FR2702110A1 - Method of automatic frequency control for receiving multi-carrier signals, and corresponding receiver - Google Patents

Abstract

The invention relates to automatic frequency control in receivers of digital signals transmitted using a plurality of carrier frequencies (FDM signals). The method of the invention, for at least one regularly transmitted frequency synchronisation symbol having characteristics known to the receivers, comprises an operation (410) of determining an estimate (411) of the frequency offset between the oscillation frequency (fr) of the reciver and the nominal frequency of the received signal (41), by means of a calculation taking account of the value of at least one synchronisation data element forming the said frequency synchronisation symbol and the value of at least one of the adjacent data elements (on the frequency axis), in the form of an arithmetic ratio making it possible at least partially to eliminate distortion induced by the transmission channel. The invention also relates to a device implementing this method.

Description

 Automatic frequency control method for receiving multicarrier signals, and corresponding receiver.

 The field of the invention is that of receiving digital signals.

More specifically, the invention relates to the reception of signals transmitted simultaneously over a plurality of carrier frequencies. In other words, the invention applies to signal receivers transmitted according to the frequency division multiplexing technique (in English: Frequency Division
Multiplex (FDM)).

By way of example, the transmission system of the invention may be of the type of the digital broadcasting system described in particular in the French patent.
FR-86 09622 filed July 2, 1986 and in the document "Principles of Channel Modulation and Coding in Digital Broadcasting to Mobile" (Mr. Alard and R.

Lassalle; U.E.R. Journal, No. 224, Aug. 1987, pp. 168-190), and known as Coded Orthogonal Frequency Division Multiplex (COFDM).

 This COFDM system, which will be described in more detail later, is based on the combination of two essential principles.

 The first principle is the distribution of the information to be transmitted on a large number of modulated carriers each at a low rate, in order to reduce the selectivity effect of the transmission channel.

 The second principle of the COFDM system consists of correlating consecutive information elements by a coding method and transmitting them at points distant from the time-frequency domain (technique of time and frequency interleaving). The distance of these points is chosen so as to ensure the statistical independence of the disturbances that they are likely to undergo during the transmission.

 The coding used is convolutional. It is advantageously associated with a decoding implementing a Viterbi algorithm in soft decision.

 The COFDM system was developed in the framework of the European project Eureka 147 "DAB" (Digital Audio Broadcasting). He is also a candidate for standardization for the terrestrial broadcasting of digital television.

 The invention applies more generally to all transmission systems, since they implement several carrier frequencies transmitted simultaneously. The number of these carrier frequencies is indifferent.

 More specifically, the invention relates to automatic frequency control (AFC) in the receivers of these multicarrier systems. This operation, known per se, consists in adjusting in a receiver the oscillation frequency of a local oscillator, so that it is as close as possible to the modulation frequency of a transmitted signal.

 The two-dimensional nature (temporal and frequency) of the signals emitted in these systems indeed imposes, in the receivers, the implementation of a sampling according to the two time and frequency axes.

 The determination of sampling times in time is, in the conventional case, the result of the time synchronization of the receiver.

Frequency sampling results from the application of a filter bank (or equivalent technique) on time samples. This filterbank associates with each sample (or symbol) received the sequence of data elements received corresponding to this sample. In the case of the COFDM technique already mentioned, this frequency sampling consists of the application of a discrete Fourier transform (whose calculation is easy using the algorithm called "Fast Fourier"
Transform ", or Fast Fourier Transform (FFT)), which is equivalent to an orthogonal filter bank.

 The position of this bank of filters in the frequency domain is determined in each receiver by the frequency of oscillation of the local oscillator. The accuracy of the frequency sampling therefore depends directly on the frequency difference between this oscillation frequency and the nominal modulation frequency of the transmitted signal.

 A frequency offset of the filterbank can have several detrimental effects on the demodulation of a particular carrier frequency of a signal transmitted using a plurality of carrier frequencies.

 Firstly, the difference between the frequency of the transmitted signal and the frequency of the local oscillator introduces a phase error with respect to the nominal phase fixed by the modulation.

 Moreover, the result of the filtering is attenuated, the position of the carrier considered does not correspond to the maximum of the associated filtering profile, as is the case when the tuning of the local oscillator is perfect.

 Finally, when the transmission system uses orthogonal carrier frequencies (for example a COFDM signal), a frequency difference can introduce noise due to the loss of orthogonality between the carrier frequencies, said interference noise. intercarrier.

 These various disadvantages will be described in more detail later, in connection with FIG.

 To avoid, or at least limit as much as possible these disadvantages, it is necessary that the frequency of the local oscillator is set very precisely. Thus, in the case of a COFDM system, the accuracy of this oscillator must be much greater than 10-5 (accuracy generally acceptable for single-carrier systems). Indeed, if one considers the case of a COFDM signal for which the difference between two carrier frequencies is 10 kHz and the modulation frequency of lGIXz (classical orders of magnitude), a precision of 10-5 of the local oscillator (ie, 10kHz accuracy) may cause a carrier offset on reception, making any interpretation of the received signal impossible.

 A first solution to these problems is of course to use local oscillators of very high precision. However, such local oscillators are very expensive, incompatible with consumer applications for which the receivers of these broadcast systems are intended.

 It is therefore necessary to use automatic frequency control means to adjust the frequency of the local oscillator based on an analysis of the received signal.

We already know, notably by the document "A new system of sound broadcasting to mobile receivers" (R. Halbert, M. Alard, B. Le Floch, D. Apple tree
Conference Eurocom 88), a first CAF application method. This known method consists of an increase to the power n (where n is the number of possible phase states of the modulation used) of the demodulated signal, followed by an integration.

 However, this method seems to be insufficiently precise, particularly in the case of orthogonal carrier systems.

 To increase the accuracy of the CAF, it was then thought to insert into the transmitted signal frequency synchronization patterns known to the receivers and recurring over time. This diffusion method is described in particular in patent FR 90 01492, filed February 6, 1990.

