• No results found

Redundant Residue Number System Based Error Correction Codes

N/A
N/A
Protected

Academic year: 2020

Share "Redundant Residue Number System Based Error Correction Codes"

Copied!
5
0
0

Loading.... (view fulltext now)

Full text

(1)

CODES

Lie-Liang Yang and Lajos Hanzo

Dept. of ECS, University of Southampton, SO17 1BJ, UK.

Tel: +44-703-593 125, Fax: +44-703-594508

Email: lh@ecs.soton.ac.uk, http://www-mobile.ecs.so ton. ac. uk

ABSTRACT

Inthispaper residue number system(RNS)

arith-meticandredundantresiduenumbersystem(RRNS)

basedcodesaswellastheirpropertiesarereviewed.

We propose a number of applications for RRNS

codes and demonstrate how RRNS codes can be

employed in global communication systems, in

or-derto simplify the associated systems by unifying

theentire encoding and decoding procedure across

globalcommunication systems.

1. INTRODUCTION

RNS-based arithmetics exhibit a modular structure that

leads naturally to parallelism in digital hardware

imple-mentations. TheRNS has twoinherent features that are

attractiveincomparisonto conventional weighted number

systems,suchas for examplethebinaryweighted number

system representation. These two features are [1{ 4]: 1)

thecarry-freearithmeticand2)thelackofordered

signi-canceamongsttheresiduedigits. Therstpropertyimplies

thattheoperationsrelatedtothedierentresiduedigitsare

mutuallyindependentand hencethe errorsoccurring

dur-ingaddition,subtraction andmultiplicationoperations,or

dueto the noise induced by transmission and processing,

remainconnedto their originalresidue digits[1,5{7]. In

otherwords, these errors donot propagate and hence do

notcontaminateother residuedigitsduetotheabsenceof

acarryforward. Theabove-mentionedsecondpropertyof

the RNS arithmeticimplies that redundant residue digits

canbediscardedwithoutaectingtheresult,providedthat

asuÆcientlyhighdynamicrange is retainedby the

resul-tantreduced-rangeRNSsystem,inordertounambiguously

describethenonredundantinformation symbols.

As it is well known in VLSI design, usually so-called

systolicarchitectures[1] areinvokedtodivideaprocessing

taskintoseveralsimpletasksperformedbysmall,(ideally)

identical,easily designed processors. Eachprocessor

com-municatesonlywithitsnearestneighbour,simplifyingthe

interconnection designand test, while reducing signal

de-laysandhenceincreasingtheprocessingspeed. Due toits

Thisworkhasbeenperformedintheframeworkofthe

Pan-EuroepanISTprojectIST-1999-12070(TRUST),whichispartly

fundedbytheEuropeanUnion. Theauthorswouldliketo

ac-knowledge the contributions of their colleagues, although the

viewsexpressedarethoseoftheauthors.

Thenancial supportof theEPSRC,Swindon,UK isalso

gratefullyacknowledged.

carry-free property, the RNS arithmetic further simplies

the computationsby decomposinga probleminto asetof

parallel, independentresiduecomputations.

The properties of the RNS arithmetic suggest that a

redundantresiduenumbersystem(RRNS)canbeusedfor

self-checking, error-detection and error-correction in

digi-tal processors. TheRRNStechniqueprovidesauseful

ap-proachtothedesignofgeneral-purposesystems,capableof

sensing and rectifying their own processing and

transmis-sionerrors. Forexample,ifadigitalreceiverisimplemented

using theRRNShavingsuÆcient redundancy, thenerrors

in theprocessedsignals canbedetectedand corrected by

the RRNS-based decoding. Furthermore, the RRNS

ap-proach [1] is the only one, where it is possible to use the

verysamearithmeticmoduleintheverysamewayforthe

generationofboththeinformationpartandtheparitypart

ofaRRNScodeword. Moreover,duetotheinherent

prop-erties of the RNS, the residue arithmetic oers a variety

of new approaches to therealization of digital signal

pro-cessing algorithms,suchasdigitalmodulationand

demod-ulation as well as the fault-tolerant design of arithmetic

units [2{ 4,7{ 9]. It also oers new approaches to the

de-sign of error-detection and error-correction codes. Let us

rstreviewthebasicmathematicalmodelrequiredforthe

conversion of operands fromthe weighted numbersystem

to the RNS, orconversely,from theRNS to the weighted

numbersystem.

