CA2311103A1

CA2311103A1 - System and method for improving compression on a telephone loop

Info

Publication number: CA2311103A1
Application number: CA002311103A
Authority: CA
Inventors: Yatish Kumar
Original assignee: Catena Networks Canada Inc
Current assignee: Catena Networks Canada Inc
Priority date: 2000-06-12
Filing date: 2000-06-12
Publication date: 2001-12-12
Also published as: AU2001268389A1; WO2001097434A3; US20020015410A1; WO2001097434A2; US6539016B2

Description

_ ii s SYSTEM AND METHOD FOR IMPROVING COMPRESSION ON A TELEPHONE
LOOP
4 The present invention relates to the field of data compression, and more particularly to data compression on an xDSL loop.
BACKGROUND OF THE INVENTION
With the increasing popularity of the Internet, there has been a corresponding increase in the demand for high rate digital transmission over the local subscriber loops of telephone companies. A loop is a twisted-pair copper telephone line coupling a user or subscriber 12 telephone to a central office (CO).
Traditional, data communication equipment, uses the voice band of the subscriber loop.
Such equipment includes voice band modems, which operate at up to 56 kbps using 16 compression techniques. On the other hand, Integrated Services Digital Network (ISDN) systems have boosted data rates over existing cooper phone lines to 120 kbps.
However traditional voice bands equipment is limited by the maximum data rate of the existing switching networks and PCM (Pulse Code Modulation) data highways.
By utilizing the frequency bandwidth of the loop outside the voiceband has enabled other high-speed systems to evolve. However because loops can differ in distance, diameter, age and transmission characteristics depending on the network, they pose some 24 limitations and challenges for designers of these high-speed systems.
Current high-speed digital transmission systems of the above type include asymmetric, symmetric, high-rate, and very high-rate digital subscriber loops, conventionally known 28 as ADSL, SDSL, HDSL and VDSL respectively. Normally these and other similar protocols are known as xDSL protocols.
Of these flavours of xDSL, ADSL is intended to co-exist with traditional voice services 32 by using different frequency spectra on the loop. In the future, it is possible that multiple different transmission schemes may be employed in different frequency bands on the , , same loop, and that these transmission schemes may include traditional analog voice services as well as current and new forms of xDSL. In today's ADSL systems, the plain old telephbne services (POTS) uses the frequency spectrum between 0 and 4kHz and the 4 ADSL uses the frequency spectrum between 30kHz and 1.lMHz for data over the telephone line. This is shown schematically in figure 1 a. ADSL also partitions its frequency spectrum with upstream (subscriber to CO) transmission in a lower frequency band, typically 30kHz to 138kHz, and with downstream transmission in a higher 8 frequency band, typically 138kHz to SSOkHz or 1.lMHz. ADSL uses a discrete multi-tone (DMT) multi-carrier technique that divides the available bandwidth into approximately 4kHz sub-channels.
12 In order to maximize the throughput on a given channel, it is important to minimize the redundancy in the transmitted data, followed by the careful addition of some redundancy, to enable the use of forward error correction. Thus far, there has been a lot of activity to improve the performance of DSL, particularly on long loops with the use of better 16 forward error correction. Reed Solomon encoding, and Trellis Coded Modulation are already part of the G.DMT specification, and further additions of Concatenated Convolutional, and turbo codes, are open issues for the G.DMT specification.
SUMMARY OF THE INVETION
One aspect of the invention provides for the limiting of the maximum compression 24 bandwidth to assist ATM provisioning.
Another aspect of the invention provides for limiting the average compression bandwidth to assist ATM provisioning.

A still further aspect of the invention provides for ATM flow control over the ADSL loop to assist ATM provisioning in the presence of bandwidth variation.
(Compression is one way to get bandwidth variation. But there are others).

2 A still fiu-ther aspect of the invention provides for the use of multiple or hybird compression algorithms to match the interleaved data traffic seen on ADSL
loops.

Figure 1 shows the location of compression function (Comp) in the ATU
reference transmitter model; and 8 Figure 2 is a schematic diagram showing a method for making compression compatible with ATM payload scrambling.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
The present invention provides for the use of compression on the data in a channel, to remove some of the inherent redundancy, in order to yield much better throughput, particularly in conjunction with the more powerful FEC schemes that are currently being 16 studied. The removal of redundancy on the transmitted data is independent of the data rate on the loop. However, for long loops with reduced data rates, the improvement can provide significant benefits to the end user of the DSL service.
20 We consider first whether Data compression should be performed at the physical layer in an ADSL modem. Performing data compression at the physical layer for a DSL
link is a practical solution to the problem. If it was deferred to higher layer protocols, and applications, it would be difficult to ensure that the union of all application programs, 24 operating systems, network protocols, and content providers would present data to the ADSL link in a compressed format. We would most likely be left with the status quo.
Some users, of some applications would use data compression. While many users of the Internet would continue to transfer files, download web pages, and exchange email 28 without the benefit of compression.
The inclusion of data compression in the DSL link does no harm, but it potentially provides a great benefit. It is relatively easy to design a data compression scheme, which 32 will not provide a degradation of throughput.

