CA2300835C - Method for packet transmission using an arq protocol on transmission channels in a digital transmission system - Google Patents

Abstract

In a method for packet transmission using an ARQ protocol on transmission channels in a digital transmission system, for channel coding, turbo coding is carried out in a turbo coder at the transmitter end and turbo decoding is carried out in a turbo decoder at the receiver end using soft-decision output signals, with a return channel being provided, using which the receiver requests the information from faulty packets once again. An RCPTC is used as the turbo code. When the information in a faulty packet is transmitted once again, at least a portion of the information suppressed by the puncturing of the RCPTC in the previous transmission is transmitted. This additional information is inserted into the already existing information at the receiver end, and this completed information is decoded once again. When the repeat transmission is made, the only bits which are transmitted are those which are additionally available at the next lower coding rate, since they are not punctured. The method is repeated until the packet is decoded without any errors or all the coded information in a packet is transmitted.

Description

Description tHFthod for packet transmission using an ARQ protocol on transmission channels in ~ digital transmission system The invention relates to a method for packet transmission using Gn ARQ (Automatic Repeat Request) protocol on transmission channels in a digital transmission system, in which, for channel coding, turbo 1Q coding is carried out in a turbo coder at the transmitter end and turbo decoding is carried out in a turbo decoder at the receiver end using soft-decision output signals, with a return channel being provided, using which the receiver requests the information from faulty packets once again.
she use of turbo codes for dieital transmission ~,,~~~=ms is investicated in F. Jung "Comparison of Turbo:-Code Decoders Applied to Short Frame Transmission ~vstem=", IEEE Journal on Selected Areas in 2G Communications, Volume 1~ (1°°6) pages 530-537, with both coders and decoders being investigated for the turbo cedes in the transmission path. Decoding of the turbo codes is based on the use of soft-inputlsoft-output decoders, which can be produced using either MAP
_' ~I~G~;imum a-postericri) symbol estimators or MAF
=ec:uence estimators, for example an estimator using an -priori soft-output Viterbi algorithm (AFRI-SOVA).
Ths publication describes four different decoder arrangements and their capabilities to process specific 3C:~ error rates. FurthErmore, the performance of these decoders is investigated for different applications. It has been found that the turbo codes and their iterative decedinc are an effective measure against packet errors.
ICC ' °5, Seattle, tvashington, June 18-2c, 1°95, "'~urbc Codes for BCS. Applications", D. Div_salar and F.
Follara, proposes turbo codes to achieve error correction ' CA 02300835 2000-02-18 virtually as far as the so-called Shannon limit.
Relatively simple component codes and large interleavers are intended to be used for this purpose.
In this publication, the turbo codes are produced in a coder using multiple codes, and are decoded in a suitable decoder. The turbo codes were introduced by Berrou et al. 1993 (see C. Berrou, A. Glavieux and P.
Thitimayshima, "Near Shannon limit area correction coding: Turbo codes" Proc. 1993 IEE International conference on communications, pages 1064-1070). On the one hand, this method allows very good error correction to be achieved.
So-called turbo equalization is known from ETT
European Transactions on Telecommunications, Vol. 6, No. 5, September-October 1995, "Iterative Correction of Intersymbol Interference: Turbo-Equalization", Catherine Douillard et al., whose use is intended to overcome the disadvantageous effects of intersymbol interference in digital transmission systems which are protected by convolution codes. The receiver makes two successive soft-output decisions, which are made in an iterative process by a symbol detector and a channel decoder. Each iteration makes use of extrinsic information from the detector and the decoder for the next iteration, as with turbo decoding. It was found that intersymbol interference effects in multipath channels can be overcome by turbo equalization.
The publication "A Novel ARQ Technique using the Turbo Coding Principle", Narayanan et al., IEEE
Communications Letters, Volume 1, No. 2, March 1997, pages 49-51 describes an ARQ-III method using punctured turbo codes, in which the bits which were punctured for the first transmission are transmitted after the occurrence of faulty data packets. The receiver then combines the punctured code and the punctured bits, and thus obtains the unpunctured code.
AMENDED SHEET

