow

random

it

error

ates

to otal

oss

of large port ions of

a

t ransmiss ion . he differen t services nd

igher

a ta

rates

const ra in h e

choice

o f error

protect ion

met hods

to

those

of

sma l l

circuit

complex i ty ,

an d

lo w

power .

T he

different

services

n d ariable

a ta

rates

ls o

equire

ew

app roaches

o

a ta

ecurity .

Encrypted

a ta mu s t

e

rotected

ga ins t

errors

s ing

higher

aye r

f error orrect ion r

error

etect ion nd

ransmiss ion . a ta ecurity c hemes ha t

an

e nified ith

a ta

protect ion are appropr ia te in this

ew environment .


7/47

Beyond

ata in k ecurity nd

eliabil i ty,

ommun ica t ing arge ata messages n ireless

networks

ba sed sy s t ems is a

key

performance,

func t iona l ,

and

cos t issue .

Add ing in te l l igence

to

wireless

commun ica t ion

sys t ems

is

required.

f

one

can

e-configure

th e mobile

nits

f wireless

a ta

ommunication

ys tem, ne

an

rovide ignificant ly

improved

unctionality.

n e an uppor t major nhancemen t s y

v

mak ing

uch ys tems

dynamical ly

e-conf igurab le ,

while

keep ing

th e

power

requirements

ow :

•

nhance ecurity f data

transfer

•

nsure

integrity

of

data

t ransfers

•

arry u t

compress ion

an d

expan s ion of

da ta ,

and

mul t i -med ia .

•

p ac e -T ime

C o d e

Divers i ty ,

espec ia l ly

over

mult ip le

services.

This

processing

an e

one

imul taneously

with

th e

asic unct ions equired

or

ireless

network

services,

as

well as with

h igher - leve l

data

operat ions

(e .g. mul t i -media ,

a ta us ion)

t

th e

source,

poss ib le

relay,

and dest ina t ion .

Future

ery-h igh andwidth ireless etworks il l mos t ikely

nvolve

ew echnologies ,

such s aAs ,

hoton ic ,

uantum

ev ices , ow-power.

arly

nvo l vemen t n

hese

t echnologies will

enable

compat ib i l i ty ,

early

insert ion,

nd

ther

ynerg ies

ha t

re nvaluab le

in

their

ow n

r ight ,

as ide

from

be ing

"cos t

shared" .

Algorithms

for

Coding and

Security

T he haracterist ics f ireless

ommunication

e tworks

resen t

oth

hal l enges

nd

opportunities

n he

re a

f

data-reliability

nd

ecurity.

ha l lenges

rise

ro m he ighly

var iable

data

rates

and

la tencies . imp le

rotocols

uch

s A R Q

(A u toma t i c ep ea t

reQuest )

require

arge

uffers o tore eal - t ime ata

or

oss ib le et ransmiss ion . ppor tun i t ies

re

prov ided y he bility o

erform

ynamical ly

econfigurable

unct ions

nd

he

communications

edundancy

uppl ied

y

ariable

ata

ates

nd

ul t iple

ndependent

t ransmiss ion

paths.

T he

pr imary

reliabil i ty

an d

security

need s to

be

addres sed

are:

•

rror

detec t ion

at high

data

rates

•

orward

error

correct ion

o

reduce

re t ransmiss ion

requirements

•

a ta

security to dea l with misrout ing and in tercept ion

T he inves t iga t ion

of

integrated

approaches

that

solve

all

three

a ta

protection

ssues

us ing

unified

c od ing

method

s

equired .

In pr inc ip le ,

el iable

nd ecure

ommunication

f

data

can

e

ccomplished t

ow

a ta

rates

by

t andard

error

protec t ion od ing nd cryptographic

echn iques

ppl ied

o

he D

bit

s t reams . n

ract ice,

we

nticipate

ha t

more

ffect ive

met hods

il l

ake

n to

ccoun t

he

var iable

nature

of

th e

wireless t ransmiss ion channe l and th e

digital

data .

Da t a

security nd eliabil i ty

an e

nhanced

s ing

ul t iple

ndependent

ata

aths.

f

data

is

t ransmit ted over a n u m b e r of differen t

routes

so that th e n t ended ece ive r

has larger

probability

f eceiving ach r an smis s ion

han

n avesdropper ,

hen

ecret -shar ing

nd

other

space- t ime

c ode

divers i ty

t echn iques can

be

emp loyed

to

obtain

rel iabil i ty nd

security.

