WO1996035284A1

WO1996035284A1 - Symbol detector device comprising adaptive symbol threshold means

Info

Publication number: WO1996035284A1
Application number: PCT/SE1996/000559
Authority: WO
Inventors: Per Ola BÖRJESSON; Per M. ÖDLING; B. Håkan ERIKSSON
Original assignee: Boerjesson Per Ola; Oedling Per M; Eriksson B Haakan
Priority date: 1995-05-05
Filing date: 1996-04-29
Publication date: 1996-11-07
Also published as: SE9501679D0

Abstract

A symbol detector device for detecting at a receiving end of a transmitting channel a sequence of transmitted, and by the channel distorted, data symbols having a symbol alphabet including Q letters (for example, for the particular case of binary symbols Q=2, e.g. +1 and -1, and for trinary symbols Q=3, e.g. +1, 0, -1) is described. A receiving means (11;11) receives the data symbols and provides an individual symbol value for each symbol to be detected. Symbol threshold means (Tr1; (Tr1)1-(Tr1)M;... TrN;(TrN)1-(TrN)M) provide at least two variable thresholds per symbol. The number of thresholds is dependent on the number of letters in the alphabet. The symbol threshold means compare the values from the receiving means with the set of threshold levels. Combination means (13-15;13i-15i,18) combine the threshold comparisons for each symbol and provide an output per symbol which either is a representation value of one of the letters in the alphabet or an additional value signifying that no decision was made for the symbol.

Description

Symbol detector device comprising adaptive , symbol threshold means

This invention relates essentially to a symbol detector device of the kind specified in the preamble of claim 1.

BACKGROUND OF THE INVENTION

Data transmission, for instance through different kinds of networks and transmission channels, is exposed to additive noise. When data is transmitted as a serial symbol-stream, the transmission may also be subjected to intersymbol inter- ference (ISI) due to multipath propagation, i.e. a smothered copy of the signal, taking another pathway than the main signal, is delayed and superimposed on the main signal.

The invention is directed to a solution of the problems concerned with detecting symbols, typically binary antipodal- ly modulated, in a symbol stream that has been subjected to additive noise and ISI. With binary antipodally modulated we mean that "1" is often represented by a specific wave form and "0" by the negated version of the same wave form. The received signals are processed by a symbol detector at the receiving end of the transmitter channel in order to repro¬ duce the transmitted symbol stream.

DESCRIPTION OF RELATED ART

When ISI is present, symbol detectors are in general either computationally complicated or performed in a non-optimal way. One frequently used symbol detector for detecting sym¬ bols distorted by ISI and additive noise is the so-called maximum likelyhood sequence detector (MLSD) . An MLSD is a processor which, given a received signal, finds the sequence of symbols that was mot probably sent given the received signal, and thus is optimal measured with the probability of sequence error. However, this detector, even when implemented

SUBSTITUTESHEET with the Viterbi algorithm, has an unattractive computational complexity in many applications.

OBJECTS OF THE INVENTION

The main object of the invention is to provide a symbol detector device for detecting symbols that has been subjected to ISI and additive noise, with a low probability of error on detected symbols.

Another object of the invention is to provide a symbol detec¬ tor device having a structure of low computational complexi¬ ty.

Still another object of the invention is to provide a symbol detector device providing the same results, regarding some symbols in a symbol stream, as the MLSD, but using less computational complexity.

Still another object of the invention is to provide a symbol detector device which is able to detect at least some of the symbols in a transmitted symbol stream and is able to co¬ operate with other kinds of symbol detector devices in order to detect the remaining symbols, such as provided as a pre- processor.

These objects are fulfilled with a symbol detector device of the kind disclosed in the characterizing part of the indepen¬ dent claim. Other features and developments of the invention are disclosed in the dependent claims.