 According to this method, the receivers extract frequency synchronization patterns from the received signal (consisting for example of a data symbol carrying a reduced number of carrier frequencies whose position and phase are known to the receivers) and compare them with the predefined characteristics. that they must present, if the agreement of the local oscillator is perfect. The result of this comparison makes it possible to adjust the frequency of the local oscillator.

 More precisely, according to this patent, the tuning of the tuning of the local oscillator must be based on an analysis of the modules of the frequency synchronization data elements delivered by the filter bank (or the Fourier transform).

 Indeed, the method of CAF must be very resistant on the one hand to the transmission noise, and on the other hand to the distortions of the channel. The person skilled in the art has always been convinced that, in order to better withstand the latter, it was necessary to base the adjustment processing of the local oscillator on an analysis of the modules (in the mathematical sense of the term, according to the decomposition technique of a complex value in its module and its phase) demodulated data elements, because the module of the frequency response of the channel varies much less rapidly depending on the frequency than its phase.

 However, the analysis of these modules involves a specific pretreatment, and therefore expensive in capacity and calculation time. It is indeed necessary to carry out a complex multiplication by sample. More precisely, each sample is multiplied by a predefined complex coefficient (a set of coefficients being stored in a read-only memory) before the application of the FFT.

 The object of the invention is in particular to overcome these disadvantages of known automatic frequency control methods.

 Thus, an essential objective of the invention is to provide such a method, making it possible to combat, on the one hand, the phase rotation problem due to frequency offsets, and on the other hand the problem specific to OFDM loss signals. orthogonality between the carrier frequencies.

More specifically, an object of the invention is to provide a method of
CAF for multicarrier broadcasting systems using frequency synchronization patterns that is simple and inexpensive (in terms of calculation time and cost) to implement.

 In particular, the invention aims to provide such a method that is compatible with consumer applications, such as the production of sound receivers and / or images, especially for mobiles.

 Thus, an object of the invention is to provide such a method making it possible to use, in the receivers, the use of conventional precision local oscillators.

 Another object of the invention is to provide such a method, which does not require the calculation of the modules of the frequency synchronization data elements, nor any other particular pretreatment of the data elements delivered by the filterbank.

 Another object of the invention is to provide such a method which does not require any modification or adaptation of the transmitters and the structure of the transmitted signals, as long as these comprise frequency synchronization patterns.

 In other words, an object of the invention is to provide such a method which is compatible with the other known methods, and in particular the methods based on an analysis of the modules. Thus, the use of the method of the invention should not be mandatory, but on the contrary be an optional feature of the receivers.

 Another particular objective of the invention is still to provide such a method, which can be used in diffusion systems using the COFDM technique.

These objectives, as well as others which will appear later, are achieved according to the invention by a method of controlling the oscillation frequency of a local oscillator of a digital signal receiver, a method of the type intended for adjusting said oscillation frequency to the nominal modulation frequency of a received signal, said digital signals being transmitted using a plurality of N carrier frequencies, in the form of a succession of data symbols, each of said data symbols comprising a set of N data elements each modulating one of said carrier frequencies, at least one frequency synchronization symbol being regularly inserted among said data symbols, said frequency synchronization symbol consisting of at least one data element of frequency synchronization modulating a frequency synchronous carrier frequency selected from the said N carrier frequencies, said frequency synchronization data element having known characteristics of said receiver, the received signal being further affected by a distortion induced by the transmission channel, the method according to which, for at least one of said frequency synchronization symbols
said received frequency synchronization symbol is demultiplexed so as to obtain N received data elements;
for determining at least one of said frequency synchronization data elements of said frequency synchronization symbol, an estimate of the frequency difference between said oscillation frequency and said nominal frequency is determined by means of a calculation taking into account the the value of the synchronization data element and the value of at least one of the adjacent received data elements, in the form of an arithmetic ratio for at least partially eliminating said distortion;
said oscillation frequency is adjusted according to said estimate of the frequency difference.

 Thus, according to the method of the invention, it is no longer necessary to calculate the modules of the data elements received, which makes it possible to strongly limit the calculations required for the CAF. This new and inventive approach clearly goes against the a priori of the person skilled in the art, who has always considered that it was necessary to use the modules.

 Indeed, the data elements, considered directly, are much more dependent on the distortions induced by the transmission channel than their modules, and it appears that no effective analysis is possible on an independently considered data element.

 The problem of this invention is overcome by relying on a novel approach of taking into account several data elements (and not their modules) and calculating a homogeneous ratio of some of these data elements (more precisely, for a frequency synchronization data element, the corresponding received data element and at least one of the data elements carried by adjacent or adjacent carrier frequencies are taken into account). It is indeed verified, mathematically, that this ratio is not (or little) affected by the distortions of the channel, although each element of data is.

 This ratio advantageously corresponds to a ratio of two weighted sums of the data elements of a symbol. Of course, some of the weighting coefficients (and in practice, most of them) may be zero. They are chosen so that the result of the report is representative of the frequency deviation that one seeks to correct.

 Advantageously, at least some of said frequency synchronization symbols comprise at least two elements of frequency synchronization data, and said arithmetic ratio takes into account the values of at least two of said frequency synchronization data elements and at least one of the elements. data adjacent to one of said frequency synchronization data elements.

 Indeed, if each frequency synchronization carrier is not affected by the distortion of the channel, it remains sensitive to the noise that is added to the samples, especially in the presence of fading.

 To limit this problem, an estimate of the frequency deviation is thus calculated using several frequency synchronization elements (for example 16), and then an overall estimate of the frequency deviation (also called "average") is determined. thereafter, although it is not a real average, in the conventional sense of the term), giving less weight to those corresponding to faded carriers.

 Note that, in practice, this "average" (global estimate) can be obtained directly, if the weighting coefficients of the two sums of the report are determined adequately.

