Prediction differential pulse code modulation system with adaptive compounding

Description Submit all comments and votes

The invention relates to a transmission system including a transmitter and a receiver for transmitting information signals by means of a pulse code, the information signal to be transmitted in the transmitter being applied through a sampling circuit to an arrangement provided with a non-uniform quantizing circuit whose output signal is transmitted to the receiver by means of pulse code modulation and to a local receiver incorporated in the transmitter, said two receivers being provided with a predictor for generating a prediction signal which is combined in an adder with a signal derived from the quantizing circuit for constituting an input signal for the predictor, the prediction signal and the sampled information signal in the transmitter being applied to a difference producer for obtaining a difference signal constituting the input signal of the quantizing circuit. Such systems are particularly used for the transmission of speech signals.

In all transmission systems in which the information signals are transmitted by means of pulse code modulation (PCM) the noise caused by the amplitude quantisation affects the transmission quality. This quantisation noise can be reduced by using a PCM coding having a larger number of code bits per signal sample; this is, however, accompanied by an increase in the pulse frequency and thus requires a larger bandwidth of the transmission path. Another possibility of enhancing the transmission quality consists in the use of a non-uniform PCM coding with, for example, a piece wise linear compression characteristic according to the CEPT standard. Although a substantially constant ratio between signal and quantisation noise is obtained over a large signal range, a PCM coding with at least eight code bits per signal sample is necessary for obtaining a satisfactory transmission quality of speech signals.

It is an object of the invention to provide a transmission system of the kind described in the preamble in which by utilizing the properties of the information signals a considerable improvement of the transmission quality is realised, which improvement makes it also possible to enlarge the dynamic range of the information signals in which a satisfactory transmission quality is obtained while reducing the bandwidth required in the transmission path.

The transmission system according to the invention is characterized in that the transmitter includes dynamic control means for controlling the dynamic range of the quantizing circuit for its input signal, said two receivers also including dynamic control means for controlling the dynamic range of the quantizing circuit for the signal applied to the adder, the transmitter and the receiver including a control generator fed by the input signal from the predictor, said generator being provided with a storage network and an averaging network for obtaining a control signal which corresponds to the average of the absolute values of the predictor input signal over a limited number of sampling periods, said control signal being applied to a control input of the dynamic control means.

The invention and its advantages will now be described in greater detail with reference to the Figures.

FIG. 1 shows the block schematic diagram of the transmitter and FIG. 2 shows the block schematic diagram of the receiver of a transmission system according to the invention;

FIG. 3 shows a characteristic to illustrate the transmission quality of the transmission system according to FIGS. 1 and 2;

FIG. 4 shows the block schematic diagram of the transmitter and FIG. 5 shows the block schematic diagram of the receiver of a modification of the transmission system according to FIGS. 1 and 2 largely composed of digital structural elements;

FIG. 6 shows the block schematic diagram of a counting coder circuit used in the transmitter according to FIG. 4;

FIG. 7 shows the block schematic diagram of a counting expander circuit used in the receiver according to FIG. 5.

The transmission system shown in a block schematic form in FIGS. 1 and 2 is arranged for the transmission of speech signals by means of differential pulse code modulation (D-PCM).

In the transmitter of FIG. 1 the speech signal derived from a signal source 1 is applied through a band-pass filter 2 having a passband of, for example, 0.3-4.0 kHz to a sample-and-hold circuit 3 fed by a sampling pulse generator 4 having a pulse frequency of, for example, 9 kHz. An arrangement 5 including a non-uniform quantizing circuit 6 whose output signal is coded in a PCM coding circuit 7 with, for example, eight code bits per signal sample is connected to sampling circuit 3. Apart from the non-uniform quantisation the relationship between input signal and output signal of quantizing circuit 6 may be linear, but for a general sake it is assumed in the further description that this relationship is non-linear and that, for example, a compression is also effected in addition to the non-uniform quantisation in quantizing circuit 6 of FIG. 1. In the non-uniform PCM coding thus realised the signal compression is effected, for example, in accordance with the piecewise linear compression characteristic according to the CEPT standard (Conference Europeenne de Poste et Telecommunication). The code bits occurring at the output of the PCM coding circuit 7 are transmitted to the receiver of FIG. 2.

