FR2811181A1 - Mobile telephone signal transmission equalisation method having quantified signals representing symbols with confidence information formed using difference value/between group value. - Google Patents

Abstract

The equalisation method has a signal transmitted producing a quantified signal, representing symbols. Each symbol represents two groups of binary information. For each symbol, a confidence level is evaluated. Channels are processed decoding quantified signals and confidence information. The confidence level is found using two values, a difference value and a value between two groups of binary information bits.

Description

Method of equalizing a signal transmitted via a channel and use in

  The subject of the invention is a method of equalizing a signal transmitted via a channel which can in particular be used in mobile telephony. It is intended to apply to all areas of transmission: by radio (radio-relay systems, radiomobiles, satellite transmissions, ...), or by cables (metallic, coaxial, optical fibers, ...), and globally in any "modem" transmission in the broad sense of the term. These transmissions can also be used to carry voice, images (still

  or animated), or any type of digital data.

  To be concise, the invention is applicable to any digital transmission chain which uses an equalization device in reception

  followed by a channel decoding device.

  In almost all digital transmission chains, the signal propagation channel is not linear and generates

  inter-symbol interference (in English ISI for Inter Symbol Interference).

  This interference comes in particular from the non-linearities of the propagation domain, and from the multi-paths encountered for example in the

  over the air and mobile telephony.

  In order to eliminate this interference, equalization devices (or "equalizers") are generally used. Such devices use an estimation of the propagation channel, considered to be constant over relatively short time intervals. This involves in particular identifying the delays of the main paths and correcting the effects of interference between symbols. The signal thus processed generally feeds a channel decoding device which uses redundancy in the digital message to be transmitted in order to correct any residual errors, and allows

  ultimately recover the information.

  In order for the overall signal reception operation to take place with the minimum of errors, it is easily understood that the information that the equalizer transmits to the channel decoder must be as reliable as possible. Rather than simply transmitting the bits identified by the equalizer (a process known as having a "hard" output), a confidence value can also be associated with them and be communicated to the channel decoder. When the equalizer is thus output "flexible", we say

  also that it is flexible. In this case, the channel decoding must itself -

  even being able to interpret these confidence values communicated to it by the equalizer, it must be "flexible entry". Such flexible input channel decoders are known. In practice, the gain brought by the use of flexible decisions can be equivalent to an increase in the signal-to-noise ratio of 2 to 3 decibels, which is very significant, with a slight increase in the number of processing operations necessary for the development of the values. of confidence. However, the construction of reliability measures must be efficient. In the following paragraphs, the different types of hard or flexible outputs that can conventionally be produced by an equalizer are highlighted, in order to better specify the interest for the designer of the chain of

  transmission to opt for a flexible decision device.

  The modulation of a carrier by a binary information sequence to be transmitted can be carried out so that several consecutive bits are simultaneously transmitted. Thus the phase and / or amplitude modulation of a carrier results in a quantified coloration of this carrier, so that each sample of the transmitted signal is representative of a "binary word" consisting of one or more bits (we denote conventionally by "symbol" or, to avoid any ambiguity when necessary, by "quantified symbol" each binary word; and the

  duration of the sample is called "symbol duration").

  For example, the so-called BPSK modulation (Binary Phase Shift Keying Phase modulation with 2 states, comparable to an amplitude modulation), very conventionally encountered, implements an amplitude modulation of the carrier to lead to the representation , for each quantized symbol, of a single bit (worth 0 or 1, or - 1 or +1 according to the conventions

adopted).

  Other digital modulations such as PSK (Phase Shift Keying -

  Modulation by phase jumps) lead to the only phase modulation. In particular, the 8-PSK modulation makes it possible to represent 8 different quantized symbols of constant energy, each of them carrying a

  binary information of 3 consecutive bits.

  Without being exhaustive, among the classic modulations, mixed amplitude and phase modulations still exist, with

  in particular QAM (Quadrature Amplitude Modulation -

  Amplitude in Quadrature) which generally use a large number of symbols. We can also represent each modulation by a "constellation" of its symbols positioned in a two-dimensional plane space, where the energy of the phase component of each symbol is on the abscissa, and the energy of the quadrature component is on the ordinate. An overview of these different modulations and of the coding that a signal to be transmitted can undergo is for example recalled in the

  preamble to document EP - A - 0 700 164.

