FR2703544A1 - Method of reducing the memory size and the time necessary for error correction in an asynchronous time-division network - Google Patents

Abstract

The method consists in attributing (1) Reed-Solomon error detector/corrector codes to each information symbol to be transmitted on the network, in performing (2) a diagonal interlacing of the Reed-Solomon coded information before they are transmitted in packets on the network, and, on reception, performing a diagonal de-interlacing (9) of the information contained in the packets in order to correct (10) the errors and the erasures in the symbols reproduced by means of the Reed-Solomon codes accompanying the reproduced symbols. Application: high-definition television and compatibles, high-fidelity sound, video telephony.

Description

Method for reducing memory size and time
necessary to correct errors in a
asynchronous time network
The present invention relates to a method of reducing the memory size and the time required to correct errors in an asynchronous time network. It applies in particular to the protection of distributive and interactive audiovisual services against bit errors and cell losses using asynchronous time networks.

In these networks1 the digital information is transmitted discontinuously in the form of packets, called "information cells" or "ATM cells", where ATM is the abbreviation of "Asynchronous Transfer".
"Mode" in the Anglo-Saxon language: the transmission mode consists, instead of transmitting byte by byte, to constitute in the sending terminal equipment a packet of bits as and when the digital data is created, and wait until you have enough information to build a complete information cell to send it to the network.

 In digital television, this mode of transmission is governed by CCITT recommendations 1.361 and 1.363 which respectively define the ATM adaptation layer, four classes of services differentiated by their rates, their connection modes and the temporal relationship levels between transmitters and receivers. as well as four types of adaptation that can be used in each class of service. Audiovisual transmissions are characterized by widely varying information rates: 64 Kbitls for the low quality videophone, 2 Mbitls for the video surveillance and the high quality videophone, 8 Mbitls at 20 Mbitls for the standard television and 30 to 140 This latter bit rate, combined with the real-time constraints inherent in the transmission of the video signal prohibit the retransmission of the information when errors transmitted in the transmitted bitstream are detected.

 This problem can still, and in many cases, be solved by masking erroneous information. However, the multiplicity of temporal or spatial masking techniques used in video and the difficulty of making masking on the audio signal are such that the sensitivity admitted to transmission errors, in terms of residual errors is very variable and commonly evolves from one to another. error every 10 minutes to an error every 2 hours. Naturally the solution to these problems is all the more difficult to find because of the perfectly random nature of these errors and the fact that they can be in packets or bursts.

Although it is in principle possible to guarantee error rates of 10-12 for HDTV, 10-11 for standard television and 10-10 for video surveillance at the user-network interface by implementing different error detection and correction codes such as Reed codes
Solomon for example, which have a very large correction power and an extreme simplicity of implementation, these codes do not however make it possible to remedy the problem raised by the loss of cells because they, by their format which conventionally comprises 48 bytes information, require error-correcting codes that are too powerful and too redundant to allow intracell corrections.

 CCITT Recommendation 1.363 proposes a method to overcome the aforementioned drawbacks. This method is based on the use of a Reed-Solomon code (128, 124) on GF (256) and an order interleaver 47 and is dedicated to the distribution of video signals with bit rates greater than or equal to 10 bits.

This method, however, has the following drawbacks:
a processing time proportional to the size of the interleaving memory and inversely proportional to the rate of the service to which it applies
At = (Memory size interleaver)
service flow which proves poorly suited to low-speed audiovisual services, especially if they are interactive (video telephony)
- the need to use a working memory twice the size of the interleaving matrix (flip-flop operation)
- misuse of Reed code correction capabilities
Solomon due to the fixed choice of an interleaving depth equal to the size of the useful information block of the ATM cell, namely 47.

 The object of the invention is to overcome the aforementioned drawbacks.

For this purpose, the subject of the invention is a correction method against bit errors and cell losses for distributive and interactive audiovisual services in an asynchronous time network of protecting each information symbol to be transmitted over the network by a code Reed-Solomon error-correcting detector, interleaving Reed-Solomon coded information prior to transmission in packets or cells on the network, performing upon de-interleaving the information contained in the packets or cells and correcting errors and deletions in the symbols returned by means of the Reed-Solomon code, characterized in that the interleaving consists of memorizing line by line the information by words of Reed
Solomon of N symbols in a capacity interleaving memory C1 (respectively C2) \ and read this diagonal information diagonally to form the SAR-SDU fields of the ATM cells.