T he

imp les t ecret -shar ing

ethod s he ol lowing .

he

nformation

equence

s

t ransmit ted s wo

ndependent

eq uences

nd , here

onsists

f andom r

sufficiently

ompl ex

seudo-random it s

generated y

he

ransmit ter n d

= X

Z s th e

bitwise

exc lus ive-or of th e data

with

this

noise

equence .

he

ece ive r can

asily

econstruct

th e

t ransmi t t ed

data from th e tw o received s t reams, s ince

X

=

(X Z) Z Z.

Because

th e

random

sequence s independent

f

th e a ta

,

th e

equences

nd

separately

provide

no

informat ion bou t

th e

da ta ,

o

n

dversary hat

ntercepts

n ly

ne

f


8/47

th e

equences earns

oth ing

bou t

he

ata .

Note

ha t

he

ece iver

an

erform he

reconstruct ion

in

real

t ime

if nd re

sent

s imul t aneous ly

and

if th e

ece iver

can

djus t

sma l l

relat ive

delays

in

th e

incoming

data .)

This

s imple

ex amp le

reaks he ecret

n to wo pieces, nd both f

which

re

needed

o

econstruct he

ecret,

and ither f

which ives

absolu te ly o nformat ion

bout

X. In fact , s

one-time-pad

ncoding

f ,

where th e

ey


9/47

2.

roposed

Work

W e

proposed

to

carry

out research

in

lgorithms

and reconfigurable architectures

or he

Adapt ive

ompu t ing

ys tems

A C S ) rogram, it h pplications o

ireless

etworks,

inc luding oding

and

ecurity.

n

ddit ion,

e

roposed

esearch nd eve lopmen t f

reconfigurable ardware

es t

bed

imed

t eployed

obi le

ireless ervices,

nder

he

Globa l Mobi l e Information

Systems

(GloMo)

program.

W e roposed

esearch n to

l gor i thms

or

ecure, obust , mar t , nd

ynamic

e-

Conf igurable Wireless

Networks .

In

rder

o

op e

with

th e volut ion nd omplex i ty

f user

and

miss ion

equirements ,

lex ib le

rchitectures with

in tel l igent nd

econfigurable

e tworks

and compu t e r

sy s t ems

nd

sub sy s t ems

are

required.

New

coding algorithms and

coding

echniques

or

digital

ata

were

invest igated

s

par t f

a

f ramework

ha t

rovides he

ecessary

etwork

nd

omputer

eliabil i ty

nd

ecurity

or

digital

a ta ransfers ,

ver ireless ommunication

inks .

u r r amework

rov ides

h e

f lexibil i ty,

ons is t ency ,

nd

ndependence

f ssur ing el iabil i ty nd

ecurity

n

arious

doma ins

and at var ious

levels

of th e

network

and compu t e r sys t ems .

Wireless communications resen t

both hal l enges nd pportunities. ur ew algorithms

take ign if ican t dvan t age f

hese

ppor tun i t ies . or

ns tance ,

nique

et f ew ata

protec t ion

l gor i thms

s

ased

n

ecret

har ing .

hese

l gor i thms

xploi t

he

va i lable

paral lel ism n he l ex ib le

ombination

f pace- ,

ime - ,

ode -

nd

requency-division

mul t ip lex ing f

wireless network nd

ompu t ing

rch i tec tures .

h is

eads o n

ntegrated

approach ha t ncludes he olut ion o hree a ta rotection

ssues:

rror etect ion, orward

error

orrect ion o educe et ransmiss ion equirements , nd ata ecurity o eal ith

misrout ing nd

n tercept ion .

he

bility

o

rade

ff

cod ing ,

etwork,

econfiguration,

nd

other system parameters

enab le s

f lexibil i ty,

and

cos t

effec t iveness .

Security

and

robus tness is

uppor ted

m o n g

ther

spec t s

of

our

rchitectures,

y

lex ib le ,

t ab le-d irec ted ethods t

he

ardware

evel

ia ynamic e-configuration,

.g .

