WO2011077430A1 - Dsm cross-talk cancellation technique for xdsl lines - Google Patents

Abstract

A technique for practical implementation of a DSM processing system/engine for cross talk cancellation, which resolves an architecture problem of implementing the DSM processing engine so as to allocate its so-called "cancellation resources" in an optimal manner and to make the engine less critical to changes in an xDSL lines system. When processing an initial plurality of crosstalk cancellation coefficients, the method distributes the operation between at least two blocks one of them being a memory and the other being a cross-talk cancellation processor, compresses the coefficients before storing in the memory, transfers the compressed coefficients to the processor, and decompresses the received compressed coefficients before applying to the cross-talk cancellation processor for implementing cross-talk cancellation in the xDSL system.

Description

DSM cross-talk cancellation technique for xDSL lines

Field of the invention

The present invention relates to a technology of cross-talk cancellation for a group of data carrying lines/channels exposed to mutual interference (coupling) in data communication systems.

Background of the invention

Digital subscriber line (DSL) technology is used to transform an ordinary telephone line (e.g., copper wire twisted-pair) into a broadband communication link. It works by sending signals over the telephone line in previously unused high frequencies.

Over the years, DSL technology has evolved into a family of specific, standardized implementations. These various implementations offer a variety of transmission speeds and transmission distances. One DSL or xDSL line (which can also be called a DSL channel or a DSL earner in the frame of the present application) is presently able to transmit information in both directions via approximately 4000 frequency bins (sub-carriers). A DSL system comprises a group of DSL lines; usually, a plurality of DSL lines of the same system form a common binder or cable.

It is common to refer to the various DSL implementations that have evolved over the years collectively as xDSL. A number of factors determine the performance of the various xDSL implementations. For example, the performance of any of the xDSL implementations is highly dependent on the local loop length (e.g., the length of a twisted-pair circuit between a central office and a customer) and the local loop condition. The local loop condition is affected by several factors such as, for example, line noise. Line noise may corrupt data- bearing signals as the signals travel along the line. As a result, the transmitted data-bearing signals may be decoded erroneously by a receiver because of this signal corruption.

Therefore, a number of noise mitigation techniques/strategies have been developed for xDSL systems.

Internally generated noise sources are mostly influenced by the transceivers' design of central office (CO) and customer premises equipment (CPE) and generally cannot be mitigated through deployment practices. Robustness to loop impairments and internal noise sources can be improved through system design guidelines and advanced DSL transmission technologies, such as adaptive echo- canceller, adaptive hybrid and programmable digital/analog filters. External stationary noise limits, such as crosstalk, are often accounted for the network design.

Crosstalk is one of the main limitations of DSL performance today. Static Spectrum Management (fixed spectral masks) ensures that DSL lines in the same cable (binder) are spectrally compatible under worst-case crosstalk assumptions.

Dynamic Spectrum Management (DSM) increases capacity utilization of the cable by adapting the transmit spectra of DSL lines to the actual time-variable crosstalk interference. DSM comprises a set of techniques for multi-user power allocation and/or detection in DSL networks to ensure spectral compatibility under crosstalk assumptions.

With DSM, crosstalk is either reduced by shaping the spectra of the transmit signals, or is (partially) cancelled within the binder. These techniques are very effective for deployment scenarios where crosstalk is the dominant source of impairment.

There are four levels of DSM coordination:

1) DSM Level 0 corresponds to static spectrum management (SSM) maximizing individual DSL line performance without considering the performance of neighboring lines.

2) DSM Level 1 deals with an autonomous power allocation management used for crosstalk avoidance.

3) DSM Level 2 is a centralized power allocation management between neighboring lines to avoid crosstalk.

4) DSM Level 3 is used for crosstalk mitigation. It can be used only when either transmitters and/or receivers are collocated; A so-called matrix of crosstalk cancellation coefficients is usually built in such a case. The matrix presents coefficients (sometimes called pre-coding coefficients), each of which is recommended for cancellation of interference of a specific xDSL line (one of the matrix columns) to a different specific xDSL line (one of the matrix rows).

Various aspects of cross-talk cancellation are discussed in the prior art.