 Furthermore, preferably, the imaginary part of said arithmetic ratios is canceled, said imaginary part being induced by the transmission noise of said signal.

 This makes it possible to suppress the non-significant imaginary term added by the noise to the calculated estimate. Thus, this estimate (and in particular its numerator, which is the critical term) is very robust vis-à-vis the noise.

 As will become clearer later, the three characteristics presented above independently (for reasons of simplification of the presentation) can of course be grouped together in a single calculation.

 Advantageously, the method of the invention comprises a preliminary step of limiting said frequency difference to a value less than or equal to fs / 2, fs being the difference between two consecutive carrier frequencies.

 Indeed, the calculation of the estimate described previously only applies in this context.

 According to an advantageous characteristic of the invention, the adjustment of said oscillation frequency takes into account the sign of said estimate of said frequency difference.

 In a particular embodiment of the invention, said frequency synchronization symbols comprise a number M of carrier frequencies modulated by frequency synchronization data elements less than the number N of carrier frequencies.

For example, said frequency synchronization symbols comprise
The set of data elements {Ck} osksN-ff where C k is a non-zero frequency synchronization data element if and only if k = 16.p + 8, p integer.

 Other types of synchronization patterns may also be used, such as those described in patent application FR 90 01492 already cited.

 Advantageously, the frequency difference between two of said M carrier frequencies is chosen so that it is greater than twice the maximum difference between said oscillation frequency and said nominal modulation frequency.

 Thus, each frequency synchronization element can be considered independently (before calculating the average of the estimates).

According to an advantageous and particularly simple embodiment, said arithmetic ratio is written (Xr -) / X0, where
XO is a received frequency synchronization data element
corresponding to a carrier frequency of frequency synchronization
X4 and X1 are the respective recruited data elements respectively
at carrier frequencies preceding and following said carrier frequency of
frequency synchronization.

 This ratio, particularly simple to calculate, gives indeed a very reliable and sufficiently accurate estimate of the frequency difference.

Furthermore, it should be noted that, although the method described above applies to all frequency division multiplexing systems, it is particularly advantageous for systems implementing orthogonal carriers, and in particular the system COFDM. Indeed, these require
CAF very precise, that the invention allows to obtain with limited calculations, compared to other known methods.

Finally, the invention also relates to receivers and
CAF implementing this method.

Other characteristics and advantages of the invention will appear on reading the following description of several embodiments of the invention, given by way of illustrative and non-limiting examples, and the appended drawings in which:
FIG. 1 is an overall block diagram of a COFDM signal diffusion system of known type;
FIG. 2 illustrates the structure of the signal transmitted in the broadcasting system of FIG. 1;
FIG. 3 illustrates the position of the carrier comb received by a receiver, in the case of a perfect match, then of an imperfect match;
FIG. 4 represents the overall structure, known per se, of an automatic frequency control device;
FIG. 5 is a simplified flow chart showing the general principle of the method of the invention;
FIG. 6 is an example of a frequency synchronization symbol that can be used in the method of FIG. 5;
FIG. 7 shows a particular embodiment of the analysis means of FIG. 4, implementing the method of FIG. 5.

 Before presenting in detail a preferred embodiment of the invention, the main characteristics of the COFDM diffusion system, to which the invention is advantageously applied, are recalled below, and then the phenomenon of fading and how to compensate for them by exploiting the time diversity of the transmission channel.

 It should be noted, however, that the invention is in no way limited to this particular broadcasting system, presented only by way of example, but can instead be applied to all broadcasting systems implementing a plurality carrier frequencies (FDM or OFDM techniques).

 FIG. 1 is therefore a block diagram of a transmission and reception chain of a system implementing the COFDM technique. The characteristics of this broadcasting system are described in particular in the article "Principles of modulation and coding channel in digital broadcasting to mobiles" de3à quoted.

 The COFDM digital broadcasting system is based on the joint use of a channel coding device and an orthogonal frequency multiplexing modulation method.

 The channel coding implements a convolutional code.

 The actual modulation method of this known system makes it possible to overcome the problems related to the frequency selectivity of the channel. It consists in ensuring the distribution of constituent digital elements of the data signal in the frequency-time space and in simultaneously transmitting sets of digital elements on a plurality of parallel broadcasting channels by means of a frequency multiplex using orthogonal carriers. In particular, this type of modulation makes it possible to prevent two successive elements of the data stream from being transmitted at the same frequency.

 Although the invention does not specifically concern the aspect of the coding canaI 119 (it is indeed the characteristics of the modulation (OFDM) which impose the use of a very precise CAF), it is described quickly, so that a complete broadcasting system comprising receivers according to the invention is presented.

 The digital source data Il to be transmitted are therefore subjected to a convolutional coding 12. The general principle of such a code is to associate with each source value a coded value dependent on this source value and of at least one of the values that above. Because of the link thus created between the coded values, it is then possible, at decoding, to reconstruct the sequence of the source values even when a received coded value is false, using a maximum likelihood decoding, such as a soft-decision Viterbi decoding (i.e., a decoding providing an estimate of the value received and a weighting representative of the confidence that can be attached to this estimate).

 Advantageously, an external code of the Reed-Solomon or CSRS type (Cyclotomatically Shortened Reed Solomon (cyclotomatically shortened Reed Solomon code)) can be concatenated with the convolutional code.

 The source data can of course be of any type, be it sound signals, image signals or data signals. They can also correspond to several sources of different origins, issued simultaneously.

Thus, for example, the patent application FR 90 16383 filed on 19.12.1990 in the name of the same applicants proposes an organization of the data into frames and into channels making it possible to ensure in particular the simultaneous transmission of several sound channels (corresponding, for example, to the channels stereophonic of several radio stations), still or moving images, teletext information, paging signals, etc.