P G A s ,

software

econfigurable

ate

rrays.

ecurity

s

nhanced

ince he

ardware

s either

algorithm

pecif ic

or

nstruct ion et

dependen t .

obus tness

nd

ecurity

re

nhanced

y

th e

ine-grained

ature f

th e

r imi t ive

func t ions

rovided

y he

ardware .

hey an

e

used

to

enhanc e

th e

security

of data

t ransfers , nsure

he

ntegrity

f

data

t ransfers ,

nd arry

out

compress ion ,

decompress ion ,

and

other

func t ions

such

as

future

serv ices ,

.g .

multi-media

or

commun ica t ion

yp e

ata .

These

ine-grained

r imi t ives

re

deally

atched

o

uppor t

ireless

ommunication

(encoding ,

ransmiss ion ,

ece iv ing ,

ecoding ,

ecurity,

nd uture unct ions) . he

r imi t ive

func t ions

ot

n ly

irect ly m p l e m e n t ower

level

func t ions equired y os t

a ta ecurity

a lgor i thms ,

u t

ls o irect ly uppor t

igher

rder

unct ions ha t

mp rove

f f ic iency nd

performance

of algor i thms

nd

assoc ia ted

compilers .

O ur s tudy

of

well

matched e-configurable architectures

nd

roper

evels

of

granularity

of

mac ro

unct ions

will

help

etermine

uture

eneration

rchitectures f

high-performance

and ow-power

commodity

P G A s

instead

f

being etermined y

yesterday's

ustomers

t yp ica l

glue- logic" )

O ur

proposed

work

was

also

synerg is t ic

with

ongoing

work

n

ecurity

nder

M D O / A R O ,

and ther o v e r n m e n t genc ies ,

or loba l igh-speed hoton ic

ack-bone

nd

atellite

networks .

he

ppea rance f

photon ic

echnology

or

F

inks

"fiber-radio",

adar , idar ,

image ,

and

ther ensors) , s

n ot

nly

e

ompa t ib l e nd omplementary n he hort

ang e ,

b u t even cruc ia l

omponent

echno logy f choice or uture

ery

i gh -bandwid th ireless

networks. hoton ics

will

play a cruc ia l role in paramet r ic )

F

an tenna rrays. e

were

and

con t inue o e

nvolved

n hoton ic

nd uan tum

evices

esearch

nder

arpa

Ultra) ,

B M D O / O N R ,

nd R O

M U R I 9 9 )

n e rms

f

otent ial ys tems pplicat ions,

nd e

re

expanding

he ynerg i sms

m o n g

hese fforts,

ncluding

he

otent ial mp ac t n ireless


10/47

network

ervices

and

rchitectures.

or

ex amp le , we

expec t

ha t

he

hoton ic nd

uantum

devices work will lead to very

effect ive

imp lementa t ions

nd

ombinations

f

very

ow power

commun ica t ion s ,

uan tum-c ryp to

sys t ems ,

us ing pace- t ime c ode

divers i ty .

O ur proposed

esearch

fforts

n he reas

f

data

ecurity

and

eliability, nd n tel l igen t

process ing fo r wireless

networks

are

desc r ibed

in th e

fo l lowing .

V

Data

reliability

and

security

We

proposed

esearch

n to a ta

ecurity

nd

eliabil i ty

will concen t ra te n eve lopmen t f

new

algor i thms

nd

oding

echn iques or

afe

nd

ccura te

elivery

f

ig i ta l

a ta

rom

diverse

sources over

highly var iable wireless

commun ica t ion s

l inks .

Security

and

reliability

via secret sharing

Previous ly

we ad s sumed

oiseless

hanne l

o

ha t

ece ived

acket s

p ieces)

ere

ident ica l o hose ransmit ted , l though

here as ossibil i ty ha t acke t ould

e

os t

(hence

we

used

of a

0-out-of-20

sys t em) .

herefore

we

considered x t end ing

he

m o d e l

o

allow errors

within

rece ived

packets .

f

only

0

acke t s

re ece ived ,

we

have n ly b it f

in format ion

bou t

he b it f

data ,

nd

we canno t

orrect

or

ven

etect )

rrors.

f

1

packet s are received, ns tead ,

we

then av e

00

edundant

it s

an d rror orrect ion

ecomes

poss ib le .