In the transmitter according to FIG. 1 the output signal from the non-uniform quantizing circuit 6 is also applied to a local receiver which includes a predictor 8 for generating a prediction signal from the previous signal samples. This prediction signal is combined in an adder 9 with a signal derived from the non-uniform quantizing circuit 6 for obtaining the input signal for predictor 8. If quantizing circuit 6 has a relationship between the input and the output signal which, apart from quantisation, is linear, the output signal from quantizing circuit 6 could be directly applied to adder 9. However, since it has been assumed that the input signal in quantizing circuit 6 is generally also subjected to an operation resulting in a non-linear relationship between the input and the output signal, the output signal from quantizing circuit 6 must first be subjected to a reciprocal non-linear operation before it is combined with the prediction signal in adder 9. Particularly in the transmitter in FIG. 1 in which a compression is effected in quantizing circuit 6, the output signal from quantizing circuit 6 must undergo an expansion, which is reciprocal to this compression, in an expander circuit 10 before it is applied to adder 9.

Furthermore the prediction signal and the sampled speech signal are applied in the transmitter to a difference producer 11 for obtaining a difference signal which constitutes the input signal for the compressing non-uniform quantizing circuit 6. In the transmitter according to FIG. 1 the quantized and coded difference between the sampled speech signal and the prediction signal obtained in the local receiver is thus transmitted to the receiver according to FIG. 2.

The receiver cooperating with the transmitter of FIG. 1 is shown in FIG. 2. The code bits transmitted through a transmission path not further shown are applied after regeneration in a regenerator 12 to a PCM decoding circuit 13 associated with PCM coding circuit 7 in the transmitter. As a result a signal is obtaind at the output of PCM decoding circuit 13 which corresponds to the output signal of the compressing non-uniform quantizing circuit 6 in the transmitter. The code bit frequency required for regeneration and the sampling frequency of 9 kHz are derived from the transmitted signals by means of a PCM synchronizing circuit 14 which can be formed in a manner conventional for PCM transmission.

The output signal from PCM decoding circuit 13 is processed in the receiver according to FIG. 2 in exactly the same manner as the corresponding output signal from the compressing non-uniform quantizing circuit 6 in the local receiver of the transmitter of FIG. 1. The elements required for this purpose are denoted by the same reference numerals in the receiver as the corresponding elements in the transmitter, but are provided with indices in FIG. 2. The prediction signal derived from predictor 8' is combined with the signal derived from expander circuit 10' in adder 9' to a signal which is applied to a reproduction circuit 16 through a bandpass filter 15 passing the desired speech band and suppressing frequencies located above this band.

In the speech transmission system of FIGS. 1 and 2 the predictors 8 and 8' are formed as storage elements controlled by the sampling frequency of 9 kHz and holding a signal applied thereto during one sampling period.

The operation of the speech transmission system described so far will now be explained. If a speech signal x(t) is applied to sampling circuit 3 and sampling is carried out with a sampling period T(T is thus equal to one period of the pulse frequency of 9 kHz) the signal samples x(nT) occur at instants t=nT (n is an integer) at the output of sampling circuit 3. A difference signal e(nT) is constituted in subtractor or difference producer 11 from a signal sample x(nT) and a prediction signal x(nT) generated by the predictor 8, which difference signal is given by:

e(nT = x(nT) - x(nT) (1)

the compressed quantized difference signal e.sub.cq (nT) is generated with the aid of the quantizing circuit 6, which signal is transmitted to the receiver after coding in PCM coding circuit 7.

The prediction signal x(nT) desired at a given instant t=nT should in theory be derived from the signal samples of the speech signal x(t) at previous sampling instants t=(n-1)T, (n-2)T . . . . However, since the prediction signal in transmitter and receiver must be obtained in exactly the same manner and since only quantized values are available in the receiver, the prediction signal x(nT) is derived from the previous quantized signal samples x.sub.q { (n-1) T }, x.sub.q { (n-2)T } . . . and in the used predictor 8 it is derived by exclusively retaining the immediately preceding quantized signal sample x.sub.q { (n-1)T } during one sampling period T, thus:

x(nT) = x.sub.q { (n-1)T } (2)

the input signal of predictor 8 is then constituted by the quantized signal samples x.sub.q (nT) for which on the basis of formula (1) there applies that:

x.sub.q (nT) = x(nT) + e.sub.q (nT) (3)

so that x.sub.q (nT) is obtained by combining in adder 9 the prediction signal x(nT) with the quantized difference signal e.sub.q (nT). The difference signal e(nT) undergoes in quantizing circuit 6 a non-uniform quantisation and a compression so that the compressed quantized difference signal e.sub.cq (nT) at the output of quantizing circuit 6 must undergo an expansion which is reciprocal with this compression in expander circuit 10 so as to obtain the quantized difference signal e.sub.q (nT).

