Note: Descriptions are shown in the official language in which they were submitted.
101520253O35CA 02264414 1999-03-05GR 98 P 1284Description of invention:Communications Receiver and Method of Detecting Data fromReceived SignalsThe present invention relates to communications receiverswhich operate to generate data from signals representative ofthe data which signals have been corrupted duringtransmission between a transmitter and the communicationsreceivers. In particular, but not exclusively, the presentinvention relates to communications receivers which operateto detect data from received radio signals representative ofthe data which radio signals have been corrupted duringtransmission via the ether. Furthermore, the presentinvention relates to methods of detecting data from receivedsignals which are representative of the data.In many known communication systems, such as mobile radiotelephone systems, the integrity of data communicated by thesystem is reduced as a result of naturally occurringdisturbances which corrupt signals representative of thedata, during transmission between a transmitter and areceiver of the system. For example in mobile radiocommunications, disturbances such as multi-path propagationand frequency shifts of radio signals which carry the data,cause inter-symbol interference and fading. In order tocommunicate data effectively, receivers of the radio signals,are therefore provided with means to mitigate thesedisturbances.To mitigate the effects of multi-path propagation, radioreceivers are known to use equalisers which operate tosubstantially cancel the detrimental effects of inter-symbolinterference and fading, caused by the multi-pathpropagation. An equaliser known to those skilled in the art,which provides a significant increase in the integrity of thedata recovered from the radio signals, is the maximum101520253035GR 98 P 1284 CA 02264414 1999-03-052likelihood, or near maximum likelihood sequence estimator,also known as the Viterbi equaliser. The Viterbi equaliseris well known to those skilled in the art as a sequenceestimator which operates to determine a most likely sequenceof symbols from all possible sequences of symbols inaccordance with a comparison between the received signals andall possible symbols which could form the symbol sequence.The Viterbi algorithm is well known, and so a detaileddescription will not be repeated here, however a moredetailed description of this algorithm, can be found in anarticle published in IEEE Transactions of Information Theory,May 1972, Vol. TI 18, No 3, Pages 366-378,âMaximum Likelihood Sequence Estimation of Digital SequencesG D Forney, Jr:in the Presence of Inter-symbol Interferenceâ.In some communications systems, such as for example timedivision multiple access mobile radio systems, an estimate ofthe impulse response of the communications channel throughwhich the radio signals have passed is required to facilitateTo this end,transmitted within a predetermined time slot,operation of the equaliser. radio signalsalso includeknown signals generated from a training sequence of knowndata symbols which when cross correlated with a locallygenerated reference version of the training sequence at thereceiver, provide an estimate of an impulse response of thecommunications channel. This estimate of the channel impulseresponse is used by the equaliser to form error signals independence upon a comparison between reference signalsgenerated from a possible sequence of symbols convolved withthe channel impulse response estimate, and the received radiosignals. By minimising the error signals, the equaliseroperates to substantially maximise a probability that aselected sequence of symbols is that which was transmitted.In rapidly varying communications channels, orcorrespondingly where the radio signals representing asequence of data symbols have a temporal length which is101520253035CA 02264414 1999-03-05GR 98 P 12843greater than a coherence time of the communications channel,the estimate of the channel impulse response from the knowntraining sequence signals at one temporal location in theradio signals may not provide an accurate representation ofthe communications channel at another temporal location ofthe radio signals. in the case of the knownmobile radio system the Global System for Mobiles (GSM), theknown training sequence signals are arranged to beFor example,transmitted in the middle of a burst of radio signalsoccupying a time slot. As such, although a version of thechannel impulse response generated from the training sequencemay be accurate for data communicated in or around the middleof the burst, the channel impulse response at temporalpositions displaced from the middle of the data sequence, maynot be accurate as a result of the rapid changes in thecommunications channel. In this situation, the integrity ofthe data detected at these displaced temporal positions willbe substantially reduced, as a result of the inaccuracy ofthe channel impulse response estimate used by the equaliser.As a solution to this technical problem, European Patent EP042545881 teaches an arrangement for a Viterbi equaliser,wherein a plurality of channel impulse response estimates areprovided, each estimate of the channel impulse response beingassociated with a corresponding state of a trellis modellingthe communications channel formed by the Viterbi equaliser.Each of these channel impulse responses is separately adaptedin accordance with error signals formed from a comparison ofcoefficients of each channel impulse response estimate andnewly determined symbols estimated by the Viterbi equaliser.As a result the adaptive Viterbi equaliser taught by EP042545881, is able to substantially maintain the integrity ofthe data generated from the received radio signals, despitevariations in the channel impulse response of thecommunications channel.As already mentioned, the Viterbi equaliser operates toprovide an estimate of the most likely sequence of symbols. .. ...............m-_u-â......._......