' CA 02300835 2000-02-18 Future transmission systems, for example the European UMTS (Universal Mobile Telecommunications System), require the support of a large number of co-existing carrier services with carrier data rates of up to 2 Mbit/s in a flexible manner, with the best-possible spectral efficiency being desirable. An MA
(Multiple Access) scheme has been developed in the ACTS
(Advanced Communications Technologies and Services) project AC090 FRAMES (Future Radio Wideband Multiple Access Systems), which is called FRAMES Multiple Access (FMA) and satisfies the UMTS requirements. As a third-generation transmission system, which covers a wide range of application areas, carrier services and widely differing scenarios, FMA must to comply with present and future developments of UMTS radio interface standards. FMA comprises two operating modes, namely WB-TDMA (Wideband Time Division Multiple Access) with and without spreading and compatibility with GSM
(Global System for Mobile Communications) and WB-CDMA
(Wideband Code Division Multiple Access). Although, essentially, a system based on FMA is considered here, it is also possible to include other transmission systems using multiple access methods, for example FDMA
(Frequency Division Multiple Access) MC-CDMA
(Multicarrier-CDMA) or combinations of the said transmission systems.
With regard to the high performance of turbo codes, it is desirable to use these in digital transmission systems. The complex requirements, for example for FMA, mean, however, that it is necessary when using such turbo codes to ensure that the data transmission makes full use of the capabilities of the turbo codes.
The invention is based on the object of providing a method for packet transmission using an ARQ
protocol on transmission channels in a digital transmission system in which turbo coding is used for channel coding, in which the channel loading from ARQ
AMENDED SHEET

car~ L~e ~;ept as low as possible by means of a new turbo code and puncturing matched to it.
According to the invention, the method mentioned initially is for this purpose characterized by the features cf claim 7.
When the RCPTC (Rate Compatible Punctured Turbo Coding) is used, the coding rate can be se' b;.T suitable puncturing of the systematic or non =~~~stematic infcrmation at the output of the turbo 1f encoder. An increase in the coding rate, that is to say rT;ora information being punctured, in this case means shat the decoding result is worse fer a given channel ouality_ This means that the bit error rate BER
_ncreases. The use of the RCPTC for channel coding _. means that it is not necessary to transmit the entire r~~ac~:et once again in packet-switching services when an ::=,P,Q is initiated. The first transmission of the packet i= carried cut using a high coding rate, that is to say ~,.ith little error protection, and if the packet is C' identified as being faulty, then an ARQ is initiated.
P.fter thps, rather than transmitting the entire packet gain, only the information which was punctured in the first tran_smis=ion, or a portion of this punctured information, is transmitted. The coding rate is thus compatibly matched to the channel as a result of which, ~v~;erall, less data need be transmitted over the cv:annel. The advantage of this method is thus that the total load or. the channel is reduced.
In this document, the term Quality of Service ~G is used as follows. Specific QcS criteria (QoS
Quality ef Service) apply to various services, and the definitions of the QoS criteria for various carrier ser~.~ices have been worked out in the course of FRAMES.
one important component of a QoS criterion is the ~" carrier data rate R. The QoS criterion also includes a ma::imum permissible error rate P~~ or a packet loss rate F;" in conjunction with a maximum failure probability F;,~;~~. In the case of line-switched services, the probability Pr{Pb > Pb~} of the instantaneous bit error rate Pb exceeding Pb~ must not be greater than PoutG. that is to say Pr { Pb > Pb~ } < Poutc .
For voice transmission, Pb~ is equal to 10-3 and Pouts is equal to 0.05. A similar condition for the instantaneous packet loss rate P1 applies to packet services:
Pr { P1 > P1~} < PoutG.
Apart from the criteria relating to Pr, there are also other conditions relating to the QoS
criterion. However, the QoS parameters Pb~, P1~ and PoutG
will mainly be considered here, which relate directly to the choice of the error correction code (ECC). For ECC, the coding rate R~ is essentially governed by the multiple access method, the modulation and the packet parameters. In other words, the coding rate R~ is directly related to the question as to whether a QoS
criterion is or is not satisfied for a specific service.
In a method in which a soft-input/soft-output symbol or sequence estimator is used at the receiver end, it is advantageous for the Quality of Service to be determined from the variances a2 of the soft-decision output signals from the turbo decoder, with the bit error rate advantageously being calculated from the variances a2, as a measure of the Quality of Service.
In a method in which an MAP symbol estimator or an MAP sequence estimator is used at the receiver end, it is advantageous for the Quality of Service to be determined from the variances a2LLR of the soft-decision output signals from the turbo decoder.
In a method in which a Viterbi algorithm for sequence estimation is used at the receiver end, it is AMENDED SHEET

