GB2336277A - Preloading for equalisers - Google Patents

Abstract

A decision feedback equaliser 50 is preloaded with filter coefficients for training purposes. A data signal is received which includes a training sequence. A sample value of the training sequence are correlated 34 with a stored equivalent training sequence and the peak correlation value is determined. A first set of correlation values is preloaded into a feedforward portion 42 of the DFE for use as initial feedforward filter coefficients before training and the outputs of a feedforward filter are used as initial feedback filter coefficients in a feedback portion 48 of the DFE.

Description

2336277 1 PRELOADING FOR EQUALISERS Field of the Invention _le -vention

relates to a preleading tech .niclue:c-al--a-:):ive eaualisers of the type used -'n radic modems w--e-- e the equaliser taps are preloaded before equalIsation Background to the Invention - I- 1 z The invention has been developed -in the context of a wire-"ess Asynchronous Transfer Mode (ATY) nezwerk--'nQ nfras-ruezure which is capable of suc:)or---ng -.ult--'-mec 1 iia data traffic at high bit rates in local and wide areas, but the "invention could also be used with other netwo-kinc:

nfrastruc,iures.

in radio communications systems, the ra-sr.ii:ted s-'gnal travels from transmitter to receive-- over a c'-,anne-' cons.-s----na of a number of different paths, '..-nown as.mu',.:ii'pat'-s. The transmitted signalcomponent- tra-,.,e--'--nci alona a particular path experiences absorption, re-'i-eci-4on =- scattering by any objects located in that pat. This causes the signal components arriving at:he rece---,,-er ic 'ave --ferent amplitudes, phases and dela,,,?s so that they 4nterfere with one another. This interference -'s known as multicath fading, and results in Intersymbol (ISI) at baseband. ISI describes the spreading out of the data symbolls so that components of past and -future sv.mbols are supe-rimposed on the current symbol, '.--us causIng svmbc-' errors.

A well-known technique for reducing the effects of iSI is adaptive equalisation. Equalisation is a process whic'- uses filters to remove ISI caused by distortion In the channel. Adaptive equalisation makes use of filters which adapt to follow any changes in the channel dis.7-orz'Lon. For channels with severe distortion, a 1 Decslon Feedback Equa' iser Is to al data I- -e e,sualiser data ie--ec--icn bejins. ha-,,,e a cer--a-'n =nimum length c c e f f _ the --a--'ning sequence depends -enc--- 3 the --'--ters used of ihe channel distortion and algor--'t'-m -sed. T'--c-7e-'--aci in ie-ms f ---;znsT.-czs svr-.bo-'s wr,--c"- would otherwise liser to ensu e t-ai equa in ec -ne len-t- of on factors sue' as the t'- e e a I i s e r, the s e v e tv --e ---i-,i-ce of --ra--n--ig se--,-ie-ce --e,,Drese-.ts an --ate, since it, uses up be use' for data s,,,mbols.

-rihe=,ore, the tIme taken for 7e e-j- a ser to tran causes a de-la,,, -n the deec-:i-ic,- of each data packet, irr,e 'or ec- -,acket.

wiz'- hit rate radio co,- 'c; ---ticn svste.ms as trans-it snor,: cat racke.:-tzz and recu---e ,:-re-,ess A Jelays data 7=_ be as low as p-ssible.

b th he data c),,,,=-head and the sys-ems, 0- I n delay caused by the use of a trainIng seque=e iin.cose severe on performance. -_herefore, ::here is a need t reduce the -'engh of the zraLn-ng se,:.-e-ce and also = reduce the de-'a,,,; in train-n- --'-e equaliser.

Preloading the e-,ma.:e of:he tap s_ as close as -,)oss-'bie that the inItial esz--mate eciualiser taps allows an coefficient values = be r-.a,--ie is orverged va--ues. reasonably ac-urate, the tap coefficients will be almost converged before ea--ial--sation begins. --his results -M reduced convergence iraining sequences may be c-f:'-e equaliser.

which means t at shorter used degrading the pe 3 1 1 C i -1 3 C However, exs---ng methods for the ta values are comeutatona'ly makes zhese methods unsuitable for a-n,-)licaticns -ec,,,- ri s'-ort -,,.irn-around times. Therefore, there is a need a sim-cle method for calculating the initial equa-l--ser coefficient values.