Bec au se

random

arity heck odes ch ieve

hannon ' s hannel apac i ty ,

he

1

eceived

packet s

allow reliable t ransmiss ion

as

ong

s

th e rror ate s sma l l

enough ha t

th e apac i ty

of th e

c hanne l

is somewha t greater than bit/1.1 b it

. 91 . f 2

acke t s

re ece ived,

they allow

reliable

t ransmiss ion

so

long

s

th e

rror

a te

s

mal l

enough

o

hat

he

apac i ty

of

th e

c hanne l is somewha t

greater than

bit /1.2

kbi t

.8 3 nd

so

on .

ence , from he

Shannon theory

poin t

of

view

(in

which

computa t ional

equirements

re

ot

onsidered) , h is

general izat ion of secret

shar ing also allows

opt imal

error correct ion f more

han

he

m i n i m a l

n u m b e r

of

packet s

is

rece ived .

Unfor tuna te ly ,

orrecting

rrors ith n rbitrary

arity

heck

ode

s

n

NP-complete

p rob lem,

o

he

omputational

equirements

f h is

therwise

p t imal

ys tem

an

e

horrendous. e

herefore ropose

nves t iga t ing

hether omputationally fficient

rror

correcting

odes

e.g . ,

C H

nd

eed -So lomon

odes)

an

e

sed

or

enera l ized

ecret

shar ing ,

and

if

not, t rying to

f ind

new

lasses of

codes

hat,

while

less

eff ic ien t

ro m purely

error correct ing poin t

of

view, also

allow genera l ized

sec re t

shar ing .

Rela ted o

his

opic s he uest ion

f

he

radeoffs e tween

e t a ta a te

nd

oding

c omp lex i t y .

W e wish to

determine ,

fo r

a

g iven

l eve l of

reliabil i ty,

he od ing ompl ex i t y s

a

func t ion f a ta ate,

s

he

ata

a te ang es

ro m

ery

mal l

o

he a x i m u m hannel

capac i ty .

n

art icular ,

t

is

impor tan t

o etermine whether

th e oding

omplex i ty al ls

off

rapidly

or s lowly

as th e

data

rate is

reduced

from

capac i ty .

Low communication-complexity error detection codes

In

ddi t ion

o

he

ppl ica t ion

f

secret har ing o etwork ata ecurity nd eliabil i ty, we

propose o

nves t iga te

op ics

elated o rotect ion f

ig h

peed, ig h

o lu me

ata

communications.

Whatever

h igher

level data recovery methods

are

used , data

packet s

m u s t

be

upp l ied

with a

checksum

ha t

uaran tees etect ion f

ll

u t

mal l

raction

f

oss ib le

rrors.

he

s tandard met hod , yclic

edundancy

heck

C RC ) ,

has

m a n y

dvan tages . In art icular, t s

op t ima l

with

respec t

to

burs t error

detec t ion ,

and

can detec t all b u t a fract ion,

'P ,

f poss ib le

errors,

where p

s

th e n u m b e r of

check bits . he tw o mos t o m m o n hift

egister

ircuits or

calcu la t ing

C R C s

a re

quite s imple and

have

ei ther

constan t or 0(log

p)

ate

delays .

owever,

these

circuits

equire ither

arge

anou t

r

an in ,

o

ha t

n ery

i gh - speed

perat ion,

he


11/47

long

wires

needed

to commun ica t e

th e

s tate of th e

c ircu i t

to

all

parts

of

th e

ne x t

s tate

logic re

th e

computa t ional bot t leneck . W e

will

study

th e

error detect ion efficiency

of s impl i f ied

R Cs

and

ther

inear

block

odes

n which

ach heck it epends n n ly mal l u m b e r

f

nearby hift

egis ter

its.

m o n g

he

ossibil i t ies

re adder- l ike t ructures

s ing

-gates

ins t ead

of shif t

registers .

Error detection

via modulation

We

p l aned o

xplore ow o se edundancy

nheren t

n arious modulation

c hemes

o

provide

uff ic ien t

error

detec t ion .

Other echn iques

ha t

re

ot e t

used

n

he

onsumer

omain ,

uch

s

ow

robabil i ty

f

in tercept

ethods

e.g.

pread-spectrum,

ltra-wide-band

ignals ,

emto- second , nd

quantum-limited echnologies)

ppear

o

be ompa t ib l e it h ur

ecurity

l gor i thms,

iven

ou r curren t

work

with B M D O / A R O

and other governmen t

agenc ies

3 . roject

Organizat ion

T he

research

suppor t

fo r this

roject

am e

ro m

wo

Darpa

rog rams :

he

l gor i thms

nd

reconfigurable

rchitectures

esearch

as

upported

y

he

dap t ive

Computing ys tems

( A C S )

program,

and

th e

research

into

appl ica t ions

to

wireless

etworks,

ncluding

oding

nd

security

was

suppor ted

by th e

Globa l

M o b i l e

Information

Systems

(GloMo)

program.