In the receiver the compressed quantized difference signal e.sub.cq (nT) is reobtained at the output of PCM decoding circuit 13, which signal is processed in exactly the same manner as in the local receiver of the transmitter. The quantized signal samples x.sub.q (nT) then occur at the output of adder 9', which samples yield a speech signal after filtering in bandpass filter 15 which, apart from quantisation noise and possible interference in the transmission path, corresponds to the speech signal x(t) at the input of sampling circuit 3 in the transmitter.

The described transmission signal uses very simple signal predictors 8 and 8' which bring about a signal extrapolation of the order of zero with the aid of a storage element retaining the last preceding signal sample for one sampling period T. In the transmission of speech signals a satisfactory transmission quality is already obtained with this simple system.

The predictos 8 and 8' may alternatively be arranged for a signal extrapolation of a higher order in which different preceding signal samples are involved in the signal prediction. It is known from the statistical communication theory how such a signal prediction can be rendered optimum, provided that the relevant signals are stationary. It is, however, known that speech signals do not fulfil this condition. Since a signal extrapolation of a higher order requires predictors 8 and 8' having a considerably more complicated structure and since the Applicant has found from extensive experiments that no considerable further improvement of the signal prediction can be expected for speech signals, the described signal extrapolation of the order of zero is preferred with which, as already stated, a speech transmission of satisfactory quality is obtained.

According to the invention a considerable improvement of the already satisfactory transmission quality of the speech signals is obtained in that the transmitter includes a dynamic control means 20 for controlling the dynamic range of the quantizing circuit 6 for its input signal and that both the local receiver and the receiver include dynamic control means 21, 21' for controlling the dynamic range of the quantizing circuit 6 for the signal to be applied to the adders 9, 9', while the transmitter and the receiver include control generators 17, 17' fed by the input signal from the predictors 8, 8', which generators are provided with storage networks, 18, 18' and averaging networks 19, 19' for obtaining a control signal which corresponds to the average of the absolute values of the input signal for the predictors 8, 8' over a limited number of sampling periods, which control signal is applied to a control input of the dynamic control means 20 and the dynamic control means 21, 21'.

The control generators 17, 17' used in the transmitter of FIG. 1 and the receiver of FIG. 2 are built up in the same manner in which corresponding elements have the same reference numerals while those in FIG. 2 have been provided with indices. The two control generators 17, 17' are arranged in such a manner that the control signal s at a sampling instant t=nT corresponds to the average of the absolute values of the input signal x.sub.q of the predictors 8, 8' at a limited number N of preceding sampling instants t=(n-1)T . . . , t=(n-N)T, so that there applies that: ##EQU1## In the transmission system according to FIG. 1 and FIG. 2 N is chosen to be 4. The storage networks 18, 18' each comprise four series-arranged storage elements 22-25, 22'-25', which are controlled by the sampling frequency of 9 kHz and each of which stores a signal applied thereto for one sampling period T. The absolute values are obtained by means of full-wave rectification and this in FIG. 1 and FIG. 2 by incorporating full-wave rectifiers 26, 26' before the storage networks 18, 18'. The averaging networks 19, 19' are formed in known manner with the aid of four resistors coupling the storage elements 22-25, 22'-25' to a summing resistor from which the control signal s according to formula (4) is derived.

Furthermore the dynamic control means 20 and 21, 21' for controlling the dynamic range of quantizing circuit 6 for the signal derived from difference producer 11 and the signal to be applied to adders 9, 9' are formed as amplifiers having an adjustable amplification factor. Amplifier 20 is connected to the input of quantizing circuit 6 while the amplifiers 21, 21' are connected to the outputs of the expander circuis 10, 10' so that the compression operation in quantizing circuit 6 is cancelled by a reciprocal expansion operation in expander circuits 10, 10' before the signal to be applied to adders 9, 9' undergoes a dynamic control. These amplifiers are so arranged that the amplification factor of amplifier 20 decreases and that of amplifiers 21, 21' increases with increasing values of the control signal. The control signal adjusts the amplification factor of amplifiers 21, 21' at the inverse value 1/A of the amplification factor A of amplifier 20. The adjustment of amplifiers 20, 21, 21' then has no influence on the quantized difference signal applied to adders 9, 9', provided that there is no limitation in quantizing circuit 6.