 As already stated, the COFDM system relies on the simultaneous use of a plurality of carrier frequencies transmitted simultaneously. The number N of carriers can be any. It is classically of the order of a few hundred (it could also be of the order of a few units).

Each of these carriers is modulated at a low rate (compared to the rate required for a corresponding single carrier system). This reduces
The selectivity effect of the channel.

 The overall transmitted signal is therefore a broadband signal (occupying for example a band of a few megahertz).

 This wide band is an advantage, in the case of systems designed to take advantage of multipath, such as COFDM. Indeed, because of the spreading of the response of the transmission channel, it is very unlikely that deep fading simultaneously affects the entire signal.

 For example, in a frequency band of 8 Mhz, it is possible to define 512 separate carrier frequencies of 15625 Hz. Of these 448 are usable, after elimination of the center frequency of the spectrum and of the side carriers (1 / 8th spectrum) to account for filtering constraints.

 In an OFDM signal, and contrary to the signals of conventional frequency multiplexing methods, the spectra of the different carriers overlap each other. However, the complete signal verifies certain orthogonality conditions, allowing the separation of the information associated with the different carriers, for example by using the technique of the Fourier transform (as specified below). In other words, the notion of orthogonality of carrier frequencies implies that the carrier spectra may overlap, provided that when one of the spectra has its maximum power, ie the frequency accurate of the carrier corresponding to this spectrum, all other spectra have zero power. The decoding is not disturbed if we consider this precise frequency.

 The intersymbol interference introduced in particular by the multiple paths during the transmission can weaken this orthogonality. To avoid this problem, a guard interval (during which no decoding is performed) is inserted between each transmitted symbol. The duration of this guard interval is chosen to be greater than the spread of the impulse response of the channel.

The convolutional coding module 12 delivers coded data elements
Ck 13 belonging to a modulation alphabet. The choice of the alphabet specifies the type of modulation used. For example, for a modulation with 4 phase states (MDP4), the alphabet used is {1 + i, 1 - i, -1 + i, - i}. Many other types of modulation can be used, such as the modulations MDP8, I6QAM or the trellis coding modulations according to the Ungerboeck method.

 The coded data elements 13 are then subjected to a distribution operation in the frequency-time space, which consists in associating with each of the carrier frequencies data elements selected in the sequence of the coded data 13 so as to break, by brewing, the correlation of the distortions experienced by the transmitted samples. By time-frequency space is meant a set of points distributed along two perpendicular axes, the time axis and the frequency axis. Depending on the frequency axis, there are as many points as there are carrier frequencies. According to the time axis, a point corresponds to the duration of a symbol.

 For example, this distribution ensures that at least two successive source data are not transmitted consecutively and / or on the same carrier frequency. More generally, the distance in the time-frequency space between two successive coded data is at least such that the statistical independence between these data is ensured.

 In practice, this distribution 14 in the time-frequency space may correspond to a time interleaving 14A consisting for example of a selective application of delays of different durations, followed by a frequency interleaving 148, consisting of a selective assignment of delayed data elements to different carriers.

 Each carrier frequency is then modulated by the sequence of data elements Ck which is intended for it after the time and frequency interleaving 14. This modulation operation can be performed by the application of an inverse fast Fourier transform ( FFT1) 16 on the sequence 15 of interleaved data elements delivered by the module 14.

The inverse transformation module 16 delivers elementary modulation symbols 17 corresponding to the simultaneous modulation of the N carrier frequencies and each intended to be transmitted during the time interval Ts = ts + A, where ts is the duration of the "useful" symbol , on which the demodulation will be carried out and where A represents the duration of the guard interval (for example: A = Tus / 4)
These symbols 17 are then emitted, in a conventional manner, using a conventional transmission module 18, which performs in particular the digital / analog conversion of the symbols 17, and then a transposition of the corresponding analog signal in the field of radio frequencies.

pour t E [O, TsJ
oÙ fk = fO + k/ts
et avec: N : nombre de porteuses du multiplex de porteuses orthogonales
f0 : fréquence arbitraire
Ck: élément de l'alphabet de modulation.Each symbol emitted x (t) can be written:

for t E [O, TsJ
where fk = fO + k / ts
and with: N: number of carriers of the orthogonal carrier multiplex
f0: arbitrary frequency
Ck: Element of the modulation alphabet.

 The signal transmitted in a transmission channel 19 (generally having multiple paths) is received in a reception module 110, which is also conventional.

où Hk représente la réponse du canal 19 à la fréquence fk.If the guard interval is longer than the impulse response of the channel, and if it varies slowly with respect to the duration Ts of a symbol (invariance of the channel during the duration of a symbol), each received symbol ( unaffected by intersymbol interference) can be in the form

where Hk represents the response of channel 19 at frequency fk.

 In the reception module 110, the received signal is demodulated on the in-phase and quadrature channels of a local oscillator of transposition at the frequency 0 d -1 / (2T) and sampled by an analog / digital converter at a rate of 1 / T, with T = tJN.

(n = O à n-l)
Ce signal 111 est soumis à une transformation (FFT) 112, symétrique de la transformation 16. Cette transformation 112 délivre les données 113 suivantes

sur l/e71semble [(-1)k z(n131, = o a N-l
Ces données 113 subissent ensuite le décodage 120, symétrique au codage canal 120. Elles sont tout d'abord démodulées (114). La démodulation peut être cohérente ou différentielle. Dans le cas d'une démodulation différentielle 114, et si l'on pose:
Xi,k = Hjk.Ci.k où l'indice j représente la dimension temporelle, la démodulation consiste à utiliser au rang j un estimateur simplifié du canal déduit du rang j-1: Hi,k = Xl4,k / Cjl,k
On obtient donc les éléments de données estimées Ci,k = Xi.k / Hj,k.The signal 111 obtained is written

(n = 0 to nl)
This signal 111 is subjected to a transformation (FFT) 112 symmetrical to the transformation 16. This transformation 112 delivers the following data 113

on l / e71set [(-1) kz (n131, = oa Nl
This data 113 then undergoes the decoding 120, symmetrical with the coding channel 120. They are first demodulated (114). Demodulation can be coherent or differential. In the case of a differential demodulation 114, and if one poses:
Xi, k = Hjk.Ci.k where the index j represents the temporal dimension, the demodulation consists in using at rank j a simplified estimator of the channel deduced from rank j-1: Hi, k = Xl4, k / Cjl, k
The estimated data elements Ci, k = Xi.k / Hj, k are thus obtained.