Summary of the Invention

A preferred embodiment of the present invention provi-----s a method for preloading equaliser taps t --e ste::)s of:

obtaining first estimates for the equaliser feedfo_rward tap coefficients from the output of a correlaicr; the first est'.ma.:es:io give Improved, generating feedback tap coeffic--'ents by feeding the ouicuz of the correlator as data sa--.i;:ies tnroug- the,c,reloade-- feedforward filter.

This and other aspects of the invention are de-'--ne--- more precision in the appended claims, to which reference s'-ould now be made.

A. jreferred embodiment of the invention w---7-11 now be described in detail by way of example, with re-ference to the figures in which:

Figure 1 shows the structure of a data packez and preamble model; Figure 2 shows a block diagram, of a typical rad--o channel model; Figure 3 shows a typical channel limpulse response; Figure 4 shows a block diagram of a correlation and svmbc-' timing extract circuit for use tocether w-t'- an equaliser. Note: only the baseband signa 1 -! processing is considered here. Frequency conversion and -f-'-'ter-Lna are omitted for clarity; 1 Figure 5 shows feedback eua-'iser; a cck of a- known dec is lon Figure 6 shows the res'--=p--nc prc:-:-les to be used in an embod-men.:

--resent -r.ve:t-c)r.; Figure 7 SCws a b-'ock diagram of an ecuai--se-Figure 8 s'nows a j--acra--i:5-f --.rie of ti'-le -::--rcuii c-' 7; and Figure 9 shows a flow cha---- of the method for on -ee(-.-o--warci in an e.--ibedii-.en.:

Detailed Description of Preferred Embodiments daza s-,,-s-zems ransmit data gr,uoe-- to--etner in packet-s.

in:-u--es a shcr: prea=ble sequence which Is known to transmitter and receiver 'see I 1). T'- e p r e amb 1 e as eaualiser irain--'ng at zhe channel of i-e Ficure 2 with sampled -m.,uise -as a transmitter 2, a receiver 6.

The cannel rece-ver.

is transr.,tted over a radio iv,e shown in t-e mode'- resconse h.

- _ve -c-,se channel 4, ad,-iit The channe-' has a samp'ied -mpu.se res3onse of i-e tvpe shcwn -'n 3. Channel response -'s give- by:

h = [h 7 '... ' _1 - 7 - The samoled _-eceived sign-a-7 at iii-.e 'j = L -1 yk= E X k-I - h; - n, 1=0 where 1 1 te i-eriod of t-e transmizted svrobols h 21, 2 z t, e the the sampled received signal at time t: Jntegers transmtted sv-nbo-l at time tsampled channel impulse response -jectcr -he sam3Die of the vector h noise sample at time t = kT matrix transDose.

The channel impulse response is characterlsed by echoes (precursors), the main pulse 'curse-), and following echoes (postcursors) as shown in Figure 3.

channel where there is a direct, or (!OS) 2ath between the transmitter 2 and receiver 6 is known as a 11'ne-cf-sight 'LOS) channel, and will have a dominant i-..a-in pulse, but no precursors in the channel response. A channel having no -1ine-o--f-s-.ght the transmItter and receiver is known as an Pat' between obstructed or non-!OS channel, and may have large precursors -Ln the channel impulse response.

At the receiver (as shown in Figure 4), sar-,-:)'ed by an analog-todigital (A/D, conver-er 60 and into a correlator 62. The corre-lator 62 computes the complex correlation product between the data sequence and a stored copy of the known preamble sequence.

the data is The magnitude of the correlator 2utput at tJ-,,e -- is:

A (t) = RJR,_ where p xx,yi (cross correlation of received d 1 =1 samples with preamble se--uence X. ith symbol of the preamble sequence V, ith samnle of the received signal n length of the preamble sequence complex conjugate operator.

