CH683307A5 - A method for transmitting digital data. - Google Patents

Description

5

10th

15

20th

25th

30th

35

40

45

50

55

CH 683 307 A5

description

Technical field

The invention relates to a method for the transmission of digital data over a fading channel, in which method a) the data present in the form of at least four-stage symbols with a given symbol rate 1 / Ts are converted into a baseband signal with a pulse shaper,

b) the baseband signal is transmitted via the fading channel using an FM method,

c) a received signal containing the transmitted baseband signal is sampled at a specific sampling rate Ta and d) symbols estimated from the sampled received signal are determined.

The invention also relates to a device for carrying out the method mentioned. In particular, the invention relates to a transmitter circuit and a receiver circuit for transmitting and. Receiving signals of the type mentioned.

State of the art

A method of the type mentioned at the outset is e.g. from the article "The Feasibility Study of the Ny-quist Baseband Filtred 4-Level FM for Digital Mobile Communications", K. Kage, Y. Sasaki, M. Ichihara, T. Sato, Globecom'85, Conference Record, pp. 200-204, 1985. In the system described there, a two-stage input signal is converted into a four-stage signal and band-limited by a premodulation filter. The transfer function of this filter corresponds to the root of a so-called raised-cosine roll-off Nyquist filter. The band-limited baseband signal is transformed with a modulator into a continuous FM signal. The FM signal is demodulated on the receiver side and filtered by a band limiting filter which corresponds to the above-mentioned root Nyquist filter. Finally, a clock signal is extracted and 4-step symbol detection and reduction to a binary signal are carried out. In order to adjust the drift of the center frequency, a decision-based automatic control of the DC offset and the amplitude is proposed.

The advantage of this known system is the high bandwidth efficiency. However, the quaternary FM is much more sensitive to circuit imperfections than the binary one. Accordingly, there are certain operating modes with large transmission losses. From the general study by K. Kage et. al. it is not clear how the negative effects of technical imperfections can be avoided.

Presentation of the invention

The object of the invention is to provide a method of the type mentioned at the outset which avoids the shortcomings present in the prior art and is particularly suitable for implementation in a modem which can be used e.g. can be connected to a walkie-talkie.

According to the invention, the solution is that in a method of the type mentioned at the transmitter, the baseband signal is at least freed of its DC component by a high-pass filter and the estimated symbols are determined using a DFE decoder which at least largely compensates for an existing intersymbol interference (ISI) will.

The high-pass filtering eliminates the signal components in the lowest frequency range. This has a "positive effect on the parameter estimation of the DC offset and the signal amplitude to be carried out in the receiver. Signal-related interference on the parameter estimation does not occur. As a result, the parameter estimation becomes more robust. With the Décision Feedback Equalizer (DFE), the ISI, that were introduced by the high pass are removed.

According to a particularly preferred embodiment, the symbols are transmitted in mutually independent blocks of a certain block length. Block-by-block transmission enables operation in time multiplex. The parameters to be assumed to be constant within a block differ greatly from block to block. According to a preferred embodiment of the invention, they are therefore determined on a digital level with a feedback-free parameter estimation.

The so-called open loop parameter estimation, which is not based on the estimated symbols, generally has a shorter settling time than a decision-based method. If the symbols are transmitted in mutually independent blocks (burst operation), a short settling time of the parameter estimator is very important. While the intersymbol interference due to DC decoupling considerably complicates the settling of the parameter estimator with decision feedback, the open loop method benefits from the removal of the DC components in the transmitter. For the rest, the

2nd

5

10th

15

20th

25th

30th

35

40

45

50

55

60

65

CH 683 307 A5

feedback-free parameter estimation has the advantage of an unlimited estimation range. After all, such a parameter estimate cannot “hang up” with an incorrect value.

The high-pass filter advantageously has a cut-off frequency of at least 20 and at most 300 Hz. A cut-off frequency of 50 to 200 Hz, in particular 50 to 100 Hz, is particularly preferred. A high cut-off frequency is advantageous with regard to a good parameter estimate for the DC component . However, the higher the cutoff frequency, the more the symbol estimate is impaired.