3 Finally DSL provides data rates and services that are quite different from dial up modems, and Ethernet based LANs. DSL specific issues need to be addressed by DSL

4 standards, and cannot be left to general networking solutions. One example of a DSL
specific issue is the use of ATM cells over the link. It is possible to exploit the redundancy in the ATM cell headers, to help compress the traffic on the loop by up to 10%. Another example is the multiplicity of services that can be run on the bandwidth of 8 a DSL link. It is possible to download a file, while browsing the web, and perhaps listening to an audio broadcast. Compression algorithms in such an environment need to be agile to the interleaved traffic.
12 The compression function is best performed between the ATM Cell TC block, and the LS 1 input to the interleaved data stream as shown in figure 1. This provides the benefit that the compression algorithm can tackle the redundancy inherent to the ATM
headers, which can account for 10% of the usable bandwidth. Furthermore, it allows the 16 compression algorithm to work in the presence of scrambling within the ATM
cell payload (more will be discussed on this concept in the next section).
ATM Cell Scrambling:
20 Most compression algorithms will perform poorly if the source data has been scrambled.
The scrambling makes it impossible for the algorithm to identify redundancy in the source data. The ATM layer is required to scramble its payload, in order to eliminate the possibility that during HEC delineation the payload might be interpreted as a valid ATM
24 header. This can be handled in the presence of compression as shown in figure 2.
It can be seen from Figure 2, that at the ATM TC level, the requirement for payload scrambling can still be met, by adding a descrambler/scrambler pair as part of the 28 compression function. The operation for the additional descrambler/scrambler pair would be the same as that of the ITU-T recommendation L432, using the self synchronizing polynomial X43 + 1.

Implications of the fact that the decompressed data rate is source data dependent ATM PVC Provisioning:
4 The unpredictable requirement for network bandwidth, at the ATU-C, required to service a DSL loop using compression, is an issue that requires some careful consideration. An analogous problem exists in systems developed using present DSL technology. If a shelf is built to support 100 full rate modems, the theoretical maximum bandwidth required 8 will be 6 x 100 = 600 Mbps of data. However depending on the statistics of the number of active subscribers, and the length of the loops connected to the 100 modems, the actual bandwidth used can be significantly less. System designers have a choice to either accommodate the peak traffic, or allow some blocking for the average traffic they feel is 12 reasonable.
Similarly, with statistically variable bandwidth, in the presence of compression, it is possible either to provision for the maximum compressed data rate, or to provision for a 16 reasonable average. There are two differences in the analogy however.
First, in the shelf example the statistics are an ensemble average over many users, and in the compression case the statistics are a time average for a single user. Secondly, in the shelf example it is possible to calculate the absolute maximum traffic that can be generated, in the case of 20 compression it is more difficult to establish this value. The following sections address these issues.
Limiting the maximum traffic generated through compression:
24 With a simple-minded implementation of data compression, it is possible to generate a very high, peak data rate at the output of the decompression circuit. For example, using run length encoding, if a user is sending all 0's, it is possible to transmit a single 0, and a count. If the count is large, a very large amount of data is instantaneously created at the 28 output. Furthermore, the latency to produce the first output is determined by the length of the burst. Such pathological cases can be avoided by placing latency, and coding gain requirements on the compression algorithm.

Once this is accomplished, it is possible to determine the peak throughput, given an uncompressed line rate. This absolute maximum can be used to determine the network bandwidth that needs to be provisioned at the ATU-C.

Provisioning less than the absolute maximum possible bandwidth:
There are two possible solutions to provision less than the absolute maximum possible bandwidth. The first is to limit the provisioning of compression to longer loops with 8 inherently less bandwidth. Let us assume that the maximum compression gain is limited to a factor of 2x. In a system designed to service 6Mbit modems, if a subscriber loop has an uncompressed bandwidth greater than 3 Mbits, then compression is disabled for that loop. A user on a long loop using compression is then indistinguishable from a user on a 12 short loop with no compression.
The second solution is to implement flow control over the DSL loop. If a subscriber's peak bandwidth exceeds the buffer space allocated in the central office, the flow of ATM
16 cells is reduced at the ATU-R until the average bandwidth matches the provisioned PVC
at the central office. A variation of this algorithm would be the use of feedback in the compression algorithm, which would ensure that the average data rate does not exceed a prescribed amount. This amounts to a cap on the average performance, rather than the 20 peak performance of the algorithm.
In conclusion, compression will maximize the channel capacity on all loops, as well as complementing the benefits that would come from more powerful error correction 24 techniques. The gains can be particularly important to subscribers with access to limited bandwidth on long loops.