2. RESIDUENUMBER SYSTEM TRANSFORM

ANDITS INVERSE

Forconvenience,wedenetheresiduenumbersystem

trans-form(RNST)asthealgorithmthattransformsany

conven-tional weighted numbersystem, such as for example the

natural binary codeddecimal (NBCD) numbersystem to

the RNS [3,7]. The inverseRNST (IRNST) is denedas

the algorithm, whichtransforms theRNS tothe weighted

numbersystem,anissuetobeelaboratedonbelow.

Let X be a non-negative integer numberin the range

of0X <% n

,where%isabaseandXisrepresentedinthe

weightednumbersystemasX =fbn

1 bn

2

::: b2 b1b0g=

P

n 1

j=0 b

j %

j

, where b

j

2 f0;1;:::;% 1g . Specically, a

weightedbinaryrepresentationisconsidered,when%=2.

Let f m1;m2;:::;mug , the set of so-called moduli

in-volvedintheRNS,beasetofpairwiserelativeprime

num-bers,wherenotwomodulihaveacommonintegerdivisor

greaterthan1. If M = Q

u

i=1 m

i %

n

,thentheRNSTis

(2)

therangeof0X <% -asau-tupleresiduedigitsequence

(r1;r2;:::;ru),whereri=X (modmi)=X b X

m

i cmifor

i=1;2;:::;u,bzcdenotesthelargestintegernotexceeding

z,and0r

i

m

i

1. Furthermore,[0;M 1]isdened

asthedynamicrangeoftheRNSandanyintegerXinthis

rangeisuniquelyrepresentedbyau-tupleresiduesequence,

whichisexpressedas:

X,(r1;r2;:::;ru); 0X<M;

ri=X (modmi); i=1;2;:::;u: (1)

AssumingthattheintegersX1andX2haveRNS

repre-sentationsofX1=(r11;r12;:::;r1u)andX2=(r21;r22;:::;

r2u),respectively,thenX1X2,wheredenotesaddition,

subtractionormultiplication,yieldsanotheruniqueresidue

sequenceX

3

,providedthatX

3

remainswithinthedynamic

rangeof[0; M). ThearithmeticoperationsovertheRNS

canbecarriedoutonaresidue-by-residuebasis,whichcan

beexpressedas:

X1X2,[(r1ir2i)(modmi)] u

i=1

; (2)

whereonthelefttheoperationrepresentsordinary

addi-tion,subtractionor multiplicationof two integers,and on

therightitrepresentsthesameoperationsperformedonthe

basisoftheappropriateresiduedigitsr1i andr2i,with

re-specttotheircorrespondingmodulusm

i

. Thisallowsusto

convertbinaryarithmeticoperationscarriedoutwithlarge

integerstoresiduearithmeticoperationsinvolvingahigher

numberofsmallerresiduedigits,where theoperationscan

beexecutedinparallelfashionandthereisnocarryforward

between the residuedigits. Thelack ofcarry forward

be-tweentheresiduedigitspreventsthepropagationoferrors

betweenresiduedigits.

Incontrast totheRNST, theIRNSTis denedasthe

algorithmthattransformstheoperandsfromtheRNS

do-maintotheconventionalweightednumbersystem

represen-tation. A well-knownimplementationoftheIRNSTisthe

so-calledChineseremaindertheorem(CRT)[5]. According

to the CRT, for any given u-tuple (r

1 ;r

2 ;:::;r

u

), where

0ri<mifori=1;2;:::;uthereexistsoneandonlyone

integerX suchthat0X<M andri=X (modmi). It

canbeshownthat the numerical value ofX can be

com-putedby[5]:

X= u

X

i=1

riTiMi (modM); (3)

where M

i

= M=m

i

and the integers T

i

are computed a

prioribysolvingtheso-calledcongruenceT

i M

i

=1(modm

i ).

ThecoeÆcients Ti are alsooften referred to asthe

multi-plicativeinversesofM

i .