  When such modulations are used, it is then a question, at the level of the receiver, of identifying each of the symbols of the constellation which have been successively transmitted. However, the symbols have undergone the effects of the propagation channel and have been altered by transmission. They then no longer have the quantified ideal values expected in phase and in amplitude. The role of an equalizer with hard outputs is precisely to find these quantized values from the sequence of samples received and to produce them at

his exit.

  Before going into more details, it is important to differentiate the meaning of the terms used below. A distinction is made between the "samples" available at the input of the equalizer and which must be processed by it, from "estimated symbols" which are each a pair of coordinates (not quantified) - in the plane of energies in phase and in quadrature from calculations performed by the equalizer. From each estimated symbol, the equalizer can in certain cases take a decision on the quantified symbol which it will supply at its output, it is here the "decided symbol". When the equalization takes place in a satisfactory manner, each estimated symbol must therefore be close (in terms of distance) to the quantified symbol of the constellation which should ideally be decided for an error-free transmission. And in such a case, each symbol decided must naturally be

  strictly equal to each symbol emitted by the transmitting device.

  Thus, conventionally, when the equalization provides a hard output, the choice of the decided symbol (the choice of its corresponding position in the constellation) is carried out from the "decision zone" in which the symbol estimated by the equalizer. This decision zone simply designates the set of points of the plane which have the same quantified symbol for the same symbol of the constellation. There are therefore as many decision zones as quantified symbols represented. In the following, the term "decision border" will be used to define the delimitation between two contiguous decision zones. Finally, each symbol estimated by the equalizer only corresponds to a point in the plane whose closest quantized symbol is a priori the sought-after transmitted symbol, and therefore the actual output

hard output equalizer.

  In FIGS. 1 and 2 presented, each cross represents a symbol Ei estimated by the equalizer and represented in a constellation of quantified symbols Sk. The latter are presented as they are constructed on transmission. The decision areas highlighted by decision boundaries represent a fraction of the symbol space. Any Ei symbol estimated by the equalizer is therefore necessarily in one of these zones,

  and is automatically assimilated to the nearest Sk symbol.

  In FIGS. 1 and 2, in examples of cases of BPSK and 8PSK modulations respectively, the estimated symbol Ei is assimilated to a symbol S1 by the equalizer. The group of bits corresponding to the symbol S1 will then be processed by the channel decoder. It is the group of bits most likely transmitted with regard to the relative distances between each

  constellation symbol and the estimated Ei symbol.

  The type of hard output device presented previously allows satisfactory performance to be obtained, but however for symbols estimated to be very close to the zone limits - and therefore almost at equal distance between two symbols of the constellation - this type of force decision-making the equalizer to commit clearly and without weighting for one or the other of the quantified symbols, and to make the choice of a decided symbol when the probability of mistaking a symbol is manifestly high. The flexible decisions are born from this observation, and from the possibility of being able to postpone the final choice of the symbol decision up to the level of the channel decoder, which has the redundancy introduced into the signal on transmission to recognize the symbols with

more certainties.

  Thus, conventionally, the equalizer with flexible output no longer provides only the decided symbol, ie S1 for Ei estimated in FIGS. 1 and 2, but also a confidence value which is attached to the decision on this symbol. This confidence value is then all the stronger as the

  estimated symbol is distant from other neighboring decision zones.

  In order to present the construction of flexible decisions with the maximum of generality without claiming to be exhaustive, we can say that each quantified symbol decided will be weighted, and that this weighting will depend on a ratio (or a numerical combination) of the distances which separate the estimated symbol from the other quantified symbol or symbols which

  are the closest in the constellation.

  In FIGS. 1 and 2, two cases of low and high confidence values are illustrated, on the left and on the right, respectively. Note that in the comments which follow, linked to these illustrations, the distance ratio has been arbitrarily chosen as being the method of evaluation of the confidence measure, but another numerical combination could be used.

  in the same way to carry out the same type of assessment.