â.-...â.........-...... -.. . .. .l0l520253035GR 98 P 1284 CA 02264414 1999-03-054corresponding to a burst of received radio signals. Howeverin some applications, a symbol estimator is preferred. Anexample of a symbol estimator is the Bahl, Cocke, Jelinek and(BCJR) algorithm,Decoding of Linear Codes for Minimising Symbol Error Rate,âby L. R, Bahl, J. Cocke, F. Jelinek, publishedin the IEEE Transactions on Information Theory, March 1974A symbol estimator such as theBCJR algorithm, to generate an estimate of the mostlikely symbol, on a symbolâbyâsymbol basis,the generation of soft decision information, which isRaviv disclosed in an article "Optimaland J. Raviv,between pages 274 to 287.operateswhich facilitatesparticularly appropriate where turbo-detection is used in asubsequent error correction decoder when the data has beenencoded in accordance with a forward error correctionalgorithm.rapidly changing,Moreover, where the communications channel isadaptation of the channel impulse responseas taught by European patent 04254588 herein before describedis also desirable. As such, providing a symbol estimatorwith a means for adapting a channel impulse response estimateto substantially mitigate the effect of changes in theimpulse response of a communications channel represents atechnical problem which is addressed by the presentinvention.According to the present invention there is provided acommunications receiver for detecting data symbols fromsignals representative of said data symbols received from acommunications channel, said communications receivercomprising an equaliser which operates to detect said datasymbols by calculating a plurality of transition probabilityestimates associated with transitions from first states ofthe communications channel to at least one subsequent stateof the communications channel from an estimate of an impulseresponse of said communications channel and said receivedsignals, wherein said communications receiver furtherincludes a channel impulse response adapter coupled to saidequaliser, which operates to adapt said channel impulsel01520253035GR 98 P 1284 CA 02264414 1999-03-055response estimate in dependence upon said plurality oftransition probability estimates, whereby a likelihood ofcorrectly detecting said data symbols is substantiallyimproved.The term state as used herein with reference to acommunications channel refers to a particular sequence ofdata symbols, the number of which symbols is determined bythe number of past transmitted symbols which affect asubsequent symbol at a receiver as a result of inter-symbolinterference caused by the channel.By adapting the coefficients of the channel impulse responseestimate in accordance with a plurality of transitionprobability estimates associated with transitions from firststates to at least one subsequent state of the communicationschannel, an increment by which the channel impulse responsecoefficients are updated is influenced by the transitionsmodelled by the symbol detector. The adaptation of thechannel impulse response estimate is thereby effected in away which avoids a requirement for selecting one of theplurality of transitions leading to a subsequent state, whichare not made with symbol estimators.The channel estimate adapter may have a data store in whichthe channel impulse response estimate is stored, and a dataprocessor which operates to adapt said channel impulseresponse estimate in accordance with an adaptation algorithmin combination with said transition probability estimates.The adaptation algorithm is provided to effect an update ofthe channel impulse response as the symbol detector advancesin time through the radio signals detecting the data symbols.Advantageously the channel estimate adapter may furthercomprise a plurality of data stores, each of which datastores is associated with a state of the communications1020253035GR 98 P 1284 CA 02264414 1999-03-056channel and serves to store a version of the channel impulseresponse estimate, and the data processor may operate toadapt each version of the channel impulse response estimate,in accordance with an adaptation algorithm in combinationwith a corresponding plurality of the transition probabilityestimates for transitions from a plurality of the firststates to the state with which the channel impulse responseis associated.By providing a Version of the channel impulse responseestimate for each state of the communications channel, andarranging for a data processor to adapt each impulse responseestimate separately, a substantial improvement in theprobability of correctly