Since in practice the adjusted amplification factors A and 1/A must be accurately reproducible for a long time, the amplifiers 20, 21, 21' are not continuously, but stepwise adjustable. To this and these amplifiers include, for example, a number of stages having a fixed amplification factor which are connected with the aid of a corresponding number of switches in such a manner that a signal applied to the amplifier input is directly passed on or is passed on through one or more stages to the output. The structure of such amplifiers is generally known and need not be explained. In connection with this stepwise adjustment the control signal at the ouput of averaging networks 19, 19' is applied to evaluating circuits 27, 27' which determine in which partial interval of the total control signal range this control signal lies and which apply at each partial interval a certain adjusting signal to the amplifiers 20, 21, 21' for operation of these switches. These evaluation circuits 27, 27' may be formed, for example, as parallel-arranged threshold circuits whose respective threshold values are adjusted at the limits of two contiguous partial intervals, each threshold circuit operating a separate switch.

In the embodiment described the total control signal range between the minimum value s=0 and the maximum value s=s.sub.max is subdivided in four partial intervals so that four different adjusting signals and hence four different values of the amplification factors A and 1/A are obtained for the amplifiers 20, 21, 21'.

To obtain a structure which is most suitable for the use of digital techniques the amplification factors A and 1/A are adjusted in steps each bringing about a variation by a factor of 2. For a control signal increasing from the value s=0 the amplification factor A of amplifier 20 successively assumes the values 8, 4, 2, 1 and the amplification factors 1/A of the amplifiers 21, 21' successively assume the values 1/8, 1/4, 1/2, 1.

If the compressing non-uniform quantizing circuit 6 has its own dynamic range D for its input signal and consequently the expander circuits 10, 10' likewise have their own dynamic range D for their output signal, the adjustment of amplifier 20 at an amplification factor A means that the combination of amplifier 20 and quantizing circuit 6 for the difference signal e(n)T derived from difference producer 11 has an effective dynamic range D.sub.eff which is given by:

D.sub.eff = D/A (5)

the associated adjustment of the amplifiers 21, 21' at the amplification factor 1/A then results in the combination of expander circuit 10 and amplifier 21 and the combination of expander circuit 10' and amplifier 21' for the quantized difference signal e.sub.q (nT) applied to adders 9 and 9', respectively, having an effective dynamic range likewise given by formula (5). In the above-mentioned series of values of the amplification factor A this effective dynamic range D.sub.eff assumes successively the values D/8, D/4, D/2, D for a control signal increasing from the value s=0.

The limits of the partial intervals of the total control signal range at which limits the amplification factor A and thus the effective dynamic range D.sub.eff undergoes a variation by a factor of 2 are now chosen to be in such a manner that

the variance of the difference signal e(nT) for the respective partial intervals is different, which variance has the lowest value in the partial interval with s=0 as a lower limit and which variance has, with increasing values of s a larger value for each next partial interval;

the probability of occurrence of overload errors, which result when the difference signal e(nT) exceeds the value D.sub.eff adjusted for a given partial interval, does not exceed a given very small value which is equal for all partial intervals.

Based on extensive investigations of a speech signal having a bandwidth of 4 kHz and a duration of 3 minutes the following values were chosen for the limits of the partial intervals in the relevant transmission system: s.sub.max /64, s.sub.max /16, s.sub.max /8. This choise is also determined by the aim for a structure suitable for use of digital techniques.

The table below summarizes the data. The first column indicates the partial intervals as fractions of the maximum value s.sub.max while the second and third columns show the associated values of the amplification factor A and the effective dynamic range D.sub.eff.

______________________________________ s/s.sub.max A D.sub.eff ______________________________________ 0 - 1/64 8 D/8 1/64 - 1/16 4 D/4 1/16 - 1/8 2 D/2 1/8 - 1 1 D ______________________________________

It is surprisingly found from these extensive invenstigations that by using the control generators 17, 17' according to the invention a control signal s(nT) is obtained which is a satisfactory measure of the order of the difference signal e(nT) to be expected at the output of difference producer 11. On the basis of this control signal s(nT) the effective dynamic range D.sub.eff of the combination of amplifier 20 and non-uniform quantizing circuit 6 can be adjusted at any instant t = nT in such a manner that on the one hand the own dynamic range D of quantizing circuit 6 is always completely utilized for processing the then occurring difference signal e(nT), while on the other hand the probability of overload errors due to exceeding of this own dynamic range D retains a very low value. Thus, in the relevant transmission system a considerable increase of the signal-to-quantizing-noise ratio is realized over a large dynamic range of the speech signals to be transmitted, which results in a considerable improvement of the transmission quality.

