WO1997038549A1

WO1997038549A1 - Method and apparatus for forward error correction of transmitted digital signals in networks

Info

Publication number: WO1997038549A1
Application number: PCT/EP1997/001500
Authority: WO
Inventors: Georg Carle; Gehard KRÜGER
Original assignee: Universität Karlsruhe (Th)
Priority date: 1996-04-05
Filing date: 1997-03-25
Publication date: 1997-10-16

Abstract

A cell-based Forward Error Correction (FEC) scheme which is located in the ATM Adaptation Layer (AAL) is designed for the Service Specific Convergence Sublayer of the standardized AAL Type 5. The FEC scheme encodes cell-level protocol information, encodes frame-level protocol information into Protocol Data Units of the Common Part Convergence Sublayer of AAL5, and encodes and decodes the required redundant information by a Double Matrix Interleaving XOR scheme. Applications are networks where reliable high performance point-to-point and point-to-multipoint communication services are required, e.g. broadband integrated services digital networks, distributed multimedia systems, computer-supported cooperative work applications, and virtual shared memory systems.

Description

Method and apparatus for forward error correction of transmitted digital signals in networks

Field of Invention

Distributed computing, distributed multimedia applications as well as advanced applications for computer-supported cooper¬ ative work depend on networks with high performance in terms of throughput, delay and reliability. Frequently, ATM networks are capable of providing adequate throughput and delay for these applications. However, bit errors and the loss of ATM cells may cause significant problems. The performance of ATM networks degrades very fast with a growing cell loss or bit error rate. There are both technical and economical reasons that may lead to non-negligible errors in ATM networks.

Most error control mechanisms that are currently used to correct errors in ATM networks are frame-based schemes. They are designed for very low cell loss probability and do not handle non-negligible cell loss well. In frame-based schemes with significant bit error or cell loss rate the achievable throughput for reliable services also degrades for a growing frame size. If a reliable multicast service is to be provided, the system throughput will be seriously degraded for a growing number of receivers. A cell-based Forward Error Correction (FEC) scheme which is located in the ATM Adaptation Layer (AAL) may improve the end- to-end Quality of Service (QoS) in all these cases. FEC schemes may be very beneficial in cases where retransmissions by a higher layer protocol lead to low QoS and an inefficient use of network resources and in cases in which delay requirements make the use of retransmissions impossible. An FEC scheme can be used for real-time services, and for reliable non-real-time service. The present invention comprises a novel FEC scheme which is designed for the Service Specific Convergence Sublayer (SSCS) of the standardized AAL Type 5 (AAL5, [1.363]) . This FEC scheme presented here supports the scaleable provision of reliable pomt-to-pomt and point-to-multipomt services. Therefore, it uses a method and an apparatus to encode cell-level protocol information. Additionally, the presented FEC scheme uses a method and an apparatus to encode frame-level protocol inform¬ ation into Protocol Data Units (PDUs) of the Common Part Convergence Sublayer (CPCS) of AAL5. Furthermore, the presented FEC scheme uses a method and an apparatus to encode and decode the required redundant information. By use of the present mvention the amount of redundant information can be adjusted according to the application requirements in a specific communication scenario.

Background of the Invention

The present invention relates to a method and an apparatus for data communication and, more particularly, to protocol mechan- isms for error control in a particular type of networks: fast packet switched networks using short packets of fixed size called cells. The Broadband Integrated Services Digital Network (B-ISDN) under standardization by the International Telecom^¬ munications Union (ITU, [1.121] ) is the most popular network of this network type. B-ISDN is based on the Asynchronous Transfer

Mode (ATM) , utilizing cells with a total length of 53 bytes, of which 5 bytes are the header and 48 bytes are the cell payload.

Upcoming applications, e.g. distributed multimedia systems, computer-supported cooperative work (CSC ) applications, and virtual shared memory systems, require reliable high per¬ formance pomt-to-point and point-to-multipoint communication services. Quality of service (QoS) issues of importance are not only throughput, delay, and delay ηitter but also differences of αelay and reliability within the group. A key problem that must be solved to provide a reliable multipoint service is the recovery from cell losses due to congestion in the switches. The probability for cell loss may vary over a wide range depending on the strategy for usage parameter control (UPC) and call admission control which is applied. So far it is an open question how low cell loss rates can be guaranteed for bursty traffic, in particular for bursty multicast traffic, while using network resources efficiently. Cell losses caused by buffer overflows do not occur randomly distributed but rather show a highly correlated characteristic [OhKi91] . If a reliable service in ATM networks is based on traditional transport protocols like TCP, severe performance degradation may be observed [Rom93] .

In addition to cell loss due to congestion in several scenarios the service quality may also be affected by bit errors. One example for networks with non-negligible bit error rates are customer premises networks with low-cost physical layers such as unshielded twisted pair (UTP) cables. The intro¬ duction of wireless ATM networks will give another example for a physical layer that may cause significant bit error rates. Bit errors in the ATM header that cannot be recovered by the CRC of the ATM Header Error Control (HEC) may lead to cell loss or cell misdelivery. This makes them subject to AAL error control procedures. Bit errors in the ATM payload are also subject to AAL error control.

For the provision of a reliable multipoint service the probability for losses increases for a growing number of receivers. However, there are no convincing concepts for reliable high-performance group communications in ATM-networks so far. Therefore, the provision of reliable group communi¬ cations requires the development of efficient protocols and of communication systems that achieve high performance even under conditions with high cell losses. Description of the related art

Error control for ATM networks based on FEC has been studied previously. Shacham and McKenny [Sha90] study the performance of FEC schemes for ATM networks that are based on the generation of 1 or 2 redundant cells in a block of k cells. They do not give any details on protocol functions such as how to distinguish different cells for error recovery.