detecting data by a data detectionalgorithm which uses the adapted channel impulse responseestimates is effected. This is particularly true for rapidlychanging communications channels.The adaptation algorithm may operate to generate at least oneadaptation factor with which coefficients of each version ofthe channel impulse response are adapted. The data processormay further weight the adaptation factor by a combination ofsaid corresponding plurality of transition probabilityestimates.The data processor may further operate to adapt each versionof the channel impulse response estimate by generating anintermediate up-dated Version of the channel impulse responseestimate for each of the transitions from the first states tothe subsequent state and combining the intermediate up-datedversions of the channel impulse response estimate weighted bysaid corresponding plurality of transition probabilityestimates.The equaliser may further operate to substantially maximise aprobability of correctly detected data symbols in dependenceupon a plurality of state probability estimates each of which..........4............«........._.â,..... , . Wl0152O253035âA n................â.........«.........-...u.-.....-..... .._. .. \ ..... /GR 98 P 1284 CA 02264414 1999-03-057state probability estimates is associated with one of theplurality of states of the communications channel andcalculated from the received signals in combination with theadapted version of said channel impulse response estimateassociated with the state.Advantageously, the data processor may operate to generateeach adapted version of the channel impulse response estimateby weighting the intermediate up-dated versions of thechannel impulse response estimate by a combination of thecorresponding transition probability estimates and the stateprobability estimate for each of the first states whichdetermine the transitions to the state with which the channelimpulse response is associated. The transition probabilityestimates and the state probability estimates may becalculated in accordance with a symbol-by-symbol estimatoralgorithm such as the BCJR algorithm.According to an aspect of the present invention, there isprovided a method of detecting data symbols from signalsreceived from a communications channel, comprising the stepsof;- calculating a plurality of transition probability estimatesassociated with transitions from first states of thecommunications channel to at least one subsequent state ofthe communications channel, from an estimate of an impulseresponse of the communications channel in combination withthe received signals;- adapting the channel impulse response estimate associatedwith the subsequent state in dependence upon the transitionprobability estimates; and- detecting the data symbols from the plurality of transitionprobability estimates.The method of detecting data may further include the stepsof;101520253035GR 98 P 1284 CA 02264414 1999-03-058- generating a plurality of versions of the channel impulseresponse estimate, each of which is associated with a stateof the communications channel; andâ adapting each of the channel impulse response estimates independence upon a corresponding plurality of the transitionprobability estimates for transitions from a plurality of thefirst states to the state with which the channel impulseresponse is associated.The step of adapting each of the plurality of versions of thechannel impulse response estimate further comprises the stepsof;(mâ)to the subsequent state associated with the channel impulse- for each transition from the plurality of first statesresponse estimate, generating an intermediate up-datedversion of the channel impulse response estimate;â weighting each of said intermediate versions with saidtransition probability estimates; andâ combining said weighted intermediate up-dated versions togenerate said adapted channel impulse response estimate.One embodiment of the present invention will now be describedby way of example only with reference to the accompanyingdrawings, wherein,FIGURE 1 is a schematic block diagram of part of a radiocommunications channel used within a mobile radiotelephone system,FIGURE 2 is a schematic representation of a time divisionmultiple access frame,FIGURE 3 is a schematic block diagram of the radiocommunications receiver shown in figure 1,FIGURE 4 is a schematic block diagram of a data detector,which appears in figure 3,l0l520253035CA 02264414 1999-03-05GR 98 P 12849FIGURE 5 is a schematic block diagram of part of thereference signal generator shown in figure 4,FIGURE 6 is an illustration of a trellis modelled by thedata detector shown in figures 3 and 4.While the present invention finds application in many areaswithin the digital communications art, and therefore toreceivers for corrupted signals generally, an embodiment ofthe present invention will be described with reference to aradio communications receiver, and in particular to a radiocommunications receiver operating within a time divisionmultiple access mobile radio communications system. Examplesof such systems are the Global System for Mobiles (GSM), anda system operating in accordance with a hybrid Time Divisionand Code Division Multiple Access system, such as thatdescribed in an article entitled "Performance of a