  In FIG. 1, a binary constellation shows on the left an estimated symbol Ei which will be assigned a low confidence value by the equalizer, taking into account a distance ratio M1 / X2 close to 1 (reflecting a high probability of make a mistake by choosing the symbol S1), and on the contrary on the right an estimated symbol Ei which will be assigned a high confidence value due to a ratio XA2 close to 0 and which translates much more certainty on the symbol S1 than we decide. Intermediate confidence values are also provided for reporting

distance between 0 and 1.

  Between FIG. 1 on the left and FIG. 2 on the left, the same observation of low confidence value is made with regard to the value of the symbol decided, here S1. On the other hand, figure 2 on the right shows a symbol S1 decided with more certainty taking into account the best report

X1 / X2.

  In all cases, the assignment of intermediate confidence values for 1 / X2 ratios between 0 and 1 or more generally for numerical combinations giving results reflecting non-extreme confidence values, may depend on different types of scales. (linear or exponential) according to the choice of the designer. The values of the ratios; 1 / X2 (or any numerical combinations used) are directly readjusted according to the scale chosen to form

  flexible equalizer information.

  In the state of the art and in the case of an equalization with flexible output, the calculated confidence values therefore weight each symbol as it was presented previously, without taking into account that, if the symbol represents several bits, all the bits of this symbol do not necessarily have to be penalized by the same low value of

  trust. The invention proposes to remedy this defect.

  Indeed, in the state of the art, the boundaries used to assess the distances are considered to be significant of the ambiguity of the meaning of a symbol. However, the invention notes that, if a symbol represents more than one bit of information - typically if it represents at least two, three or more (denote by X this figure) - identifying a symbol in the constellation amounts to equivalent way to identify those

  values of the X bits which constitute it.

  Some terms need to be introduced here. In the following, a "group of bits" designates a set of one or more bits in given locations, constituting part of the binary meaning of the quantized symbols, the positions of which are not necessarily consecutive on the X possible positions. In the example of a 16 - PSK modulation where each symbol is significant by 4 bits, a group of bits can for example group together the single third bit, or the set formed by the first and the third bit, or even the set formed by the last three bits at positions two to four. Two distinct groups of bits grouping bits together in distinct positions can each group together a number of different bits. In addition, groups of bits may not be disjoint, a

  same bit can belong to several groups.

  To repeat the description of the invention, the process of taking

  decision on a single symbol is therefore analogous to a decision process on each group of bits identifiable among the X bits. It follows from this observation that, just as it was possible to distinguish decision zones relating to each quantified symbol, it is also possible to distinguish decision zones specific to the value of each group of

  identifiable bits among the X bits that make up the quantized symbols.

  According to the invention, a border is thus significant of the ambiguity of a group of bits, carried in each symbol, the value of which is different on either side of the border. In other words, a border typically leaves on either side a group of one or more symbol values whose group of bits related to each symbol has the same

  value near and on the same side of the border.

  The number of traceable boundaries, and to be considered in the overall decision process, increases with the number of groups of bits represented and identified by the designer of the receiving device. In all cases, this same designer must have identified enough groups of disjoint bits so that each of the X bits which constitute the symbols is included in at least one group of bits. For each group of bits considered, one or more boundaries of decision zones can be highlighted, from which the constitution of two or more decision zones is derived. The number and the distribution of these decision zones depends, in addition to the proper structure of the constellation of symbols, on the allocation - carried out by the modulation designer - of the

  values of the X bits for each of the symbols of the constellation.

  Note that the relevant designation of these boundaries, and of the different groups of bits associated with them, obviously implies the

  responsibility of the designer of the receiving device.

  According to the invention, the boundaries must be significant of the ambiguity of a group of bits associated with a symbol, but not of the ambiguity of all the bits of the symbol taken as a whole. Rather than penalizing the whole symbol, therefore all the bits of the symbol by a single confidence value conventionally developed from decision zones per symbol, the invention thus proposes to assign as many confidence values as there are. bit groups by symbol designated by the receiver designer, based on the boundaries relating to the ambiguities of the different bit groups. In practice, it will therefore be assigned to each group of bits of each estimated symbol a confidence value

  different from that of other groups of bits of the same estimated symbol.

  The invention proposes to carry out the calculation of the confidence value assigned to a group of bits by measuring the distances of the symbols estimated at the different boundaries of zones relating to the ambiguity of this group. For each group of bits to be evaluated, it is in all cases the distance from the estimated symbol to the zone boundary - relative to the decision on

  this group - the closest that determines the trust value to assign to it.