Ohta and Kitami [0hKi91] study the performance of an FEC scheme that is applied on Virtual Paths. In this scheme the payload of cells remains untouched. They arrange cells into matrices generating one redundant cell per row and one red¬ undant cell per column. In this scheme the redundant cells per row are used for error detection. The present invention differs from the NTT scheme by the fact that it uses Cell Sequence Numbers for error detection while the NTT scheme uses redundant cells per row instead as well as by the fact that the NTT scheme uses only a single matrix for cell loss recovery.

McAuley [McA90] describes an FEC scheme which is applicable for any packets size, including AAL frames, and ATM cells. This scheme is based the generation of a fixed number of redundancy packets for a given number of user data packets applying a modified Reed-Solomon burst erasure code(RSE) . The code is able to recover any h cells lost out of k+h cells. For this FEC scheme, Biersack [Bie93] presents a performance evaluation. Zhang and Sarkies [ZhSa91] have studied the performance of a two-dimensional FEC scheme which is applicable for virtual paths. In this scheme Reed-Solomon codes are used for detection of bit errors and for recovery of lost cells. The performance of this scheme under bursty cell losses is investigated in [ZhSa91] . Zhang and Sarkies claim that this scheme could be used for either virtual paths or virtual channels. Therefore, they do not specify whether cell sequence numbers are to be placed into the cell header or into AAL fields. The way in which Reed-Solomon-codes are used for the recovery of lost cells in this scheme is equivalent to the Bellcore scheme. The present invention differs from the scheme reported in [ZhSa91] and the Bellcore scheme by the fact that it uses a Matrix-XOR- Coding-Scheme which has a slightly lower error correction capability for identical amount of redundancy but is signi¬ ficantly simpler to be implemented on a processor.

When is an AAL level FEC scheme beneficial? Using an FEC scheme for reliable data communication in ATM-Networks allows improvements of the following QoS aspects: (1) throughput;

(2) latency;

(3) data-transmission reliability, i.e. robustness; and

(4) scaleability for large distances and a high number of receivers . Using an FEC scheme for reliable data communication in ATM- Networks also allows to decrease the following costs:

(1) data transmission costs which are determined by bandwidth, duration of call, and by the number of transmitted cells; and

(2) protocol processing costs. The transmission performance in terms of throughput and latency are degraded due to cell loss and bit error in the ATM networks. The actually provided quality of cell loss ratio and bit error ratio depend on the following considerations:

(1) The physical layer that is used below AAL/ATM. Some media may have a larger bit error ratio or may have a bit error characteristic of higher burstiness than the physical layers currently specified. In the case of wireless Local Area Networks (LANs) for example tolerating bit error rates higher than currently specified for the physical layer allows to reduce implementation costs significantly.

(2) The type of service that is provided by the ATM network. In ATM networks QoS parameters associated with the cell loss ratio may be negotiable and depend also on the service type. For some service classes, e.g. UBR, a cell loss rate suffi¬ ciently low to achieve a satisfying application level QoS may not be expected.

(3) The congestion status of the network. Certain conditions exist for the occurrence of congestion in ATM networks which lead to buffer overflows and cell losses. For some applica¬ tions, e.g. for mission critical applications, it s desirable to limit the degradation of service quality in cases of congestion. A reduction of the effective bit error ratio and cell loss ratio for the upper layer process, e.g. IP, may improve the overall service quality significantly. This is especially true for the application that requires the provision of a highly reliable service quality. As shown in [Bιe93] an FEC scheme may be used advantage¬ ously for multiplexing of VCs with different QoS requirements. If data streams with redundancy and data streams without red¬ undancy are multiplexed, different QoS requirements may be satisfied even for a switch that does not distinguish the data streams.

The current AAL types for data transfer, i.e. AAL5 and AAL3/4 [1.363], do not provide an error correction scheme. Rather, they only perform error detection and rely on the error correction capability of the transport layer protocol, e.g. TCP. When error correction is performed by TCP or other transport layer protocols, complete AAL PDUs are discarded if a bit error or a cell loss has been detected. The error correc¬ tion capability of the transport layer is typically based on retransmissions from the sender, for example TCP uses a variant of the go-back-N retransmissions scheme.

In general, the transport layer or network layer, e.g. IP, does not have a cell-based error correction capability. There¬ fore, complete packets must be retransmitted even if received packets have only a single bit error. In go-back-N schemes not only the packet in error but a full -ransmission window has to be retransmitted. For example, a data-unit of 64 Kbytes, i.e. the maximum αata-unit size of AAL5, the probaoility that the received data-unit has any bit error is expected to be about 5x10^" , for uncorrelated errors with a bit error ratio (BER) of 10^~Q. For packets of 9180 Byte, i.e. the default Maximum Transmission Unit (MTU) size for an IP packet defined in [RFC1626] , the resulting packet error rate is approximately 1x10 . Moreover, in the reliable multicast service, the expected packet error probability due to bit errors will be linearly increased for a growing number of receivers. This means that for a large scale reliable multicast service, e.g. interactive games over the Internet, it is difficult to provide a service with satisfying throughput and latency performance without an FEC scheme.

In the ATM layer, the issue of cell loss due to buffer overflow must be considered. Since transmission service data units (TSDUs, e.g. IP packets) will be segmented into multiple cells, a complete packet is assumed to be an error packet even when only one cell within the received packet is missing. For example, the default MTU size defined in [RFC1626] is 9180 Byte corresponding to approximately 160 cells. The maximum size of an AAL5 CPCS-PDU corresponds to approximately 1330 cells (= 64 Kbytes) . For example, for a frame size of 64 KBytes the probability that the received frame experienced cell loss is approximately 1.3x10^" when the cell loss ratio (CLR) is 10^" . For 9180 Bytes it is about 2xl0^~4. Moreover, in the reliable mult-cast service the expected packet error probability due to cell -oss will increase linearly according to the number of receivers. This means that for the large scale reliable multicast service high throughput and low latency together with an efficient use of network resources cannot be achieved without an FEC scheme. Advantages for the use of FEC in ATM networks have previously been investigated in [Sha90], [0ht91] , [Bιe93] , and [Ayan93] .