Stanford niversi ty

as

he r ime ontractor, esponsib le

or

esearch n l gor i thms

nd

reconfigurable rchitectures

or

dapt ive

Computing

ys t ems

A C S ) , it h ppl ica t ions

o

wireless

networks ,

inc luding coding

and security.

en V ia was

a

subcon t rac tor ,

esponsib le

or

esearch

nd

evelopment

f

econfigurable

hardware test bed a imed at deployed mobi le wireless services (e .g. A M P S , P C S )

Late

n to

he

ont rac t

nVia

as eorgan ized, u t

he eorganized nVia /Morph ic as

transparent o

he esearch ontract ,

s

ll

work

was erformed

nder nVia 's ubcontract;

however ,

th e f inal report

is

organ ized recogn iz ing this

change

of

responsib i l i ty .

Work

and

Funding

Timeline:

Work

actually

star ted

on

this

projec t

on S e p t e m b e r 6 , 996 .

T he

ubcontract it h n

V ia

as

ina l ized

n

id

December

996 ,

herefore

he

adio

subcontract

it h

n V i a tarted

hree onths ate .

h is

ubcontracted

ork

oughly

main t a in ed his

three mont h e lay , om e f th e work augh t p

it h

he r ig ina l chedu le ,

an d om e ther or k ot urther elayed.

om e

f he ddi t ional

elay

as

aused

y

commerc ia l

par tner (Bel l

outh ) unding

f en V ia

t ha t

was ei ther

delayed

r

inally

id not

mater ia l ize ,

hence en

V ia

had to

wait

fo r

other

funding

sources .

A B M D O ecurity

ontract

ffort

har ing

ha t

was

n

lace

it h

Stanford niversi ty,

as

rescinded

b y ongress

n

October

996 , fo r

non- techn ica l easons ; he

ompensatory t uden t

suppor t was

later

approved , and

later

funded

in

part.


12/47

4 .

Research

T he tanford rchitecture nd

r i thmet ic

roup

arr ied

u t esearch

n to

lgor i thms

nd

reconfigurable

rchitectures

or

dap t ive

ompu t ing

ys tems

A C S ) ,

it h

pp l ica t ions

o

wireless

networks ,

nc lud ing

coding and

security.

e n V i a carr ied out research and

deve lopmen t

of

a reconfigurable ardware

es t

bed

ime d t

dep loyed

mobi le

wireless

services

(e .g.

A M P S ,

PCS. )

4.1 tanford Research

Algorithms,

Reconfigurable Architectures,

and Security

Introduction

T he

tanford econfigurable

rchitecture esearch

n

C S

s

ased

n ur

amBlox

app roach

o

PG A rogramming,

n

bjec t r ien ted es ign

ethodology ha t as

hown

some

s ign if ican t advan t ages .

arious

arithmet ic

an d security

a lgor i thms were

imp l emen t ed

n

eff ic ien t low-power

F P G A

des igns , nc lud ing

D ES

and

IDEA

encrypt ion

s t andards .

T he ecurity

ork

as

emonstrated

ul t i-path

pproach

o

ecure

ireless

ata

t ransmiss ion n 99 8

s ing

hree

aptop

omputers , ased n wired,

m o d e m ,

R ,

or

Ethernet

l inks ,

and c ommerc i a l

wireless

Ricochet or

Freewave

l inks .

W e

nves t iga ted ew

pace- t ime

od e

iversi ty ased

l gor i thms

or

ireless

communication

n he ontex t f oftware

efined adios,

ith

he

oal f roviding r

extending

imul taneous ecurity

pr ivacy ,

ntegrity,

eliabil i ty,

vailabil i ty,

nd

rotection

from everse ngineering)

y

xploiting aral lel ism

nd calable

ierarch ica l multiplexing

s c hemes to

bu i ld oftware

def ined

ad io rch i tec tures .

e

deve loped

ew

Space-Time-Code

Diversity S T C D )

ased

ecurity

rch i tec tures

ha t

uppor t ll

hese

equirements nd

map

well

onto oftware def ined radio t echnologies .