aza _ne peak value = '-.-e outpu-z is L.a,--kei bv a timinc ex--r-c-c--rcu--:: wh- s sho,.,7n in Figure i--n,-lnc is the data Feedback 7-,2ual--ser -8 of the z,.,,pe sl-.,-w- Feedback 7-L-'ter associated adde--s, and sift registers.

zlcnc whic input signal --'-e input sam.nies are fed the Input to each ---1e-'a,,,, 10 to a respecz--ve multiplier 12 -n ;eig'-zed a c. and then L 1 14 whose out--,:)u-- -'s denoted z_ The nurnose the ---:'F fIlter is a--)prcx-4T,a,ie a matched filter for the rece--ve,- s--gna--, and seco-dly to mcve non-m--'nii-.u.m phase zer-cS This ensures in3--'--e ihe unit hat the cascaded channel and 7-'E' filter have a combined i-m--u--se response d hase.

Sv-,r,bol Interfe--e-ce IS'-', S ymb -- _ S.

Z - S m 1 n -- Mu, m combined ----i:)ulse rescense is:

d = h 0 c convoluticn.

F3 filter -'s to remove Inter- due tc crev-icus-',, detected It cor7,-r4ses a set- --f delavs 10 w-ich re--e--'-.,e the t, - - outiDu-sa=)7Les of the DFE. These s--mt,les are,e- Ghted in.

respective multipliers 16 by -i'-er, s,,i=e-- n an =dder 18.

W'-en the vector of sampled data y is input to the -)FE, M samples are nrocessed to cro,--Juce one Cutput symbcl, - z - E w',ere -,- is t e number of taps in the FE' f ilter Th e nur- lter is 3.

of taps in the FB f- The outcuts of adders 14 and 1-8 of the "."F and FE te-s are combined in a further adder 20 befcre --c a de--ec-zcr 22.

The de--ector uses the output sar.ple z. of.7i-e adde-r make a decision on the transmitted syrr.bol at tIme Th-'s decision is An error signal e, =z... Is ceneraze,-- by subtracting the equalised value z,... fro.m the out--ut of the training sequence 26 during the no-de and -from the output of the detector 2' during decis-on-directed mode.

known preamble tr n"ac:ed ai aining sequence 7s the be----'nning of each data packet, to prcv'a-de ce-- fect values for _R., instead of using the detector CILitput decisions. T-- 's is known as the training mode, and allows the eaualiser taps to at least partlally converc:e before the, actual data begins. training mode, a switch 24 switches the ou--nui, and hence the input of the FB filter to a training se--uen:e store 26 which stores values equ--valeni ic zhose received in a data packet preamble.

once the preamble sequence is finished the detect-or 22 decisions are used for 9,. This is known as directed mode.

The error signal is used to update the- za-c f' 'en.Es o' the FF and FB filters using an,-:n---ate cce --c: - L - algorithr. such as Least Mean Squares (LMS) or Recursive Least Sauares (RLS). The time taken for the coefficient ta:)s to settle within a specified range of tl-e--'r c;:t-i-T.u-n values is known as the convergence time. convergence time -is critical in packet data transmission systems since longer convergence times necessitate longer preamble sequences.

of te preferre,:i e-rbc(--.ment is shown 7gure -1, and the cc)era--cn of: 7he preferreei coeration --f te Ficure data Qacket.

ea:.- - - d----',z-;---" values v, bv te SamPler and correlaicr 31 com-cutes the product -211, be,,,jee- sample vector y, and the known tra--n-,-g se=ence, and the values A, to the Exirac.: 3-..

T'-e Extrac-7i Circuit, 3-'- se-rc-es the larues.:

t'-e largest:)eak in -the data - R:, -..R,:,,. a nd R-. te a Ta;: Resnaoe Circuii 38, and the set f a' S R t 0 B u f f e 0 and echceS, and hen uses an;zcnrc--riate res'-a,,--)i-na cro-file to res'-ape -the.,-a-"ues R_. to prcd,.,:e as described in the flow char-- of -7i,2:ure -0.