CellularHybrid CâTDMA Mobile Radio System Applying Joint Detectionand Coherent Receiver Antenna Diversityâ by G. Blanz, A.Klein, M. Naï¬han and A. Steil published in the IEEE Journalon Selected Areas in Communications, Volume 12,1994 at page 568.no. 4, MayIn figure 1, a block diagram is shown, which isrepresentative of parts of a radio communications channeloperating within a mobile radio telephone system. In figure 1digital data s(n) to be communicated over the radiocommunications channel is fed to a modulator and digital toanalogue converter 1, via a data encoder 2.The encoder 2,operates to transform the digital data s(n) to encoded datasymbols x(n) in accordance with aalgorithm.applies a half rate convolutionalforward error correctionIn the GSM system for example, the encodercode to the most importantbits of a frame of digitally encoded speech. Analogue signalsgenerated by the modulator l are fed to an input of atransmitter 3, which serves to amplify and up-convert the101520253035GR 98 p 1284 CA 02204414 1999-03-0510analogue signals to a radio frequency. The radio frequencysignals are thereafter radiated by an antenna 4. The radiosignals radiated by the antenna 4, propagate between theantenna 4, and a receive antenna 6 via the ether.A characteristic of mobile radio communications systems, isthat radio signals travel between the transmit antenna 4, andthe receive antenna 6, via a plurality of paths, such as adirect path represented by the line 8, and indirect pathsreflected from objects, such as a house 10. The exampleInter-symbolinterference is generated in the received radio signals,indirect path is represented by the line 12.as aresult of the radio signals travelling between the transmitThe receivedand the receive antennas 4, 6, by multi-paths.signals detected by the receive antenna 6, are fed to a down-conversion and base band unit 14 which operates to convertthe received radio signals from the radio frequency to a baseband frequency, and furthermore the base band unit 14,operates to analogue to digital convert the base bandsignals. As such digital samples of the base band signalsrepresented as z(n),band unit 14.signals z(n),are generated at an output of the baseThe digital samples of the received base bandwhichoperates to generate an estimate of the digital data fed toare thereafter fed to a data detector 16,the modulator 1 which is designated as £(n). The estimate ofthe digital data is thereafter fed to a decoder 18 whichserves to operate in accordance with a decoding algorithm toeffect a reverse of the encoding operation performed by theIn the case of GSM the decoder 18,in accordance with a Viterbi decoding algorithm to decode theThus,estimate of the data communicated over the channel isdata encoder 2. operatesconvolutional code. at an output of the decoder 18, angenerated which is represented as §(n).Time division multiple access systems are characterised inthat signals transmitted on a radio frequency carrier signalare divided into a plurality of time slots in which each slot10l520253035 GR 98 P 1284 CA 02264414 1999-03-0511represents a channel allocated to a different mobile unit.This is illustrated in figure 2, where the transmission ofdata on a radio frequency carrier is represented by the lines22, 24. As is illustrated in figure 2, where time isrepresented by the lines 22, 24, going from left to rightacross the diagram, the carrier signal is divided into aplurality of time slots 1 to M. As illustrated in figure 2 amobile communicates a burst of radio frequency signals in oneof the time slots, which is illustrated for the time slot 2,by the burst of radio signals 28. The burst of radiofrequency signals 28, comprises data bearing parts 30, and apart in which known signals corresponding to theaforementioned training sequence are transmitted. Thetraining sequence 32, provides the data detector 16 with ameans for estimating an impulse response of the channelthrough which the burst of radio signals 28, has passedbetween the transmitter 3 and the receiver 14.in a situation where a mobile unit isthat is,in the frequency of the communicated radio signals withAs already explained,travelling at high speed, a Doppler shift, a changerespect to time, is large enough to cause a significantchange in the impulse response of the communications channelwith respect to time which is experienced by the radiosignals. As such data symbols transmitted at temporalpositions removed from a temporal location of the trainingsequence in the middle of the burst 28, will experience asubstantially different channel impulse response than thechannel impulse response estimate determined at the temporalposition of the training sequence. Therefore, for example,the channel impulse response estimate generated from thetraining sequence will no longer be representative of thechannel impulse response at either end of the burst of radiocommunications signals 28. The equaliser which forms part ofthe data detector 16, must therefore mitigate these effects.Furthermore,in some applications, such as in a case wherethe decoder 18, is required to operate in accordance with a... ....2..............ul01520253035GR 98 P 1284 CA 02264414 1999-03-0512a symbol detector such as theBCJR-detector is preferred as a means for equalising theturbo detection