The CRTis a classical algorithm for the

implementa-tionoftheIRNST.However,thereal-timeimplementation

oftheCRTisnotpractical,sinceitrequiresmodular

oper-ationswithrespecttoalargeintegerM. Inordertoavoid

processinglarge integers,especiallyinthe contextoferror

control invokingtheRRNS, afrequentlyusedapproachis

theso-calledbaseextension(BEX)operationinconjunction

withthemixedradixconversionapproach[7].

3. REDUNDANTRESIDUENUMBER

SYSTEM ANDRRNSENCODING

If a residue number system is designed not only for the

representationofdata,butalsofortheprotection ofdata,

usuallywedesigntheRNSusingso-calledredundant

mod-uli, in order that the system has the capability of

self-checking, error-detection and error-correction[1]. Inthis

case the operand X is limited to the so-called

informa-tion dynamic range of

0; M= Q

v

i=1 mi

, where v u,

andm

1 ;m

2 ;:::;m

v

arereferredtoastheinformation

mod-uli, while mv+1;mv+2;:::,muare theso-called redundant

moduli. We express the product of the redundant

mod-uli as M

R =

Q

u v

j=1 m

v+j

. The interval[0;M) constitutes

the legitimate range of the operand X, and the interval

[M;MM

R

) is the associated so-called illegitimate range.

Any integer belonging to the legitimate range will be

la-belledaslegitimateandthosebelongingtotheillegitimate

rangeasillegitimate,sinceitdoesnotrepresentalegitimate

numberor operand,directlyaccruing fromthe

nonredun-dantinformation-bearingresiduedigits.

AnimportantpropertyoftheRRNSisthataninteger

representedbytheresiduesoftheRRNScanberecovered

byanygroupofvnumberofmoduliandtheircorresponding

residue digits. This property constitutes the basis inthe

designofRRNSbasederror-detectionanderror-correction

schemes.

A RRNShaving v numberof information moduliand

u vnumberofredundantmoduliisdenotedasaRRNS(u;v)

code. Theoperationassociatedwithgeneratingthe

redun-dant residue digits of an integer operand X can also be

viewedorinterpreted,asanencodingoperationgenerating

theparityresiduedigitsofaRRNS(u;v)code. TheRRNS

encodingoperationcanbeimplementedbasedonagroupof

codewordshavingtypicalintegervalues,byinvoking

exclu-sivelyadditionoperations. Inordertoaugmentthis

state-ment,belowweprovideanexampleshowingtheassociated

groupsof codewords fortheRRNS(7,3)codehaving

infor-mationmoduliofm

1 =4; m

2 =5; m

3

=7andredundant

moduliofm4=9; m5=11; m6=13; m7=17. The

legit-imaterangeofthiscodeis[0,139],since457=140.

Ta-ble1portraysarangeoftypicaldecimalintegermessages,

whicharebasedontheintegers`1,2,5,10, 20,50,100'

of-tenusedinmonetarysystems. Itiswellknownthatonthe

average any integer inthe legitimate range of [0,139] can

beexpressedby thesumofasubsetofthelowestpossible

numberofthespecicintegersgivenabove. InTable2we

summarised a range of binary integer numbers and their

correspondingRRNScodewords.

Example 1 Forthe encoding of the decimal integer X =

123, since X can be expressed as 123=100+20+2+1,

the RRNS codeword of X can be simplyobtainedfrom the

codewordscorrespondingtoX

7 , X

5 ,X

2

inTable1andX

1

onthebasisofmoduloaddition. Itcanbereadilyshownthat

thecodewordofX=123is(3;3;4;6;2;6;4). Similarly,for

a binary integer X =1011011, the associated codeword is

simplyconstituted bythecorrespondingmoduloadditionof

X

7 ; X

5 ; X

4 ; X

2 andX

1

of Table 2, where the subscripts

7;5;4;2;1 are simplythe indices of the binary 1 positions

in X. The associated codeword of X =1011011 is hence

(3)

messageX residuedigits residuedigits

r1 r2 r3 r4 r5 r6 r7

X

0

=0 0 0 0 0 0 0 0

X

1

=1 1 1 1 1 1 1 1

X

2

=2 2 2 2 2 2 2 2

X3=5 1 0 5 5 5 5 5

X4=10 2 0 3 1 10 10 10

X

5

=20 0 0 6 2 9 7 3

X

6

=50 2 0 1 5 6 11 16

X

7

=100 0 0 2 1 1 9 15

Table 1: RRNScodewords ofsometypicaldecimalinteger

messagesX in the RRNSwithmoduli m

1

=4; m

2 = 5,

m3 =7,m4 =9; m5 =11, m6 =13and m7 =17, where

r

i

=X (modm

i

)andM =457=140.