The values .. 1 are t'-en -oade--- i'-e ia,.:,s ef the FF fJ ter 42 from 'he 7,a-., Res-ape Circui.:

S-,7- 4i is then moved nt-3 oosto- --2 and Swicn is moved inte position p! The values R, are shifted out of Buffer B 0 t-ro,,.:ch Sw_1-,--'- A 4- and i'--- -7F filter -2 in the same wav as data samcles. TI- e out:)ui 'rc.-n R,1-,... are the values b=Llb., 42 resu-L7ing from input l,-z-",-ies f are Icaded into the tacs cf: the FB filter 48 t'--rouc'i Switch 5 4'0 in position pl. Switch A 4 is then moved to position pl, and Switch B 46 is moved to position p2. At ihis:D,-JLnt, '-ct'- the -7F filter 42 and --'-e FB filter 455 are)re-'caded an-- the e.-ual-ser -'s ready to accept data. T'-e sampled rece--ve-- through f rom rest c)f ihe data packet.

The Preferred embodiment of the in7jention cc.-n-cr-ises n - for adaptive eauai-st:-is 7Gr use in :ecn, _Laue mc--;ei-.s as described with reference to r---cure a -I-- c e.

The ejualiser taps are preloaded before ebegins, using:he following method:

---st estimates for the feedforward tac coefficienz values are obtained us-'.-g a channe! estimate z.he correlator of Figure 4 as acts.--ncn the z--ain--ng sequence; the first estimates for the -=ee---crwar--- ta;D are..reshaped to give improved es-.ir-L.ates cf oct-irr.u-.

-,.7a-l,es; tap values are generated by runn--'-nci the ccrrelator output through the preloaded feedforward taps.

m-his suboptimum approximation is sufficiently accurate to reduce the equaliser training time, vet co,n,--u--ational-',v simple enough to allow high speed oc)eration. Accordingly, the convergence tIme is by preloading the equaliser tap coefficients wit' initial values -w'nich are obtained using a simple ac--rcx--'-mat-on method.

In an alternative implementation en-ly the coefficients are preloaded usIng the --es'-aDed esi-'-,--ia-.es. The feedback coefficients are t',en derived in the usual:-,,arner using the training sequence. This is not as good a method but may be used if the sma-'I time delay required to calculate the FB filter taps is unacceptable.

s gn a 11 y i s t h e ri s h i _f t e d c f the Modified DFE SC, and t'---s time onwards as in a ec-ven--i'ona'L ---7E for the u a I -- s a.:: - i - Subolptimum Approximation for the FF Filter Taips met-'-,.ds in the art the F--- filter --a-- c from an estimate c-an-e-- the -,;.a-',--es -for tne ---F filter iaiD ece:---1cie-ts c filter with Impulse res,-,r.se fellowed bl, a resha,.:-4.nu filter g.

7he r-,.at-c'-ed --es:)c.nse is ce-rivea irjm tne --crre' outputs as fc11c.,,;s:

These esti-.a.:es are reliable prc-,i--ed the alzocrrera--ion function =f the tral-ning se----ence is -mu"a.zion -es-.:'ts have shown that te res'-a;Ding --r -a-,,; a--prox- ma7:ed bv a sc -j- 'a 7_ zo, r a pp ed o each c_f tne matched fil--e-. The scaling factors can be d---,,ided Into two groups: scaling factors -CS cnannel and scaling fac-crs.0o- a -on-L-S -2r a - - -h a r -e-. The vector of scaling factors t, may be,.ised may be used -for anv non-----S c-anne.' '-av-'-g sign-'L-fican-: in --'is channel ir-,,p.-,lse response. are a set of heuristic r,--les that are.. ised to reshaping rofiles. however, t-ev are onl,,; gene-ra' rules and therefore do not give a un-jaue solution for the reshaping profiles. The rules are summarilsed here:

(ai The FB filter of the --FE can TZ1 from past symbols (th-'s corresponds to the impulse response). Therefore, zo nave a-v precursors in the combined response of c'-annel and F-7 filter d = h o c. This can be achieved '--v -orcinc a-"! the FF filter taps before zhe reference tan c zero, thus all taps of the reshaping profiles be-fore the reference tan are zero.

The reference tap should be a scalar of the largest tap of the matched filter w'-ich is achieved by setting the re-ference tan of the reshaping profile to a value of un-,.zv a -i h ou a h t'-is 7,a-ue mray be scaled if necessary).