SUMMARY OF THE INVENTION

According to the invention, a detector device is provided for detecting a sequence of symbols distorted by ISI and additive noise. The device comprises a symbol detector device for detecting at a receiving end of a transmitting channel a sequence of transmitted, and by the channel distorted, data

SUBSTITUTE SHEET symbols having a symbol alphabet including Q letters (for example, for the particular case of binary symbols Q=2,e.g. +1 and -1, and for trinary symbols Q=3,e.g. +1, 0, -1). The device is characterized by receiving means receiving said data symbols and providing an individual symbol value for each symbol to be detected; symbol threshold means providing at least two variable thresholds per symbol, the number of thresholds being dependent on the number of letters in the alphabet, and for comparing the values from said receiving means with said set of threshold levels; and means for com¬ bining the threshold comparisons for each symbol and provid¬ ing an output per symbol which either is a representation value of one of the letters in the alphabet or an additional value signifying that no decision was made for the symbol. The symbol threshold means provides preferably at least one pair of variable thresholds per symbol, the number of pairs being dependent on the number of letters in the alphabet. When adapted to detection of binary, antipodally modulated data symbols, the symbol threshold means comprises preferably one threshold means per symbol comparing the output of said receiving value providing means to a first and a second threshold level, and providing an output having a first decided symbol value if the output is higher than the first level, a second decided symbol value if the output is lower than the second symbol level, and a third symbol value, denoting "no decision", if the output lies between the first and the second level.

The first and second levels of each threshold means can be calculated separately for each symbol. Comparing of the stored output of the provided received value, some symbols may be detected. The first and second levels of each thres¬ hold means can then be recalculated separately for each symbol. The first and second levels of each threshold means may then come closer to each other and new symbols might be detected. This can be redone several times in an iteration process.

SUBSTBTUTΞ SHEET The detector according to the invention will make the same decisions as an MLSD on the symbols that are detected, and it is simple in structure. Some of the symbols in each data sequence may not be detected. The number of detected symbols and their positions are in general different for each receiv¬ ed sequence. The thresholds in the threshold means are depen¬ dent on the received signal and are calculated using an iter¬ ative method, in which the number of iterations is stochastic but never exceeds the number of transmitted symbols. The number of iterations is dependent on the intersymbol inter¬ ference and the received signal. The computational complexity is typically much less than the computational complexity of the MLSD.

An example of a possible transmission system of which the detector according to the invention could be a part, is sketched in what follows. A sequence of symbols is trans¬ mitted as a waveform on an analog channel. This waveform is distorted by the channel, typically by time dispersion and additive noise. Often in the waveform received as the output of the channel, the individual symbols are not separable in time. The detector then comprises preferably a whitened matched filter where filtering first is made in analog form on the received analog waveform. The output of the analog filter is then sampled where care is taken for the respective location for each symbol in a transmitted sequence. The sampled signal is then possibly filtered again using a dis¬ crete-time filter, for instance as prescribed for a whitened matched filter. The discrete-time stream of numbers could be input to the detector according to the invention. The detec¬ tor would use a threshold means, which for binary symbols would have two variable thresholds, for each symbol to be detected. The output of the detector would, for each symbol to be detected, consist of one value out of the symbol alpha- bet (for example +1 and -1) or an additional value signifying that no decision was made for that symbol.

SUBSTBTUTE SHEET BRIEF DESCRIPTION OF THE DRAWING

For a more complete understanding of the present invention and for further objects and advantages thereof, reference is now made to the following description taken in conjunction with the accompanying drawings, in which:

FIG 1 shows a schematic block diagram of the essential part of the symbol detector according to the inven- tion,

FIG 2A-2D shows the output of the whitened matched filter in FIG 1 plotted together with the thresholds of the detector in an example with four iterations,

FIG 3 shows a block diagram of a first embodiment of the symbol detector according to the invention,

FIG 4 shows a block diagram of a second embodiment of the symbol detector according to the invention,

FIG 5 shows a block diagram of the symbol detector used as a pre-processor, FIG 6 shows a 3D-diagram illustrating the percentage of sequences that needed a certain number of iterations plotted versus signal/noise ratio (SNR) and the number of iterations,

FIG 7 shows a 3D-diagram illustrating the percentage of sequences where a certain number of symbols were detected plotted versus SNR and the number of sym¬ bols.

DETAILED DESCRIPTION OF THE DRAWINGS

Referring to FIG 1, an incoming sampled signal stream y re¬ presenting a series of symbols from a communication link with ISI and noise is fed to an input of a filter 1, which makes a summation of each symbol signal while weighting it in depen- dence of the ISI derived from monitoring the communication link. The filter 1 is shown as a whitened matched filter but this is not necessary. This filter is here presented in a discrete-time form. This function could also be obtained by

SUBSTITUTE SHEET sampling the output of a continuous-time filter operating on a continuous-time observation of the received signal.