Binary Nonredundant Redundant

messageX residuedigits residuedigits

r

1 r

2 r

3 r

4 r

5 r

6 r

7

X0=0000000 0 0 0 0 0 0 0

X1=0000001 1 1 1 1 1 1 1

X

2

=0000010 2 2 2 2 2 2 2

X

3

=0000100 0 4 4 4 4 4 4

X4=0001000 0 3 1 8 8 8 8

X5=0010000 0 1 2 7 5 3 16

X6=0100000 0 2 3 5 10 6 15

X

7

=1000000 0 4 1 1 9 12 13

Table2: RRNScodewordsofsometypicalbinarymessages

X inthe RRNSwith moduli m1 = 4; m2 = 5, m3 = 7,

m

4

= 9; m

5

= 11, m

6

= 13 and m

7

= 17, where r

i =

X (modm

i

)andM =457=140.

4. ERROR-DETECTIONAND

ERROR-CORRECTION IN RRNS

Inthissection,webrieysummarisesomepropertiesofthe

associated error-detection and error-correctionprocedures

inthecontextoftheRRNS.Error-detection/correction

de-coding of RRNS(u;v) codes has been discussed in depth

in [3,5{7]. Let us invoke two simple examples, in order

togaininsightintotheerror-detectionanderror-correction

mechanismoftheRRNS.

Example2 Letusconsiderthemoduli3;4;5;7,where3;4

and 5 are the information moduli and 7 is the redundant

modulus. Theinformationdynamicrangeis[0;M =345)

=[0;60). Uponconsideringan integer decimalmessage of

X=21,thecorrespondingresiduevaluesareX =(0;1;1;0).

IfthereisanerrorintheRNSrepresentationdueto

trans-missionorprocessing, forexampler

3

ischangedfrom 1to

3,thenthereceivedRNSrepresentation becomes(0;1; ~

3 ;0).

Uponfollowingthegeneralapproach ofthe CRTinEq.(3)

andusingtherstthreeresiduedigitsandtheirmoduli,we

obtain:

M1=45=20; M2=35=15; M3=34=12;

T1=2; T2=3; T3=3;

X =[0220+1315+3312]mod60

=33:

However, where X = 33 (mod 7) =5 6= r4 =0, and we

canconclude thattherewereerrors intheRNS

representa-tion. Therefore,upondesigningtheRNSusing one

redun-dant modulus,theresiduedigiterrorof r3 canbedetected.

Example 3 Letusnowinvokeanadditionalredundant

mod-ulus,namely11intheaboveexample,whichresultsina

to-taloftworedundantmoduli,namely7and11intheRRNS.

Letusalso considertheinteger messageX =21,now

hav-ing corresponding residue digits of X =(0;1;1;0;10) and

that r3 is inerror andit waschanged from 1to 3,i.e the

receivedRNSrepresentationbecomes(0;1; ~

3;0;10).

Accord-ingtotheCRT'sapproach,theintegerXintherange[0;60)

canberecoveredbyinvokinganythreemoduliandtheir

cor-responding residue digits, if no errors occurred in the

re-ceivedRNSrepresentation. Letusnowconsiderallpossible

cases andattempt torecover the integer X represented by

(0;1; ~

3 ;0;10), upon retaining all possible combinations of

threeoutofve residuedigits,whichresultsin:

(r

1 ;r

2 ;r

3

)=(0;1;3) , X

123

=33(mod60);

(r1;r2;r4)=(0;1;0) , X124=21(mod84);

(r1;r2;r5)=(0;1;10) , X125=21(mod132);

(r1;r3;r4)=(0;3;0) , X134=63(mod105);

(r1;r3;r5)=(0;3;10) , X135=153(mod165);

(r

1 ;r

4 ;r

5

)=(0;0;10) , X

145

=21(mod231);