For a non-LOS channel, ii Is des-rable to --he channel.

capi:ure signal energy from precursors of T-t is obvious that for any particular channel the -nii---al %,alues -found using these reshapIng profiles wi-', However, on the average, the iniiial tap -ct be exact. val-,es are close enough to the f inal converged -,,-a-', ues to sign--'fican.zlv reduce the training time.

The exa= values chosen for the sca-"--'nc: factor vect,,:rs f, -,-iav vary slightly according to the part--'cu-"a-ecrua--ser configuration, but once chose- t-ev remaln ccns--an'i. Examples of scaling factor vec-iers ú. and ú. for an eaual--ser wlith nine feedforward taps are given be-',c,,,; :

2 _5 LOS CHANNEL f, = [0.3, 0.0, 0.0, 0.0, 1.0, 0.2, 0.2, --2, 0.2] Accordingly, the reference (centre) tan is multipli by a scale facior of 1.0, taps before the reference ziap are mult-4plied by a scale factor of 0.0 and taps after reference tan are multiplied by a scale factor of 1'-2.

NON-LOS CHANNEL ce r t --e, t ap -4 s rpu, t,' P, e d b S -- a 1 e f a c -- j --esu-'ts have s'-own that t-e -actor ú, and f, as 11sted abo-,,-e,:Tood nerr.-n.arce -or a range of different radio channels. T rem07:e ISI due 7-c prev--'cdsl-,.7 de--ece-- s-j--7,bols. Thus, th.e 7B iaps can cnlv r-:-,r-L.,-ve zost-,-,--sors of te -r-.P"-se _response d = h s c. Tis --,Lean-cz that -for radio c'-an-e' s exhihit severe jh-=se and a-,-i-liucie sca"e factor vector s'-cul,-4 attemet ic reshaue ihe ma-_ced creicursors, in the --cmb-,ned imnu-'se respon-se d. -7 c r _ channe-" with a main Pulse, it Is necessar,,.7 to to decrease mac.,n-Liu,--'e cf precursors an_ Postcursors of the rratc'-ed --'-i-'ter.

L-S ThIs 1eads to t-e numerical va-7--,es z---jen above for tl-e s--ale -factor -,.,ector f,. For a nen-L--S 2hannel w-iih sic:n-i-f--cant power in the precursors of the channeI --'mr-u-7--e response, ---- is necessary ic increase the magni7-u---Je of precursors of t-e matched filt-er, and --c decrease --'-e of pos.cursors. This leads to the numerIcal v-;:1-',ues give.- abo7.7e for The scale factor -,,ec-tors f, ancl f, may be theuch-of as reshaping profiles for:he matched filter as shown in F-'gu-re 6.

Preloading t'-e initial values for t'-e ewual-,'ser -tap takes Diace at the s--ar-- c-f e-,,ery received burst of data.

However, the values cf t',-e resha-----g profiles f_. and f. are fixed. The ecua-'--'ser selects the appropr-ate reshacing pro-file, either f. or on a burst --v-'-urst basis.

s-m:),e method for realis-ng t'h_-s res,aping is to start w-'-th the matched filter response R...(-.) = h-T'--- zor--e'Lal--or output RX, gives reliable estilmates of hprovided --',at the autocorrelation function of t-e traIning ceuence is flat in the region of interest arcund the central r:)eak. The next task is to into one of two groups, LOS or non-LCS. Using the estimated matched filter values measured bv tne correlazor. This is done by calculating --'-e power --- each matched filter tap coefficient value where:

the c-annel 1 z 3 _ Power, = lc.' f or -4 7, = ' 1... 1. ' Note t-a-. the values of icl' teiv ava-'a'-!e are ---imed-a sin'ce t.-iev are equal to the correlator va-',-ies and t-herefcre the values 1 C112 need not be calculated.

Denote -he laraest tap as c,, where 1 < m, < il. T'-e c_ referred to as the reference tap, or 7main taj,. '!'-.e ccrrelator should usually adjust the bit timing so that the reference tap is positioned at the centre c-E the flter. However, the reference tap may alternatively be moved to a position offset from the centre of the FF filter -'if -i-- is required to minimise the:-,,,-irber of FF 'i' er tans.

