WO2001052424A1

WO2001052424A1 - Method for effecting error protection during the transmission of a data bit stream

Info

Publication number: WO2001052424A1
Application number: PCT/DE2001/000022
Authority: WO
Inventors: Wen Xu
Original assignee: Siemens Aktiengesellschaft
Priority date: 2000-01-13
Filing date: 2001-01-05
Publication date: 2001-07-19
Also published as: DE10001147A1

Abstract

The invention relates to a method for effecting error protection during the transmission of a data bit stream in a digital message transmission system according to which a recursive convolutional code is used to code the data bit stream block-by-block during channel coding. In order to protect the information-carrying bits located at the end of a data block, a predetermined number of tail bits that are known beforehand are appended to the end of the respective data block before channel coding.

Description

description

Procedure for error protection during the transmission of a data bit stream

The invention relates to a method for error protection in the transmission of a data bit stream in a digital message transmission system, in which ^{: in} the case of channel coding, the data bit stream is coded in blocks using a recursive convolutional code. It also relates to a device for performing the method and a device for decoding a data bit stream transmitted using the method.

Such methods and devices are used, among other things, for the transmission of signals, such as data, voice, audio or video signals. An example of an area of application is in particular voice transmission in mobile radio networks.

For the transmission over a cellular route z. B. the speech signals first in a source encoder, in this case a language encoder, broken down into various parameters by which temporal language sections can be described. These real-valued speech parameters are then quantized and thus represented by a certain bit combination. The speech parameter coded digitally in this way is then forwarded. In the case of other source signals, there is a similar breakdown into so-called source-coded parameters or also source-coded coefficients, which are then passed on. Other data signals to be transmitted, e.g. SMS signals in the GSM standard are usually already available in digital form.

Since disturbances and losses can be expected when transmitting the parameters via a communication transmission channel, redundant information is added to these coded signals in a channel encoder, from which conclusions can be drawn on the receiver side about the correctness of the received signal.

Convolutional codes are often used in the transmission systems to generate the redundant information during channel coding. When a convolutional code is used, code bits are generated from the information bits in such a way that, according to certain rules specified by so-called generator or code polynomials, each information bit is linked to certain other information-carrying bits arranged within the data bit stream. In terms of circuitry, these rules specified by the code polynomials are usually implemented by means of a shift register with several delays connected in series and a binary addition of the bits located at the inputs or outputs of the delays at a particular point in time. The number of delays in the shift register is referred to as the so-called “memory” m of the code.

Various links can be used to generate several different code bit streams from one primary information bit stream. The ratio of the number of input bit streams to the number of codes generated

Code bit streams is the so-called "rate" of the code. A code with rate 1/2 accordingly generates two code bit streams v ₁ ⁽¹⁾ v ₂ ^<1) v ₃ ⁽¹⁾ from an information bit stream uι u ₂ u ₃ ... . and Vι ^<2) v ₂ ⁽²⁾ v ₃ ⁽²⁾ ... These code bit streams become a common bit stream in the form V ^{1) v _\ ^(2> v ₂ ⁽¹⁾ v ₂ ⁽²⁾ . .. which is then transmitted over the channel, two code bits are output in the transmitted bit stream for each information bit of the incoming bit stream.

When using convolutional codes in transmission systems, for various reasons, e.g. B. the synchronization or the disruptive influence of decoding delay, only finite long code sequences are used. This means that it is necessary to split the data bit stream into individual data blocks for coding. The simplest way to do this is to abort the coding after a certain number of time cycles without further precautions. However, the information bits in the end area of the code blocks generated in this way are then significantly less well protected against transmission errors than information bits at other positions, e.g. B. at the beginning or in the middle of a block of code.

In order to avoid this effect, the code is terminated. The aim of the termination is that the shift register of the encoder is in a known state after the transmission of a block, for example is reset to "0".