(r2;r3;r4)=(1;3;0) , X234=133(mod140);

(r

2 ;r

3 ;r

5

)=(1;3;10) , X

235

=153(mod220);

(r

2 ;r

4 ;r

5

)=(1;0;10) , X

245

=21(mod308);

(r3;r4;r5)=(3;0;10) , X345=98(mod385);

where X

ijk

represents the recovered result by using

mod-uli m

i ;m

j and m

k

as wellas their corresponding residue

digits ri;rj and rk. From these results we observe that

X

134 ;X

135 ;X

234 ;X

235 and X

345

are all illegitimate

num-bers,sincetheirvaluesareoutofthelegitimaterange[0;60).

In the remaining ve cases, except for X123, all the

re-sults are the same and equal to 21. Moreover, all these

results were recovered from three moduli without including

m

3

, i.e. from X

124 ;X

125 ;X

145 and X

245

, which are equal

to 21. Hence,we might conclude that thecorrect result is

21andthattherewasanerrorinr3,whichcanbecorrected

bycomputing^r

3

=21(mod5)=1.

RRNS codes exhibit similar coding properties to the

well-knownReed-Solomon (RS) codes [5,7]. RRNScodes

constituteaclassofmaximumminimum-distanceseparable

codes. An RRNS(u;v) code - where the information

dy-namicrangerepresentedbytheRRNS(u;v)is[0; Q

v

i=1 mi)

andtheRRNScode'sdynamicrangeis[0; Q

u

i=1 m

i )-hasa

minimumdistanceof(u v+1)andhenceitiscapableof

detecting(u v)orlessresiduedigiterrorsandcorrectup

tob(u v)=2crandomresiduedigiterrors. AnRRNS(u;v)

codeiscapable ofcorrectingtrandom residuedigiterrors

andsimultaneouslydetecting(>t)residuedigiterrors,

if and only ift+ (u v). Moreover, anRRNS(u;v)

codeiscapable ofcorrectingtrandom residuedigiterrors

andsimultaneouslycorrectingresidueerasures,provided

that 2t+(u v).

Accordingtothecarry-freepropertyandduetothelack

of weighted signicance of the residue digits in the RNS

arithmetic, in RRNScodes some of the channel-impaired

residue digitscanbediscardedasanerrorcorrection

(4)

re-ouslydecodetheresult. Theabovestatement canbe

aug-mentedasfollows. Letfm

1 ;m

2 ;:::;m

u

gbeasetofu

mod-uli ofan RRNS(u;v) code, where m1 <m2 <::: <mu.

Furthermore, let X be the integer message, which is

ex-pressedwiththeaidoftheresiduedigitsas(r

1 ;r

2 ;:::;r

u )

withrespecttotheabovemoduli. Ifthedynamicrangeof

X is [0; Q

v

i=1 m

i

), wherev u,thenX canbe recovered

fromany voutoftheunumberofresiduedigitsandtheir

relevantmoduli. Thispropertyimplies thatd(du v)

numberofresiduedigitscanbedroppedbeforetheIRNST

stage, while still recovering the message X using the

re-tained residue digits, provided that the retained residue

digitsarethecorrect,ieuncorruptedresiduedigits.

Moreover,ifweletdbethenumberofdiscardedresidue

digits,where d u v, then, aRRNS(u;v) codeis

con-vertedtoaRRNS(u d;v)afterdoutoftheuresiduedigits

arediscarded. Hence,thereducedRRNS(u d;v)codecan

detect up to (u v d) residue digit errors and correct

up to b(u v d)=2c residue digiterrors. This property

suggeststhattheRRNS(u;v)decodingcanbedesignedby

rstdiscarding d (d u v) outof the u residue digits,

whichis followed by RRNS(u d;v) decoding. Since the

discardedresiduedigitsandtheircorrespondingmodulido

nothavetobeconsideredintheRRNS(u d;v)decoding,

thedecodingprocedureisthereforesimplied.

5. APPLICATIONSOFRRNS CODES

Inadditiontothe abovementionedapplications ofRRNS

codesforself-checking,error-detectionanderror-correction,

ausefulpropertyoftheRRNSisthatanRRNScodeword

doesnotchangeitserror-detectionanderror-correction

char-acteristics after arithmetic operations, such as addition,

subtraction and multiplication [7]. This property can be

employedinordertosupportfault-tolerantsignal

process-ing. Let us invoke an example, in orderto augment this

property.