 These data elements 115 are then subjected to a deinterleaver module 116, performing the inverse operations of the module 14, so as to reconstitute the original order of the symbols, which are then directed into a decoding module 117, performing a decoding posterior likelihood, such as soft decision Viterbi decoding.

Indeed, in practice, there is always noise during the transmission of signals. The received signal must then be written
Xj, k = Hi, k Cj, k + Nisk where Njk is a complex Gaussian noise of which each component has a 2 variance #jk. j, k
The decoding according to the criterion of maximum likelihood a posteriori then consists in minimizing the expression:

 The decoding module thus provides, after a possible decoding of the concatenated code, if such a code has been implemented on transmission, the signal 118 corresponding to the source signal 11.

 In the COFDM system, the transmitted symbols are advantageously organized into symbol frames. Figure 2 shows, by way of example, such a structure. More precisely, FIG. 2 illustrates a frame consisting of M successive symbols.

Each frame starts advantageously with four particular symbols
S1, S2, S3 and S4 whose role is specified later. It then comprises a number of useful symbols S5 to SM, each comprising N modulated orthogonal carriers 21.

 The symbol SI is a null symbol, allowing on the one hand to carry out an analog synchronization, and on the other hand to perform the spectral analysis of the broadcast channel. The symbol S2 is a second synchronization symbol constituted by an unmodulated multiplex of all the carrier frequencies, with a substantially constant envelope. It makes it possible to readjust the synchronization more precisely by analyzing the impulse response of the channel. The role and the embodiment of these symbols SI and S2 are described in patent FR SS 15216, filed on 18.11.88, in the name of the same applicants.

 Of course, these synchronization symbols are not mandatory vis-à-vis the invention.

 The symbol S3 is itself a wobble symbol, giving a phase reference for the demodulation of each carrier of the following symbols, when they are modulated differentially.

 Finally, the symbol S4 is a frequency synchronization symbol, used in the method of the invention to adjust the frequency of the receiver's local oscillator. Again, these symbols S3 and S4 are optional, and not mandatory for the implementation of the invention.

 By frequency synchronization is meant the transposition from the time domain in the dual domain of the frequencies of the known time synchronization operation (by insertion of synchronization words) in the single-frequency systems.

As already indicated, such a frequency synchronization symbol comprises at least one data element having characteristics known to the receiver. Three examples of frequency synchronization symbols are given in patent FR 90 01492 already cited.
- a symbol in which only a few carrier frequencies are
retained, the others being set to zero, these frequencies conserved
constituting the frequency synchronization pattern (this type of
symbol is shown in Figure 6, discussed later)
a symbol whose frequency synchronization pattern is
constituted by the deletion of at least one carrier frequency;
a symbol whose frequency synchronization pattern is
constituted by a service word known to the receiver, modulating
least one of the frequencies carrying the symbol.

 More generally, a frequency synchronization symbol must be such that at least one of the carriers constituting it is identifiable by the receiver. The data element modulating this carrier must therefore be known to the receiver, and different from the data elements that modulate the adjacent carriers (so as not to confuse the reference carrier with its neighbors).

 FIG. 3 shows the position of the carrier comb received in a receiver, with respect to the frequency response of the filter bank produced by the Fourier transform, in the case of a perfect match 32 of the local oscillator, then imperfect agreement 33.

 The frequency response 31 of the filter bank consists of the superposition of the frequency response 34A, 34B, 34c of each filter. Each of these responses 34c comprises a main lobe 35, which is the useful part of the filter, and a series of secondary lobes 36A, 36B, 36c which create inter-carrier interference (the secondary lobes of a filter influencing the main lobes of the filters neighbors).

 However, the signal is designed so that the interfering interference is canceled at the frequencies F1, F2, F3, ... corresponding to the maximum of the main lobe of each filter (principle of the orthogonality of the carrier frequencies).

Thus, when the agreement is perfect (32), each of the carriers 37 is filtered by a single filter 34c
On the other hand, when this agreement is imperfect (33), a difference in frequency Sf appears between each carrier 38 and the maximum 39 of the corresponding filter.

The filtering therefore reproduces an interfering interference noise, due to the sidelobes 31OAX 31OB 31OC * BE
Moreover, the result of the filtering is attenuated, the carrier 38 no longer coinciding with the maximum 39 of the main lobe. In addition, this frequency difference adds to the nominal phase fixed by the modulation an error term A 27r = 2 Sf Ts, the observation being made over the duration Ts of a symbol.

 Figure 4 shows the general structure (known per se) of an automatic frequency control device.

 The signal 41 received by the receiver is first demodulated, conventionally, by multiplication 42 by an oscillation frequency fr delivered by a voltage controlled oscillator (OCT) 43. The objective sought is of course that the frequency fur be as close as possible to the nominal modulation frequency fi of the received signal 41.

 The demodulated signal 11 is then filtered, using a low-pass filter 45, delivering the useful part 46 of the signal and then sampled, using an analog / digital converter 47.

Digital samples 48 are subjected to a transformation of
Fourier (FFT) 414, corresponding to the application of the filter bank described in connection with FIG. 3. This transformation associates with each received symbol the corresponding data elements received 415.

 These data elements 415 are on the one hand decoded (49), and on the other hand transmitted to a module 410 of numerical analysis, responsible for estimating the phase difference = = fr - fO. The invention essentially relates to the operation of this analysis module.