technique,received signals. The BCJR-detector is an example of a symbolestimator which operates to maximise the likelihood of asymbol on a symbol-byâsymbol basis as opposed to a Viterbiequaliser which operates to maximise the likelihood of asequence of symbols. However a problem with the symbol-by-symbol type equalisers is that there are no decisions made asto which of the plurality of transitions leading to asubsequent state is selected for further processing of thealgorithm because the symbol estimated at the latest stepthrough the trellis diagram is determined from a combinationof predecessor states. However, a data detector including anequaliser which operates as a symbol-by-symbol detector andwhich adapts the channel impulse response in accordance withan example embodiment of the present invention is presentedin figure 3 where parts also appear in the figures 1 and 2bear identical numerical designations.A receiver for detecting a burst of radio communicationssignals and for recovering the data conveyed by the radiosignals, is shown in block diagram form in Figure 3. InFigure 3 the radio signals 8, 10, are detected by the antenna4 and fed to the down conversion and baseband processing unit14.thereafter fed to a data detector 19 which operates toThe digital samples of the burst of radio signals aredetermine a sequence of digital data symbols corresponding tothe digital samples z(n) by processing the digital signalto the effect that the inter-symbol interferenceand fading effects are substantially mitigated. At an outputof the data detector 19, the estimates of the detecteddigital symbols £(n) are presented on a conductor 34 to thedecoder 18.is presented on a conductor 36.samples z(n)A final estimate of the communicated data §(n)Also generated at the output of the data detector 19 is a setof soft decision metrics associated with each of the detected101520253035GR 98 P 1284 CA 02264414 1999-03-0513data symbols £(n) which are fed on a further conductor 38from the data detector 19. the datadetector 19, generates a vector of error signals é(n)comprising an error signal em(n) associated with each of theOn a conductor 40,states m modelled by the data detector,described.â The error signals are fed on conductor 40âA furtherthe channel estimate adapter 42 is provided by theas will shortly beto aninput of a channel estimate adapter 42. input tosoftdecision data from the conductor 38â. The channel estimateadapter 42 operates to generate at an output conductor 44, aset of channel impulse response estimates which are fed to areference signal former 46. The signals generated on aconductor 44 from the channel estimate adapter 42, arerepresentative of a set of channel impulse responseeach state of the channel modelled by the datadetector 19, being provided with one of the set of separatelyThetherefore operates to convolveestimates,adapted channel estimate, as will shortly be explained.reference signal former 46,each of the set of sequences associated with a correspondingstate of the channel modelled by the data detector 19, withan adapted version of the channel impulse response providedby the channel estimate adapter 42. The reference signalformer 46, a set ofgenerates at an output conductor 70,reference signals each of which is associated with acorresponding state transition of the trellis modelled by thedata detector 19 and which is formed for each transition froma first state of the channel to a subsequent state of thechannel. In order to provide an initial estimate of thechannel impulse response, channel initialiser 50, is coupledto the output of the down conversion and base band unit 14,and serves to extract the midâamble training sequence fromthe sequence of signal samples z(n), and to convolve thiswith a locally generated version of the training sequencewhich serves to yield an initial estimate of the channelimpulse response, in accordance with a technique well knownto those skilled in the art.estimate is fed via conductor 51 to the channel estimateThis channel impulse response101520253035GR 98 P 1284 CA 02264414 1999-03-0514adapter 42. Also fed to the channel estimate adapter 42, viaconductor 52, is a set of adaptation tap weight coefficientsï¬mu=[pm;q,."â;g] provided to an adaptation algorithm which updates the channel impulse response estimate. The adaptationcoefficients ï¬mu=[pMpâ.u",pJ are fixed for the operation ofthe data detector 16,operating parameters.in accordance with preâdeterminedAs will be appreciated, the embodiment of the presentinvention may be understood principally with reference to theoperation of the data detector 19, the channel estimateadapter 42 and the reference signal former 46 shown in Figure3. To facilitate a better understanding a more detailedblock diagram of these blocks 19,Figure 4, where parts also appearing in Figure 3 bear42 and 46 is presented inidentical numerical designations.In Figure 4 the channel estimate adapter 42, is shown to becomprised of a data processor 41, and a plurality of datastores 43, Two vectors of data from thefirst and second conductors are 40â, 38â, are fed to the dataOn the first input 40â, is a vector comprisedof the soft decision data SD am, with a soft decision datum$%n(n) associated with each of the states m modelled by thecoupled thereto.processor 41.data detector 