Example4 Let us consider the RRNS codes based on a

RRNSusingnonredundant moduliof m

1 =4; m

2

=5and

m3 = 7 as well as redundant moduli of m4 = 9; m5 =

11, m6 =13 and m7 = 17. Table 1 summarised a range

of typical decimalintegers and their corresponding RRNS

codewords. Letusnowconsideranoperationinthedecimal

integereld:

Y =5Y1+Y2Y3; (4)

whereY1,Y2 andY3arepossiblyfromdierentsourcesand

mayincludethetransmission andprocessingerrors.

How-ever,duetotheinherentpropertiesoftheRRNScodes,the

operations inEq.(4) can be carriedout asfollows. Firstly,

before we evaluate Eq.(4), Y1, Y2 and Y3 are rst

error-correction decoded. Since four redundant moduli are

in-cluded, up totwo residue digit errors in Y1, Y2 or Y3 can

be corrected. Secondly, Table 1 is used in order to

eval-uate Eq.(4). Thirdly, since an RRNS codeword does not

change its error-detection/correct ion capability, when the

codewordsaresubjectedtothearithmeticoperationsof

addi-tionand multiplication, theresultantRRNS codeword

cor-responding toY in Eq.(4) can be further error-correction

decoded, inorder toremove theerrors that mayhave

hap-penedintheevaluationofEq.(4). Notethatnoconventional

codes, such as Hamming codes, BCH codes and

convolu-tional codes have this property, since all these codes will

subjectedtothemultiplicationoperation.

Using theoperand valuesof Table 1, let Y

1 =X

1 =1,

Y2 = X2 = 2 and Y3 = X6 = 50. For example, the

RRNS codewordcorresponding to5is (1;0;5;5;5;5;5)

ac-cording toTable 1. It can be readily shownthat the

code-word corresponding to the operation in Eq.(4) is given by

Y =51+250=105,yielding theresidue-sequence of

(1;0;0;6;6;1;3) - provided that no more than two residue

digit errors occurred in Y1, Y2 or Y3. Moreover, even if

tworesiduedigiterrorsoccurredduringthelastoperationof

Eq.(4),andhencetheabovecodewordchangedto(1; ~

1;0;6; ~

3;

1;3), i.e., r

2 and r

5

are in error, by using RRNS

error-correction decoding, thevalue ofY =105 can still be

cor-rectlyrecovered, sincefour redundant moduliwere invoked.

Let us nowdemonstrate anotherpotential application

of the RRNS codes inthe context of the generic

commu-nication system depictedinFig.1. Inthis framework,

re-dundancyhastobeaddedfortheimplementationof

fault-tolerantcomputing(orsignal-processing),protectionof

in-formation overboth wiredand wireless channels, inorder

to achieve reliable processing and communications.

Con-ventionally, the encoding and decoding at dierent stages

hasbeentreatedseparately. Theencodedcodewordsofthe

wired channels have beenhistorically decodedinorderto

remove the related redundancy, before forwarding the

in-formationtotheencoderoftheairinterfaceofFig.1. The

informationattheairinterfaceisthenoftenre-encoded

us-ing adierentchannelcode,inordertoimplementreliable

transmissionovertheassociatedwirelesschannels. Further

additionalencoding/decodingoperationsmaytakeplaceat

other interfacesof a globalcommunications system, as

il-lustratedinFig.1.

However, with the advent of RRNS codes, the above

mentionedencoding/decoding procedures canbe

substan-tiallysimplied,sincetherequiredredundancyoftheentire

globalsystemcanbejointlydesigned. Letusinvokean

ex-ampletosupportthisargument.