In a frequently used method for decoding a transmitted code, a so-called "trellis diagram" is run along a path using a Viterbi algorithm. A trellis diagram of a convolutional code consists of branches (state transitions) and nodes, with several branches at each node A node represents a state of the memory of the convolutional code, and the termination of the code at the end of a data block has the advantage that the path in the corresponding trellis diagram ends in a previously known state.

In order to achieve such a termination, with a normal convolutional code without a feedback, ie without a feedback branch, a number of so-called tail bits "0" corresponding to the memory of the code can simply be appended. After m clocks, each is at the exit of each Delays the state "0" so that the entire shift register is in the "0" state and thus the convolutional code is terminated in the "0" state. In addition, there are also recursive convolutional codes in which a certain information bit is encoded not only with information bits preceding it in the data bit stream, but also with bits behind it, in that bits present between the delays of the shift register are fed back to previous positions of the shift register. Such recursive convolutional codes are particularly safe in poor channel conditions. They are therefore largely used in the existing mobile radio standards.

However, the simple way of terminating the code just described is not possible due to the feedback. Instead, the attached tail bits have to go through a special feedback branch in a recursive process

Encoders themselves are generated in order to terminate the shift register or the code to “0”. The tail bits generated in this way are therefore not only dependent on the final state of the shift register but also on the code itself and are therefore not known before the encoding of the entire information bits.

A schematic block diagram of an encoder working with a recursive convolutional code, which is terminated at the end of each data block, is shown in FIG. This is an example of a classic method for terminating a so-called RSC (recursive systematic convolutional) code. Such RSC codes are already used in GSM AMR channel coding and also as component codes in turbo codes.

As can be seen from the diagram, the associated shift register has three delays D1, D2, D3. The code therefore has a memory of m = 3. The information bit stream ui u ₂ u ₃ ... arriving at input 1 is here passed to the encoder via switch 2 and looped through line 3 to a first output 14 without change. The first code bit stream v * _. ⁽¹⁾ v ₂ ^αι v ₃ ⁽¹⁾ ... therefore corresponds to that incoming information bit stream ui u ₂ u ₃ .... The code is thus a so-called "systematic code" since the incoming information bits are contained identically in the outgoing data bit stream.

Furthermore, the incoming bit stream uι u ₂ u ₃ ... is passed through the shift register consisting of the three delays Dl, D2, D3, the respective bit at the input and output of each delay Dl, D2, D3 via the data paths 5, 6, 7, 8, 9 is tapped and fed to the binary adders 10, 11, 12, whereby the code is generated according to the desired rule. Finally, at the output 13 of the binary adder 11, the code ι ⁽²⁾ v ₂ ^<2) v ₃ ⁽²⁾ ... encoded according to the specified rule is present. To generate a single output code, the individual bits of the two code bit streams Ui u ₂ L_ ₃ ... and Vι ^<2) v ₂ ⁽²⁾ v ₃ ^(2> ... are placed one after the other in a series converter 16 as desired.

Since two outgoing bits are generated for each incoming bit in the RSC code shown, the code has a rate 1/2. The generator polynomials technically implemented in the form shown are [1, n (D) / d (D)], with d {D) = l + lf + D ³ and n {D) = 1 + D + D ³ . In order to terminate the code or the shift register to the "0" state, the

Information bits ui u ₂ u ₃ ... of a data block then send the bits ti, t ₂ , t ₃ via the line 15 shown in dashed lines and the correspondingly flipped switch 2 to the input of the encoder. Since these tail bits t _x , t ₂ , t ₃ are decoupled from the feedback line before the input of the first binary adder 12 and thus identical bits are present at both inputs of the binary encoder 12 at all times, there is inevitably always a “0” at the output ". Consequently, the shift register changes to the" 0 "state in = 3 cycles. The additional code bits to be transmitted corresponding to the termination are then the tail bits at the output 14 ti, t ₂ , t ₃ and at the output 13 the code bits generated from these bits.

A disadvantage of such a termination of recursive codes is that the information bits ui u ₂ u ₃ ... and the tail bits ti, t ₂ , t not only have to be specially treated during coding, but also require separate handling during decoding is.