Since the basic data-unit in ATM networks, i.e. a cell, is typically very small compared to application level data-units, e.g. IP packets in the Internet protocol suite, one data-unit in application level corresponds to dozens, hundreds or sometimes thousands of cells.

If reliable services are to be provided in ATM networks using packet-based error control or frame-based error control schemes, respectively, the loss of a single cell in a packet requires the recovery of a complete packet. Therefore, through¬ put is severely limited for packet-based error control schemes either when the cell loss ratio is high or the packet length is large, the peer-to-peer throughput is severely limited.

The loss rate of higher layer packets, e.g. TCP packets, may grow linearly for a growing number of cells composing a packet as shown in [Rom93] . The response time of a transport protocol that provides a reliable service in ATM networks using packet-level error control is also degrading very fast for growing cell loss.

The performance degradation under the influence of cell loss or bit errors is even more serious for a large scale reliable multicast service. Examples for applications that require a large scale reliable multicast service are inter¬ active games, conferences, and video on demand. In some en¬ vironments due to the lack of error correction capability in lower layer services it cannot be expected that the bit error ratio is sufficiently small.

Summary of the invention

The present invention which describes a new FEC scheme for AAL5 comprises the following parts. The first part is used for encoding and decoding of redundant information. This is the FEC method and apparatus of the present invention which will be referred to as Double Matrix Interleaving XOR scheme. This scheme allows for the efficient generation of redundant inform¬ ation at the transmitter and for efficient error recovery at the receiver. The second part which will be referred to as protocol encoding method and apparatus, allows for distingui¬ shing of different cells, for decoding of protocol information, and for error detection at the receiver.

The present invention allows to exploit the potential benefits of an AAL level forward error correction (FEC) scheme for reliable data transmission services in ATM networks. The following conditions for successful integration of FEC schemes into the Service Specific Convergence Sublayer of AAL5 are fulfilled by use of the present invention: (1) Compatibility with existing AAL type 5 is important.

(2) AAL-SDUs of variable length, e.g. IP packets, should be supported.

(3) It should be possible to adjust and negotiate the parameters of the FEC method and apparatus. The parameter of the FEC would be the size and the structure of the appended redundant information. The purpose of this requirement is to give a possibility for optimization of the transmission efficiency, i.e. to minimize redundant data transmission and to achieve a media/service independence. (4) It should be possible to segment large AAL-SDUs into several smaller CPCS-PDUs and to protect the CPCS-PDUs individ¬ ually by the FEC scheme. Pipelining should be supported to reduce the peer-to-peer data delivery latency. (5) Tr.e processing costs of the FEC scheme in lossless state, i.e. without cell loss or bit error, should be as small as poss ole. Without loss or error in the CPCS-PDUs carrying the user αata no FEC decoding should be required. Similarly, reordering of data should be avoided in the lossless state. O 97/38549 PCΪ7EP97/01500

10

(6) Typically, frames will be transmitted without ceil losses or Dit errors with a very high probability . This case has to be optimized for low processmg costs and low overhead.

(7) To simplify CRC and FEC calculation it is preferable to send CRC and FEC cooe after the corresponding data is sent.

(8) The numbering space of the cell sequence numbers has to be selected according to the capabilities of the FEC and other error control mechanisms. If CRC-32 is used for error detection and if no cell-based retransmissions are planned, the cell sequence numbering space can be selected accordmg to the FEC capabilities. This means that an FEC scheme which is able to detect G, e.g.G = 4, missing cells should use cell sequence numbers with a numbering space of more than G. Sequence numbers of SNS = [ld(G+D ] bit, in this example 3 bit, whereby [ld(G+l) ] equals the Ceiling of the logarithm dualis of (G+1) , would be sufficient. For an FEC scheme with up to 16 redundancy cells sequence numbers of at least 5 bit are appropriate. A sequence number space which is slightly larger than the maximum number of cells that can be corrected is useful m order to m- crease the probability of a correct cell loss detection.

While substantial parts of the invention are for general application, some aspects of the present invention do partic^¬ ularly exploit special properties of ATM networks. In partic^¬ ular, special properties of the data formats of the ATM header, where one bit called AUU is used, and of the trailer of AAL5 frames are exploited.

With the ATM cell header 3 bits are used for identific^¬ ation of the payload type. A particular bit of these 3 bits, called the AUU-bit, allows to distinguish two types of cells. For cells carrying user data the AUU-bit can be used to mark the last cell of a sequence of cells thus defining a so-called Adaptation Layer Frame (AAL-frame) .

Within the AAL5 trailer as standardized in [ .363] two bytes, the AAL5 ϋser-to-User-Indication (AAL5-UU) byte, and tne Common Part Indicator (CPI) byte are available, which can be used for frame-level protocol information by protocols residing above the AAL5 Common Part Convergence Sublayer (AAL5-CPCS) . The present invention is designed to use the available two bytes of the AAL5 trailer efficiently.

The FEC method and apparatus for encoding and decoding of redundant data could be based on the well known Reed-Solomon- Codes [McAu90], [Fel93] . Reed-Solomon Codes have the advantage of a nigh error correcting capabilities but the disadvantage of a fairly high implementation complexity, i.e. high processing costs. Therefore, the present invention comprises a novel method and an apparatus for an FEC scheme which is referred to as the Double Matrix Interleaving XOR scheme. This scheme allows simple implementation both in hardware and in software. The FEC scheme of the present invention uses either one or two redundancy blocks. If two redundancy blocks are used, the first block is called the primary redundancy block and the second block is called the secondary redundancy block. If only a single redundancy block is used, this block is called the redundancy block. Each redundancy block consists of a number of redundancy cells which are calculated from the data cells of an ATM Adaptation Layer Service Data Unit (AAL-SDU) . For the evaluation of the redundancy cells of a redundancy block the data cells are arranged into a matrix. When two redundancy blocks are used the matrices used for encodmg the redundancy are different.