4.1.1

FPGA—Adaptive

Computing

Systems

Summary

O ur

D a r p a

C S

projec t pecif ic

ontributions

or

he PG A

des ign

nvironment

nc lude

se t of

tools

nd

met hods :

AM-Blox,

onsis t ing f

evels f

bstract ion

P a m B l o x

nd

PaModu l es ) .

ur PAM-Blox

software

vl.O

s

ava i lable

to

he D A R P A

A C S

commun i t y nd

can

e

ownloaded

ro m he web t

ttp://umunhum.stanford.edu/PAM-Blox/

h is

software

inc ludes a n u m b e r of

PAM-Blox

des ig ns

or

r i thmet ic mult ipliers ,

matrix-multiply,

J a c ob i

elaxat ion) , nd t andard ncryption

D E S ,

D E A ) ,

hat ompare

avorab ly

ith he

R A W

enchmark s ,

ee

ur

ubl ica t ions

e low.

enera l ly ,

e

ch ieved

erformance

improvemen t s n

throughput ,

area, and /or MOPS/Wat t , ranging ro m factor f tw o

to ix f

our

AM-Blox

ased

P G A

es igns

v er icro

rocessors,

S P,

n d

SICs , n atency

tolerant

app l ica t ions . AM-Blox provides a

c lean interface

to share F P G A des igns .

herefore

we encourage

con t r ibu t ions

in

th e

form of

P a m B l o x ,

P a M o d u l e s

and

ent ire

ppl ica t ions

sing

these odu les . ont r ibu t ions

hould

ollow he

tyle

nd

orma t

iven

n

he ase

distribut ion,

nd

ontain

t

east

ne

amp le

ppl ica t ion

here

he

modu le s

re

sed .

ll

contributions

o

PAM-Blox

ave

o e

ist r ibuted

it h

he N U

enera l

ubl ic

i cense .

Ser ious cont r ibu t ions

will be reviewed and ,

if

accep table , inc luded

in

further

releases

f

PAM-

Blox.

Programmabil i ty ,

Performance

and

Power Metric

fo r

FPGAs

PAM-Blox

consists

of

open,

bject-oriented

ircuit enera tors n

op

f

he C I amet te

des ign

environment ,

P a m D C .

PAM-Blox as

first in t roduced

in

M e n c e r

[ M M F

Apr98] .


13/47

PAM-Blox re n t ended

o

e

art

f n pen

eposi tory hat

nables

es ign har ing

between

m e m b e r s

f

th e

dap t ive

compu t ing

commun i t y .

High-performance

F P G A des ign

or

dap t ive omputing

s

s impl i f ied

y s ing

hierarchy

of

pt imized

ardware

bjects escr ibed n ++ . rogrammability

s

mproved

y

s ing

object-oriented

echn iques uch s

empla tes ,

irtual unct ions , unct ion

ver load ing ,

nd

inheritance,

in

th e specif ic

way

out lined n his ect ion.

s

with

object-oriented

oftware,

th e

des ign

f

th e

ardware

bjec t

nterface

s

rit ical

o

he

sefu lness

f

he

ys tem.

n

act,

more ffort was p end

n

he

terative es ign

f

he

nterface,

han as ater ecessary o

des ign

th e

hardware

objects

t hemselves .

In

rder o

r idge

he

pace etween he l gor i thmic

epresen ta t ion

f

n

ppl icat ion nd

th e gate level

( lookup

table

level)

we

add

evels

of

abstract ion, tar t ing

t

th e

egis ter

transfer

level

( R T L ) which

s

equivalen t

o he a m D C

evel .

he big

icture

s

hown n

igure

below.

IflL-iUmiJ'iaSx

S -

|

InsSnaütion

Set-1

IS«,'

svi

-

sv t

s\v

S |PA

I

L V V

H A

- -

-

t

lk"hii»ral JIM.

'l\E>-tdl lö

Jävs V:M

PliMA'ÄIf

IK A

ASIC

Th i s igure hows th e

Hardware/Software

in terface or

ifferent

echnolog ies :

av a micro-

processor

J a v a V M ) ,

onven t iona l

ic roprocessor ,

P G A s , nd SICs .

n

enera l ,

he

prog rammer has

to

br idge

th e gap

between

algor i thms

and t ransistors.

T he

boundary

between

hardware

and

software

fo r

processors ,

F P G A s ,

and

A S I C s ,

def ines

he

interface

e tween

rogrammer/compiler

nd

he

omputing

lement s .

he

ow-level

boundary

e tween oftware

and ardware

or P G A s

requires he oftware

to

r idge large

space

from

algorithm down

to

th e

gate

level .