 The originality of the invention consists in the read-write technique of the interleaving matrix which makes it possible to divide by 2 the processing time of the data, with code length N and interleaving depth R equal to and by about 4 memory size required for the interleaving mechanism.

Other characteristics and advantages of the invention will appear below in the light of the following description made with reference to the appended drawings which represent:
- Figure 1 the different steps of the method according to the invention in the form of a flowchart.

 FIG. 2 is an organization diagram of an interleaving device enabling the method according to FIG. 1 to be implemented.

 - Figure 3 a diagram illustrating with a simple example the interleaving method.

 FIG. 4 is a diagram illustrating with a simple example the deinterlacing method.

 FIG. 5 is a diagram describing the organization of the memory space of the interlacing devices and the deinterlacing of FIGS. 3 and 4 as well as the addressing sequences of these memories.

 FIG. 6 is a diagram comparing the processing delays in the case of conventional interleaving and in that of interleaving as described in FIGS. 3 and 4.

 FIG. 7 an ATM cell segmentation format, for type 1 adaptation according to CCITT Recommendation 1.363.

The method of protection against ATM cell errors and losses according to the invention is illustrated by steps 1 to 10 of FIG. 1. It consists of step 1 to convert the representative symbols or bytes of the video signal into codewords. Reed-Solomon in the manner described, for example, on page 207 of G. Cullmann's book entitled "Detector Codes and Error Correctors" published by DUNOD 1967 in the "Introduction to the Novelties of Science" series or in the manner described. pages 304-308 of the book of
Bernard Sklar entitled "DIGITAL COMMUNICATIONS Fondamentals and
Applications published by Prentice Hall 1988. The coded information obtained is code words of length N in number of symbols and containing K symbols of useful information.The code is denoted C (N, K, D) where D is the distance of Hamming.

This information is interleaved in step 2 by means of an interleaving device arranged in accordance with the block diagram of FIG. 2. According to this device, the words of N symbols are stored horizontally in the interleaving memory of a device. capacity equal to:
C1 = aR (R + 1) + b (b + 1)
2 2 with N = aR + b and b <R, where R is the interleaving depth and where a and b are respectively the quotient and the remainder of the Euclidean division of N by R. The SAR-SDU information fields ATM cells are obtained by diagonally reading the interleaving memory as described in the example of Figure 3.

 The calculation of the insertion of sequence numbers which is carried out in step 3 leads to a segmentation of the cells (SAR-PDU) in a format shown in FIG. 7 in accordance with the CCITT recommendation 1.363.

This format has 40 bits of Header header, 384 bits of SAR-PDU including 4 SN bits of sequence SNP 4-bit SN field protection against errors and a field of 376 bits of information forming a cell 5 + 48 = 53 bytes. The sequence numbering allows the tracking of the cells and a lost cell can thus be detected at the receiver and replaced by a "ghost" cell in the deinterleaver.

 In reception the unpacking of the cells is carried out in step 6. The cells are restored in the SAR-PDU format of FIG. 7 and a correction of the numbering of the sequences takes place in step 7 followed in step 8 of FIG. possible insertion of "ghost" packets or destruction of inserted packets.

De-interleaving of the cells then takes place in step 9 by means of a symmetrical deinterleaving device of the device of step 2 using a capacity deinterleave memory: C2 = aR (R + 1) + b (b-1)
2 2 + bR-2 and which consists of the diagonal writing of the SAR-SDU fields of the received cells followed by the horizontal reading of the reconstituted Reed Solomon codewords.

 These codes are then analyzed in step 10 to perform the error and erase corrections. Naturally, the deinterleaving operation of step 9 must be in synchronism with the insertion of the ghost packets and the destruction of the inserted packets of step 8.

 The method and the device of the invention which have just been described apply to very different code configurations.

 An example is shown in FIGS. 3 and 4 which makes it possible to understand the originality of the interlacing device. In this example the values of the various parameters are deliberately low, for the sake of simplicity of presentation of the method.

 If N is the length of the code, R the interleaving depth, L the length of the packets from the interleaving memory and intended to be transmitted, N = 10 has been selected here, R = 4 and L = 5.