Consider the transmission of sequences of binary symbols, typically interspersed with sequences of symbols already known to the receiver, through the transmission link. The resulting transmission system can be represented by the matrix notation y = Hb + n (1) where y is a vector of the incoming signal comprising a sequence of antipodally modulated data, N being the number of symbols in the sequence and L the length of the channel memory, H is deterministic and known (N+L-l)xN matrix repre¬ senting the ISI, b is an (Nxl) vector containing the N binary antipodally modulated symbols, and n is a jointly Gaussian, zero mean random vector with a N(0,R_n) distribution, where R_n is the covariance matrix E[n*n ] of the noise vector. The matrix H may be derived by monitoring the behaviour of the transmission channel in moments when known data are sent.

In a linear, time-invariant and causal system, the matrix H becomes o ••• ••• ••• u i h₀ 0 hi 0

H = h L_τ.--l 0

• • hτ._-l ho

where [ho,hi .h_L_ι] is ^the impulse response of the system.

Defining the matrix M - H^TR_n~ H and using the model of the equation (1) , it follows that z Δ H^TR_n ^-1y = Mb + H^TR_n ^-1n (4)

SUBSTITUTE SHEET This is the output from a preferred embodiment of the filter 1 which is a whitened matched filter.

The symbol detection is done blockwise, so that one sequence of a data stream is detected at a time. However, different sequences could have different symbol lengths within a pre¬ determined maximum sequence length.

The output z of the filter 1 is fed to a threshold device 2 having at least as many parallelly arranged threshold units as there are symbols to monitor in the same data sequence.

The matrices H and R_n are provided by a device monitoring the channel and is provided as an input to the threshold device. This input comprises control information, together with the signal z from the filter 1, for deriving two threshold values for each threshold unit in the device 2.

Preferably, the output vector b from the device 2 should be the same as if it had been processed by an MLSD.

The log-likelyhood function for the k:th symbol in the symbol stream to be detected could be written as

L'(b,z) = 2b_k(Δ(b,k)-z_k)+g(z,b,k) (5) where

Δ(b,k) 4 ∑ biirii.k and g(z,b,k) -=. Σ biirti._jb_j - 2∑ bj_Zi + m_kfk + z^TMz(7) where the sums in (6) and (7) are taken over all i≠k. Note that the last two equations are independent of b_k and that |Δ(b,k) I < Σ Ibimi._k I = Σ lm_i/k|, (8) where the sums are taken over all i≠k.

If the output of the filter 1 at the instant k is above the possible maximum of Δ(b,k) , i.e. z >∑ lι&i,_k 1 then the like¬ lihood function L'(b,z) is maximized by b_k=+l, independently

SUBSTITUTE SHEET of the other symbols. Hence, if the value in position k of the output of the filter 1, preferably a whitened matched filter, is above the positive of the threshold, the corres-

^Λ MTJSD ponding output of an MLSD is positive, b = +1. Similar- ly, if z_k<∑ lmi._k I, then the

_ _+1# This is obviously true for all positions k, and the tests can be done independ¬ ently for all symbols.

Assume now that some of the symbols in b have been detected as described above. These decisions on detected symbols, once made, does not change due to any subsequent decisions on still-undetected symbols. Using detected symbols as con¬ stants, we proceed by minimizing L'(b,z) with respect to the remaining (undetermined) symbols. With the same technique as described above, the function Δ(b,k) can be bounded from above by

Δ(b,k)+ < Σ bjmi._k + Σ lm_ifk| and from below by

Δ(b,k)- > Σ bim_ifk - Σ lmi._k| where the first sums in the above expressions are taken over those i corresponding to detected symbols and the second sums are taken over those i corresponding to symbols not detected. We can again check whether z_k > Δ(b,k)+, and if it is, then fa^MLSD = +1. Analogously, if z_k < Δ(b,k)-, then b^⁰ = __1# Thus, by comparing the output of the matched filter, z_k, with these new and tighter thresholds, additional symbols could possibly be detected. This procedure can be repeated until no more symbols are detected.