The primary redundancy block consists of h_p redundancy cells. These redundancy cells are generated using all data cells of a frame, XOR-operations, and matrix interleaving. For a consisting of D_x D_m data cells, where the number of data cells m can be expressed by the equation = (L-h_D) + r cells, h_p redundancy cells (R_j R_h) are generated according to the following calculation: R, = Z⁾, θ n_t+, Θ _Jtl ©/>_{( I+1 1}

(ι=i r)

.sing the redundancy cells of the primary redundancy block, a maximum of h_p lost cells can be recovered. However, this maximum can only be recovered if not more than a single cell per column is lost.

Figure 1 shows a suitable FEC encoder for the h cells calculated over the columns. Figure 2 depicts a suitable decoder. Please note that m all Figures D(ι) is used as a synonym of D_x. Both encoder and decoder are based on simple XOR-functions described by Φ.

For decoding an arbitrary lost cell per column can be decoded by performing an XOR-operation on all but the missing cell of the column together with the redundancy cell of this column. In case that more than a single cell per column is lost a secondary redundancy block can be used to regenerate the original data cells. For generation of the redundancy cells of the secondary redundancy block the data cells D_λ D_m are arranged into a Matrix with h_s columns.

The matrix parameters L_s and r_s of the secondary redundancy block can be expressed by the equation m = (L_s-h_s) + r_s. Usmg this matrix additional redundancy cells are calculated either per row or per column. If the redundancy cells are calculated per column, it is possible to generate h_s redundancy cells. If the redundancy cells are calculated per row, it is possible to generate L_s+1 redundancy cells. However, the recovery capabil¬ ity of the present invention is only reduced by a sma_^l amount if less than h_s redundancy cells or L_s+1 cells, respectively, are generated and transmitted to the receiver.

If the number of columns in the primary redundancy block is odd, a simple and efficient dimensioning for the secondary redundancy block are two columns, where only a single redundancy cell is evaluated for the first column.

If the number of columns in the primary redundancy block is even, a simple and efficient dimensioning for the secondary redundancy block are two rows, where only a single redundancy cell is evaluated for the first row. In this case, the second¬ ary redundancy matrix has [m/2] columns, whereby [m/2] equals the Ceiling of m/2.

Decoding for a column of the primary redundancy block with two missing cells (D_x and D_y) is performed comprising the steps of:

(1) Recover all cells in columns where only a single cell was lost by using the primary redundancy block.

(2) Evaluate an auxiliary cell M₁ by performing an XOR- operation on all but the missing two cells of the column together with the primary redundancy cell of this column.

(3) If less than h_s redundancy cells were transmitted in the secondary redundancy block, identify the first missing cell (D_x) which was used for generation of a transmitted secondary redundancy cell R_Ξa. Subsequently, regenerate the missing cell D_x by performing an XOR-operation on all received data cells which were used to generate R_sa together with R_sa.

(4) Recover D_y by performing an XOR-operation on all received data cells of this primary column together with both the D_x and the primary redundancy cell of this column.

Comparing with known schemes for error recovery at the receiver the Double Matrix Interleaving XOR scheme has a superior ratio of the error recovery capability versus the number of operations required for this error recovery. Further, in comparison with known schemes for protocol encoding the present invention is particularly efficient in partitioning the information required by the receiver for error detection as well as recovery in protocol fields per cell and protocol fields oer frame. Description of the Drawings

The above mentioned and other features and objects of this invention, and the manner of attaining them, will become more apparent and the invention itself will be better understood by reference to the following descriptions of embodiments of the invention taken in conjunction with the accompanying drawings, wherein

Fig. 1 is a block diagram of an FEC encoder; Fig. 2 is a block diagram of an FEC decoder; Fig. 3 is a block diagram of a frame format with an FEC code of data from a previous frame;

Fig. 4 is a block diagram of Cell Format Type 1; Fig. 5 is a block diagram of Cell Format Type 2; Fig. 6 is a block diagram of Cell Format Type 3 with AUU=0; Fig. 7 is a block diagram of Cell Format Type 3 with AUU=1 in the case of Variant A;

Fig. 8 is a block diagram of Cell Format Type 3 with AUU=1 in the case of Variant B;

Fig. 9 is a block diagram of an AAL5 CPCS trailer; Fig. 10 is a block diagram of a format for the CPCS-UU;

Fig. 11 is a block diagram of another format for the CPCS- UU;

Fig. 12 is a block diagram of an extended format for the CPCS-UU; Fig. 13 is a block diagram of an Error Protocol Format with Cell Format Type 1

Fig. 14 is a block diagram of an Error Protocol Format with Cell Format Type 2

Fig. 15 is a block diagram of an Error Protocol Format with Ceil Format Type 3 in the case of Variant B

The drawings are not necessarily to scale and certain features may be exaggerated in order to better illustrate and exυlain the present invention. Description of the Present Invention

The embodiments disclosed below are not intended to be ex¬ haustive or limit the mvention to the precise forms disclosed in t ne following detailed description. Rather, the embodiments are chosen and descrioed so that others skilled in the art may utilize their teachings.

One embodiment of the present invention comprises a method and an apparatus for encodmg an arbitrary number of redundancy cells, i.e. h primary redundancy cells, from cells of an AAL5- CPCS-PDU of variable length. This has the advantage that the apparatus based on the invention can generate redundancy cells with .Little delay. Another advantage of this embodiment is that a relatively small integrated circuit can be used to implement the apparatus for generating primary redundancy cells. Another embodiment of the invention comprises a method and an apparatus for recovering missing data cells by using the primary redundancy cells. This has the advantage that a receiving apparatus can start to regenerate corrupted or missing cells directly after the first redundancy cell is received, thus allowing error correction with little delay.