- Sa -advantageous for the Quality of Service to be determined from the variances a2~IT of the soft-decision output signals from the turbo decoder.
AMENDED SHEET

In a method in which o Viterbi algorithm f.or sequence estimation is used at the receiver end, it is advantageous for the Quality of Service to be determined from the variances a'viT of the soft-decision output signals fxom the turbo decoder.
Since the method according to the invention can be used not only with MAP estimators but also for estimation using a Viterbi algorithm, there is <<irtually no limitation with regard to the most important methods for sequence and symbol estimation.
This is true even though this statement is used only in conjunction with an MAP symbol estimator in the following specific description.
According to one advantageous refinement, the method according to the invention is characterized in that se-called Berrou's puncturing is used for puncturing, in which only the non-systematic information i= punctured. This type of puncturing is Gd~-antaaeou= for relatively low signal-to-noise ratio ~0 values.
According to one advantageous refinement, the method according to the invention is characterized in that so-called UKL (University Kaiserslautern) puncturing is used for puncturing, in which both systematic information and non-systematic information are punctured. This type of puncturing is advantageous for relatively high signal-to-noise ratios, and thus for bit error rates of <10-4.
According to one advantageous refinement, the method according to the invention is characterized in 3f. that when the repeat transmission is made, the only bits which are transmitted are those which are additionally available at the next lower coding rate, since they are net punctured. In consequence, only the minimum required information is transmitted in order to achieve the aim of reasonable transmission quality.

According to one advantageous refinement, the method according to the invention is characterized in that the method is repeated until the packet is decoded without any errors or all the coded information in a packet is transmitted. The entire potential of the coding is thus always fully utilized to correct the errors in a packet.

Exemplary embodiments of the invention will now be des cribed with reference to the attached drawings, lU in which:

Figure _ shows a block diagram of a turbo coder;

Figure 2 shows a block diagram of an RSC (Recrusive Systematic Convolutional) coder, as is used in a turbo coder as shown in Figure 1;

Figure 3 shows a block diagram of a turbo decoder;

Figure 4 shows a graph of the performance of an RCPTC

in an AWGN (Additive White Gaussian Noise) channel as a function of the number of iterations of turbo decoding;

Fia,ure ~ shows a graph of the performance of an RCPTC

in ~ Rayleigh channel for a different number of iterations of turbo decoding;

:~~;:~_~re 6 shows a graph of the performance of an RCPTC

in an AWGN channel as a function of different coding rates;

Fioure i chews a graph of the performance of an RCPTC

in a Rayleigh channel as a function of different coding rates;

Figure 8 shows a graph of the performance of an RCPTC

in an AWGN channel for various coding rates;

.;cure G shows a graph of the performance of an RCPTC

in a Rayleigh channel for various coding rates;