The PAM-Story

P A M t ands

or

rogrammable

ctive

M e m o r i e s .

he

irs t

A M ,

eRLe-0 ,

eve loped t

D EC R L n

rance ,

s

ne

f

he

irst P G A ased

omputing

ach ines .

ex t o

he

hardware

fforts he A M eam ls o eve loped ++ lass ibrary, a m D C , or reating

des igns

fo r Xilinx

F P G A s . T he

m o s t impress ive

result

obta ined

with

a m D C nd he eRLe-1

board is R S A

encrypt ion

at

Cray speeds .

T he

mos t

curren t

P A M

is

th e

P C I

Pame t t e

board

deve loped

b y

M a r k

S h a n d

t

th e

DIGITAL

Sys t ems

Research Center .

he P C I Pamet t e consis t s

of

5 Xilinx X C 4 0 0 0 series

F P G A s .

PAM-Blox:

PamBlox and PaModules

PAM-Blox onsis t

of tw o

major

ayers f

abst rac t ion .

irst,

amBlox

e e parameterizable

s imple l ements uch s ounters nd

dders .

utomat ic l a cemen t f

arry

ha ins

nd

f lexible shapes

are

uppor ted . a M o d u l e s re more ompl ex

l emen t s

oss ib ly

nstan t ia t ing

P a m B l o x .

P a M o d u l e s eneral ly ave

f ixed

hapes

nd re

sually

pt imized

or

pecif ic


14/47

data-width . Exampl es or

a M o d u l e s

re

ult ip liers ,

oordinate otat ions (CO RDICs )

[ M M F Jun98 ] ,

pec ia l

r i thmet ic

uni t s

or encrypt ion , nd commun ica t ion modu le s .

T he

ey

ifference f

u r

pproach

o

ost

ther

es ign

ools or

P G A s

s

ha t he

des igner

has

ptional

ontrol over p l acemen t

at

each eve l

of

th e es ign ierarchy ,

which

s

th e ey o igh-performance P G A design . s

mentioned

efore, the bjec t

nterface

as

chosen

areful ly

o

encourage

code-reuse

an d

s impl ify

code-sharing

between

designers .

A

su bse t

of th e

P a m B l o x

h ierarchy is

shown

in th e

f igure

below

PamBlox

PBtopi

ConstReg

Register

MuxReg

\

SerToPar

\ParToSer

Counter

r—

RCounter

LCounter

^RLCounter

\ LECounter

Add

FlexAdd

LEShifter

represen t ing a

sub se t

of th e

P a m B l o x

hierarchy.

he op bjec t Btop onsists f

a m i n i m a l

a m o u n t

of

logic n d

set

of placement unct ions

ha t re nher i t ed y ll P a m B l o x bjects .

Pref ix R '

s tands or Resetable , L ' fo r Loadable ,

nd E

fo r n Enable . he op

object ,

Btop,

consists

of

a vec tor

of registers , nd se t

of

placement unct ions ha t

andle

different arry-chain onfigurations.

s

n

xamp l e

f

od e euse , very hild f

PBtop

inherits

th e

p l acemen t

func t ions

an d

can

overwrite

them

if

necessary .

PaModules

re

ore

ompl ex , enera l ly

i x ed

ircuits

mplemented

s ++

bjects .

P a M o d u l e s

an

nclude ul t iple

a m B l o x

nd re

ptimized

or

pecif ic

ata-width .

Exampl es

re

onstan t (K)

oefficient

ul t ipl iers

K C M s ) ,

ooth

ult ip liers ,

Coordinate

Rotat ion

Dig i t a l Computer

( C O R D I C )

ircuits

(mentioned

above) ,

and

spec ia l

purpose

ar i thmet ic units uch

as

a constant mul t ip ly

modu lo

(2^ +

)

operat ion fo r

encryp t ion .

T he table below

shows th e

code-s ize of AM-Blox ersion .0 c ircu i t enera tors . ode-s ize

is g iven in

P a m D C

/

C ++

l ines

necessary

to

i m p l e m e n t

th e

objects .