Let b⁽¹⁾ = [b⁽¹⁾ι, b⁽¹⁾ ₂, b⁽¹⁾ _N__!, b⁽¹⁾ _N]^T denote the output of the inventive symbol detector device after the l:th iter¬ ation and let b_Δ denote its final output. Furthermore, let

Δ_k ⁺(b^(1-1)) and Δ_k ^~(b^{( _1)}), respectively, denote the upper and the lower thresholds at iteration 1. The detecting opera- tion can thus be done in the following way

SUBSTITUTE SHEET 1 . 1 = 0

2 . b . (^ι 0^υ )' _k. ₌ = 0 , V k REPEAT

3 . 1 = 1 + 1

4. Calculate the upper threshold:

Calculate the -lower threshold:

A; fε"-") '

»*sr^> - i Σ≠k hui (i - lϋ*-" I) . v

UNTIL (||b^ω||f = N or b^W = b⁽'^_1))

7. b_Δ = b⁽

Note that the output b_Δ of the detector, when implemented in the way described above, belongs to {-1,0,+1} , where a zero signifies that the detector did not make a decision on the corresponding symbol. The iterations of the algorithm would typically continue until no more symbols could be detected or until all symbols are detected. One alternative way to termi¬ nate is to always iterate N times.

The threshold units are controlled in sequence in order to provide the individual thresholds for each of them. Thus, according to the invention, the output of each of the thres¬ hold units is ternary, "-1" if the symbol value is below the lower threshold, "+1" if the symbol is above the upper thres- hold, and "no decision", typically represented by "0", if the symbol value is between the thresholds.

The thresholds of each unit are recalculated in each itera¬ tion taking into account the output of the surrounding thres- holds, output of filter 1 and control input, H and R_n. The

SUBSTITUTE SH T stored signal from the filter 1 is fed through the threshold device 2 again and new outputs from the threshold units are obtained.

This procedure is repeated in order to bring the percentage "no decision" output to a minimum. Thus after each iteration, if this percentage is higher than a predetermined value, or if two consecutive iterations have not given the same result, then a new iteration of the result is made. It is to be noted that the predetermined value preferably will be set to 100%.

To illustrate how the detector device according to the inven¬ tion calculates the thresholds, and makes decisions, an example of the outcome of four iterations by the threshold units is presented in FIGs 2A to 2D, in which sequences of binary data having a length N=20 are transmitted across a time-invariant channel with ISI. The noise is white, station¬ ary and Gaussian with an SNR of 10 dB. The solid lines are the upper and lower thresholds, Δ_k (b* ^~ ') and Δ_k ^~(b ' ) , the stars (*) and circles (o) are the stored outputs z after each iteration, where o denotes determined symbols and * the unde-termined symbols left after the preceding iterations. The output z at the beginning of the iterations is the output of the whitened matched filter 1. The upper and lower thres- holds in each unit will merge when a sufficient number of symbols in a sequel are detected.

FIG 2A shows the result of the first iteration in which the same thresholds were provided for all the threshold units except for the two at the beginning and the end of the data sequence. The calculation of the threshold values is done taking the surrounding data values into account. Therefore, the threshold values at the ends are nearer to the zero-value than the others. Eight symbols were undecided after the first iteration.

SUBSTITUTESHEET FIG 2B shows the second iteration in which already some of the thresholds have merged particularly in the beginning of the sequence of values where a sequence of symbols were already determined at the first iteration. The upper and lower thresholds of most of the threshold units are now rearranged and the appearance of the threshold design is changed. However, the thresholds are determined to be at a few different levels. Also, the upper and lower thresholds for a particular symbol will merge when this symbol and some sequential symbols around it have been determined.

FIG 2C shows the third iteration and illustrates that symbols 15, 17 and 19 are not yet detected. Note also that symbol 12 was detected as positive although the output from the whiten- ed matched filter 1 was negative.

FIG 2D shows the fourth iteration and illustrates that the symbols not detected in the third iteration are still not detected. Then there is nothing to gain by making another iteration. Therefore, a new iteration is not made.

DESCRIPTION OF A PREFERRED EMBODIMENT

An embodiment of a device for implementation of the invention is illustrated in FIG 3. The received signal, which is an analog signal representation of the sequence of symbols radi¬ ated to the receiver having the detector device according to the invention, is received by a circuit 11. The circuit 11 is provided with information about the transmission channel or link and corresponds with the filter 1 in FIG 1. It processes the incoming signal such that a series of samples represent¬ ing the received symbols and illustrated in FIG 2A are pro¬ vided on its output. Due to distortion in the transmitting channel, the received samples differ from the transmitted symbols. As apparent from the above, further processing must be done to provide the value symbols "l":s and "-l":s from the package in such ratio that the information of the package is understandable.