19.error data §(n), is fed to the data processor 41. The errordata comprises an error datum em(n) associated with each ofOn the second input 38â, a vector of thethe states m modelled by the data detector 19.third conductor 51, to the data processor 41, is a furthervector comprising the set of adaptation coefficients ï¬,Fed on thewhich may be switched between two values in dependence uponwhich of the mid-amble and the data symbols are beingprocessed by the data detector. The adaptation coefficientsare otherwise fixed for the operation of the equaliser. Thedata processor 41, adapts each of the channel impulseresponse estimates stored in the data stores.101520253035CA 02264414 1999-03-05GR 98 P 128415The reference signal generator 46, comprises a plurality cffinite impulse response filters 56, an example of which 58,is shown in Figure 5, where parts also appearing in Figure 4bear identical numerical designations. As shown in Figure 5,the finite impulse response filter 58,with possible data symbols be{-l,possible values of the latest symbol, resulting inis fed on an input 59,+1} corresponding to atransitions from a first state of the channel to a subsequentstate of the channel modelled by the trellis. In thisexample, we assume a binary modulation scheme, so that, thepossible data symbols are -1, +1. Thus for each possiblesymbol that could form the latest detected data symbol, bthere is generated a corresponding reference symbol )3 j,associated with the i-th symbol for the j-th state. Theexample reference signal generator shown in Figure 5corresponds to state 0 of the channel. The finite impulseresponse filter 58, is formed from a shift register 60,having a plurality of symbol delays 62, and a correspondingset of multipliers 64 and adders 66, which operate to scalesymbols present at the corresponding output of the symboldelays 62 of the shift register 60, with a set of impulseThe data symbols multiplied by theimpulse response coefficients w; are thereafter summed by theadders 66, inlj )nLa at anoutput 68 of the finite impulse response filter 58.response coefficients wi.and form the reference signalseach state of the states of the channelmodelled by the data detector 19,response filter thereby providing at each output thereof aCorrespondingly,has a finite impulsenumber of reference signals corresponding to the number of(mâ) entering a subsequentmodelled by the data detector 19.signals are fed to the data detector 19,transitions from a first statestate (m) The referencevia parallelconductors 70 as shown in Figure 4.As already mentioned, a state of the channel is determined bythe number of symbols which have an influence on a detectedsymbol. Thus, if the inter-symbol interference causes a1O1520253035GR 98 P 1284 CA 02204414 1999-03-0516current symbol to be influenced by VÂ¥l earlier symbols, thenthe number of possible states of the channel will be BVâ%where B is the alphabet size of the data bearing symbols.Thus for a binary modulation scheme B = 2. For the presentexample the finite impulse response filter models a channelwith inter symbol interference corresponding to V = 5symbols. Thus the number of states modelled by the datadetector is 24 = 16.An example of a trellis diagram modelled by the channelequaliser is shown in figure 6. In figure 6 one stage in thetrellis diagram is represented with example transitions withbetween each state. The example trellis diagram has 16 statescorresponding to a channel which has a memory V of fivesymbols. On the left side of the trellis diagram first statesassociated with a time t-l or correspondingly n-1 in terms ofsignal samples are denoted mâ with new or subsequent stateson the right side denoted m. Between each of the states ofthe trellis is a corresponding transition associated with alatest symbol to be detected by the equaliser. It is thelatest symbol which determines the new or subsequent state mprovided that the old or first state is mâ.The exemplified embodiment of the present invention will bedescribed for a symbol-by-symbol detector which operates inaccordance with the BCJR-algorithm well known to thoseskilled in the art. In the aforementioned article,a fulldescription of the BCJR algorithm is presented with referencepublishedin IEEE Transactions on Information Theory,to the decoding of convolutional codes. For completeness ofthe description of the embodiment of the invention, a briefout line of the BCJR algorithm will be given in the followingparagraphs. However, a full explanation is provided in theaforementioned prior art disclosure which is incorporatedherein by reference.1O152O2530CA 02264414 1999-03-05GR 98 P 128417The BCJR algorithm is based on the calculation of the a-posteriori probabilities of the communications channel beingin states I{Sn==m/:00} (with Sn representing the states atsymbol sample n, and m representing one subsequent state) andthe probability of transitions between statesP{Sn_1==mï¬Sh==m/200}, where mâ is a first or old state atsample n-l, and z(n) = 2(1),digital signal samples,2(2), 2(3),and for the present explanation,..... , z(n) are theeachsample corresponds to a transmitted symbol. The detectionalgorithm is therefore based on the calculation of the joint(2),(1) is the probability of the channel being in state m,probabilities given in equations (1) and where equationandequation (2) is the probability of a transition from state mâat symbol sample n-1 to state m at symbol sample n. Forsimplicity, the following explanation is given in terms ofsamples n, instead of time t, and with one sample per symbol,n also corresponds to detected symbols.in (m) = P{Sn = m;z(n)} (1)§n(mâ,n)=P{Sn_1=mâ;Sn =m;z(n)} (2)The probabilities associated with equations (1) and (2) arecalculated using equations (3), (4) and (5):an(m) = P{Sn = m;z(n)} (3),6ân(m)=P{z(n+1)/Snzm} (4);/n(mâ,m) = P{Sn = m;z(n)/Sn_1= m'} (5)l0l5202530GR 98 P 1284 CA 02264414 1999-03-0518It can be shown that equations (1) and (2) can be describedfrom equations (3), (4) and (5) as shown in equations (6) and(7):2,,<m> = a,,(m)-ï¬,,(m) <6>0,,(mâ,m)=an_1(mâ)~7,,(mâ,m)-ï¬,,(m) <7)The function given by equation (3) is calculated in a forwardrecursion procedure of the BCJR algorithm, that is to say,moving recursively through the corresponding trellis diagramof the trellis from states at symbol samples n-1 to states atsymbol samples n. The forward recursion may be represented byequation (8):an(m)= an_1(m')-7n_1(mâ,m) (8)all statesmâThe function given by equation (4) is calculated in abackward recursion procedure of the BCJR-algorithm, that isto say, moving recursively through the corresponding trellisdiagram of the trellis from states at samples n to states atsamples n+1. The backward recursion may be represented byequation (9)./3,,(m') = Zï¬n+1(m)-7n+1(mâ,m) <9>all statesmTo effect calculation of the functions according to equations(8) and (9), it remains to define the function ynonï¬m) whichis the conditional probability of state m at sample n and thereceived symbol z(n), under the condition that thepredecessor state is mâ.(10).This is calculated from equation1015202530CA 02264414 1999-03-05GR 98 P 128419;/n(m',m)= Z P{,z(n)=b/Sn =m;Sn_1=mâ}-P{Sn =m/Sn_1=mâ}-P{z(n)/g(n)=b}all,z(n)(10)In equation (10) the factor P{;((n)=b/Sn=m;Sn_1=mâ} will beeither 0 or l, in dependence upon whether or not a branchtransition in the trellis diagram is present or not. If the(b E {-1, +l}) , with the effectthat the modulating symbols are binary (such as Binary PhaseShift Keying or Gaussian Minimum Phase Shift Keying) thenthere will only be two values per state Sn=nn which are nonmodulation scheme is binaryZGIIO .The factor I{S;==m/Sn_1=nn} is determined in accordance witha priory information of the transmitted data sequence.If a part of the data sequence is known, then this can beused to determine PY%{=m/Sz_1=nf}. For example,P{%{=m/Sn_1=nf}=1%x00=â4} for a transition from state mâto m where a known data symbol is -1, andP{%f=m/Sn_1=nf}=I%xO0=44} for a transition from state mâto m where a known data symbol is +1. if a modulatingdata symbol is known to be x(n) = then P{x(n)} willbe 1. Otherwise these probabilities are set to 1/B where B =Thus,-1 or +1,2 for a binary modulation scheme. Finally the factorP{dn)/zOï¬==b} is a factor which depends on the channel.As an approximation the error signal anon is used as anapproximation for the latest detected symbol, formed by acomparison between the received signal sample per symbol z(n)and the reference signal y(n),(11):which is expressed by equation(z(n) â yb,mI)Z (z<n>ây,,,,,,.>all mâP{z(n)/ z(n) = 1;} =1O152O2530GR 98 P 1284 CA 02264414 1999-03-0520where as before (b E {-1, +1}) for a binary symbol sequence.The function ynonï¬m) can be split into two parts, where eachpart is associated with the latest symbol being one of two(-11 +1)(12):values for the binary case. This is given inequationr,,<m'.m> = 7,,,_1(mâ,m,z(n) = -1)+7n,+1(mâ,m,x(n) = +1) <12>Finally soft decision detection is effected using the BCJR(13), where the top part of thefunction is generated by summing the probabilities oftransitions associated with the symbol 100 being -1,algorithm by equationand thebottom part of the function is generated by summing theprobabilities of transitions associated with the symbol 100being +1.2 (-1 - transitions/In (m))SD(n) = logââ . . < 1 3 >2 (+1 - transztzons/1n(m))mHard division output from the BCJR algorithm is determinedso that the harddecision output is estimated in accordance with equation(14):201) = {with reference to a threshold at zero,â 1 if SD(n))0+1 ifSD(n)(O (14)It is however the soft decision outputs which makes the BCJRalgorithm particularly useful for turbo decoding used in asubsequent error detection stage, such as that shown infigures 1 and 3 as decoder 18.As will be appreciated from the above discussions, symbo1-by-symbol detectors such as the BCJR algorithm are characterisedin that there are no decisions made as to which of the1015202530CA 02264414 1999-03-05GR 98 P l28421plurality of transitions leading to a subsequent state isselected for further processing of the algorithm because thesymbol estimated at the latest step through the trellisdiagram is determined from a combination of predecessorThus,survivor path is selected,states. unlike the Viterbi algorithm no uniqueso that the new metric is formedas a combination of rather than a selection from twocandidates.Returning to the description of the example embodiment of thepresent invention, adaptation of the channel impulse responseestimate so as to substantially mitigate effects associatedwith Doppler frequency shift causing changes in the channelimpulse response with time, will now be described. Asalready explained, the new state metric during the forwardand also the backward recursions in the BCJR algorithm is thesum of contributions from several candidatessymbols).symbol-by-symbol detector is effected by calculating the newtap coefficients using contributions from transitions from(mâ) and the probability of thesecandidate first states as a weighting factor. As such, the(two for binaryAdaptation of the channel impulse response, for aall possible first statesadaptation principle is expressed by equations (14), (15),and (16):wmâ_1 = LWm,,_1+£m,,_1 (14)Wm,+1=LWm,â+l+Km.