As a rule, a high pass is already implemented at the input of a two-way radio. So that the quality of the data transmission is independent of the radio device used, it is advantageous if a high pass with a relatively high cutoff frequency is provided at the output of the transmitter circuit of the modem. The cut-off frequency of the modem's high pass is e.g. five times larger than that of the radio.

Since the ISI energy for a given amplitude response is strongly dependent on the phase response of the high-pass filter, a first-order filter is advantageously used as the high-pass filter.

A block equalizer is advantageously connected upstream of the DFE decoder, which specifically compares the intersymbol interference caused by adjacent blocks. The DFE decoder itself only compensates for the block-internal ISI. In this way, the settling of the DFE decoder can be significantly improved, particularly in time-division multiplexing.

According to a preferred embodiment of the invention, the received signal is sampled with a free-running oscillator with a sampling rate 1 / Ta which is greater than the symbol rate 1 / Ts. The clock phase error is determined with a clock phase estimate, according to which the oversampled received signal is interpolated once per symbol interval. The advantage of this embodiment is that feedback from the digital to the analog part is avoided. In addition, the samples can be processed in blocks with a digital signal processor (DSP) in time sharing mode.

In accordance with a particularly preferred embodiment of the invention, clock phase error-dependent coefficients are used for the interpolation to weight the sampled received signal, which optimize the error between the interpolated and the ideal signal value at the sought sampling time, taking into account the entire transmission path according to the MMSE criterion. Because the interpolation takes into account not only the sample values as such, but the entire transmission system, it can manage with a very small number of taps. It has been shown that, with a sampling rate that is four times greater than the symbol rate, and an interpolation that takes into account at least two and at most three values of the sampled received signal, high-quality signal detection can be achieved despite the very low implementation effort.

It is particularly advantageous if, in the clock phase estimation, the values of the oversampled received signal are filtered with an FIR pre-filter, the support points of which are spaced apart according to the sampling rate Ta and whose coefficients are at least approximately according to cos (9k / 2) g (dTA) (I)

are dimensioned, where g (.) is the transfer function of the overall system. Such a pre-filter is also characterized by a particularly low number of tapes.

To reduce the effort involved in implementing the pre-filter, the optimal impulse response is approximated according to the LSE principle. Such a pre-filter delivers very satisfactory results with just 3-5 taps. It should be emphasized that such a pre-filter can also be used for clock phase estimation in systems that are not characterized by high-pass filtering on the transmitter side, DFE decoding on the receiver side or block equalization.

A transmitter circuit, which is particularly suitable for use in a modem that can be connected to an FM radio, is characterized by a pulse shaper for converting at least four-level symbols into a baseband signal, a signal output for connecting the transmitter circuit to the FM radio, and a high-pass filter with a cut-off frequency of at least 50 Hz and at most 200 Hz between the pulse shaper and the signal output.

A receiver circuit suitable for performing the part of the method on the receiver side has a signal input for connecting the receiver circuit to an FM radio, a sampling circuit for sampling the received signal at a given sampling rate Ta and a symbol detector for determining estimated symbols from the sampled received signal . According to the invention, such a circuit is characterized in that the symbol detector comprises a DFE decoder which at least largely, if not completely, compensates for an ISI.

A transmission system according to the invention thus also includes FM radio devices which are equipped with connections for a modem according to the invention, the modem having the transmitter and receiver circuits mentioned above according to the invention. So that several modems can exchange data at the same time, the modems run in time multiplex mode.

Further advantageous embodiments of the invention result from the totality of the dependent claims.