  According to the structure of the constellation and the values of groups of bits carried by the different quantized symbols, the calculation of a confidence value thus passes by the prior estimation of as many distance measurements as there are boundaries of group ambiguity areas

bits processed.

  In a preferred variant, it is possible to replace each distance measurement of the symbol estimated at a zone border by the distance measurement of the symbol estimated at the nearest quantified symbol of the constellation positioned on the other side of the zone border. This variant provides performance comparable to the version presented more

  above, and it is briefly illustrated later in this document.

  The subject of the invention is therefore a method of equalizing a signal transmitted via a channel, in which - this transmitted signal is processed to produce quantized signals representative of symbols transmitted by this signal, - each symbol representing at least two groups of binary information bits, - an associated confidence information is evaluated for each symbol, - and the quantized signals and the associated confidence information are processed in a channel decoder to produce binary information bits corresponding to these symbols , characterized in that - for each symbol at least two confidence values are evaluated which can be associated, respectively and in a differentiated manner, with

  each of the at least two groups of binary information bits.

  It also relates to a method for equalizing a signal transmitted via a channel, in which - this transmitted signal is processed to produce quantized signals representative of symbols transmitted by this signal, each symbol representing at least two groups of bits d binary information, - an associated confidence information is evaluated for each symbol, - and the quantized signals and the associated confidence information are processed in a channel decoder to produce the binary information bits corresponding to these symbols, characterized in what - a confidence value corresponding to a group of bits is measured by evaluating a distance from a symbol measured at an area boundary, - this area boundary separating, at least locally, the value space of the symbols into two zones, a first zone where the group of bits equals a first binary value and a second zone where the group of bits

is worth a second binary value.

  The subject of the invention is also a method of equalizing a signal transmitted via a channel, in which - this transmitted signal is processed to produce quantized signals representative of symbols transmitted by this signal, - each symbol representing at least two groups of binary information bits, - an associated confidence information is evaluated for each symbol, - and the quantized signals and the associated confidence information are processed in a channel decoder to produce the binary information bits corresponding to these symbols , characterized in that - a confidence value corresponding to a group of bits is calculated by evaluating a distance from an estimated symbol to another quantized symbol of a constellation, - a zone border separating, at least locally, I ' value space of the quantized symbols in two areas, a first area where the group of bits equals a first binary value and a second area where the group of bits is worth t a second binary value, - the said other quantized symbol belonging to one of the two zones

  thus formed does not include the measured symbol.

  The invention will be better understood on reading the description which follows

  and examining the accompanying figures. In all cases, the example cases treated are presented for information only and in no way limit the invention. The figures show: - Figures 1 and 2 (already commented on): representations of BPSK and 8-PSK type constellations; - Figure 3: a schematic representation of a transmission chain, useful for illustrating the method of the invention; - Figures 4a to 4d: the implementation of the invention in the context of an 8-PSK type modulation; FIGS. 5a to 5d: the implementation of the invention in the context of a 16-QAM type modulation; - Figure 6: a representation of decoding results with the various known processing solutions and with the invention;

  - Figure 7: a preferred form of flexible value assessment.

  FIG. 3 shows a transmission chain which can be used to illustrate the method of the invention. A digital signal S (t) to be transmitted is preferably admitted to an input of a source encoder 1. The encoder 1 performs a compression of the signal S (t) to limit the bit rate to a useful part only. An output of encoder 1 is connected to a channel encoder 2 which adds redundancy to this compressed signal to make it more resistant to the noise and faults of a transmission channel to be used. An output of the encoder 2 is connected to a modulator 3. The modulator 3 is connected to a transmission channel 4, the output of which is connected to a reverse architecture, comprising a demodulator 5, a channel decoder 6 and a source decoder 7. The different coders and decoders are shown here in a differentiated manner, but, the signal to be transmitted being digital, all the processing of coders and modulators 1 to 3 and or of decoders and demodulators 5 to 7 can be carried out by a processing processor implementing a program of

  treatment. Optionally, the processor is assisted by co-

  processors which are wired circuits produced inside a microprocessor integrated circuit. In one example, organs 5 to 7 are contained

in a mobile phone.