Another advantage of this embodiment is that a relatively small integrated circuit can be used to implement the apparatus for recovering errors by decoding redundancy cells.

Another embodiment of the invention comprises a method and an apparatus for encoding additional redundancy cells, the secondary redundancy cells, of an AAL5-CPCS-PDU of variable length. This has the advantage that a receiver can correct certain errors of corrupted or missing cells which could not be recovered without. Another advantage of this embodiment is that additional redundancy cells can be qenerated with little delay and a relatively small integrated circuit can be used to imple¬ ment the apparatus for generating secondary redundancy cells. another embodiment of the mvention comprises a method and an aDparatus for recovering corrupted or missing data cells ov using both the primary redundancy cells and the secondary red¬ undancy cells. This has the advantage that certain errors can be corrected now which cannot be corrected without the present invention. This also has the advantage that very little delay is introduced at the receiver in case there are no errors or in case all errors can be corrected using only the primary red¬ undancy cells. Another advantage of this embodiment is that this error correction can be performed with little delay and a relatively small integrated circuit can be used to implement the apparatus at the receiver.

In order to disclose a suitable data format the following four examples to be addressed by an FEC scheme for AAL5 to¬ gether with a number of respective solutions are presented:

Example 1: Which fields per cell are required for an FEC-SSCS scheme?

An FEC Scheme for AAL5 needs an appropriate protocol format for cells and frames to allow error detection and error correction. The protocol format to be defined can be partitioned into a cell format and mto a frame format to be called as 'per-cell protocol information' . Two possible solutions for an FEC scheme for the SSCS of AAL5 are presented. Thereby, a scheme is de¬ fined which allows simpler implementation by fewer functions and fewer fields per cell as well as higher efficiency by better error control properties and less protocol overhead.

Solution 1.1

One requirement for the scheme is that for simplicity all user data of an SSCS-SDU has to be in the same CPCS-PDU. In this case no per-cell protocol information needs to be used to identify the first cell or the last cell of an SSCS-SDU.

Solution 1.2

Solution 1.2 is mdependent from the previous Solution 1.1. For additional simplicity, it is possible to use no special b t for indication of the begin or the end, respectively, of a sequence of data cells or redundancy cells an SSCS-PDU, if one of the following conditions is true (sucn a bit could be called B/E bit) :

(a) If individual CPCS-PDUs are used for transmission of data and transmission of redundancy (FEC code), the last data cell or FEC cell, respectively, is identified by the length field of the AAL5 CPCS trailer. (b) If both data and FEC cells are transmitted in one CPCS-PDU, they need to be distinguished. They are be distinguished by using one or more of the following methods: (bl) By a cell type identifier (D/F) ;

(b2) If the number of data cells or FEC cells, respectively, is known in advance, for mstance by signaling; and/or

(b3) If the number of data cells or FEC cells, respectively, is given by one field of the frame.

The B/E bit allows to detect a loss of the last data cell or last redundancy cell. However, this loss is also detected by the cell sequence number and the length field of the AAL5 trailer.

Example 2 : The last cell of the AAL5 PDU will contain 40 bytes of padding when the CPCS-SDU ends with an FEC cell. How could this be avoided?

Solution 2.1

If FEC cells are sent in a separate CPCS-PDU which is not of AAL5 format the AUU-bit in the ATM cell header, which usually indicates the last cell of a AAL5 CPCS-PDU, is used to mark the last FEC cell.

This cell does not contam an AAL5 trailer. Interpreting the _ast four bytes as CRC-32 indicates a CRC violation. If CRC is automatically performed by hardware, the method is still used as long as the AAL5 implementation passes the complete payload of all cells including the cell with AUU=1 to the SSCS mstance .

Solution 2.2

In one AAL5 CPCS SDU one or several FEC cells of an FEC frame are followed by data cells of the subsequent FEC frame. This leads to the data structure depicted in Figure 3.

In this solution a padding field of 40 bytes is required in the last FEC frame of a burst. In most cases this requirement does not matter. If the user data has the appropriate length, the other frames have zero padding fields. In many cases, the user data can be segmented in this way.

This solution is most appropriate and is therefore proposed as the standard method embodied in the present invention.

If solution 2.2 is combined with the requirement that all data cells of one FEC frame are in the same AAL5 CPCS-PDU, the following three Cell Formats are possible:

Cell Format Type 1 (CF=1; cells without individual CRC) :

For Cell Format Type 1 as depicted in Figure 4, CRC is only performed on a AAL5 CPCS-PDU basis using CRC-32 m the trailer. This case is appropriate for low bit error rates.

Cells do not need a CF field to indicate their Cell Format. Rather, the cell format can either be constant per VC, e.g. indicated by signaling, or indicated by a Cell Format field per frame.

Cell Format Type 2 (CF=2; cells with individual CRC) This format is shown in Figure 5 and appropriate for relatively high bit error rates. For simple processmg at the receiver cells in which bit errors are detected by CRC violation are simply discarded. Th s format is relatively easy to implement. However, t has a certain disadvantage because the CRC code nas to De sent prior to the rest of the data, thus complicating the implementation.

Cell Format Type 3 <CF=3; cells with individual CRC) The previous cell format has the disadvantage that CRC field is located in the cell header. Cell format Type 3 (CF=3) which is presented m Figure 6 is trailer-oriented. This allows a streaming mode processing of individual cells and reduces processmg costs. In combination with AAL5 Cell Format CF=3 which is shown in Figure 6 is used only for cells with AUU=0. Cells with AUU=1 must have the standard AAL5 trailer of 8 bytes with CRC-32 as last 4 bytes. However, they do not need additional fields for SN, D/F or CRC-10. This format for the last cell of the CPCS-PDU is shown in Figure 7 and called Cell Format Type 3 Variant A.