_ 8 _ Figure 10 shows a graph of the relationship between the bit error rate BER and the variance a2LLR of the output of the second decoder; and Figure 11 shows a graph of the variance a2LLa plotted against the signal-to-noise ratio for incorrectly and correctly decoded packets.
With regard to economic use of the hardware, the ECC circuit should be usable as universally as possible, while the ECC configuration should allow high flexibility by means of software control. The RCPTC
used here allows this, since it has the necessary flexibility. The RCPTC can be produced using the turbo coder 2 shown in Figure 1. The turbo coder 2 has Ne = 2 binary, recursive, systematic convolution coders 4, 6 (RSC) with a small constraint length, for example between 3 and 5, which are connected in parallel using a turbo interleaver 8. The input sequence a is supplied to the coder 4 (RSC, Code 1) and, via the turbo code interleaver 8, to the coder 6 (RSC, Code 2 ) as well as to a puncturing/multiplexer device 10. The puncturing/
multiplexer device receives a further input C1 from the coder 4 and a further input C2 from the coder 6. The output of the puncturing/multiplexer device 10 is the output sequence b.
In the turbo coder 2, the minimum code rate R~, min is equal to 1/ (Ne+i) - 1/3. The minimum coding rate R~, min could be further reduced by using additional RSC coders.
The binary input sequence u, which has a finite time duration, is entered in the coder 9, at whose output it produces the redundant sequence C1 with the same finite duration as u. The sequence uI, which represents the sequence a after interleaving, is passed to the coder 6. The coding in _ g _ the coder 6 results in the redundant sequence C2. The redundant sequences C1 and C2 and the sequence a are punctured and multiplexed, in order to form the output sequence b. The turbo coder is a systematic coder, with a being the basis of the systematic information contained in b.
An RSC coder, as can be used for the coders 4 and 6, is shown in Figure 2, using the example of the coder 4. The sequence a is applied as systematic information to one input of the coder 4. The sequence a passes via an addition element 12 to a delay stage 14 and to a further addition element 16. The output signals from the delay stage 14 pass to a second delay stage 18 and to the addition element 12. The output signals from the second delay stage 18 pass to the addition stage 12 and to the addition stage 16. The output from the addition stage is then the redundant sequence C1.
The hardware costs are a significant factor in the choice of this coder, and should be kept as low as possible. For this reason, the two RSC coders for use in the context of FRAMES are identical and have a constraint length of 3. Although these RSC coders have only four states, their performance is good when the values of the signal-to-noise ratio Eb/No are low. The performance of the RCPTC using these coders is thus advantageous at low signal-to-noise ratios.
The output sequence b of the turbo coder 2 passes via the transmission channel and a demodulator to the turbo decoder 22 (Figure 3), which has an RSC
coder 24 and a second RSC decoder 26. A turbo code interleaver 28 is provided between one output of the decoder 24 and one input of the decoder 26. A turbo code interleaver 30 is provided between one output of the decoder 26 and one input of the decoder 24. The decoders 24, 26 are soft-input/soft-output decoders.
The demodulator (not shown) supplies estimated values xn of the systematic information un which is contained in u, as well as estimated values yl,n and y2,n of the transmitted redundant bits, which have been produced by the coders 4 and 6, respectively. The two decoders 24, 26 require channel state information (CSI), which comprises the instantaneous signal amplitudes and the noise variance. Each of the decoder s 24, 26 processes the systematic information, the redundant and a-priori information Lei,n and Le2," while processing the CSI, as a result of which the extrinsic information Lez,n and Lel,n is produced, which is then used as a-priori knowledge in the downstream decoder.
The decoding process is iterative, and the result of the decoding process is improved with each iteration.
However, the amount of improvement gradually decreases with further iterations. after a certain number of iterations, the output signal from the turbo decoder 22 is supplied to a detector (not shown), as is normal for such transmission systems.
In order to match the use of the RCPTC to the existing service requirements, it would be conceivable to adapt the RSC coders, but this would lead to an undesirable additional load with regard to the hardware costs. The matching of the interleaver size to the specific services is known per se and, is also a problem when using an RCPTC, owing to its flexibility.
Furthermore, the number of iterations for decoding can be set in accordance with the QoS
criterion, taking account of the overall coding complexity. There are two options at the receiver for utilizing this characteristic of the turbo code. For a given QoS