-L - - Now sum the nower of the coefficient taos occurrina before the largest tap, which is the estimated power due to leading echoes (precursors):

Power in leading echoes M-1 p EI ' leading Ci 12 i=l cower cfi tans wrich is zne eSt- --ea power d-e _o = 'v Po.wer --'n echoes p PreloadincT FB Filter Taps Th-e -7E f-4-'ter z-aps sho uld converge ic b - h 0 c which of t'-p:hanne' i=,u__se response with t-e -7F f_1ter. _s possible tc an estimate for -.-ectcr b by prelca,--iin,-- the F-F f-J-"t-er --ap values [ C, and then running the se- of --onjucated matched channel imp._-Ise -response h t-r.-igh the -7F in t-e same manre-r as --;aia -- a ' u e s c -1 th -- F-E' _f _ 1 -- e r w _ 11 t- e n iter e s. 7-he ci-,tpu-- - i s e d ve r v s v '- -v a t.E a c, i n c: t- e c o n j---a a - e d Ir. a -: c e,' rea- f -i -I t e r packet estIma be cr i ns - CY'] at the -front of the data t-her. t-rough --he 77 filter tans. These FB taps are the- used to preload the 7E b = 1, 1-, ' L -: 1....._ before equallsation Algorithm for Findincr Initial FF and FB Tap Values ^ow-ng alcorJ7:hrr, su-Lmar-'ses the --roce,-.-re for -c-- -F:4ndina the Jmi-a_,,,aues of the F7 and FB filter ---ns:

-, z calculate P-e,=-., and 2. _. F i EI > p:ND > THEN channel = non-LOS (lead-ng ec'-c' EISE channel =!OS (Ifollowing echo) J non-!OS (leading ec-o channel): mu-L--ip-v the mac,ed fiter tans ',-v the vector of leading echo scale 'Lacic--s f, to obtain ^ 5 - !OS -fo!-7ow-ing echo channel): t-e taps c.,, hy the ve,,::z-- of ec'-c scale factors f, f-A to c' pre'cad filter taps with the resha c c 1,41 the values R, f---"-,-er. Output b [b,, 5.

ced -,,a-",-,es te.77 G preload 7B filter taps using calculated values b 'b., The values of Threshold, and components of f, and f, are sma-l real numbers, which are chosen -for t-e.oar.i,cu ar equaliser configuration, and remain cons-an:

Sampling Rate 1 The embodiment described herein includes, '--uz -'s nct nited zo, sy.Mbel-spaced sampling (whe-re t,-period Is ecrual to the symbol period,. Fra--.i--'onal'L%,spaced sampling may also be used where Q samples are taken cer svr-,,bol ceriod, and where Q is an integer.

Computational Complexity order to quantify the computational advantages of --he oresent invention over existing techniques, the n.,mber of mu-.iol--'caiions and additions are totalled f--r each listed s method. In a'! cases, the number of operations that required to calculate the initial values for the F9-7 taps cnlv, where Y is the number of FF tacs. -7cr nurnoses of comnarison, it is assumed that the number of FF ta; :,s i':, is equal to the length of the channel impulse response In the suboptimum approximation method of the present nven.--'on, M additions are required to sum the cower in 1;5 :he and -- 1 e:- ces, and 1,c 11 Eat lons ae --e(u---ed each ---,zer -,a' -,ie c)nstan- scal a r e sh ap- n g Method for finding Complex Multiplies Complex Adds initial tap values -0d, on Table 1:

It can be seen fror.

aop--:x-ma.::--o- mezhcd of T ab- 1 e -- i 1 a i t e s u b c i. t i mul m te presenE in-,,e-tin reuu-- 're-c -nve2:s-,cn Le7,,inson -:he z-e sub3ctir-,.lim appr,--ximaiic-- --ec--'sze--s and adde-vectors ú. and f, 2onsist of -he scale factor ---ethod and --'ne me.zhod -nav be zrees because are used const-an'z, rea' numders.