Another method for the uniform protection of information bits within a data block is the so-called “tail-biting”, which is described, for example, in the article “Binary unequal error-protection block codes formed from convolutional codes by generalized tail-biting” by H. Ma in the journal “ IEEE Trans. Information Theory ", vol. 32, pp. 776-786, 1986. In this way, an" endless chain "is formed from the data blocks in a certain way, in which the start bits of a data block are used in a certain way as" tail bits " Although this method does not require additional tail bits, quite a lot of effort is required to encode and decode the information bits.

It is an object of the present invention to provide a simple and inexpensive alternative to this known prior art.

This object is achieved in that a predetermined number of known tail bits are appended to the end of the respective data block in order to protect the information-carrying bits located in the end area of the data blocks from the channel coding. Because the code is recursive, these tail bits generally do not lead to termination. This means that the code is normally not terminated, since at the end of the decoding the code is generally in an undefined state, depending on the respective code and on the respective information bits of the data block. It has surprisingly been found that this termination can be dispensed with by using known bits as tail bits, without this leading to an increase in the bit error rate for the bits of the data block located in the end region. In contrast, the alternative method according to the invention has the significant advantage that the information bits and the tail bits are treated uniformly by the encoder and decoder, so that it is no longer necessary to distinguish the information bits and tail bits in the encoder and decoder. Therefore, it is particularly possible, depending on the requirement, for. B. with regard to the performance of the transmission channel, also use more or fewer tail bits without changing the coding or decoding method. The method therefore allows a very quick adaptation to the different transmission options, for example to the available channel rate and the current transmission quality of the respective channel. In particular, in contrast to the known methods in which the code is terminated, a number of tail bits which are below the memory m of the code can also be used.

The procedure can be used for all recursive codes for which a termination was previously necessary. However, the method offers particular advantages when using a systematic code in which the information bits are also transmitted as code bits. Since the tail bits are known, they do not have to be transmitted, which can reduce the amount of data. These systematic recursive codes include e.g. B. the RSC or turbo code already mentioned.

Furthermore, it is also expedient to link the respective code with puncturing. Here, the code bits are punctured after the proposed method has been applied. In conjunction with the proposed method, decoding methods such as source-controlled channel decoding can be used, the maximum (absolute) apriori knowledge, for example the log-likelihood ratio in the case of the so-called apri-viterbi algorithm, being set on the reception side for the known bits ,

A corresponding device for carrying out the method has a channel encoder for recursively encoding a data bit stream divided into data blocks and means for appending a predetermined number of known tail bits to the end of the respective data block.

To decode a data bit stream transmitted using the method according to the invention, a device is used which preferably has source-controlled channel decoding, in particular for executing an Apri-Viterbi algorithm or MAP algorithm, and a storage device with a database for the values of the log-likelihood ratio for has the known tail bits.

Of course, a common coding and decoding device can also be used, which can be used both at the transmitter and at the receiver. Such a combination device preferably also has a source-controlled channel decoder, in particular for executing an Apri-Viterbi algorithm or MAP algorithm.

The proposed method and devices for the error-protected transmission of the source signals mentioned at the outset, such as voice, audio or video signals, are of particular practical importance. They are therefore particularly suitable for use in mobile radio systems.

The invention is explained in more detail below using an exemplary embodiment with reference to the accompanying drawings explained. The features shown can be essential to the invention not only in the combinations mentioned, but also individually or in other combinations. Furthermore, it is expressly pointed out that features which are only shown in connection with the method according to the invention can also be essential to the invention with regard to the devices for carrying out the method and vice versa. Show it:

Figure 1 is a schematic block diagram of a RSC code according to the classical, in the prior ^'art known methods with a termination of the code by the

tail bits;

Figure 2 is a schematic block diagram of the same RSC as in Figure 1, but when using the inventive method;

FIG. 3 shows an example for the encoding and decoding of a block of N information bits X _x with three tail bits “0”;

Figure 4 is a diagram of the simulation results for an RSC code with memory m = 8 and rate = 1/2 over an AWGΝ channel with a signal-to-oise ratio of -1 dB.