â+1 (15)Wm : Wm,â1'*m,ââ1+Wm,+1'Wm,+1 (16)are given by-1Where weighting factors Wâm, +1and Wâmâequations (17) and (18):152O2530CA 02264414 1999-03-05GR 98 P 128422W, _ an:%(m)ynf4(mï¬n0 (I7)"*1 an,_1(m).7nâ_1(m',m) + an,+1(m).7nâ+1(mâ.m)01 (m)-7 (m',m)W, : n54 rL+1 (la)mï¬dan,_1(m)'ynâ_1(m'>m) + an,+1(m)âyn,+1(m'am)and W are intermediatem -1 m,+19In equations (14) and (15) Wupâdated versions of the channel impulse response estimate,generated by up-dating the versions of the channel impulseresponse estimate Wm,_1, Wm,+1, associated with the firststates mâ, in accordance with adaptation factors Am,_1,3 .mIcalculated according to equations (19) and (20). The+1adaptation factors Amu_1, Amu+1,_.are calculated by the dataprocessor 41, of the channel impulse response adapter 42, independence upon the data symbol (-1, +1) associated withtransitions from the first states mâ to the subsequent statem. The intermediate up-dated versions of the channelimpulse response estimate Wâr_1, Wm+4, are formed by addingadaptation factors Amu_1, Amu+1, to the versions of thechannel impulse response estimate associated with the firststates mâ, in some way. For example, one such way is given(14) (15). (14)and (15), correspond to the well-known leaky Least Meanas an example in equations and The equationsSequences equation, with a leakage factor L, for example withL = (1 - lleast significant bit) being used for improvedstability. However, as will be appreciated, other adaptationalgorithms may be used. The adaptation factors Am._1,..A in equations (14) and (15) correspond to changes tomï¬+1âthe channel impulse response coefficients associated with theadaptation step size p, for a corresponding error signal, inaccordance with the least mean squares algorithm, asexpressed by equations (19) and (20):, ....,.m.4..............................-.101520253035GR 98 P 1284 CA 02264414 1999-03-0523â =*.â 19AmZ_1 ;wm,,zmz_1 ( )â =*â 20Amï¬+1 4£nï¬ Zmï¬+1 ( )Where E is the error signal, fm, is a symbol vector-1associated with the first state mâ where the new symbol is-1, and 2m, is a symbol vector associated with the first3,+1state mâ, where the new symbol is +1.of the channel impulse response estimate WmThe adapted versionis formed inaccordance with equation (16), by scaling the intermediateup-dated versions of the channel impulse response estimatesw by the weighting factors W%_4, FV andwm,ââ1' m,+1â m,+1âsumming the result.As will be appreciated by those skilled in the art othersymbol-by-symbol detectors may also be used with the presentinvention. A further example of a symbol-by-symbol detectoris a detector which operates in accordance with the Maximum(MAP) algorithm. symbol-by-symbol detectors arecharacterised in that transitions in a trellis diagramin that allpossible paths into a subsequent state contribute to theA-Posteriorimodelling the state of the channel are retained,subsequent state metric. Adaptation of the channel impulseresponse used in the detection of the data with a symbol-by-symbol detector in accordance with the example embodiment ofthe present invention can be interpreted as calculating anaverage coefficient vector that results from averaging over"all candidate old states. The probability ratio(17) (18))adaptation, as this provides an average value calculated from(equationsand is proposed as a weighting factor forthe probability of occurrence. The two extreme cases are thefollowing: If one candidate is much more likely than theother, the influence of the other is suppressed and theresult is equivalent to the case for an adaptive Viterbiequaliser as herein before described. If both candidates101520GR 98 p 1284 cAmn2644141999wnâ0s24(differing in the latest impulse response tap) are equallylikely no effective update is performed for the latestdetected symbol,the (unweighted)and the new coefficient vector just becomesaverage of both contributions FKn_1 andmï¬J'As will be appreciated by those skilled in the art, variousmodifications may be made to the aforementioned embodimentwithout departing from the scope of the present invention.For example, although the example embodiment has beendescribed with a separate channel impulse response estimateit will beappreciated that a single channel impulse response estimateper state modelled by the data detector,could be adapted in accordance with the principles of theinvention,symbols.application for detecting data received from a variety ofand used by the data detector to estimate theFurthermore, the present invention findscommunications channels, where a limitation in the bandwidthof the channel or other effects cause interâsymbolinterference.