Example 5 Withreference toTable1, letus assumethat

the fault-tolerant terminals use only one redundant

modu-lus forthe self-checking of theassociated arithmetic units,

employingforexampletheRRNS(4;3)code. Letusassume

furthermore thattworedundant moduliare requiredfor the

protection of information over the wired channel section

of Fig.1, employing the one-residue digit error-correcting

RRNS (5;3) code. Finally, four redundant moduli have

to be employed for the protection of information over the

low-reliabilitywirelesschannels,using thetwo-residuedigit

error-correction RRNS(7;3) code. Let m1, m2, m3, m4,

m5,m6andm7bethemoduliinvokedintherequiredRRNS,

where m

1 , m

2 andm

3

are thenonredundant moduli,while

m4,m5, m6 andm7 aretheredundant moduli.TheRRNS

codes protecting theentire system ofFig.1canbe designed

asfollows.

Step 1: The fault-tolerant terminalNo.1 uses moduli

m

1 ,m

2 ,m

3 andm

4

,inordertoformtheRRNS(4;3)

code-words,whichareexpressedas(r1; r2; r3; r4),wherer1,r2

and r3 arethe nonredundant residuedigits, while r4 isthe

redundant residue digit. Following all the associated

sig-nal processing operations tobecarried out bythe terminal

-whichmayinvolveadditions,subtractionsand

(5)

the fault-tolerant terminal forwards the RRNS(4;3)

code-wordstotheencoder of thewired channels. For simplicity

theoutputcodewordsofthefault-tolerantterminalNo.1are

still expressed as (r1; r2; r3; r4), which will also be used

throughoutourfurther discourse.

Step2:WiththeaidoftheBaseExtensionalgorithm[2],

thewiredchannel's encoder- namelyEncoder 1inFig.1

-computesanadditionalredundant residuedigitr

5

based on

(r1; r2; r3; r4)and(m1;m2;m3;m4;m5),inordertoform

theRRNS(5;3)codewords(r1; r2; r3; r4; r5). Then,these

codewordsaretransmitted overthewired channelinFig.1.

Step3: Thewiredchannel'sdecoder-namelyDecoder

1 in Fig.1 - receives the codewords (r

1 ; r

2 ; r

3 ; r

4 ; r

5 ),

whichareerror-correctiondecoded. Thecorrectedcodewords

are expressed as (r1; r2; r3; r4; r5), whichare forwarded

totheencoderof thewireless channel.

Step 4: Similarly to Step 2, with the aid of the base

extensionalgorithm[2],thewirelesschannel'sencoder

com-putestheadditionalredundantresiduedigitr

6 andr

7 ,based

on (r1;r2;r3;r4;r5) and (m1;m2;m3;m4;m5;m6;m7), in

ordertoformtheRRNS(7;3)codewords(r

1 ; r

2 ; r

3 ; r

4 ; r

5 ;

r6; r7). Then, these codewords are transmitted over the

wirelesschannelof Fig.1.

Step 5: After the wireless channel's decoder received

the codewords (r1; r2; r3; r4; r5, r6; r7), the codewords

areerror-correctiondecodedandtheredundant residue

dig-itsr6andr7areremovedfromthecorrectedcodewords. The

wireless channel decoder then passes the RRNS(5;3)

code-words to the wired channel's encoder, namely to Encoder

3in Fig.1. These RRNS(5;3) codewords are expressed as

(r1; r2; r3; r4; r5).

Step6:TheRRNS(5;3)codewordsaretransmittedover

thewiredchanneltothewiredchannel'sdecoder, namelyto

Decoder 3inFig.1.

Step 7: After thewired channel'sdecoder receivedthe

RRNS(5;3)codewords(r1; r2; r3; r4; r5),error-correction

decodingisinvoked,inordertocorrecttheresiduedigit

er-rors and then the redundant residue digit r5 is removed.

The resulting RRNS(4;3) codewords (r

1 ; r

2 ; r

3 ; r

4 ) are

thenforwardedtothefault-tolerant terminalNo.2.

Step8: Thefault-tolerantterminalNo.2nallycarries

outitstasksusingtheRNSrepresentationofitsoperandsin

theformoftheRRNS(4;3)codewords,invokesself-checking,

andnallyoutputs therecovered information.

In the upper branch of Fig.1 the wired channel's

en-coderonlyhadtocomputeoneadditionalredundantresidue

digit,incontrasttoinvokinganindependentencoder,asin

aconventional system, where eachsection of the

commu-nications links is protected independently. Similarly, the

wirelesschannel's encoderonly hadto computetwo

addi-tionalredundantresiduedigits,incontrasttocomplete

re-encodinginindependentlyprotectedconventionalsystems.