  The modulator 3 modulates a carrier by the signals coming from the encoder 2. The modulation can be in baseband, in which case the central frequency of the carrier is 0. The processing according to the method of

  The invention is essentially implemented in the demodulator -

  equalizer 5, with a slight modification to decoder 6.

  On reception, in the demodulator - equalizer 5, the signal transmitted via the channel 3 is processed to produce quantized signals representative of symbols transmitted by this signal. In the case where the decoder 6 is a decoder with flexible inputs, the demodulator-equalizer 5 classically evaluates for each symbol a confidence value which is associated with it. These quantized signals and these associated confidence values are processed in the decoder 6 to produce the binary information bits corresponding to these symbols, and to the compressed signal S (t). This last compressed signal is decompressed (if it has been previously compressed)

in decoder 7.

  The modulation according to the invention is such that each symbol of the modulated signal represents more than one bit of the signal from the channel encoder 2, for example at least two bits. In general, each symbol will represent many more bits, and will comprise at least two groups of bits, a group of bits which may comprise one bit, and another group of bits which may also comprise one bit. In the case of an 8-PSK type modulation, a first group of bits of the symbol could correspond to one bit and another to two bits, or else there would be three groups. In the case of a 16-QAM type modulation, there can be two groups of two bits or four groups of one bit. Any other distribution is also possible. According to the invention, each differentiated group of bits will then be assigned a confidence value specific to this group of bits, and no longer as previously specific to the entire symbol. For this purpose, the channel decoder 6 will be slightly modified. In fact, during a preliminary step, a flexible input channel decoder assigns to all the bits of an estimated symbol a confidence value equal to the unique confidence value transmitted to it by the equalizer 5. In the invention, this preliminary step (or of the wired circuits which carry it out) is changed so that the assignment of the confidence value is no longer the same for all the bits, but so that it is differentiated and so that the values of confidence to assign to each bit are those that the equalizer 5 communicates in correspondence of each of the bits or groups of bits. For groups of bits, in the invention, all the bits of the same group of bits of a symbol will be assigned the confidence value proper to the group of bits. Since there are at least two groups

  bits, there are at least two confidence values.

  FIGS. 4a to 4d represent a constellation with eight phase states corresponding to expected symbols SO to S7. They show, by comparison with FIGS. 2a and 2b, and in the case of an 8-PSK type modulation, the splitting modes and the differences in evaluation of the confidence values according to the invention. Each symbol represents three bits. It was arbitrarily chosen in this example to consider three groups of bits comprising only one bit each; which implies a division highlighting zones specific to each of the 3 bits which encode the symbols. Figure 4a shows the division of the areas of belonging of

  symbols estimated to clarify the ambiguity of the first bit of these symbols.

  FIGS. 4b and 4c show the divisions corresponding to the second and third bits. In FIG. 4a, a straight line 8 inclined to 22 5 with respect to the horizontal axis separates the space in two: above the straight line 8 the first bit is worth 0, below it is worth 1. For Figures 4b and 4c, similarly lines respectively 9, and 10 and 11, inclined to 112 5, and 157 5 and 67 5 make it possible to divide the space into zones where the second and third bits are worth 0 or 1 respectively: the affected values are indicated in ovals. Depending on the position of an estimated symbol 14 with respect to each of these lines, a value bi of an ith bit is assigned for the estimated symbol corresponding to the measured symbol.

  According to the position in phase and in amplitude of the estimated symbol, and according to the invention, one can, moreover, consider as representative of its confidence value the distance which separates this estimated symbol 14 from the line 8 for the first bit, of line 9 for the second bit, and lines 10 and 11 for the third bit. In the latter case, we will take as a precaution

  the lesser of the distances to any one of these two lines 10 and 11.

  In fact, the smaller the distance, the more ambiguous the bit. It is therefore necessary to take the shortest distance to avoid errors. In this 8-PSK modulation example, the measurement of a confidence value can thus pass

  by measuring four distances and retaining only three.

  Each of the three measured distances is quantified (possibly treated as will be seen below on examining FIG. 7) to be associated in

  as long as the confidence value ci has a bit bi of rank i in the symbol.