Claims (6)

1. CLAIMS:

   A method for preloading filter in a dec-L.s-lcn feedback equaliser (DFE) comprisIng the s-ie;--s cf:

   recelving a data signal which a tra-ining sequence for the DFE; correlating sample values of the -irair--'ng seauence with a stored equivalent training secluence; detecting the peak c:--rre-'az--or. -.7alue derived b,,., he and leading a first set of corre-'a.ion values a feedforward cortion of the DFE for use as filter
2. 2. A method according to claim 1 --'nc--ud-'na ihe further s--e:)s of:

   passing a second set of corre-7aiicn values zhis loaded feedforward portion cf the DFE; and fee..1:orward loading the output values of.7-he filter into a feedback portion of the D-77- for use as coefficients.
3. 3. A method according to clair. 1 or 2 inCluding he s.ze:) of reshaping the first set of corre'a-on -,,,a-"ues efore load-ina them into the feedfcrwar,- Dorticn of the DFE.

   Z,l - 1 -,
4. 4. A method according to claim 3 -'- which the reshapIng step comprises weighting the first set of correlation values in dependence on the amount of energy in the received signal.

   5. Apparatusto preload filter coefficients in a decision feedback equaliser (DFE) comprising:

   means for receiving a data signal includes a zra-inina seauence for the DFE; se,juence; according 5, further means for passing a second set of correlating -:a-!ues throug -'-e thus loaded feedfcr,..j;-:,rd filter P-ortien; and means for:f te --"ter Into a -eeciba--.'- f t-e C'FE fcr -;:-,t,-s ccordn.g to or AwQ a means -for res-.ao-n-- the firs-i sei of co--relation -,ja-',ues -- 'em. Into:he feedforward 3ort-n ---f -:he 5. according to - in -which:he resha---4-a means we--'zhts the set of --c-rre-'at-'.on -,,-alies in dependence on the amouni c.: energy n the recei7,ed Signa-'.

   1 Amendments to the claims have been fled as follow 1.

   decision -raining A method for preloading -filter coefficients in a feedback equaliser (DFE) comprising the stens of receiving a data signal which includes a sequence for the DFE; correlating sample values of the training sequence with a stored equivalent training sequence; detecting the peak correlation value derived b,,, the correlation; and - c 2 _z loading a first set of correlation values into a portion of the DFE for use as initial filter coefficients before training begins.

   2. A method according to cia-'mr-,i 1 including the further steps of: passing a second set of correlation values hrough this loaded feedforward portion of the DFE; and loading the output values of the feedforward - 4 1 f-lzer i ni ti a I begins.

   into a feedback portion of the DFE for use as feedback filter coefficients before training 3. A method according to claim 1 or 2 i=luding step of reshaping the first set of correlation values before loading them into the feedforward portion of the D--E.

   A method according to claim 3 in which the reshasing step comprises weighting the first set of correlation values in dependence on the amount cl' energy in the received signal.
5. 5. Apparatus to preload filter coeff-Jcie---s in decisio- feedback equaliser (DFE) c:omor-s--'-g:

   ihe 2-0 means for receivinc a data s-iwnal w',- i, c includeS tralni-o sea,,.ence for -.he DFE; means for correlating sample val, trai-inaseauence with a stored ec.j..va-'en-- train-nc seauence; means for detecting the peak correlation value derived bv the correlation means; and .oad.-'ng a first set c-- correlation values i-.ic a portion of the DFE for use as initial feedcrwar-- filter coefficients before training begins.
6. 6. Apparatus according to claim 5 further including:

   means for Dass--9 a second set of correlatina values through the thus loaded feedforward Dortion; and loading the output values c--- the feedforward fizer i-.zo a feedback portion of the DFE for use as n-t-al feedback filter coefficients before training becins.

   7,nnaratus according to cia--m 5 or 6 comprising first set of correlation values means for reshap ng the b-efore loading them into the feedforward portion of the DF --- - 3. Apparatus according to clai.m -7 ' 1..

   in which the -Z,l - res 1 naping means we-'ghts the f--rs-1 sez of correlation values in cier:)endence on the ar-L.--uni of energy in the received sicnal.