Since the method according to the invention has a main area of application for the transmission of signals in mobile radio systems, such a transmission is assumed in the following example. However, it is expressly pointed out once again that the use of the method and the device is of course not limited to this area of application, but that in principle the method can be used in any transmission systems in which the data bit streams to be transmitted are coded by a recursive convolutional code become. Figure 2 shows an example of a relatively simple RSC code with a memory of m = 3 and rate 1/2. The generator polynomials used here are [1, n (D) / d {D)], with d {D) = l + rß + D ³ and n {D) = 1 + D + D ³ . Consequently, this is the same code that was already used as an exemplary embodiment in FIG. 1 to explain the classic method according to the prior art. Accordingly, the same reference numbers are used for the same function blocks and data links.

The essential difference according to the invention from the classic method is that at the end of a block of information-carrying bits Xi x ₂ ... x _N the tail bits ti, t ₂ , t _{3 are} not recursively generated, as shown in FIG. 1, via the feedback line 15 are, but a predetermined number of "0" bits are added (for example via switch 2) as known tail bits. These "0" bits then pass through the shift register as end values and, unlike the tail bits generated in the conventional recursive method, lead t ₂ t ₃ ^x usually not to terminate the shift register. This means that the state of the shift register, and thus also a possible random "0" setting of the register, is not known after the coding of a data block has ended.

Since this is a systematic code, the incoming information bits xi x ₂ • • • _{N are} forwarded via the data line 3 to the output 14 of the encoder and there by means of the in-series converter 16 with the code bits zi at the output 13 z ₂ ... z _N z _{N +} ι z _{N +3} z _{N +3} interleaved into a common output bit stream, which is transmitted over the channel.

The bits of a data block immediately before channel coding are shown again in Figure 3 (a). Figure 3 (c) also shows the transmitted by this encoder and on Channel decoder received data x ^* ₁₇ z (i = 1, 2.3 ... N + 1), the marked three systematic code bits "0", resulting from the three previously known tail bits "0", not being transmitted, since these Values set apriori anyway.

Since the known systematic code bits do not have to be transmitted, the data rate is lower with the same number of tail bits than with the classic method of terminating the code. In general it applies that in the case of classic termination when using a code with the rate 1 / r and the memory m, r-m code bits must also be transmitted. For the proposed method with K known tail bits, on the other hand, only additional (r-1) 'K code bits have to be transmitted. When using a systematic code with rate 1/2, twice as many tail bits can be appended to a data block as with the classic method in order to arrive at the same overall data rate to be transmitted.

The following methods are possible for decoding, for example, in which the prior knowledge about the known tail bits can be used.

In the case of a normal Viterbi algorithm, e.g. B. the prior knowledge when selecting the possible paths in

Trellis diagram can be used. This means that paths with which the known tail bits are incorrectly decoded are discarded.

When using an Apri Viterbi algorithm, as described for example in DE 42 24 214 C2, or a similar algorithm such. For example, the MAP (maximum a posteriori probability) decoding algorithm, the a priori "L values" (values of the log-likelihood ratio) for the known tail bits can be used as (possible) maximum values. For example, for a bit = "0" a value L = "+ ∞" and for a bit = "1" a value L = "- ° o" be set. FIG. 3 (b) shows the corresponding L values for the channel decoding of the bits shown in FIG. 3 (a) before the channel coding (including the known tail bits). If there is no prior knowledge, a value Li = 0 (i = 1 to N) is set for the information bits Xi x ₂ ... x _N.

FIG. 4 shows simulation results for an RSC code with a memory of m = 8 and a rate 1 / r = 1/2 when transmitting a block of 200 bits via a so-called AWGN (additive white Gaussian noise) channel. The channel has a signal-to-noise ratio E _s / N ₀ = -1 dB. The generator polynomials used are [1, n (D) / d (D)], with n (D) = 1 + D ² + D ³ + D ' ⁱ + D ⁸ and d (D) = I + D ¹ + D ² + D ³ + D ⁵ + D ¹ + D ⁸ .