3rd

CH 683 307 A5

Brief description of the drawings

The invention will be explained in more detail below on the basis of exemplary embodiments and in connection with the drawings. Show it:

5

1 shows a schematic block diagram of the overall system according to the invention;

2 shows a schematic block diagram of a symbol detector according to the invention;

3 shows a schematic block diagram of a DFE decoder;

4 shows a schematic block diagram to illustrate the effect of block equalization; 5 shows a schematic block diagram of a synchronization circuit;

6 shows a schematic block diagram to illustrate the selection of the optimal interpolator coefficients; and

7 shows a representation of the equivalent impulse response of an interpolator with three interpolation points. 15 In the different figures, the same parts are basically provided with the same reference numerals. Ways of Carrying Out the Invention

1 shows a block diagram of a data transmission system according to the invention. A first Mo-20 6.1 is connected to a first two-way radio 7.1. With these two devices, data is transmitted via a channel 3 with a loss of signal to a second radio device 7.2 and a second modem 6.2 connected to it.

The modem 6.1 has a transmitter circuit 1.1 which generates a baseband signal which, with a transmitter circuit 2.1 of the walkie-talkie 7.1, is converted into an FM-modulated high-frequency signal with e.g. 25 several hundred MHz carrier frequency is implemented. In the second walkie-talkie 7.2, a receiver circuit 4.1 demodulates the fuzzy and noisy HF signal and emits a reception signal in the frequency baseband to the receiver circuit 5.1 of the second modem 6.2. This determines the symbols transmitted from the received signal.

In order to enable two-way communication, in addition to the transmitter circuit 1.1, the first modem 6.1 also has a receiver circuit 5.2, which is of the same design as the receiver circuit 5.1 of the second modem 6.2. The walkie-talkie 7.1 also has a receiver circuit 4.2, which is designed in the same way as the circuit 4.1. The same applies mutatis mutandis to the modem 6.2 (transmitter circuit 1.2) and the walkie-talkie 7.2 (transmitter circuit 2.2).

Binary digital data is typically available at the input of the modem. These are converted into four-stage symbols using a 35 coder 8 in a manner known per se. These symbols on have a certain symbol duration Ts (symbol rate 1 / Ts) and, according to a preferred embodiment, are now transmitted in mutually independent blocks of a certain block length (of e.g. 100 symbols).

A pulse shaper 9, the impact response of which is at least approximately a root Nyquist pulse, converts the symbol sequence into a suitable baseband signal. The transfer function of the pulse shaper 9 is preferably a root-raised-cosine function, as it is e.g. from the publication by K. Kage et al. is known. The roll-off factor of the raised cosine function in the frequency domain is e.g. 0.2. For practical reasons (in particular to minimize the memory requirement), this pulse is filtered with a raised cosine window defined in the time domain, which 45 e.g. has a roll-off factor of 1 and an exponent of 1. By cutting off the pulse has become broadband and no longer corresponds exactly to a root Nyquist pulse.

As already mentioned, the invention is characterized on the transmitter side by at least one high-pass filter which removes the DC component in the baseband signal. As a rule, this high-pass filter is provided in the walkie-talkie 7.1 and directly at its signal input (cf. high-pass filter 50 11). According to the invention, this has a cut-off frequency of 20 Hz or more. In order not to impair the detection quality too much, however, it should not exceed 300 Hz.

According to a particularly preferred embodiment of the invention, however, two high-pass filters 10 and 11 can also be provided. The cut-off frequencies should then differ greatly, however, in that only the high-pass filter 55 10 arranged according to the invention at the modem output is noticeable. It is then this high-pass filter 10 whose cut-off frequency should be in the range from 20 to 300 Hz. The high pass filter 11 can then also have a much lower cut-off frequency of e.g. Have 10 Hz. So that only the high-pass filter 10 arranged in the modem is relevant, its cut-off frequency should be a multiple (e.g. five times) that of the high-pass filter 11.

It should be emphasized again that in principle a single high pass is completely sufficient. The cut-off frequency is then preferably in a range between 50 and 100 Hz. If, however, the modem according to the invention is to be connected to an already existing radio device whose input high pass cannot be changed, then a second one is arranged in the modem itself High pass (at a suitable distance of the cutoff frequency) a desired transmission behavior independent of the radio set.