PamBlox

PaModules

No. f

Objects

28

6

Lines of C o d e

1370

750

A v .

ines

per

Object

~50

20

Hierarchical Naming

Commerc ia l

ynthes is ools

m a k e t difficul t

o

pt imize

ircuits

n ow eve l

e.g.

fter

place-and-route)

y lat tening he

es ign nd

hanging

he

a m i n g

f

he

wires.

P a m D C


15/47

enab le s irect ontrol ver

th e

aming f wires. AM-Blox re mplemen t ed

o uppor t

hierarchica l

naming

s c heme

that

crea tes

unique

n a m e or ach

wire

in he

es ign

imi la r to

pa ths

n

f i le - sys t em.

he

am e

f

each wire

conta ins

ll

he ncestors paren t bjects)

f

th e

wire.

he

to p

n a m e

can

be spec if ied

b y

th e

des igner , .g .

P a M o d u l e

mul t ip l ier

with th e

n a m e multy con ta in ing ne d d e r with carry -chain , esul ts

n ,the

fol lowing am e or he

third

e l emen t of th e carry -chain ,

e.g. multy/'addOi'carry


16/47

opera t ions

'shift and

add '

m ap

easily

onto th e Xilinx X C 4 0 0 0 E

l ibrary

used

by

F P G A xpress

II . herefore here s not m u c h oom or re a mprovemen t . lock requency f

th e AM-

Blox

es ign

s

n ly

bou t

5%

igher

han

he

es ign

pt imized

y

ynopsy s D L

compiler .

However ,

th e

improvemen t

in area

is

abou t

2 0 % .

B. DES

Encryption

D ES

encryp t ion

is

well

suited

fo r

implementa t ion

in

hardware .

T he

bas ic

pr imit ives

are

ixed

permutations

nd

xclus ive-or .

he

esults

or

he

PAM-Blox

ES

es ign

n

able

elow

show a

3 0% ncrease

in per formance

(clock

f requency)

us ing

n ly

alf

of

th e

rea.

T he

uper ior

esults

bta ined it h PAM-Blox re ue o

art ia l ly

anua l

lacement

nd

t echnology

map p ing ,

i .e. he carefu l

design of

logic that fits

into 4 b it

lookup

tables .

C .

nteger Matrix Multiply

T he

M at ri x

M ul t ip ly

enchmark

mul t ip l ies

wo

4 x 4

mat r ices

with

4 2 6 mult ipl iers nd

an

dder

ree.

P G A

Express I

ses imp le

i t -ser ial

hift-and-add

ul t ipl icat ion.

u ll

matr ix

mul t ip ly

therefore

t akes more han

0 lock

yc les .

or

his benchmark e

chose

o

create a

more

eff ic ien t

computa t ional

st ructure,

.e . an

arithmet ic

un i t

o how how

PAM-Blox

can

be

used

to

adap t he r i thmet ic nits

o

he

pecif ic

equirements f th e

ppl ica t ion . y

implementing

mul t ip le

i t -ser ial

mul t ip l iers

s ing

ooth

ncod ing ,

e

are

b le

o

rade

rea

for

performance.

bviously

Booth mul t ip l iers

are more

eff ic ien t

fo r

this pecif ic

pplication.

T he idea

is

to

use th e

PAM-Blox

nv i ronmen t

to

choose

th e ar i thmet ic n it ~

n ur ase he

mul t ip l ier ~ that is

bes t suited

fo r

th e specif ic app l ica t ion .

RAW

Benchmark

Results

T he

able

elow

hows

tw o P A M - B l o x

es igns ,

A M - B l o x nd A M - B l o x ,

iffering

only

in

th e select ion

of

th e

mul t ip l ier

P A M - B l o x

ult ip lies

he

mat r ices n

7

lock ycles

while

P A M - B l o x ake s

9

lock

yc les or full

x 4 mat r ix multiplication

ncluding

ata

t ransfer .

lock

ycle

imes

or

he

A M - B l o x

es igns

re

round

3

Hz. he riginal

des ign

syn thes ized with

F P G A

Express

II

runs t

5

M Hz

nd equires

9 lock yc les or

full mat r ix

mul t ip ly .

i i

Comi ' ik Time

Arra

C L B ]

Fi^picricv"

JMXifil 'M

i 'SHjüj |

FEU

I

SO s 16 4

30

M Hz

} PAM-Blox

i 45 S

12 9

m

M H*

\

D ES 11

i

i

F E

II

Documents

Re-Configurable Multimode, Multiband Information Transfer Systems