SUBSTITUTE SHEET In accordance with the invention, the series of receiving values are provided parallel to each other in a series/paral¬ lel converter 12 and stored in individual cells in a memory 5 13. A controllable threshold unit Trl, Tr2, ....TrN is con¬ nected to each individual memory cell for the receiving values, two thresholds, a first high level and a second low level, in each threshold circuit being individually control- lably set by a processor 14, based on information of the

10 distortion of the transmitting channel and the outputs of all the threshold circuits stored in the memory 15. During the iteration process, the stored values in the memory 13 are fed to the threshold circuits, which each provides an output +1, -1 or "no decision".

15

The iterations for providing fewer and fewer outputs repre¬ senting "no decision" from the threshold circuits, shown in FIG 2A to 2D, are provided by cooperation between the proces¬ sor 14, the memory 13 and the threshold circuits Trl to TrN.

20 After each iteration the results of the threshold comparisons for all threshold circuits are stored in a memory 15.

The appropriate condition for acceptance of when further iterations are to be prohibited is checked in a checking unit

25 16 connected to the memory 15 to activate output of the stored receiving values to the threshold circuits as long as the interrupt condition is not fulfilled. The output of the unit 16 is also fed to a receiving device (not shown in FIG 3) for the output from the detector to inform that device

30 when the output is in form to be received for further proces¬ sing of the received data package, preferably when all the necessary iterations have been done and the amount of "no decisions" has been reduced to a minimum.

35 The embodiment shown in FIG 3 has one threshold unit Trl per symbol setting two thresholds each which is enough if each symbol to be received is binary. FIG 4 shows an embodiment of a device for implementation of the invention for symbols out

SUBSTITUTESHEET of a symbol alphabet having more that two symbol values per symbol, for instance four values. It is however to be noted that the essential feature here is to extract the symbol values for the received sequence of symbols by comparing them with several threshold levels and that this could be done in other ways than what is disclosed in this embodiment.

The signal S being a train of symbols to be processed is hereby parallelly received by a number of M different cir- cuitries through a filter 11' having the same feature as the filter 11 in FIG 3 to provide a particular value per symbol, for instance by a sampling operation. Each circuitry is essentially of the same kind as shown in FIG 3 and therefore each element has got the same reference as in FIG 3 but for a 1 or an M at its back depending upon to which circuitry it is a part. Each circuitry sets at the beginning of an iteration operation mutually different threshold levels in their thres¬ hold circuits (Trl)i, where i is any value between 1 and M. However, only two thresholds are used in each threshold circuit, but they are at least initially provided on differ¬ ent levels for the different circuitries, for instance each having two levels between two different neighbouring of the different symbol values in the particular symbol alphabet.

The outputs from each of the circuitries are fed to a com- binatory network 18 after that the iterations are finished in all the circuitries. In order to accomplish this simultane¬ ously for the outputs of each circuitry, the outputs from all the checking units 161 to 16M are fed to an input each of an AND-gate 19 enabling the combinatory network 18 to receive the information of the inputs from the circuitries only when an anabling signal is fed from the AND-gate 19 to the network 18. The network 18 makes a combination of the result of the threshold comparisons provided on its inputs. It provides a decided symbol value for each symbol to be detected on a separate output for the symbol. It can also provide the symbol values serially on a serial output (not shown) . Each decided symbol value consists of either one value out of the

SUBSTITUTESHEET symbol alphabet or an additional value signifying that no decision could be made for the symbol in question.

The symbol detector device according to the invention is well suited to be used as a pre-processor in combination with other receiving and processing devices for restoring a trans¬ mitted data sequence at the receiving end of a transmission channel. Two principle methods will be discussed.

The first method is to run the symbol detector device accord¬ ing to the invention in parallel with some other receiver and use the symbols detected by the symbol detector device when available and symbols from the complementary receiver other¬ wise.