65 The two-way radios 7.1 and 7.2 are commercially available FM devices. So you have a transmitter

4th

5

10th

15

20th

25th

30th

35

40

45

50

55

60

65

CH 683 307 A5

FM modulator 12 and a microphone circuit 13 connected to it and an FM demulator 14 and a loudspeaker circuit 15 connected to it on the receiver side. Microphone circuit 13 and loudspeaker circuit 15 are used in a manner known per se for the transmission of voice signals.

At the signal output of the receiver circuit 4.1, a received signal corresponding to the transmission signal is emitted in the frequency baseband to the receiver circuit 5.1 of the modem 6.2.

The received signal is first filtered using a matched filter 16. This filter is adapted to the pulse shaper 9. Its transfer function thus also corresponds to a root-raised co-sine function. The filtered received signal is then sampled with a scanner 17. The sampling rate 1 / Ta is preferably at least four times greater than the symbol rate 1 / Ts. The sampling is carried out with a free-running oscillator. This means that the phase errors of the sampling must be determined from the sampled values and an in-phase signal value must be estimated. These two functions are performed by the synchronization circuit 18. From e.g. A signal value (per symbol interval) is determined four sample values per symbol interval. The subsequent symbol detector 19 determines estimated symbols ccn. Finally, the symbols ccn are converted into binary data using a decoder 20, the function of which is inverse to the coder 8.

This describes the rough course of the data transmission. The following is about explaining the details of the method on the recipient side in detail.

2 shows a block diagram of the symbol detector 19 according to the invention. The central element is the DFE decoder 21. In principle, it eliminates the ISI, which was introduced by the high-pass filtering on the transmitter side. As a result of the high-pass filtering, the impulse response of the overall system has a relatively high main impulse value and slowly decaying after-runs. As the cut-off frequency of the high-pass filter increases, more and more energy is shifted to the pulse wake. It has now been shown that, despite the infinite signal-to-noise ratio, the simple threshold value decoder no longer works properly even at a low cut-off frequency of the high-pass filter and produces high error rates.

In the preferred block-wise transmission, the ISI can be divided into an intra-block and a non-block-related part. The block-external ISI is generated exclusively by symbols that are transmitted in other blocks. A block equalizer 22 is provided for their elimination. In principle, the parameters of the signal, namely the DC offset and the amplitude, must be known for a correct estimate in the DFE decoder 21. However, these parameters are not known in advance because they change slowly due to the circuit and must therefore be estimated in the receiver using the parameter estimator 23. Due to the influence of temperature and the scatter of specimens, a frequency offset occurs between the transmitter and the local oscillator of the receiver, which causes a data-independent DC component behind the frequency discriminator. In extreme cases, this DC offset can amount to approximately 30% of the eye opening of the transmission signal. Otherwise, the fluctuation in the modulator constant (± 5%) or. the demodulator constant (± 40%) to a fluctuation in the signal amplitude of the baseband signal in the receiver. The temporal change of the frequency offset and the modulator resp. However, the demodulator constant is very slow compared to the symbol duration Ts. The parameters mentioned can therefore be assumed to be constant within a block. In time-division multiplexing, however, they fluctuate greatly from block to block. The parameter estimator 23 thus estimates the DC offset and the amplitude in blocks and forwards these values to the DFE decoder 21.