The Apri Viterbi algorithm channel decoder according to DE 42 24 214 C2 was used for the decoding. The bit error rate BER is plotted in the diagram above the respective bit number, only the last 80 bits of the block being shown in each case. As can clearly be seen, it works

Termination of the block back to approximately the last 50 bits.

Curve IX corresponds to coding without the insertion of tail bits, i. H. without any termination. Here, the bit error rate at the end of a data block rises up to a factor of 50 compared to the average bit error rate in the middle area of the data block due to the abort. On the other hand, curve I shows that in the case of coding with classic termination, with m tail bits generated recursively in the encoder, the bit error rate in the end region of the error block can even be reduced below the average bit error rate.

An equally good result in reducing the bit error rate can be achieved by appending 16 known tail bits according to the inventive method (curve II). There these 16 tail bits are known and need not be transmitted, this addition of 16 tail bits results in exactly 16 additional code bits being transmitted. The number of additional data to be transmitted in this simulation for the protection of the information bits located at the end of the data block

Bits therefore corresponds exactly to the number of code bits {r-m, as already explained above), which are required when using m = 8 tail bits according to the classic method (i.e. 8 unknown tail bits as systematic code bits and 8 further code bits).

The further curves III to VIII lying between the two curves each originate from simulations during coding with the method according to the invention, with different numbers of known tail bits being appended. 14 tail bits were used for curve III, 12 tail bits for curve IV, 10 tail bits for curve V, 6 tail bits for curve VI, 4 tail bits for curve VII, and 2 tail bits for curve VIII. As these curves show, in principle 12 known tail bits (curve IV) are sufficient to protect the information bits at the end of the data block against transmission errors as well as the information bits at other positions.

It can thus be seen that in the present exemplary embodiment with r = 2 and m = 8 in the case of the classic one

16 code bits must always be terminated, whereas in the method according to the invention K = 12 code bits are sufficient for uniform protection of all information bits. The transmission rate can thus be reduced overall.

Another advantage is that a further reduction or increase in the number of tail bits used is possible without any problems during operation, so that a quick adaptation to the performance of the transmission channel can take place in each case.

Claims

claims

1. A method for error protection during the transmission of a data bit stream in a digital message transmission system, in which the channel coding is carried out

Data bit stream is coded block by block using a recursive convolutional code, characterized in that, in order to protect the information-carrying bits in the end region of a data block (xi, x ₂ , x ₃ , ..., x _N ), a predetermined number of previously known tail bits (tι , t ₂ , t ₃ ) at the end of the respective data block (x _x , x ₂ , x ₃ , ..., x _N ).

2. The method according to claim 1, characterized in that a systematic code is used in the channel coding.

3. The method according to claim 1 or 2, characterized in that a turbo code is used in the channel coding, which contains at least one recursive convolutional code as a component code.

4. The method according to any one of claims 1 to 3, characterized in that a decoding of the data bit stream as source-controlled channel decoding, in particular by means of an Apri-Viterbi algorithm or MAP algorithm, is carried out.

5. The method according to any one of claims 1 to 4, characterized in that the data bit stream comprises source signals.

6. The method according to any one of claims 1 to 5, characterized by the application in a mobile radio system.

7. Device for performing the method according to one of the preceding claims, characterized by a channel encoder for recursively encoding a data bit stream divided into data blocks and by means for appending a predetermined number of known tail bits (tι, t ₂ , t ₃ ) to the end of the respective data block (x _if x ₂ , x ₃ , ..., x _N ).

8. The device according to claim 7, characterized by a source-controlled channel decoder, in particular for executing an Apri-Viterbi algorithm or MAP algorithm.

9. Device for decoding a data bit stream transmitted with a method according to one of claims 1 to 6, characterized by a source-controlled channel decoder, in particular for executing an Apri-Viterbi algorithm or MAP algorithm and a storage device with a database for the values of the log Likelihood ratio for the known tail bits (ti, t ₂ , t ₃ ).