criterion, the number of iterations can be increased as the signal-to-noise ratio Eb/No rises. This is particularly advantageous in fading channels, for example in transmission channels. On the other hand, the number of iterations can also be varied with a QoS
criterion that changes with time. The capability to adjust the number of decoding iterations is available only when using turbo codes, in particular an RCPTC.
A further option for improving the performance in a system with an RCPTC is to set the puncturing so that an RCPTC with varying code rates R~,min <= R~ <_ Rc,max can be provided, by which means the coding characteristics can be changed, without changing the turbo code interleavers or RSC coders.
In principle, the sequences u, cl and c2 are available for puncturing. If two of the sequences are completely suppressed by puncturing, the maximum code rate R~,max - 1 is assumed. In this case, the coding characteristics depend on which of the sequences are punctured. If, for example, the redundant sequences cl and c2 are completely punctured, with only the sequence a being passed through unchanged, no ECC is available and time diversity gain cannot be achieved in receivers for fading channels. In this situation, the turbo decoder is reduced to a simple threshold-value detector.
If one of the redundant sequences C1 or Cz is completely suppressed by the puncturing process, in which case only the second redundant sequence together with the sequence a can pass through, the turbo coder becomes a conventional RSC coder. The turbo decoder is reduced to an RSC decoder, which is designed to carry out half an iteration. In this case, there is no a priori knowledge based on extrinsic information. The coding rate R~ may be varied between '~ and 1, depending on the QoS
criterion. Since Ne = 2, the RSC coders can be based on two different codes, and the QoS criterion and the coding complexity can be varied by suppressing a specific redundant sequence C1 or C2, without changing the coding rate R~.
However, the options mentioned above prevent turbo code operation, which is available only when bits of the two redundant sequences C1 and C2 are transmitted and un # ul,n where un and ul,n are contained in a and ul, respectively. In this case, Rc,min <= Rc < 1.
The minimum coding rate R~,min - 1 / (Ne + 1 ) is achieved if no puncturing is carried out. In this case, either the conventional RSC decoding or turbo decoding may be provided, depending on the QoS criterion and transmission channel state, with both factors varying with time for transmission applications.
The following versions are possible for real turbo code operation. The sequence a is not punctured, the redundant sequences cl and c2 are partially punctured. In this case, operation as RSC code or as turbo code is possible, the number of decoding iterations can be adjusted, and the coding rate may be between 1/3 and 1. This type of puncturing is called Berrou's puncturing.
An alternative option is for the sequence a and the redundant sequences cl and c2 to be partially punctured.