  When the invention is implemented according to the preferred variant which was mentioned above, and if we are still interested in this same estimated symbol 14, we can also consider as representative of its confidence value a distance which separates this estimated symbol 14 of the quantized symbol SO for the first bit, of the quantized symbol S3 for the second bit, and of the quantized symbols S1 and S4 for the last bit. In the latter case, the lower of the two distances is the only one retained. Concerning the first bit, the symbol SO is the quantified symbol closest to the estimated symbol 14 because it belongs to a decision zone - highlighted by the related border - not itself comprising the estimated symbol 14. In this In this case, the estimated symbol and the quantified symbol, here SO, to which the distance is calculated are necessarily in separate decision zones, with respect to the same zone border. For a bit of rank given of an estimated symbol 14, bit of rank one, two and so on, one can then find the quantized symbol with respect to which the distance must be measured by finding which is the closest quantized symbol. Then, for each of the quantified symbols, a correspondence table is established between this closest quantized symbol and another quantized symbol (with respect to which the distance will be measured). This other quantized symbol is also differentiated according to the rank of the bit for which we want to calculate a

trust worth.

  In all the cases of implementation of the calculation of the confidence values, the decoder 6 thus receives, for each bi bit value a confidence value ci, during the preliminary step 12. According to the invention, one of the values ci for a bit bi is a priori different from another value cj for

one bit cj.

  Other measurements of the confidence values for the first bit or for each of the bits would also be possible. For example, we could measure the area of a triangle formed by a height passing through an estimated symbol, lowered on a straight border, by this straight border, by the center C of the constellation and by the estimated symbol. The principle is to involve in the calculation of the confidence value a straight line (or a curve) border separating the space of the values of the symbols in two zones: a first zone where a group of bits is worth a first binary value and

  a second zone where the group of bits is worth a second binary value.

  This group of bits does not include all the bits of the symbol. Then we do the same for another group of bits different from the first group of bits. In all cases, this measure will be called "distance"

  representative of a confidence value made with reference to a border.

  It will be noted that the distribution of the binary meaning of the symbols of the constellation was carried out according to the application of a GRAY code in which, two symbols immediately contiguous in the constellation of the modulation differ only by the value of one bit. The invention does not however impose the implementation of a GRAY coding. Failure to use GRAY coding would simply have the effect of creating different decision areas, possibly in different numbers. The number of distances to be assessed could vary depending on the distribution of the decision area boundaries, but the basic principle would remain the same. Note that the invention can also be applied in the context of phase and amplitude modulations such as QAM modulations for example, and that in

  all these cases the GRAY coding remains optional.

  For the calculation of the confidence values, it has been shown in FIG. 4a that the distance 13 from an estimated symbol 14 to the right 8 was representative of the confidence value of the first bit (here 0) that we are looking for. In the example, we will tend to assimilate the estimated symbol 14 to the quantized symbol S2 (or S1), with in all cases a first bit at 0. Measuring the distance 13 can lead to performing a series of operations comprising several complex treatments. Rather than doing these calculations which can be long (on sixteen bits), it is preferable in the present case to rotate the symbol 14 around the center C of the constellation by an angle A equal to the angle of the direction of the straight line 8, so as to assimilate the distance 13 to the height 15

  of the symbol 14 once it has been turned to reach position 16.

  Since at the end of such a rotation we only retain the ordinate of the point, the calculations linked to the rotation (a complex multiplication) can, in addition, be reduced by a factor of 2 and lead to a reduced number

  instructions including two multiplications and one addition.

  Figure 4d also shows for a measured symbol Ei the calculation of three distances; 1 to X3. By using the segmentation by zones thus specified, it therefore becomes possible to assign a specific confidence value to each of the bits i (i = 1, 2 or 3) of the decided symbol, this being all the stronger as the height? i from the symbol Ei is large. The example of FIG. 4d is also taken up there by showing the limits of zones per bit. Although there remains a low confidence value on the first bit of the symbol S1 linked to its proximity to the decision boundary axis of the first bit, the other two bits are assigned a significantly higher confidence value , U2 and; 3 highlighting margins appreciably

more comfortable.