This feature can be explained as follows. A cell sequence number in the cell with AUU=1 is helpful to identify the loss of the last cell(s) with AUU=0. However, the cell sequence number is not necessary because the length field of the AAL5 trailer allows to detect such a loss. Including a field with CRC-10 mto the last cell of a AAL5 CPCS-PDU is not very useful either because CRC-32 would indicate bit errors with very high probaDility.

The cells with AUU=1 do not contain FEC code because their maximum payload is only 40 bytes versus 46 bytes maximum pay¬ load cf type 3 cells with AUU=0. Therefore, the field D/F is not required in the last cell of an FEC frame wnen cells with AUU=1 are of Cell Format Type 3 Variant A as shown in Figure 7. T^herefore, Cell Format Type 3 Variant A leads to a lower per-frame overhead, i.e. 2 bytes less, than Cell Format Type 2 and Ce_l Format Type 3 Variant B which is described below. In this case the last cell of an AAL5 CPCS-PDU has a different data format and certain error cases cannot be identified.

For higher error detection probability, Cell Format Type 3 Variant B as depicted in Figure 8 uses a cell format with the fields SN, D/F, CRC-10 also in the cell with AUU=1. These fields are located in the last two bytes of the AAL5 CPCS-PDU payload. In this case the last cell of an AAL5 CPCS-PDU has a slightly different data format compared with cells with AUU=0. For optimal performance it is proposed to use Cell Format Type 1 in the case of low bit error rates and Cell Format Type 3 Variant B in the case of high bit error rates.

Example 3: The most common case will be a lossless system. How can processing be simplified for this case?

Solution 3.1

Send data cells in one AAL5 CPCS-PDU called 'data frame' followed by FEC cells in an independent AAL5 CPCS-PDU called 'redundancy frame' . Use an identifier (one bit) per PDU that distinguishes data frames from redundancy frames. This ident¬ ifier is preferably placed in the AAL5 CPCS-PDU trailer in the fields CPCS-UU or CPI or in an SSCS header or trailer. The AAL5 trailer format is shown in Figure 9.

The CPI field currently has no defined purpose except to pad the trailer to 8 bytes. It is proposed to use this field of 1.363 in future versions. Therefore, it is preferable to use the CPCS user-to-user (CPCS-UU) indication byte for additional identifier.

The solution of sending FEC cells in a separate AAL5 PDU has the consequence that all redundancy frames have a padding field of 40 bytes. In other schemes, e.g. Solution 2.2, the probability of a large padding field will be much lower. Solution 3 . 2

Send one AAL5 CPCS-PDU which comprises both FEC cells and data cells. Use either a predefined number of FEC ceils cr use a field per PDU indicating where to separate data and FEC cells. A predefined offset is part of a parameter agreement by signaling during call set-up.

An identifier is placed in the AAL5 CPCS-PDU trailer in the fields CPCS-UU or in an SSCS header or trailer. Using the CPCS- UU to indicate the number of cells at the beginning of the frame which are FEC cells of the previous CPCS-PDU allows for minimal padding. This method is considered as a particularly preferable solution.

If primary and secondary redundancy blocks are used, the receiver needs to be informed about the number of primary redundancy cells and the number of secondary redundancy cells. The application and the number of secondary redundancy cells could be indicated to the receiver during call setup by using special signaling messages. It is also possible to specify that certain bits in the AAL5-CPCS-trailer represent the number of secondary redundancy cells directly encoded.

In an alternative solution it is assumed that a certain set of useful combination of primary and secondary redundancy blocks are implicitly known by the receiver together with identifiers which could be used by the transmitter and placed into dedicated bits of the AAL5-CPCS-trailer. Typically, the number of primary redundancy cells per frame is less than 16. In these cases, 4 bits of the CPCS-UU byte are sufficient to directly encode the number of primary redundancy cells.

The results in the data format for the CPCS-UU byte are shown in Figure 10. In many cases the number of FEC cells per frame will be the same for subsequent CPCS-PDUs. This allows for a fast implementation of the most common case. Example 4 : How can large AAL-SDUs be segmented and reassembled*⁷

Solution .1

A special b t for synchronization m every cell allows to identify the first cell of a larger unit. Such a bit allows to identify the beginning of an AAL SDU and to segment this SDU mto several CPCS-PDUs.

Solution 4.2 The last CPCS-PDU of a large AAL-SDU, also called a block, is identified by a special frame type. For identification of this frame type a special bit, e.g. called LastF, is used per CPCS- PDU. This bit is placed into the CPCS-UU byte as depicted in Figure 11.

Solution 4.3

A sequence number per frame is provided together with an identification of the last CPCS-PDU of a block. The sequence number is called 'Frame Sequence Number' (FSN) . The FSN allows to detect and identify missing CPCS-PDUs. Sequence numbers are used in one of the following ways:

(a) Incrementing FSNs modulo the FSN numbering space avoids a limit for the maximum number of segments of an AAL SDU. In this case an additional indication is required m order to identify the last CPCS-PDU of a block.

(b) Decrementing FSNs from the value which represents the number of CPCS-PDUs of a block to the value which represents the _^ast CPCS-PDU of a block. In this case no additional indication is required in order to identity the last CPCS-PDU of a block. However, the maximum number of CPCS-PDUs of a block is limited to tne FSN numbering space.

The CPCS-UU byte of the AAL5 CPCS-Trailer is used for the sequence number. If 4 bits of the CPCS-UU byte are used for the fιe_α #RedCells, a FSN field of 4 bits is possiole. The CPI-byte is used for identifying the last frame of a blocκ either oy reserving one bit (LastF) or Dy reserving a special value of the CPI-byte for this purpose. In combination witn solution 3.2 this leads to the data format for the CPCS-UU byte shown in Figure 12.

If a larger numbering space for FSNs is required, either the full CPCS-UU byte is used for this purpose or an SSCS header or trailer with a larger FSN field is applied.