In this case, operation with RSC code is impossible, and is possible only with turbo code. The number of decoding iterations can be adjusted, and the coding rate may be between 1/3 and 1. This type of puncturing is called UKL puncturing (UKL - University Kaiserslautern). Finally, the situation may also be considered in which no puncturing takes place. In this case, operation with an RSC code and a turbo code is possible. The number of decoding iterations can be adjusted, and the coding rate is around 1/3.
The advantageous feature of RCPTC is the capability to vary the coding rate R~ adaptively, in which case the required information for an ARQ can be transmitted without having to transmit the entire, coded packet. It is sufficient to transmit an additional part of the information, which compensates for the difference in the coding rate.
Now that the options for adaptation of the coding in the case of RCPTC have been described, the effects of the matching options on the performance of the system using RCPTC will be described, on the basis of simulations.
Figure 4 shows the performance of the RCPTC as a graph showing the bit error rate BER plotted against the signal-to-noise ratio Eb/No for voice transmission via an AWGN channel. The packet size was 150 bits, and the coding rate was roughly 1/3. The carrier data rate for voice transmission was 8 kbit/s. The uncoded transmission is shown as a reference line. The parameter for these simulations is the number of decoding iterations, which varies between 1 and 5.
After the first decoding iteration, the minimum signal-to-noise ratio which is required to achieve a bit error rate of < 10-3 is approximately equal to 3.5 dB. After the second decoding- iteration, approxirnately 1.3 dB less is required. The next decoding iteration allows a further gain of 0.2 dB. The next iteration allow gains of less than 0.1 dD. After five iterations, the minimum signal-to-noise ratio which is required for a bit error rate of less than 10--.= approximately equal to 1.8 dB. It can thus be seen tat the performance improvement becomes less as the number of iteratic:ns increases. By comparison, o conventional NSC (Non-Systematic Convolutional) code with a constraint length of 9 requires approximately 1.9 dB to achieve the same bit error rate of <10-3. The RCPTC is thus somewhat more powerful than conventional codes, even with packet sizes as small as 150 bits.
Figure 5 shows a graph of the performance of 1' the RCFTC, in which the bit error rates BER are plotted against the signal-to-noise ratio EEIN~ for narrowband I~DP~3 with a carrier data rate of 199 kbit/s, a packet size of 672 bits, a code rate of about 1/2 and a Raj~leiah fading channel which is completely subjected ~-t? t.~ interl=awing. The simulation parameter is once again t!~e number of decoding iterations. After four decoding i t.erations, a bit error rate of less than 10-' requires minimum signal-to-noise ratio of 3.8 dB. Only about -.4 dB is still required after ten iterations. A
ccn~%entionai I4SC code with a decoding complexity similar to that of four decoding iterations has a constraint length of 8, and requires a signal-to-noise ratio that is 1.1 dB greater.
Figures 6 to 9 show graphs of the performance 3C? when using the RCPTC, with the bit error rate BER or '--tie frame error rate FER, respectively, being plotted a,ainst the signal-to-noise ratio EE/No. Figure 6 shows the bit error rate plotted against the signal-to-noise rGtic for a packet size of 672 bits, ten decoding 3, iterations and one AWGN channel. Figure 7 shows the bit error rate plotted against the signal-to-noise ratio for a packet size of 672 bits, ten decoding iterations and a Rayleigh fading channel which is completely subjected to interleaving.
Figure 8 shows the frame error rate FER plotted against the signal-to-noise ratio for a packet size of 672 bits, ten decoding iterations and one AWGN channel.
Figure 9 shows the frame error rate plotted against the signal-to-noise ratio for a packet size of 672 bits, 10 decoding iterations and a Rayleigh fading channel which is completely subjected to interleaving. The graphs in Figures 6 to 9 use two different puncturing methods, namely Berrou's puncturing and UKL puncturing, which have been mentioned above. As can be seen, Berrou's puncturing has better performance for lower values of the signal-to-noise ratio, while UKL puncturing is advantageous for a high signal-to-noise ratio, and thus for bit error rates of <10-4. The intersections move in the direction of lower bit error rates as the coding rates increase.
In Figure 10, the bit error rate is plotted against the variance of the log-likelihood ratios (LLR) at the output of the second decoder, with an RCPTC, a packet size of 372 bits, ten decoding iterations and one AWGN channel being assumed. It can be seen from this figure that the coding rate has no effect on the relationship between the bit error rate and the variance 62LLa. since these two variahl PC aYa cirri ~ err, <.
dependent on the signal-to-noise ratio EB/No. Thus,if ~~LLR 1S known, the bit error rate can easily be estimated, and this as a basis for result can be used an action, for example adaptation of the number of decoding iteratio ns or adaptation of the coding rate in order to improve the transmission quality the or, in case of ARQ, to request a repeat tran smission an of incorrectly coded packet.

20365-4199 ' Finally, Figure 11 shows the variance a'LLa of the log-likelihood ratio LLR of the output of the second decoder plotted against the signal-to-noise ratio EB/No when using an RCPTC with a packet size of 600 bits, a code rate of about 5/9, ten decoding iterations and one AWGN channel. The RCPTC was designed for a 64 kbit/s carrier service. From Figure 11, it can be seen that a similar consideration to that in connection with Figure 10 also applies to the lU dependency ~ of the variance a2LLR on the occurrence of packet errors. 62LLR for incorrectly decoded packets is always greater than ~Q2~,Lp for correctly decoded packets .
Thus, if the signal-to-noise ratio EH/No and a2Lr.a are known for a packet which is currently being checked, a soft-decision variable, which is related to the probability of a packet error, can easily be produced ~:nd used for control purposes.
In accordance with this invention, there is provided a method for packet transmission using an ARQ
(Automatic Repeat Request) protocol on transmission channels in a digital transmission system, in which, for channel coding, turbo coding is carried out in a turbo coder (2) at the transmitter end and turbo decoding is carried out in a turbo decoder (22) at the receiver end, with a return ~5 channel being provided, via which the receiver can request incorrectly coded packets once again, in which case, when an incorrectly coded packet is transmitted once again, at least a portion of the information suppressed by the puncturing of the turbo code in the previous transmission is transmitted, and the additionally transmitted information is inserted into the already existing information at the receiver end, and this completed information is decoded once again characterized in that a punctured turbo code with a variable coding rate (RCPTC) is used, and in that the coding rate of the turbo coder (2) is chosen as a function of the Quality cr Service (QoS) of the transmission channel.