  We observe on examining this figure 4d that with the rotations, we access, with the ordinates and the abscissas of the points 16, the distances from the symbol 14 to two of the lines 8 and 9, or 10 and 11 each time. Ultimately by rotating the positions in the measurement space of the measured symbols, by plus or minus 22 5, and by reversing the sign of one or the other or of the two coordinates, the four distances sought are obtained. Distance comparisons are then made to calculate the confidence values of the bits whose determination area is located between two lines. We then obtain the three desired confidence values. All of the considerations and types of processing set out for the example case of an 8-PSK modulation can be generalized to any modulation of the PSK type, the number of zone boundaries to be taken into account can of course vary, depending on the structure. of the constellation

  chosen and the binary allocation which is performed on the different symbols.

  The illustrations in Figures 4a to 5d show that it is possible to

  highlight decision areas per bit for 16- modulation

  QAM, and by extension for any QAM type modulation, in order to develop flexible decisions per bit. In this case, the bit decision areas can be organized differently. Typically, straight borders

  no longer pass through the center C of the constellation and are organized into strata.

  FIGS. 5a to 5d which follow show the divisions of the zones to be used for such constellations. The example presented is that of a 16-QAM modulation with GRAY coding. It shows six lines 17, 18 and 19, 20 and 21 and 22 which allow respectively to give a value to each of the first, second, third and fourth bits

  of a symbol obtained by a modulation respecting this constellation.

  Depending on the position of the symbol with respect to these lines, for a bit of given rank, the estimated value of the bit is very easily obtained. The processing operations to be carried out again relate to the calculation of distances of estimated symbols with respect to zone boundaries (specific to the different groups of bits). It can be seen here, moreover, that the calculation of the confidence values by the calculation of the distances of the symbol measured at these lines 17 to 22 is particularly simple. It comes down to taking into account the abscissas, or ordinates, symbols measured for the distances at lines 17 and, and the same elements added or subtracted from an offset value

  for the distances to lines 18 and 19 or 22 and 21 respectively.

  Furthermore, if the use of GRAY coding reinforces the interest in the invention (since it makes it possible to improve performance more significantly by drawing larger decision areas per bit), it is however not essential to use it so that the proposed solution finds an advantage to be implemented. In addition, do not use GRAY coding to further fragment the symbol space by increasing the number of zone boundaries, at the same time as it decreases the extent of each decision zone. Finally, the number of distances to calculate to determine a confidence value can significantly increase

  and bring more complexity to the receiver.

  Through these examples, and whatever the modulation used in a digital transmission system, it is easy to understand that it is always possible to highlight decision zones grouping together one or more symbols which have the same group of bits in common. The invention can therefore be applied in all cases of

  figure and finds particular interest in non-digital transmissions -

  binaries that have more than two symbols in their constellation.

  FIG. 6 is an illustration of the performances of the method of the invention obtained in simulation on a DFE (Decision Feedback Equalizer) type equalizer. It is generally recognized that a DFE equalizer of the MSE type (Mean Square Error - linear equalizers or even with maximum mean square error) displays, after detection, gross equalization performance less good than an MLSE equalizer (Maximum Likelihood Sequence Estimation - maximum likelihood equalizer) which is based on a VITERBI algorithm. We were able to verify this difference in performance by means of simulations on current developments. However, these simulations also showed that with flexible decisions generated bit by bit as proposed in the invention, a DFE equalizer made it possible finally to obtain, after channel decoding, performances comparable or even superior to those offered by an MLSE equalizer. with undifferentiated flexible decisions. This provides proof that the gain brought by the flexible decisions developed bit by bit is very real. Note that an MLSE equalizer to which the invention would be applied, therefore with differentiated flexible decisions

  on bits, should provide even better results.

  In FIG. 6, the raw BER denotes the Bit Error Rate on the data just after equalization, and the post-decoding BER denotes the bit error rate after the channel decoding device. The curve relating to the DFE with flexible decisions bit by bit, but before channel decoding, presents a less good gross BER than that relating to the MLSE, with flexible decisions. We can see, however, that after channel decoding, the flexible bit-by-bit decisions allow the DFE to regain the advantage, while the plotting of the performance curve of the same DFE with flexible decisions, but elaborated symbol by symbol, would have shown

  performance noticeably worse than that of MLSE.