However, simulation experiments have indicated that a prin¬ cipally different technique for combining the inventive device and the other receiver is preferable. One is illustra¬ ted in FIG 5. The model of the inventive detector is given in equation (1) . Assuming that the estimated symbols are correct their influence could be subtracted from the received signal. In order to detect the remaining, non-decided symbols, consi¬ der the random vector x = y - Hb_Δ (9) where y is the channel output according to equation (1) and bΔe{-l,0,+l} N denotes the ternary output of the i.nventive symbol detector 1, 2. The circuitry for making a processing according to the equation (9) is illustrated in FIG 5, in which the output b_Δ is fed to a multiplier unit 20 making the calculation Hb_Δ. The signal y is fed to a (+)input and the output of the unit 20 to a (-)input of an adder 21. The output x of the adder is fed to a post-processor 22.

Let G be a matrix containing the columns of H corresponding to those symbols that were not detected. It follows from the structure of G and the equation (9) that the random vector x given b, is Gaussian with x~N(GC,R_n) , where C is a column

SUBSTITUTE SHEET vector containing the undetermined symbols. Equivalently x can be modelled as x = GC + n (10) where n is the same noise vector as in equation (1) , thus n~n(0,R_n) . Note that if no symbols were detected, then all elements in Hb_Δ are equal to zero, G=H and x=y. To detect the undetermined symbols, equations (9) and (10) can be used by giving the vector x and the characteristics of the channel model (10) , i.e. G and R_n to any receiver detector intended for sequence transmission systems. Note that G in general will represent a time-variant system even when the original channel model H represents a time-invariant channel.

Many different kinds of detectors could be applied to the reduced system according to equation (10) as post-processors to the proposed detector in order to detect the remaining symbols, for example, linear equalizers or decision-feedback equalizers.

When a decision made by the symbol detector device according to the invention is erroneous, i.e. when a detected symbol differs from a corresponding transmitted symbol, then the reduced system according to equation (10) is an incorrect model of the signal x. As a consequence, the complementary post-processor 22 will be inclined to make additional erro¬ neous decisions on the remaining symbols. However, this error propagation is likely to be less severe than, for instance, the error propagation in decision-feedback equalizers, be¬ cause the decisions, made by the symbol detector device according to the invention, have a lower symbol error rate. Experiments using the zero-forcing linear equalizer, the minimum-mean square error equalizer, the zero-forcing deci¬ sion-feedback equalizer, the minimum-mean square error deci¬ sion-feedback equalizer as post-processors have given satis- factory results.

SUBSTITUTESHEET The symbol detector device according to the invention has proved to come up to expectation. The potential of the inven¬ tive symbol detector device is being dependent on its ability to make decisions. FIGs 6 and 7 show results obtained in an experiment with the transmission of several consecutive data sequences and using simulations with a two-tap channel model hand a sequence length N=20. The impulse-response of the channel was [ho,hι,h₂] = [1,0,1].

FIG 6 presents the number of iterations needed to calculate the output of the symbol detector device in the experiment. The percentage of sequences that needed a certain number of iterations is plotted versus SNR (signal to noise ratio) and the number of iterations. With this channel the symbol detec- tor device most frequently used three iterations and rarely more than five.

FIG 7 presents the number of symbols detected in the same experiment as illustrated in FIG 6. The percentage of sequen- ces with certain numbers of symbols detected are plotted versus SNR and the number of detected symbols. The complete sequence is quite often detected when the SNR is low for the two-tap channel used. Note that if nineteen symbols are detected, the twentieth is always also detected.

As illustrated in FIG 6, only a few iterations are needed in most cases. Thus, the symbol detector device according to the invention makes fast decisions. Also, an important property of the symbol detector device, determining its potential, is the fraction of symbols on which a decision is made. The percentage of symbols detected in a sequence is stochastic and varies between 0% and 100%, with an average dependent on the SNR and the channel characteristics. Simulations with a two-tap channel model and a sequence length N=20 always give a decision rate better than 65%, while channel models with a high signal-to-noise ratio and strong intersymbol interfer¬ ence render decision rates in the order of a few percent.

SUBSTITUTESHEET While the invention has been described with reference to specific embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted for elements thereof without departing from the true spirit and scope of the invention as it is defined in the accompanying claims. In addition, modifica¬ tions may be made without departing from the essential teach¬ ings of the invention as defined in the claims. One example of such a variation would be to use other modulation tech- niques, for example, 4-QAM or M-PAM. Also, variations for complex-valued channels and continuous transmission do not depart from the essential teaching of the invention as de¬ fined in the claims.