For the purpose of illustration FIG. 3 shows for the speech transmission system according to FIG. 1 and 2 the signal-to-quantizing-noise ratio S/N.sub.q plotted as a function of the loading of the system by the speech signals to be transmitted (curve a). The root mean square value .sigma..sub.x /x.sub.max of the speech signal has been taken as a measure for loading. For speech signals this value is equal to the square root of the variance .sigma..sub.x.sup.2 because the mean value is equal to zero; furthermore the root mean square value is normalized on the maximum amplitude x.sub.max of the speech signal at which just no overload occurs. Both values S/N.sub.q and .sigma..sub.x /x.sub.max are shown in dB. Furthermore FIG. 3 shows the variation of S/N.sub.q for a system in which a non-uniform PCM coding with a piecewise linear compression characteristic according to the CEPT standard has been used for the speech signal itself, namely a PCM coding in which the number of code bits per signal sample likewise as in the relevant system is 8 (curve b).

A comparison of both curves a and b in FIG. 3 shows that in the relevant system a considerable gain in the signal-to-quantizing-noise ratio is reached relative to the known system and this gain is increased as the loading of the system decreases. This gain can be used to reduce in the relevant transmission system the bandwidth required in the transmission path -- that is to say, the number of code bits per signal sample -- and yet reach a transmission quality comparable with that in the known system. In the relevant system the remarkable advantage is obtained that in spite of the reduction in the required bandwidth the loading range with a satisfactory transmission quality if considerably increased relative to that in the known system.

For completeness' sake FIG. 3 also shows the broken line curves a', and b' which represent the variation of S/N.sub.q calculated from the said investigations in the present and the known system for the physically impracticable case where the non-uniform quantizing characteristics would be continued in an unlimited manner.

As stated in the foregoing the probability of overload errors as a result of exceeding the adjusted dynamic range is very small in the relevant transmission system and it is already by this fact that their influence on the transmission quality is very small. In addition the extensive investigations show that surprisingly an overload error in the relevant transmision system practically does not give rise to a propagation of errors although such a propagation of errors may be expected on the ground of the manner of generating the prediction signal.

An overload error at the instant t = nT does not only cause an error in the quantized signal sample x.sub.q (nT) in the transmitter and the receiver and hence an instantaneous increase of the quantisation noise, but at the next sampling instant t = (n+1)T this overload error also causes a prediction signal x{(n+1)T } which is equal in the transmitter and the receiver but is erroneous. In the most unfavourable case an overload error may again be produced in the known transmission system at the instant t=(n+1)T which in turn causes an overload error at the instant t=(n+2)T and so forth.

The extensive investigations referred to hereinbefore show that after an overload error at the instant t=nT the dynamic range in the present transmission system is automatically adjusted at the subsequent higher value so that the probability of exceeding this larger dynamic range is so drastically reduced that already at the next sampling instant t=(n+1)T practically no overload errors occur any longer. Thanks to this self correcting property of the present transmission system a propagation of errors after an overload error is extremely rapidly interrupted or even entirely prevented, even in the most unfavourable case. The overload errors occurring with a small probability are thus found to have no noticeable influence in practice on the transmission quality obtained by using the steps according to the invention. relationship between

By adjusting the effective dynamic range to an optimum value in each partial interval of the total control signal range a considerable improvement of the transmission quality is realized in the present transmission system, which improvement is also due to the self-correcting behaviour of the system after overload errors just for low values of the signal loading. Due to these favourable properties the transmission quality can be further improved by not only adjusting the optimum effective dynamic range in each partial interval of the total control signal range, but also by adapting the non-linear relationship between input and output signal of the non-uniform quantizing circuit in an optimum manner to the probability density function of the difference signal to be transmitted in this partial interval.

If the non-uniform quantizing circuit 6 in FIG. 1 is formed in a conventional manner so that the non-uniform distribution of the decision levels for the input signal corresponds to a uniform distribution of the associated representative levels for the output signal, the different non-linear characteristics can be realized in a simple manner by selecting for each partial interval of the control signal a suitable non-uniform distribution of the decision levels and by leaving the uniform distribution of the associated representative levels unchanged. When in the quantizing circuit 6 the network from which the decision levels are derived is formed in an adjustable manner, other piecewise linear compression characteristics and, if desired, also expansion characteristics than the mentioned piecewise linear compression characteristic according to the CEPT standard can be adjusted with the aid of the control signal. Similarly the associated reciprocal non-linear characteristics can be realized in the expander circuits 10, 10' in FIGS. 1 and 2.