3 now shows a block diagram of the DFE decoder 21 according to the invention. As can be seen from the figure, the previously decided symbols ân-i, ân-2, ... ân- become an estimated value sn according to n

K *

= S an_ ± gj i = 1

(II)

the ISI influence of the previous pulses on the current sample rn generated at the output of the reception filter and subtracted from rn. K is the degree of decision feedback, gi denotes the time-discrete shock response of the transmission system consisting of pulse shaper 9, matched filter 16 and high-pass filter 10. For the compensated signal rn rn = Tri - Sn (III)

(determined by the summing element 26), a simple threshold value detection (threshold value decider 27) is then carried out. The prediction of the ISI influence is therefore realized with a Ts-spaced FIR filter (delay elements 24.1, ... 24.K, summing element 25) with K coefficients (gi ... gx). _

When the DFE decoder 21 is switched on, no compensation signal sn is available. In most

5

5

10th

15

20th

25th

30th

35

40

45

50

55

CH 683 307 A5

In some cases, the DFE decoder 21 settles in relatively quickly after a few correct decisions. In the event of an unfavorable ISI disturbance, however, it is possible that long «error bursts» occur due to frequent wrong decisions and correspondingly incorrect compensation. An entire block can therefore be lost in block operation.

The impulse response of the system depends on the cutoff frequency and the order of the high pass filter. It has now been shown that the order of the high-pass filter is a much more critical parameter than the cut-off frequency.

Finally, it was found that the ISI energy increases strongly with increasing order of the high-pass filter at the same cut-off frequency and thus drastically deteriorates the detection performance of the DFE decoder. For these reasons, when implementing the system, care should be taken that the order of the high-pass filter is as small as possible.

During decoding, as many (ideally all) symbols as possible have to be extracted from a block without being able to access state information. With the DFE decoder, the status register containing the past symbol decisions will typically be set to zero for initialization. The first symbols of a block are then exposed to the full ISI and the probability of errors is correspondingly high. More reliable symbol decisions are available deeper in the block.

The block equalizer 22 now serves to improve the transient response, which in principle represents a time-variant pre-filter for the DFE decoding. This pre-filter makes a minimum mean square error equalization (MMSE) of the respective sample value. However, it is not the transmission symbol that is assumed as the target variable, but the amplitude value that would be obtained if only interference of symbols within the block would occur. The remaining inner block ISI is resolved by the DFE decoder 21 in the manner described above. If the block length is greater than the length of the total impulse response, then with this method the last samples of a block are not changed at all by the prefilter. The first sample value is always equalized in the conventional sense (minimizing the overall ISI).

Fig. 4 illustrates the principle of block equalization. aact denotes the symbol sequence in the current block. Correspondingly labeled spast the symbol sequence in the last block. Gi and G2 are matrices that are defined on the basis of the impulse response and describe the influence of the corresponding symbols on the received signal, n describes the additive disturbance. The received signal £ can be described by the following equation:

£ = £ 1 + £ 2 + n (IV)

The block equalizer, which manifests itself in matrix G in the present FIG. 4, is dimensioned such that the effect of §2 on £ is minimized in the sense of the MMSE principle:

E {.} Denotes the expected value of the given argument. (The calculation of the coefficients of the matrix G based on the specified optimization criterion is routine work for the person skilled in the art.)

The open loop parameter estimation according to the invention will now be described below. In the nth interval, the sampled received signal rn is given by

A0 and d0 denote the unknown but constant signal amplitude and the DC offset. The Gaussian disturbance process with the variable nn is free of mean values and has the variance an2. The sequence {«n} consists of four-stage, equally probable symbols that are uncorrelated and mean-free:

E {£ T (£ 1 - a)} = 0 (V)

00

rn = Ao <* an-i gi + do> + nn i = 0

(VI)

(VII)

6

5

10th

15

20th

25th

30th

35

40

45

50

55

60

65

CH 683 307 A5

According to the invention, the estimated value of the amplitude  and the DC offset d are formed as follows:

A =

° r

- [Z * (VIII)

Hfo92i a r an, °°

a = -S -! -— I s i \ (ix)

ar li = 0

In practice, the expected values rn and ar2 are obtained by averaging the received signal rn and. of the squared received signal rn2 approximated over L successive symbol intervals (i.e. over an estimation window LTs):

1 L-1

= r ", -2Q r-i <X)

L 1 = 0

2 ^ -2 'r = ", 2" (rn-i - - rn

| L 1 = 0 L-1

) (XI)

The averaging according to equations X and XI corresponds to a low pass averaging of the received signal with the bandwidth 1 / (LT). Since the noise line spectrum after the FM demodulator increases quadratically with the frequency, the interference power is largely suppressed by this low-pass filtering. Choosing a wide estimation window (LT) is therefore a very effective means of suppressing jitter. With regard to the burst operation, however, L must not be too large, since in this case the current signal parameter must be followed quickly when the burst changes. The window width is in the order of 100 symbol intervals.