SUBSTITUTESHEET

Claims

1. A symbol detector device for detecting at a receiving end of a transmitting channel a sequence of transmitted, and by the channel distorted, data symbols having a symbol alphabet including Q letters (for example, for the particular case of binary symbols 0=2,e.g. +1 and -1, and for trinary symbols 0=3,e.g. +1, 0, -1), characterized by receiving means (11;11) receiving said data symbols and providing an individual symbol value for each symbol to be detected; symbol threshold means (Trl; (Trl)l-(Trl)M; ... TrN; (TrN)1- (TrN)M) providing at least two variable thresholds per sym¬ bol, the number of thresholds being dependent on the number of letters in the alphabet, and for comparing the values from said receiving means with said set of threshold levels; and means (13-15;13i-15i,18)for combining the threshold comparis¬ ons for each symbol and providing an output per symbol which either is a representation value of one of the letters in the alphabet or an additional value signifying that no decision was made for the symbol.

2. A symbol detector device according to Claim 1, character¬ ized in that said symbol threshold means (Trl; (Trl)l-(Trl)M- ;... TrN; (TrN)1-(TrN)M) provides at least one pair of vari¬ able thresholds per symbol, the number of pairs being depend- ent on the number of letters in the alphabet.

3. A symbol detector device according to Claim 1 or 2 for binary, antipodally modulated data symbols, characterized in that said symbol threshold means comprises one threshold means (TRl-TrN) per symbol comparing the output of said receiving value providing means to a first and a second threshold level, and providing an output having a first decided symbol value if the output is higher than the first level, a second decided symbol value if the output is lower than the second symbol level, and a third symbol value.

SUBSTITUTE SHEET denoting "no decision", if the output lies between the first and the second level.

4. A symbol detector device according to Claim 1 or 2, characterized in that said symbol threshold means comprises a number of sets of threshold means (TRl-TrN) per symbol, the number being in dependance of the letters in the alphabet, each set having an individual pair of variable threshold levels and comparing the output of said receiving value providing means to a first and a second threshold level, and providing an output having a first decided value if the output is higher than the first level, a second decided value if the output is lower than the second symbol level, and a third symbol value, denoting "no decision", if the output lies between the first and the second level.

5. A symbol detector device according to Claim 4, character¬ ized by a combinatory network (18) combining the outputs from each circuitry to provide the symbol alphabet value or said no decision value for each of the detected symbols.

6. A symbol detector device according to anyone of the preceding Claims, characterized in that the setting of the threshold values in each threshold means is dependent on observed channel behavi¬ our.

7. A symbol detector device according to anyone of the preceding Claims, characterized in that each respective threshold means (Trl-TrN; (Trl)1-TrN)1, (Trl)M-TrN)M) is adapt¬ ed to set the thresholds individually for its respective symbol in said sequence of symbols to be checked.

8. A symbol detector device according to Claim 7, character- ized by means (14,15;141,14M,151,15M) for amending the in¬ dividual thresholds of each threshold means (Trl-TrN; (Trl)1- TrN)1, (Trl)M-TrN)M) making iterations on each sequence of receiving values until successive iterations regarding thres-

SUBSTITUTE SHEET hold amendments provide outputs from the threshold means having a predetermined ratio of first and second symbol values to all the values in the sequence of symbols, or two successive iterations provide the same result.

9. A symbol detector device according to Claim 8, character¬ ized in that for each threshold means (Trl-TrN) the thres¬ holds are recalculated taking account of the outcome from the preceding iteration regarding decided first and second symbol values surrounding the symbol the threshold means in question is adapted to detect.

10. A symbol detector device according to any one of the preceding Claims, characterized in that said value providing means (11;11') is a sampled whitened matched filter.

11. A symbol detector device according to Claim 10, charact¬ erized in that the value providing means (11;11') is adapted to make its calculation depending on the received sequence of symbols and the observed channel behaviour.

12. A symbol detector device according to any one of the preceding Claims, characterized in that it is adapted to be connected as a pre-processor in combination with some other receiving and processing device.

13. A symbol detector device according to Claim 12, charact¬ erized in that its output is multiplied in a multiplier means

(20) with the observed channel behaviour, and the output of its value providing means (11;11') is added in an adder means

(21) to the output of said multiplier means.

SUBSTITUTE SHEET