 In the case of ATM networks, the length L of the SAR field SDU of the ATM cell is equal to 47 and N and R can be selected, for example, respectively equal to 235 and 24. K, the number of bytes used in the Reed-Solomon code and therefore D, Hamming distance of the Reed-Solomon code do not alter the principle of interleaving.

Remember that in the case where R <L (the cell is then spread over more than one interleaving depth), N can not be chosen by chance. Indeed, R x N must be multiple of L, therefore in ATM, N multiple of L since
R <L and L = 47 is a prime number.

Formally, in steady state:
the successive code words are denoted Mj = {i, j)} j = 1 1 to N
the successive cells to be transmitted are denoted Ckl, I = 1 to N / L and Ck = {k, l, m)} m = 1 to L
At the level of the interleaving memory, the operations are as follows:
At the encoder
* step 1: MI writing
Step 2: CII readings for I = 1 to N / L using the code words Mi for i = I-R + 1 to
* step 3: 1 = 1 + 1 and loop back to step 1.

At the decoder
* step 1: write the Cki for I = 1 to N / L
Step 2: Reading of MK-R + 1 using CkI cells for k = K
R + 1 to K
Step 3: K = K + 1 and loopback in step 1.

At the decoder, as the encoder to have the Ck = {(k, l, m)} as a function of M1 = {(i, j)}, and vice versa, we have the following relations (1):
j = (l-1) L + m, m <L + 1

 At the encoder we search for (i, j) which correspond to (k, l, m) to constitute the cells Ckl and to the decoder les (k11, m) which correspond to (i, j) to reconstruct the RS code words, Mid.

In our example, in steady state at the encoder:
we write M4 = {(4,1) (4,2) ... (4,10)}
read C41 {(4, 1, 1) ... (4, 1, 5)} and C42 = {(4, 2, 1) ... (4, 2, 5)}
we write M5, we read C51 and C52, etc.

pour C41: (4,1,1) = (4,1) pour C42: (4,2,1) = (3,6) (4,1,2)=(3,2) (4,2,2)=(2,7)
(4,1,3)=(2,3) (4,2,3)=(1,8)
(4,1,4) = (1,4) (4,2,4) = (4,9) (4,1,5)=(4,5) (4,2,5)=(3,10)
Les figures 4 et 5 illustrent ces calculs.To read C41 and C42 we use M1, M2, M3, M4. We then have according to equations (1) which become:
j = 5 (1-1) + m, L <6

for C41: (4,1,1) = (4,1) for C42: (4,2,1) = (3,6) (4,1,2) = (3,2) (4.2, 2) = (2.7)
(4,1,3) = (2,3) (4,2,3) = (1,8)
(4,1,4) = (1,4) (4,2,4) = (4,9) (4,1,5) = (4,5) (4,2,5) = (3, 10)
Figures 4 and 5 illustrate these calculations.

In practice, it suffices to choose an interleaving memory size equal to C1 (respectively C2) and a deinterleaving memory size equal to C2 (respectively C1), with C1 and C2 defined as before, or to the total between encoder and decoder a memory size equal to:
C = C1 + C2 = N (R + 1)
The usual values of N and R give an interleaving memory size equal to about half of that currently used.

Some examples are given in the table below.

<Tb>

<SEP> Code <SEP> N <SEP> R <SEP> a <SEP> b <SEP> Size <SEP> Memory <SEP> Size <SEP> Memory <SEP> Gain
<tb><SEP> traditional <SEP> method <SEP> new <SEP> method
<tb> RS (235,227) x24 <SEP> 235 <SEP> 24 <SEP> 9 <SEP> 19 <SEP> 11280 <SEP> 5875 <SEP> +1.92 <SEP>
<tb> RS (128,127) x47 <SEP> 128 <SEP> 47 <SEP> 22 <SEP> 34 <SEP> 12032 <SEP> 6144 <SEP> +1.96 <SEP>
<tb> RS (235,219) x6 <SEP> 235 <SEP> 6 <SEP> 39 <SEP> 1 <SEP> 2820 <SE> 1645 <SEP><SEP> .1,71 <SEP>
<Tb>
The size is given in bytes and corresponds to the size memory encoder + decoder.