With a window width LT there is a residual data-dependent DC component which influences the estimate of the actual DC offset (d0). Due to the AC coupling, this data-dependent DC component has already been removed in the transmitter. This is why the DC estimation in particular becomes very robust. The higher the cut-off frequency of the high-pass filter, the more reliable the parameter estimates are likely to be.

According to equation VI, the received signal is proportional to the signal amplitude A0, and thus the mean rn and the scatter or are also proportional to A0. It follows that the amplitude estimate ätz according to Formula VIII is proportional to A0 and the DC estimate d according to Formula IX is independent of A0. The performance of the open loop parameter estimation is therefore not affected by the actual level of the signal amplitude. Apart from the reduced reception useful energy and the correspondingly reduced signal-to-noise ratio, which is due to the IF filtering as a result of a frequency offset between the transmitter and the receiver, the same statement applies to the level of the DC offset.

The most important aspects of symbol detection have now been sufficiently explained. A specific synchronization method is discussed below.

According to the invention, the calculations for determining the clock phase are carried out with a DSP (digital signal processor), i.e. that is, on the digital level. Clock synchronization is only one of the tasks of the DSP. The corresponding calculations are then carried out in time sharing mode. An entire block of samples is therefore buffered before processing of the block begins. The resulting delay makes tracking the sampling clock problematic. Therefore, in accordance with the invention, a free-running scan is carried out, with the result that an offset e arises between the scan and symbol clock.

7

5

10th

15

20th

25th

30th

35

40

45

50

55

60

65

CH 683 307 A5

Since only digital values of the received signal are available to the digital synchronization circuit, it must first carry out a clock phase estimation and then a corresponding interpolation.

Fig. 5 shows a block diagram of the synchronization circuit. The values supplied by the scanner 17 are supplied on the one hand to a pre-filter 28 and on the other hand to an interpolator 32. After the pre-filter 28 is followed by a squarer 29, a DFT calculator 33 (DFT = Discrete Fourier Transformation), an averager 30 and a phase extractor 31. At the output of the phase extractor 31 there is an estimated value è for the phase error, which is used in the interpolation finds.

The special features of the synchronization circuit 18 according to the invention lie on the one hand in the prefilter 28 and on the other hand in the interpolator 32.

The optimal pre-filter is time-variant when the receiver observes a section of the received signal. The self-noise is suppressed by this pre-filter. For reasons of complexity, however, a time-invariant prefilter is used. To avoid intrinsic noise, this filter Ts has to force periodic zeros in the impulse response of the overall system, which is formed by transmission filter HF (f), matched filter Hs * (f) and pre-filter HE (f). To do this, the resulting spectrum

Hs (f) = | Hs (f) I2 • HE (f), fs = 1 / Ts (XII)

the condition exp (- jo) Hg (fs / 2 + Af) = exp (jo) Hg * (fs / 2- Af) (Af) <fs / 2 (XIII)

for at least a value of ®. It was assumed that HG (f) on | f | <fs is band limited. The symmetry condition mentioned above does not provide a clear solution for HE (f). According to the invention, He (0) is therefore defined as follows:

HE (f) = [Hs (f-fs)] 2 + [Hs (f + fs) J2 (XIV)

The filter defined above has an asymmetrical bandpass frequency response with respect to f = fs. For this reason, it cannot be implemented as an FIR filter with support points spaced apart by Ts with real coefficients.