Solution 4.4

A sequence number per frame is used. This number is decremented and the last CPCS PDU carries sequence number 0. In this case an AAL SDU may be segmented mto a maximum number of CPCS-PDUs according to the FSN numbering space. FSNs of 4 bit would allow segmenting into 8 frames; FSNs of 8 bit would allow segmenting into 256 frames. In the framework of the present invention, however, Solution 4.3 is recommended.

Summary of Examples A combination of the solutions 1.1, 1.2, 2.2, 3.2, and 4.3 defines a new error protocol format which is particularly suitable for the purposes of the present invention and which is based on one of the following data formats:

- Cell Format Type 1: Using cell format type 1 (cells without individual CRC, CF=1) leads to the frame format presented in

Figure 13.

- Cell Format Type 2: If cells of cell format CF=2 (cells with individual CRC) are used, the results CPCS format is in Figure 14. - Cell Format Type 3: If cells of cell format CF=3 (trailer- oriented, with individual CRC) are used, the resulting CPCS format is depicted in Figure 15. Finally, preferred emoodiments of the present invention as previously described are summarized as follows:

One embodiment of the present mvention provides for a method and an apparatus for transmitting redundancy cells m combination with AAL5-PDUs which contam user data by using Solution 2.1. It has the advantage that no additional bandwidth overhead is required for transmission of redundancy cells. Further, receivers which are able to process AAL5-PDUs but which do not have a separate apparatus for processmg of the redundancy cells automatically discard all redundancy cells and therefore only deliver user data to the higher layer, as re¬ quired. Thus, the present invention is capable to be applied in combination with existing apparatus for the processmg of AAL5- PDUs. Another embodiment of the present invention provides for a method and an apparatus for transmitting redundancy cells in AAL5-PDUs m combination with user data cells by use of Solu¬ tion 2.2. This has the advantage that the user data is seg¬ mented into frames in a way that very little or no padding is required thus leading to a very little bandwidth overhead. Another advantage of the present invention is that it is very easy for the receiver to separate the redundancy cells and tne cells with user data. Further, the user data and the corres¬ ponding redundancy cells are transported in different AAL5-PDUs thus allowing to check the integrity of the user data and the redundancy data independently. Another advantage is that the apparatus for decoding at the receiver introduces only little additional delay and it can be implemented with a small integrated circuit . Another embodiment of the invention provides for a method and an apparatus for encodmg per-cell protocol information for detection of missing cells and identification of redundancy cells oy using Cell Format Type 1. This has the advantage that only very little bandwidth is required for transporting the per-cell protocol information to the receiver. Furthermore, the apparatus for encodmg introduces only little additional delay and if can be implemented with a small integrated circuit.

Another embodiment of the present invention provides for a method and an apparatus for encodmg per-cell protocol informa¬ tion for the detection of corrupted cells and missing cells and for the identification of redundancy cells by the application of Cell Format Type 2. This has the advantage that the receiver easily detects corrupted cells and very little additional band- width is required for this capability as well as for the per- cell protocol information. Further, the apparatus for encoding introduces only little additional delay and it can be imple¬ mented with a small integrated circuit.

Another embodiment of the present invention provides for a method and an apparatus for encodmg per-cell protocol informa¬ tion for the detection of corrupted cells and missing cells as well as for the identification of redundancy cells by the use of Cell Format Type 3. This has the advantage that the sender can start transmitting a cell prior to completing the evalua- tion cf the cyclic redundancy check CRC-10 thus reducing the overall delay. Further, the receiver easily detects corrupted cells and very little additional bandwidth is required for this capability as well as for transporting the per-cell protocol information to the receiver. This has the further advantage that the apparatus for encoding introduces only little addi¬ tional delay and it can be implemented with a small integrated circuit.

Another embodiment of the present invention provides for a method and an apparatus for distinguishing data cells from red- undancy cells by applying Solution S3.1. This has the advantage of reducing the processmg requirements the absence of errors .

Another embodiment of the present invention provides for a method and an apparatus for transmitting redundancy cells by applying Solution 3.2. This has the advantage of reducing the processing requirements of a lossless system. Further, only lιtt_e additional bandwidth is required for padding bytes.

Another embodiment of the present invention provides for a method and an apparatus for segmenting and reassembling large AAL-SDUs by application of Solution 4.1. This has the advantage that the method and the apparatus for segmenting and reassemb¬ ling large AAL-SDUs is easy and the apparatus can be implement¬ ed with a small integrated circuit. Another embodiment of the present invention provides for a method and an apparatus for segmenting and reassembling large AAL-SDUs by applying solution 4.2. This has the advantage that the method and the apparatus for segmenting and reassembling large AAL-SDUs is easy and very little bandwidth is required for this method. Further, the apparatus can be implemented with a small integrated circuit.

Another embodiment of the present invention provides for a method and an apparatus for segmenting and reassembling large AAL-SDUs by the use of solution 4.3. This has the advantage of introducing additional error detection capability which allows to detect and identify missing and missequenced CPCS-PDUs. Further, the method for segmenting and reassembling large AAL- SDUs is simple, little bandwidth is required, and the apparatus can be implemented with a small integrated circuit. Another method for segmenting and reassembling large AAL- SDUs makes use of Solution 4.4. This has the advantage of introducing additional error detection capability which allows to detect and identify missing and missequenced CPCS-PDUs. Furtner, the method for segmenting and reassembling large AAL- SDUs is simple, little bandwidth is required, and the apparatus can oe implemented with a small integrated circuit.

Finally, it is mentioned that for a scaleable provision of reliable unicast and multicast services in ATM networks the integration of FEC into the Adaptation layer offers various benefits. For low bit error rates a simple cell format with an overhead of 1 byte is proposed. For higher cell loss rates two frame formats with a per-cell CRC are proposed. All three cell formats allow the integration of FEC into the SSCS of AAL5 with

5 very little overhead.