 The estimate 411 of the frequency difference is then transmitted to a converter 412, which supplies the oscillator 43 with a control voltage 413 that is a function of the estimation of the frequency difference.

 FIG. 5 presents the general principle of determining the phase error according to the invention, as performed by the analysis module 410.

 Firstly, a first estimate 51 of the frequency difference Sf, intended to reduce it to a value less than half of the distance f between two consecutive carrier frequencies of the received signal, is made as a function of the data elements. received Xk and Ck data elements known to the receiver. This first estimation is done according to a conventional technique, such as for example the detection of the data element having a maximum power, in the case of a synchronization pattern comprising a fraction of the number of available carriers.

 Then, when the difference Sf is less than the half-distance between carriers fs, a second estimate 52, more accurate, of the difference Sf is determined.

 Indeed, a system based on an OFDM signal needs a much more precise demodulation frequency than the half inter-carrier gap.

 To do this, the received data elements Xk are directly taken into account, and not their module (as recommended by previous methods).

 The first step 53 of this second estimation consists, for at least one frequency synchronization data element (but preferably for all) of a frequency synchronization symbol to calculate a ratio of two weighted sums of the set of data elements constituting this symbol (some of the weighting coefficients may of course be zero). In practice, account is taken of the data element corresponding to the carrier frequency modulated by a frequency synchronization element and data elements corresponding to carrier frequencies adjacent thereto (two, four (or more) closer carrier frequencies).

 The essential feature of the invention is based on this principle of calculating a ratio of data elements, which makes it possible to dispense with the need to calculate the modules of these data elements, while offering a very good resistance to distortions of the transmission channel.

 Indeed, as will become clearer later, this estimation report is not affected by the distortion of the channel, although each data element is, because of the simplifications induced by the report.

 Of course, the weighting coefficients applied to each of the two sums must be chosen so that the result of the report is representative of the frequency difference Sf, an example of coefficients being given later.

 If this ratio is not affected by the distortion of the channel, it remains however sensitive to the noise that is added to the data elements received, especially in the case of fading of the signal.

 To overcome this problem, a second step 54 consisting in calculating an overall estimate of the ratio, based on the data 551 to 55L corresponding to each frequency synchronization carrier (values of the frequency synchronization element and adjacent elements), is performed. .

 In this way, more weight is given to the data items corresponding to non-faded carriers. In addition, the noise samples polluting the different estimation ratios are independent and partially destroyed during the calculation of the average, which results in a significant reduction in the relative power of the noise.

 In practice, step 54 is performed directly, as will be shown below, without calculating (53) the ratios independently for each frequency synchronization element. In other words, steps 53 and 54 are grouped, and consist of a calculation of a ratio of two weighted sums of the set of data elements of a symbol, some of the weighting coefficients being zero.

 Finally, in the case of complex data elements (in the mathematical sense of the term), noise adds to the estimation reports an imaginary term, which of course does not have to be taken into consideration. The method therefore comprises a third step 56 of deleting these imaginary terms.

 In this way, the estimation ratios obtained are not affected by the distortion of the channel, and very robust vis-à-vis the noise.

 It should be noted that the method of second estimation 52 described above in three steps 53, 54 and 56 has been thus decomposed to better show the various features of the invention. However, it is clear that these three steps can be combined in a single calculation step, as will appear in the more detailed implementation example presented later.

 Furthermore, it should also be noted that the essential characteristic of the invention is the step 53 of calculating ratios, so as to overcome the problems of distortions. The steps 54 of averaging (overall estimate) and 56 of deletion of the imaginary part are optional steps, making it possible to limit the disturbances due to noise.

 A preferred embodiment of the invention is now described in greater detail, based on the use of a frequency synchronization symbol consisting of a reduced number of carrier frequencies (frequency-synchronous carrier frequencies), as illustrated in FIG. figure 6.

 In the example of Figure 6, we kept only one carrier frequency on 16. Any other value can of course be chosen. Preferably, this value is selected so that the frequency synchronization frequencies kept are isolated, and therefore do not interfere with each other. In the case of FIG. 6, this amounts to considering that the frequency error Sf is always, in absolute value, less than 8 inter-carrier gaps (8 f). Thus, each of the carrier frequencies can be considered independently to determine an estimate of the frequency deviation Sf, and an average can then be calculated.

 The frequency synchronization symbol of FIG. 6 corresponds, like any symbol, to a sequence of data elements {Ck} o <k <N4, Ck being non-zero if and only k is equal to 16p + 8, p being integer. In other words, only the carrier frequencies f8 (61), f24 (62), f (63), 16p + 8 are retained. This symbol, i.e. this sequence {Ck} is known to the receiver. For example, it is stored in a ROM.

As already stated, in the ideal case, the local oscillator of the receiver delivers a carrier at the frequency used to demodulate the signal above. Sampling at the N / T rate (at times tn = mT + nT / N) then gives the sequence (n varying from 0 to N

 Considering only one symbol in the following, we will lighten the notation by omitting the exponent (m).

 By assimilating the sequence CZ, to its twin {(-l) nZn}, we find that the {Ck} represent the Discrete Fourier Transform of {Zn}. In the ideal case where the frequency of the local oscillator is exactly f0, the Zn FFT thus restores the data Ck.

 In order to specify the effect of a frequency deviation on a single carrier, it is now assumed that all the data elements Ck of a symbol are null, with the exception of one, Cko.