According to the invention, the pre-filter is now based on the sampling rate 1 / Ta (e.g. 4 / Ts). According to one embodiment of the invention, the coefficients Iie (9) are as follows:

he (9) = cos (3it / 2) g (9TA) (XV)

g (3TA) are the coefficients of the impulse response of the series connection mentioned above (see formula XII).

The complexity of this filter is preferably reduced by an LSE approximation (LSE = Least Square Error). The searched approximation gE (9) is chosen so that the energy of the error signal

2 e2 = E {gE (<5) - gE (0)} 2 (XVI)

0 Ö

becomes minimal. Deriving the equalizer coefficients g e (9) and setting zero results in a linear system of equations with which the pre-filter coefficients can be determined.

The interpolator 32 determines from e.g. 4 samples lying within a symbol interval Ts an estimated sample corrected in phase. According to the invention, the interpolation problem is not considered independently of the transmission system.

6 shows a block diagram to illustrate the optimal interpolar coefficients. The symbol sequence a is pulse-amplitude modulated by the transmission filter hs (t) and is disturbed during the transmission of white Gaussian noise. After the reception filtering hr (t) there is a sampling. This leads to the discrete reception vector i. A segment, starting at the lower index Nu and ending at the upper index N0 with consequently N0-Nu + 1 coefficients, is passed on to the (optimal) interpolator g0. The MMSE estimate 09 of the transmission symbol ad is a linear combination of the elements of ib. Since the useful signal is cyclostationary, a0 does not depend on index 3 if the observation window (Nu: N0) is shifted accordingly. Of course, g0 depends on the waste set e. According to the invention, g0 is therefore chosen such that the error e0 between the symbol ad actually sent and the estimated symbol ad is minimal. With the orthogonality tore you get the interpolator coefficients (vector notation).

8th

5

10th

15

20th

25th

30th

35

40

45

50

55

60

65

CH 683 307 A5

-1

So = b A (XVII>

rBrB

(XVIII)

A rßrß denotes the autocorrelation matrix of the reception vector tb:

ArBrB (0, U) = Rmm E (J-p.) TÄ]

OO

+ E h (^ TA + £ + kTg) h (] JLTA + s + kTg)

ct = - °°

Rmm ^) = No / 2 (hr (t) * hr (t)] t = z (XIX)

Rmm denotes the autocorrelation function of the filtered noise m (t). Finally, h (t) denotes the convolution of the shock responses of the send and receive filters:

h (t) = hs (t) * hr (t) (XX)

b (9) = h (9Ta + e) (XXI)

An equivalent filter can be assigned to the interpolator operation described above. This filter has a continuous impulse response g0 (t). If the samples are filtered with this filter, the optimally interpolated continuous-time received signal is produced.

7 shows the impact response of the equivalent filter in an interpolation with 3 interpolation points. The maximum is t = 0. At t = -Ta and t = + Ta the filter has a zero crossing. At t = +/- 0.5Ta the impulse response has a jump. It should be noted that the phase error e is always smaller than Ta / 2.

For a given phase error e, the interpolator coefficients from FIG. 7 are obtained in such a way that the values at e, e-Ta and b + Ta are read off. The three values closest to the phase clock to be interpolated are then always taken from the sampled values of the received signal and weighted according to the coefficients.

It has been shown that with an interpolation with only two coefficients, the linear interpolation is optimal in a good approximation.

In summary, it can be stated that the invention, taking into account relevant practical imperfections, allows a modem to be designed which has almost twice the bandwidth efficiency compared to a conventional binary FM modem. For example, 8 kb / s transmission in 12.5 kHz channel grid with 60 dB adjacent channel attenuation. Furthermore, the invention enables AC coupling with more than 100 Hz. Variations in the modular and demodulator constants and the DC offset are almost perfectly compensated if they are more or less constant over only about 100 symbols. The clock recovery requires very little taps both for the pre-filter implementation and for the interpolator. Finally, the transient response of the DFE decoder is significantly improved by the invention.