While this invention has been described as having an exemplary design, the present invention may be further modified withm the spirit and scope of this disclosure. This applica¬ tion is therefore intended to cover any variations, uses, or

10 adaptations of the invention using its general principles. Furthermore, this application is intended to cover such departures form the present disclosure as come with known or customary practice in the art to which this invention pertains.

15 References

[Ayan93] Ayanoglu, E. ; Gitlm, R.D. ; Oguz, N.C.: "Performance Improvement Broadband Networks Using Forward Error Correction", Journal of High Speed Networks, Vol. 2, 1993, pp. 287-304

20 [ATMF3.0] ATM Forum: "UNI Specification Document Version 3.0", PTR Prentice Hall, Englewood Cliffs, NJ, 1993 [Bιe92] Biersack, E. W. : "Performance Evaluation of Forward Error Correction in an ATM Environment", IEEE Journal on Selected Areas in Communication, Volume 11, Number 4, pp. 631-

25 640, 1993

[Fel93] Feldmaier, D. : "An Overview of the TP++ Transport Protocol Project", Chapter 8 in "High Performance Networks - Frontiers and Experience", Ahmed Tantawy (Ed.) , Kluwer Academic Publishers, 1993

30 [1.121] ITU-T Study Group XIII (formerly CCITT, Study Group XVIII) Draft Recommendation 1.121, "Broadband Aspects of ISDN", Geneva, 1988

[1.363] ITU-TS Draft Recommendation 1.363: "BISDN ATM Adaptation Layer (AAL) Specification", Geneva, 1993 [ITU94] ITU-T SG13 Q.6 Rapporteur meeting report, TD-41, SG-13, WP2, Q.6 (AAL1&2) , 1994

[McAu90] McAuley, A. J.; "Reliable Broadband Communication Using a Burst Erasure Correcting Code", ACM SIGCOMM90, 1990. [OhKi91] Ohta, H., Kitami, T.: "A Cell Loss Recovery Method Using FEC in ATM Networks", IEEE Journal on Selected Areas in Communications, Vol. 9, No. 9, pp.1471-1483, 1991 [Q.2931] ITU-TS Draft Recommendation Q.2931 (former Q.93B) : "B-ISDN User-network Interface Layer 3 Specification for Basic Call/Bearer Control", Geneva, 1993

[RFC1626] Atkinson, R. : "Default IP MTU for use over ATM AAL5", RFC1626, 1994

[Rom93] Romanov, A. : "Some Results on the Performance of TCP over ATM", Second IEEE Workshop on the Architecture and Implementation of High Performance Communication Subsystems HPCS'93, Williamsburg, Virginia, 1993

[Sha90] Shacham, N. ; McKenny, P.: "Packet recovery in high^¬ speed networks using coding", in Proceedings of IEEE INFOCOM ^■90, San Francisco, California, pp. 124-131, 1990 [ZhSa91] Zhang, L. ; Sarkies, K. : "Modeling a Virtual Path and its Application for Forward Error Recovery Coding Schemes in ATM Networks", In Proc. of the 1991 Singapore International Conference on Networks SICON'91, Sept. 5-6, Singapore, pp. 259- 264, 1991

Claims

1. A method for forward error correction of transmitted digital signals in networks, the method comprising the steps of: a) constructing data cells of constant length, each cell having a sequence number field, a data/FEC identification field and one of a user data field and an FEC field; b) generating a first frame and a second frame said first frame having a first set of said data cells and a first FEC code, said second frame having a second set of said data cells an a second FEC code, wherein said second FEC code corresponds to said first set of said data cells; and c) encoding and decoding said first frame by reordering said first set into a first matrix of said data cells having h first columns and L + 1 first rows and by performing and XOR operation on each of said first columns to produce a first redundancy block of length h.

2. The method of claim 1, wherein said sequence number field has a length of 7 bits.

3. The method of claim 1 or 2, wherein said data cells comprise a CRC code field.

4. The method of claim 3, wherein said data cells have a sequence number field length of 5 bits and a CRC code field length of 10 bits.

5. The method of any one of the preceding claims, wherein one of said user data field and said FEC field is located at a beginning of said data cells.

6. The method of claim 3, wherein said data cells have a sequence number field and a CRC code field with combined length of 15 bits.

7. The method of claim 1, further comprising the step of

5 encodmg and decoding said frame by reordering said first set mto a second matrix of said data cells having second columns and second rows and by performing an XOR operation on each of one of said second columns and said second rows to produce a second redundancy block. 0 8. An apparatus for forward error correction of transmitted digital signals in networks, the apparatus comprising: a) means for constructing data cells of constant length, each cell having a sequence number field, a data/FEC identification field and one of a user data field and 5 an FEC field; b) means for generating a first frame and a second frame said first frame having a first set of said data cells and a first FEC code, said second frame having a second set of said data cells and a second FEC code, wherein 0 said second FEC code corresponds to said first set of said data cells; and c) means for encoding and decoding said first frame having means for reordering said first set into a first matrix of said data cells having h first columns and L + 1 5 first rows and means for performing an XOR operation on each of said first columns to produce a first redundancy block of length h.

9. The apparatus of claim 8, wherein said sequence number field has a length of 7 bits.

30 10. The apparatus of claim 8 or 9, wherein said data cells comprise a CRC code field.

11. The apparatus of claim 10, wherein said data cells have a sequence number field length of 5 bits and a CRC code field length of 10 bits.

12. The apparatus of any one of the preceding claims, wherem 5 one of said user data field and said FEC field is located at a beginning of said data cells.

13. The apparatus of claim 10, wherein said data cells have a sequence number field and a CRC code field with a combined length of 15 bits.

10 14. The apparatus of claim 8, wherein said means for encoding and decoding said first frame include means for reordering said first set into a second matrix of said data cells having second columns and second rows and means for performing an XOR operation on each of one of said second

15 columns and said second rows to produce a second redundancy block.