In this manner the control signal obtained by using the steps according to the invention can be used for completely utilizing the available dynamic range for processing the difference signal to be transmitted, but also for adapting the compression and expansion characteristics to be used in this dynamic range to the probability density function of this difference signal. With the use of this extra degree of freedom an extra gain in the signal-to-quantizing-noise ratio in the order of 6 to 8 dB can be obtained so that the already satisfactory transmission quality of the transmission system according to the invention can be further improved considerably.

In the transmission system described so far it has been assumed that no errors occur in the transmission channel. Although in the conventional PCM transmission channels the probability of transmission errors is very low and consequently their influence on the transmission quality is likewise very low in practice, these errors may under very unfavourable circumstances give rise to a propagation of errors in the receiver which may affect the transmission quality.

The result of a transmission error is that the corresponding quantized signal samples x.sub.q (nT) in transmitter and receiver deviate from each other so that the prediction signal x{(n+1)T} in the receiver is likewise wrong. In the most unfavourable case this erroneous prediction in the receiver may lead sooner or later to an erroneous adjustment of the dynamic range whereafter in transmitter and receiver different prediction signals are generated and the relationship between the difference signal to be transmitted and the quantized signal samples in the receiver is lost.

In the transmission system of FIGS. 1 and 2 the continuation of an erroneous prediction in the receiver after the occurrence of transmission errors may be interrupted by transmitting, instead of the different signal e(nT), the signal sample x(nT) itself at regular instants.

To this end the relevant transmission system includes two synchronously operating maximum-length shift register series generators 28 and 28' which are incorporated in the transmitter of FIG. 1 and in the receiver of FIG. 2, respectively. These generators 28, 28' are constituted as a shift register having p stages which are controlled in the transmitter by sampling pulse generator 4 and in the receiver by a local sampling pulse generator 29 present in a PCM synchronizing circuit 14, which shift register is provided with a modulo-2-feedback circuit. As is generally known such a generator 28, 28' may generate a pseudo-random series of binary pulses with a period L=(2.sup.P -1)T where T is the sampling period. State detectors 30, 30' are connected to these generators 28, 28' which supply one pulse in each period having a length of L for a given content of the shift register, for example, when all stages are in the 1 state. The output pulse from state detectors 30, 30' brings the contents of all storage elements 22-25, 22'-25' in storage networks 18, 18' to their maximum value and brings the contents of predictors 8, 8' to the value of zero. The control signal defined by formula (4) is then brought to its maximum value s.sub.max and thus the effective dynamic range in transmitter and receiver is likewise adjusted to its maximum value D. Furthermore the prediction signal has the value of zero at the next sampling instant so that at that instant the signal sample instead of the diference signal is transmitted. If this transmission is not interfered, the relevant transmission system operates in the normal manner at the next sampling instant. Thus, the continuation of an erroneous prediction in the receiver due to a transmission error is interrupted.

For the required synchronisation of generators 28, 28' the techniques described in the Article "Common bandwidth transmission of information signals and pseudonoise synchronization waveforms", IEEE Transactions on Communication Technology, Vo. COM-16, No. 6, December 1968, pages 796-807 can advantageously be used because then no extra bandwidth or extra time is required for the synchronization transmission. The transmitter according to FIG. 1 includes an adder 31 in which the output signal from generator 28 (having, for example, a level of -20 dB) is added to the output signal from PCM coding circuit 7 without frequency separation and time separation. In the receiver of FIG. 2 the transmitted output signal from adder 31 and the output signal from generator 28' are applied to a cross-correlator 32 for producing a control signal which is applied through a smoothing filter 33 to the local sampling pulse generator 29 in the form of a voltage-controlled oscillator. With the aid of this closed control loop an accurate and fast synchronisation of generator 28' is realised. To further reduce the already slight influence of the synchronizing signal on the transmission of the output signal from PCM coding circuit 7, the output signal from generator 28' is subtracted in a difference producer 34 from the input signal to the receiver.

It is to be noted that in the present transmission signal the steps for interrupting the erroneous prediction after a transmission error ma