Of course, if the agreement is perfect, that is to say if the local oscillator delivers a carrier at the frequency f the FFT will directly restore Cko. On the other hand, IF the signal is demodulated by a carrier at the frequency (fo, -Rfs,), where R is a dimensionless real of the order of a few units (R is directly representative of the frequency difference: Sf = Rfs), everything happens as if fk was no longer equal to k / T, but to (k + R) / T. So we have:

soit

The FFT then gives the samples ::

is

 In the case of the frequency synchronization symbol of FIG. 6, and if the frequency error sf is less than 8 inter-carrier gaps f, we can consider that we are dealing with a series of isolated carriers that do not interfere with each other. not between them: the expression (1) remains valid around each carrier. It is therefore possible to perform the same analysis for each carrier P16p + 8> analysis relating to the symbols X16p +, 6. Each analysis will give an estimate of Sf, an averaging being then carried out to reduce the effects of the noise and to protect itself from the hazards of the channel. We first consider the case of a given carrier 16, with R <8fs and ko- 8 <k <ko + 8.

Repeating the expression (1), we find ourselves in a case where k + Rk remains very small in front of N; (1) simplifies itself by:

 It is easy to show that | Xk | is maximal when ko + R-k is minimal, that is, when k is the nearest integer to kotR.

 A first estimate of Sf is therefore easily obtained by identifying the maximum of Xk. This method, known, allows the CAF system to reduce the frequency of the OCT in the interval IEo-fs / 2 + f, / 2 [.

 This first estimate corresponds to step 511 of FIG.

However, as already mentioned, a signal-based system
OFDM needs a much more precise demodulation frequency than f0tf, / 2. Further processing of the received samples must therefore be performed once the OCT frequency has been "squeezed". The adopted ripening method must be very resistant to noise and channel distortion.

 According to the invention, a direct processing of the complex symbols is carried out without first searching for the module.

 For simplicity, we will refocus later on Xko = Xl6p + 8, posing Xk (P) = Xko kX -8 <k <8. When no ambiguity will be possible, we will omit the exponent (p) by reasoning on a given carrier.

For a carrier modulated by the symbol C0, the distortion introduced by the channel is summarized, in known manner, to multiply C0 by HO = H (f, -fk () - N / 2T), where
H (f) denotes the transfer function of the canaI.

We therefore receive the samples (according to (2)):

certains des coefficients pondérateurs a et Pj pouvant bien sûr être nuls.The main feature of the invention is to calculate an estimate of the frequency deviation using a ratio of data elements Xi of the following form:

some weighting coefficients a and Pj can of course be zero.

 It then becomes clear that such an estimate is not affected by the distortion introduced by the channel (H (f)), although each element Xk is. Indeed, the terms Ho disappear from the report, for simplification.

 In a particularly simple embodiment, it is possible to consider only the two data elements (renumbered Xt and X 1) flanking each element X0 corresponding to a conserved carrier frequency.

soit:

The basic estimate used in this case according to the invention can then be:

is:

 We therefore obtain an estimate Ro directly representative of the frequency difference (R), especially since, R becoming rapidly low, the denominator 1-R2 tends to 1. Other estimates can of course be retained, according to the same principle. Preferably, this estimate is chosen so that it is an increasing odd function as a function of sWf, and if possible linear or substantially linear.

 In practice, an exact value of the difference R is not mandatory, the correction voltage controlling the OCT taking into account only the sign of the estimate Ro.

It may be noted, however, that the exact value of R can be determined by solving the equation of the second degree corresponding to (4):
R2 + 2 / Ro.R -1 = 0
If the estimate Ro is not affected by the phase of the channel, it is on the other hand very sensitive to the noise which is added to the samples, if Hko is very small (case of a fading). It is therefore necessary to "average" (calculation of an overall estimate) on several carriers, giving more weight to the carriers that have not faded. In addition, the noise adds to Ro an imaginary term that does not have to be taken into consideration.

Reintroducing the exponent (p), we obtain the complete estimate

In reality, these complex samples Xk (P) each represent two real samples: Xk (P) = Ak (P) + jBk (P), where O (P) Ak (P) and a Bk (P) are real. R1 is written then:

 The critical term of this estimate is the numerator, since it is it that gives the sign of the correction voltage of the O CT. It is indeed very robust vis-à-vis the noise, because the samples of noise polluting the samples Ak (P) and Bk (P) are independent and partially destroyed during the summation: in this example, the summation divides by 16 the relative power of the noise.

More precisely, each term of the numerator can be written in the form:
&alpha; p. # f / f + bp + bp
where: (Xp2 is the attenuation coefficient of the channel (substantially
Xo (P) 2), and
b is the noise.

ap2 peut s'écrire: ap2 = E[&alpha;p] ap
avec: E[a2J: espérance mathématique de Cp2, constituant la partie utile
de ap;
termes résiduels indépendants et de moyenne nulle.The sum of the numerators is therefore:

ap2 can be written: ap2 = E [&alpha; p] ap
with: E [a2J: mathematical expectation of Cp2, constituting the useful part
from ap;
independent residual terms of zero average.

 It is easy to verify that the useful part of Qp2 (E [a2j) consists of terms that add constructively, while the terms p, which are often negative or positive, are at least partially subtracted from one another. It is the same for noise terms bp, which are also independent and of zero average.

 FIG. 7 illustrates a particular embodiment of the module 410 of FIG. 4, implementing the method described above.

 The data elements 71 determined by the FFT are transmitted, in the form of two real samples Ak (P) and Bk (P), to a module 72 of first estimation, which aims to reduce the value of R under 1 / 2. This first estimate consists in finding the maximum of Xk for each carrier of a frequency synchronization symbol, as a function of the actual values of this symbol {Ck} contained in a ROM 73.

 As long as the frequency difference corresponds to a value R greater than 1/2, the first estimate module 72 delivers to the converter a first estimation signal 74, rounded to the nearest integer.

 On the other hand, when R becomes less than 1/2, the samples A, (PI and are transmitted (75) to a module 76 of the second estimate, implementing the method of the invention, more specifically, this module 76 calculates for each frequency synchronization symbol the estimate R1 according to equation (5) and delivers this estimate Ri to the converter.