- 16a -Although the present description relates mainly to use of the invention with digital mobile telephones, the invention is not limited to this but can in general k~~e used for digital transmission systems, for example power-based systems, optical transmission systems (infrared and laser transmission systems), satellite radio systems, deep-space transmission systems, radio-link transmission systems and broadcast-radio transmission systems (digital radio or TV), with the said advantages.

Claims (11)

Claims

1. A method for packet transmission using an ARQ
(Automatic Repeat ReQuest) protocol on transmission channels in a digital transmission system, in which, for channel coding, turbo coding is carried out in a turbo coder (2) at the transmitter end and turbo decoding is carried out in a turbo decoder (22) at the receiver end, with a return channel being provided, via which the receiver can request incorrectly coded packets once again, in which case, when an incorrectly coded packet is transmitted once again, at least a portion of the information suppressed by the puncturing of the turbo code in the previous transmission is transmitted, and the additionally transmitted information is inserted into the already existing information at the receiver end, and this completed information is decoded once again characterized in that - a punctured turbo code with a variable coding rate (RCPTC) is used, and in that - the coding rate of the turbo coder (2) is chosen as a function of the Quality of Service (QoS) of the transmission channel.

2. The method as claimed in claim 1, characterized in that a repeat transmission of incorrectly coded packets (ARQ) is initiated as a function of the Quality of Service (QoS) of the transmission channel.

3. The method as claimed in claim 2, characterized in that - the Quality of Service (QoS) of the transmission channel is determined, in that - the Quality of Service (QoS) of the transmission channel is compared with a nominal value, and in that - a repeat transmission of incorrectly coded packets (ARQ) is initiated if the nominal value is not achieved.

4. The method as claimed in claim 3, in which a soft-input/soft-output symbol or sequence estimator is used at the receiver end, characterized in that the Quality of Service (QoS) is determined from the variances (.sigma.2LLR) of the soft-decision output signals from the turbo decoder.

5. The method as claimed in claim 4, characterized in that the bit error rate (BER) is calculated from the variances (.sigma.LLR2) as a measure of the Quality of Service (QoS).

6. The method as claimed in one of claims 3 to 5, characterized in that the number of data packets transmitted once again is chosen as a function of the Quality of Service (QoS) of the transmission channel.

7. The method as claimed in claim 6, characterized in that the number of data packets transmitted once again is chosen in such a manner that, after the repeat transmission, the Quality of Service (QoS) of the transmission channel reaches the nominal value.

8. The method as claimed in claim 3, with an MAP
symbol estimator or an MAP sequence estimator being used at the receiver end, characterized in that the Quality of Service (QoS) is determined from the variances (.sigma.2LLR) of the soft-decision output signals from the turbo decoder.

9. The method as claimed in claim 3, with a Viterbi algorithm for sequence estimation being used at the receiver end, characterized in that the Quality of Service (QoS) is determined from the variances (.sigma.2VIT) of the soft-decision output signals from the turbo decoder.

10. The method as claimed in one of claims 1 to 9, characterized in that so-called Berrou's puncturing is used for puncturing, in which only the non-systematic information is punctured.

11. The method as claimed in one of claims 1 to 9, characterized in that so-called UKL puncturing is used for puncturing, in which both systematic information and non-systematic information are punctured.