Claims (14)

Claims 1. Method for the transmission of digital data via a channel with loss, in which method a) the data present in the form of at least four-stage symbols with a symbol rate 1 / Ts are converted into a baseband signal with a pulse shaper (9), b) the baseband signal is transmitted by means of an FM method (12, 14) over the fading channel (3), c) a received signal including the transmitted baseband signal is sampled (17) at a specific sampling rate Ta and 9 5 10th 15 20th 25th 30th 35 40 45 50 55 CH 683 307 A5 d) symbols estimated from the sampled received signal are determined, characterized in that e) at the transmitter end the baseband signal is at least freed from its DC component by a high-pass filter (10) and f) the estimated symbols are used using a DFE which at least partially compensates for an existing intersymbol interference -Decoders (21) can be determined. 2. The method according to claim 1, characterized in that the symbols are transmitted in mutually independent blocks of a certain block length and that parameters such as signal amplitude and DC offset are determined on the receiver side with a feedback-free parameter estimate (23). 3. The method according to claim 1 or 2, characterized in that the high-pass filter (10) has a cutoff frequency of at least 20 Hz and at most 300 Hz, in particular from 50 to 100 Hz. 4. The method according to any one of claims 1 to 3, characterized in that the high-pass filter (10) is a first order filter. 5. The method according to any one of claims 2 to 4, characterized in that in order to improve the transient response of the DFE decoder (21) with a block equalizer (22) upstream of the DFE decoder (21), the intersymbol interference caused by adjacent blocks is specifically compared and that only the intra-block inter-symbol interference is compensated with the DFE decoder (21). 6. The method according to any one of claims 1 to 5, characterized in that a) the received signal is oversampled with a free-running oscillator with a sampling rate 1 / Ta, which is greater than the symbol rate 1 / Ts, that b) a clock phase estimate (18 ) is carried out to determine the clock phase error and c) that the oversampled received signal is interpolated once per symbol interval Ts in accordance with the clock phase error determined. 7. The method according to claim 6, characterized in that clock phase error-dependent coefficients for weighting the sampled received signal are used for interpolation, which optimize the error between the interpolated and the ideal signal value at the searched sampling time, taking into account the entire transmission path according to the MMSE criterion. 8. The method according to claim 7, characterized in that the sampling rate 1 / Ta is four times greater than the symbol rate 1 / Ts and that at least two and at most three values of the sampled received signal are used for interpolation. 9. The method according to any one of claims 6 to 8, characterized in that at the start of the clock phase estimate, the values of the oversampled received signal are filtered by means of a pre-filter (28), the support points of which are spaced apart according to the sampling rate and whose coefficients are at least approximately according to cos (cht / 2) g (9TA) be dimensioned, where g (.) is the transfer function of the overall system. 10. The method according to claim 9, characterized in that the coefficients of the pre-filter (28) approximate the optimal impulse response according to the LSE principle. 11. transmitter circuit for sending digital data according to the method of claim 1 with a pulse shaper (9) for converting at least four-stage symbols into a baseband signal and a signal output for connecting the transmitter circuit to an FM radio (7.1 or 7.2), characterized that a high-pass filter (10) with a cut-off frequency of at least 50 Hz and at most 200 Hz is provided between the pulse shaper (9) and the signal output. 12. Receiver circuit for receiving digital data in the form of received signals according to the method of claim 1, with a signal input for connecting the receiver circuit to an FM radio (7.2 or 7.1), a sampling circuit (17) for sampling the received signal with a given Sampling rate (Ta) and a symbol detector (19) for determining estimated symbols from the sampled received signal, characterized in that the symbol detector (19) comprises a DFE decoder (21) which at least partially compensates for intersymbol interference. 13. Modem characterized by a transmitter circuit (1.1) according to claim 11 and a receiver circuit (4.1) according to claim 12. 14. Device for carrying out the method according to claim i, characterized by at least two modems according to claim 13 and at least two radio transmission devices (7.1, 7.2) for transmitting voice signals, the modems being connectable to the radio transmission devices (7.1, 7.2). 10th