The method of the invention is of course not limited to the embodiment which has just been described. Indeed, one can consider many other embodiments, including:
- the number of data elements taken into account for the calculation
an estimation report may take into account more than two
data elements neighboring the element considered
- specific weighting factors (other than 1) may
to be chosen
- the number of data elements taken into account and / or the value
weighting coefficients can be variable (for example
depending on additional information, such as the level
noise, the importance of the frequency deviation, the spread of the
response of the channel, ..).

The invention is also applicable to other types of frequency synchronization symbols. Such a symbol may for example consist of reference carriers inserted among "useful" carriers (carrying useful data, that is to say constituting the signal to be transmitted). This useful data is seen
as a noise (at 0 dB) by the estimator. This noise can be eliminated by making a
averaging on many estimates (possibly on several symbols).

Moreover, besides the automatic frequency control, the method of
the invention can also be used to correct the defects of a receiver
and / or the effects of a dispersive transmission channel.

Claims (12)

 said oscillation frequency (fur) is adjusted (412) according to said estimate of the frequency difference (411; Ri).  determining, for at least one of said frequency synchronization data elements (410 ;; fig.5; fig.7), an estimate (411; R1) of the frequency difference between said frequency of said frequency synchronization symbol; oscillation and said nominal frequency, by means of a calculation taking into account the value of the synchronization data element and the value of at least one of the adjacent data elements, in the form of an arithmetic ratio (551 55L) for at least partially eliminating said distortion  said frequency synchronization symbol received is demultiplexed (414) so as to obtain N data elements received (415; Xk) I. A method for controlling the oscillation frequency (fd of a local oscillator (43) of a digital signal receiver, a method of the type for adjusting said oscillation frequency (f,) on the nominal modulation frequency a received signal (41), said digital signals being transmitted by means of a plurality (21) of N carrier frequencies, in the form of a succession of data symbols (S1 to SM), each of said symbols comprising a set of N data elements each modulating one of said carrier frequencies, at least one frequency synchronization symbol (S4; fiv.6) being regularly inserted among said data symbols (S1 to SM), said symbol of frequency synchronization consisting of at least one frequency synchronization data element modulating a frequency synchronization carrier frequency (61, 62, 63) selected from said N carrier frequencies, the said frequency synchronization data element having known characteristics of said receiver, the received signal being further affected by a distortion induced by the transmission channel, characterized in that for at least one of said frequency synchronization symbols (fiv.6 ) 2. Method according to claim 1, characterized in that said arithmetic ratio (551 to 55L) is a ratio of two weighted sums of said N received data elements (Xk). 3. Method according to any one of claims 1 and 2, characterized in that at least some of said frequency synchronization symbols (fig.6) comprise at least two frequency synchronization data elements, and in that said arithmetic ratio (551 to 55L) takes into account the values of at least two of said frequency synchronization data elements and at least one of the data elements adjacent to one of said frequency synchronization data elements. 4. Method according to any one of claims 1 to 3, characterized in that cancels (56) the imaginary part of said arithmetic ratio, said imaginary part being induced by the transmission noise of said signal. 5. Method according to any one of claims 1 to 4, characterized in that it comprises a prior step of limiting (51; 72) said frequency difference to a value less than or equal to f2, fu being the difference between two consecutive carrier frequencies. 6. Method according to any one of claims 1 to 5, characterized in that the adjustment (412) of said oscillation frequency (f) takes into account the sign of said estimate of said frequency difference (411; R1) . 7. Method according to any one of claims 1 to 6, characterized in that said frequency synchronization symbols (fig.6) comprise a number M carrier frequencies modulated by frequency synchronization data elements less than the number N of carrier frequencies. The method according to claim 7, characterized in that said frequency synchronization symbols (fig.6) comprise the set of data elements {Ck} 0sksN-> where Ck is a non-zero frequency synchronization data element if and only if k = 16.p + 8, p integer. 9. Method according to any one of claims 7 and 8, characterized in that the frequency difference between two of said M carrier frequencies is chosen so that it is greater than the maximum difference between said oscillation frequency and said nominal modulation frequency. 10. Method according to any one of claims 1 to 9, characterized in that said arithmetic ratio (551 to 55 is written s (Xi + 1 - X4) iX., Where:  X is a received frequency synchronization data element  corresponding to a carrier frequency of frequency synchronization  X, -1 and Xj + 1 are the respective corresponding received data items  at the carrier frequencies preceding and following the said frequency  frequency synchronous carrier corresponding to Xi, and  i is an index varying from 1 to M, where M is the number of frequencies  Frequency synchronization carriers taken into account. 11. The method according to claim 1, wherein said N carrier frequencies are orthogonal carrier frequencies. 12. Receiver of digital signals, of the type comprising means (412) for controlling the oscillation frequency (fur) of a local oscillator (43), said control means (412) being intended to adjust said frequency of oscillation (fur) on the nominal modulation frequency of a received signal (41), said digital signals (41) being transmitted by means of a plurality (21) of N carrier frequencies, in the form of a succession of data symbols (S1 to SM), each of said data symbols comprising a set of N data elements each modulating one of said carrier frequencies, at least one frequency synchronization symbol (S4; fiv.6) being regularly inserted among said data symbols (S1 to SM), said frequency synchronization symbol (S4; fiv.6) consisting of at least one frequency synchronization data element modulating a frequency synchronization frequency (61, 62, 63) one of said N carrier frequencies, said frequency synchronization data element having known characteristics of said receiver, the received signal being further affected by a distortion induced by the transmission channel, characterized in that it comprises:  means (414) of demultiplexing processing of at least one frequency synchronization symbol, delivering N received data elements (415; Xk);  means (410) for calculating an estimate (411) of the frequency difference between said oscillation frequency (fur) and said nominal frequency, comprising calculation means taking into account the value of the element of synchronization data and the value of at least one of the adjacent data elements, in the form of an arithmetic ratio for at least partially eliminating said distortion; and  means (412) for adjusting said oscillation frequency (f 1) as a function of said estimate (411) of the frequency difference.