Inthe bottom branch of Fig.1, the encoder of the wired

channel,andtheencoderofthefault-tolerantterminalNo.

2-whichisnotshowninthegure-canbetotallyremoved,

while inconventional independentlyprotected systemsall

theseencoderswouldhavetobeincluded. Basedonthese

facts,wecan concludethat inconjunction with

appropri-ate joint system designusing RRNS codes, the

complex-ity of theerror-correction/detection sub-systems inglobal

telecommunicationsystems can bedecreased. Our future

workinthiseldwillbebasedondesigningsystems

invok-ingtheprinciplesproposedinthiscontribution.

Fault-tolerant

terminal

Air

interface

Wired

transmission

2

Decoder 2

Decoder 3

Encoder 3

Fault-tolerant

terminal

Wired

transmission

Air

interface

1

Encoder 1

Decoder 1

Encoder 2

channel

Wireless

Figure1: Atransmissionandcomputingsystemframework

includingfault-tolerantterminals,wiredtransmissionsand

wirelesstransmissions.

6. REFERENCES

[1] R.W.WatsonandC.W.Hastings,\Self-checked

com-putation usingresidue arithmetic," Proceedings of the

IEEE,vol.54,pp.1920{1931,December1966.

[2] W.K.Jenkinsand E.J. Altman,\Self-checking

prop-ertiesofresiduenumbererrorcheckersbasedonmixed

radix conversion," IEEE Transactions on Circuit and

Systems,vol.35,pp.159{167,February1988.

[3] L.-L.YangandL.Hanzo,\Performanceofresidue

num-ber system based DS-CDMA over multipath fading

channelsusingorthogonalsequences,"EuropeanTrans.

onTelecommunications,vol. 9,pp.525{536,November

-December1998.

[4] L.-L.YangandL.Hanzo,\Ratiostatistictestassisted

residue number system based parallel communication

systems," inProc.of IEEE VTC'99, (Houston,USA),

pp.894{898,May1999.

[5] H.Krishna,K.-Y.Lin,andJ.-D.Sun,\Acodingtheory

approachtoerrorcontrolinredundantresiduenumber

systems - Part I: theory and single error correction,"

IEEE Trans.CircuitsSyst.,vol. 39, pp.8{17, January

1992.

[6] J.-D.SunandH. Krishna,\A coding theoryapproach

toerrorcontrolinredundantresiduenumbersystems

-PartII:multipleerrordetectionandcorrection,"IEEE

Trans.Circuits Syst.,vol.39,pp.18{34,January1992.

[7] L.-L.Yang and L.Hanzo, \Coding theoryand

perfor-manceofredundantresiduenumbersystemcodes."

sub-mitted to IEEE Transactions on Information Theory,

1999.

[8] L.-L. Yang and L. Hanzo, \Residue number system

arithmeticassistedM-arymodulation,"IEEE

Commu-nications Letters,vol.3,pp.28{30,February1999.

[9] L.-L. Yang and L. Hanzo, \Residue number system

based multiple code DS-CDMA systems," in Proc. of

IEEE VTC'99, (Houston, USA), pp. 1450{1454, May

1999.

1

1

References

Related documents

A Multiple-Criteria and Multiple-Constraint Levels Linear Programming Based Error Correction Classification Model..

[3] proposed a high level error detection and correction technique known as Horizontal, Vertical and Diagonal (HVD) code for multiple error detection and

Predicted subcarrier SNR (continuous line) for an given OFDM symbol and the associated residue error probabilities (vertical bars) for four selected ARRNS code words for

Raptor Codes for use in Opportunistic Error Correction Timon Zijnge Introduction LDPC codes Fountain codes Results Future work. Error decoding using

“Biff (Bloom Filter) codes: Fast error correction for large data sets,” in

The proposed technique maintains high code rate, provides multiple bit error correction capability and can best be implemented as hybrid automatic repeat

In this paper a proposal for a new coding scheme for error detection and correction using RRNS instead of the current error control codes used, where RRNS

1) Study of Digital filters and its parallel design over error detection and correction. 2) Designing of FIR filter with same response. 3) VHDL implementation and verification of