 On the other hand, thanks to the principle of reading continuous writing it is no longer necessary to wait for the complete writing of the matrix before starting reading. The principle of "flip-flop" is no longer necessary with this dynamic management. The size of the interleaving and deinterleaving memory is divided again by 2.

Overall the gain is a division by 4 of the memory. Rigorously the memory size is divided by
R = 4 (R + 1)
Regarding the treatment time, the observation of Figures 3, 4 and 6 shows that it is divided by 2 compared to the conventional method. This processing time in the interleaving memories corresponds to the difference between the instant when the code word Mi is extracted from the deinterlacing and the moment when it was written in the interleaver.

 At = KR / D instead of At = 2KR / D in the classical case with D the rate of useful information before entering the Reed-Solomon encoder.

This method requires more complex dynamic memory management. In practice things happen in the following way, for example to the coder: one writes the code word Mi, one reads the cells Ci1, Ci2,
Ci N / L, then the code word Mi + 1 is written, then the cells Ci + 1.1, Cj + 1.2 ... Ci + 1, N / L, etc. are read.

au codeur et égal à

au décodeur, où E(a) est la partie entière de a. At the memory level, after reading the N / L cells, the next RS code word is written in the memory at the location left free by the cells read. By operating in this way it is easy to see that the memory addressing pointer for the read-write operations describes a sequence cyclically. This cycle is also very long (= R * PPCM (R, R-1, R-2, ..., 1)). The sequence of addresses is however not made at random. If we represent the memory in pages each corresponding to a column of our interleaver and in lines within each page corresponding to the cells of each column i:
In our example we have 10 pages which have respectively 1, 2, 3, 4, 1, 2, 3, 4, 1, 2, lines to the encoder and 4, 3, 2, 1, 4, 3, 2, 1, 4 and 3 lines to the decoder. In general, the page i has a number of lines equal to:

to the encoder and equal to

to the decoder, where E (a) is the integer part of a.

The read-write address sequence of the memory is thus the following:
* reading and writing of the line jj of the page i
* i = i + 1 modulo N, number of pages
* ji = jj + 1 modulo the number of lines in the page
FIG. 5 describes the organization of the memory and the sequencing of the addresses on the example treated in FIGS. 3 and 4.

Claims (8)

 A method of protecting against bit errors and cell losses in an asynchronous time network for audiovisual services consisting of protecting (1) each information symbol to be transmitted over the network by a Reed-Solomon standard error-correction code performing (2) interleaving the Reed-Solomon encoded information prior to being transmitted in packets or cells on the network, performing on the receiving end a deinterleaving (9) of the information contained in the packets or cells and correcting (10) the errors and deletions in the symbols returned by means of the Reed-Solomon code, characterized in that the interleaving consists in memorizing line by line the information by Reed-Solomon words of N symbols in a capacity interleaving memory C1 and to read this diagonal information diagonally to form the SAR-SDU fields of the ATM cells.  2. Method according to claim 1 characterized in that the deinterlacing is to diagonally diagonally store the fields SAR-SDU ATM cells in C2 deinterleave memory and read this information line by line to reform the Reed-Solomon codewords of method 2.  3. Method according to claims 1 and 2 characterized in that C1 + C2 = N (R + 1) with C1 = aR (R + 1) b (b + 1) and C2 = aR (R + 1) + bR- b (b-1) where N = aR + b and b <R  2 2 2 2  where R is the interleaving depth and where a and b are respectively the quotient and the remainder of the Euclidean division of N by R.  4. Method according to any one of claims 1 to 3 characterized in that the Reed-Solomon codes are RS (N, K, D) code length codes (N), useful information (K) and distance from Hamming (D).  5. Method according to any one of claims 1 to 4 characterized in that it consists in using the correction capacity erasures.  6. Method according to claims 1 to 5 characterized in that it consists in choosing an interleaving depth R less than or equal to the length L = 47 of the SAR-SDU field of the ATM cell.  7. Method according to any one of claims 1 to 6 characterized in that each information symbol has the size of a byte.  8. Method according to any one of claims 1 to 7 characterized in that it consists in inserting ghost cells in a deinterleaving array by marking each byte to be interpreted as an erase by Reed-Solomon decoding.