  Figure 7 highlights the correspondence curve implemented to generate the confidence values (on the ordinate) from the distance measurements (on the abscissa). In the example illustrated, the confidence values are quantized over 3 bits and take values 1 to 8. Preferably, this curve is not simply linear but of the root type "n" th of d (d being the distance) as shown. This is intended to increase the discrimination of bits with a low confidence value. Indeed, the distance evaluated is all the more critical on the confidence that it must be granted that the distance is itself small. On the contrary, the bits associated with much greater distances, reflecting high confidence values, do not need a high level of discrimination. When the distances evaluated are greater than a threshold (set by the designer), it is also possible to systematically assign the maximum confidence value without this having any effect on performance. In an example the confidence values will be developed on 8 levels, three bits, while the distances

will be evaluated on 16 bits.

  Finally, it should be noted that, in a transmission chain - such is the case for example in GSM-type mobile telephony systems - signal equalization can be carried out optimally using an MLSE equalizer using the VITERBI algorithm. However, the implementation of such an algorithm may require a very large number of treatments (even prohibitive), which obliges the designer to choose an implementation solution which is costly in terms of resources as much as from a financial point of view. Other well-known equalization solutions exist, such as linear equalizers or DFE equalizers which have the advantage of offering a very interesting performance / complexity ratio of implementation, but at the expense of the performances obtained. The tests carried out so far have shown that by using the invention jointly with a linear device or DFE, the performance obtained could be very significantly improved, which further increases the interest in solutions for equalization of

relative low complexities.

Claims (8)

  1 - Method for equalizing a signal transmitted via a channel, in which this transmitted signal is processed to produce quantized signals representative of symbols transmitted by this signal, - each symbol representing at least two groups of bits of binary information, - an associated confidence information is evaluated for each symbol, - and the quantized signals and the associated confidence information are processed in a channel decoder to produce binary information bits corresponding to these symbols, characterized in that - on evaluates for each symbol at least two confidence values which can be associated, respectively and in a differentiated manner, with   each of the at least two groups of binary information bits.   2 - Method for equalizing a signal transmitted via a channel, in which this transmitted signal is processed to produce quantized signals representative of symbols transmitted by this signal, - each symbol representing at least two groups of binary information bits, - an associated confidence information is evaluated for each symbol, - and the quantized signals and the associated confidence information are processed in a channel decoder to produce the binary information bits corresponding to these symbols, characterized in that - on calculates a confidence value corresponding to a group of bits by evaluating a distance from an estimated symbol to a zone border, - this zone border dividing, at least locally, a space of value of the symbols into two zones, a first zone o the group of bits is worth a first binary value and a second area o the group of   bits is worth a second binary value.   3 - Method for equalizing a signal transmitted via a channel, in which this transmitted signal is processed to produce quantized signals representative of symbols transmitted by this signal, - each symbol representing at least two groups of binary information bits, - an associated confidence information is evaluated for each symbol, - and the quantized signals and the associated confidence information are processed in a channel decoder to produce the binary information bits corresponding to these symbols, characterized in that - on calculates a confidence value corresponding to a group of bits by evaluating a distance from an estimated symbol to another quantized symbol of a constellation, - a zone boundary separating, at least locally, the value space of the quantized symbols in two areas, a first area where the group of bits is worth a first binary value and a second area where the group of bits is worth a second binary value, - the said other quantized symbol belonging to one of the two zones   thus formed does not include the measured symbol.   4 - Method according to one of claims 1 to 3, characterized in that   that - we evaluate for each symbol as many confidence values as   there are binary information bits per symbol.   - Method according to one of claims 1 to 4, characterized in that   that - the transmitted signal comes from a signal to be transmitted, - the signal to be transmitted undergoes a phase-type modulation and or   in amplitude, for example 8 - BPSK or 16-QAM.   6 - Method according to one of claims 1 to 5, characterized in that than   - the signal to be transmitted is coded according to a GRAY code.   7 - Method according to one of claims 1 to 6, characterized in that   that - the quantized signals are processed in a DFE type channel decoder.   8-Method according to one of claims 1 to 7, characterized in that   we evaluate confidence values with a non-linear scale.   9 - Use in a mobile telephone of the method according to one of claims 1 to 8.