US2006029147A describes a method and an apparatus for training a multiple channel communication device, and for the processing utilizing a transmit signal power control operation. Multiple channel communication devices utilize inter-channel crosstalk mitigation techniques, such as MIMO processing modules or MIMO precoding systems, to cancel unwanted crosstalk coupling across active and provisioned channels. Upon provisioning and activation of a new channel that connects to the multiple channel communication device the signal on the new channel couples into other channels and the existing MIMO filtering or processing structure is untrained to mitigate the crosstalk from the new channel. A power control training operation prevents crosstalk from hindering operation on other channels during the training operation. During one example training operation, the power level of a transmit signal is incrementally increased as the crosstalk cancellation filter(s) are trained. The concept of US2006029147A is to apply channel crosstalk estimation and power control during training in a multiple channel device.

US2009257581A describes some embodiments for reducing alien crosstalk. At least one embodiment includes receiving noise data associated with a first user signal on a first tone, receiving noise data associated with a second user signal on the first tone, and receiving at least one alien crosstalk canceller coefficient for the first user on the first tone. Some embodiments include applying the at least one alien crosstalk canceller coefficient to the second user signal to reduce alien crosstalk for the first user signal. US2009257581A is related to cancellation of alien crosstalk only.

However, the problem of cancellation of in-domain (self) crosstalk, where it is assumed that the cancellation algorithm has access to the disturbing signals, is difficult for solving in xDSL systems with a great number of modems which simultaneously operate at high bit rates.

In modern high bit rate systems, such as VDSL, the conventional task of crosstalk cancellation encounters serious problems.

At DSM L3 (DSM level 3), cross-talk cancellation comprises operation of pre-coding, which means' pre-multiplication of all users' signals, viewed as a vector on each frequency bin, by a pre-coding matrix. The elements (pre-coding coefficients) of this matrix are set to mitigate or minimize the crosstalk.

For instance, when implementing the DSM L3 pre-coding, a matrix— of very large volume of cross-talk coefficients' data has. to be retrieved from a memory to the actual pre-coder in order to perform the pre-coding (multiplying coefficients in real signals to mitigate crosstalk). To give an example, when talking about pre-coding a matrix of coefficients for 48 users (lines), the speed at which the memory has to be read should be 1 180 Gbpsb (48*2* 16*4000*4000*48). With the state of the art memory technology of DDR (Double Data Rate memory), and at 250Mhz, the above task would require a 4700 bit bus memory - such a solution is not feasible today.

There is a further problem known for DSM crosstalk cancellation engines, which is especially critical in case of high bit rate, multiline xDSL systems. Usually, a DSM cross-talk cancellation engine has some finite processing power; if, for example, it supports 8 lines - it can cancel all 7 of them that interfere to a specific victim line. However, when the number of lines to be handled increases and the DSM engine remains the same, the engine may not have the power to cancel cross-talk at all the interfering lines. Prior art techniques which are presently known for example (Cendrillon et al., "Partial Crosstalk Cancellation for Upstream VDSL", EURASIP Journal on Applied Signal Processing 2004) statically allocate all the recourses required for full line-by-line cancellation and describe how to "cancel" ("burn") a constant number D of lines being the most disturbing ones per victim line. Object and Summary of the invention

One object of the following invention is to enable a practical— implementation of a DSM processing system/engine for cross talk cancellation. There is also an object to resolve an architecture problem of implementing the DSM processing engine so as to allocate its so-called "cancellation resources" in an optimal manner and to make the engine less critical to changes in xDSL lines.

The first object can be achieved by the following novel features of the proposed solution:

A) Splitting/distributing the processing of the plurality of crosstalk cancellation coefficients between at least two architecture blocks of the DSM engine hardware and software;

These two blocks may be, say, a memory of coefficients (block 1) and a vectoring engine in the form of a processor, external or located on the same chip, and constituting a pre-coding, or cross-talk cancellation processing block (block 2).

B) Reducing rate of transferring the plurality of the coefficients from block 1 in the DSM architecture to block 2 in the DSM architecture. The Inventors propose performing that by compressing the plurality of coefficients before transferring thereof for implementation, and by suitably de-compressing the plurality of the coefficients at the time of implementing thereof at the vectoring engine (block 2). In other words, the Inventors propose that pre-coding is performed at the vectoring engine after restoring the coefficients.

C) The solution may comprise dynamic resource allocation for the DSM cancellation engine. The proposed technology is especially advantageous for VDSL systems, preferably to downstream transmission of video information along VDSL lines from VTU-C (central office) to VTU- R(transceiver units at clients premises). However, it can be applied for downstream and upstream services simultaneously, by DSM engines which may be co-located but may serve different sets of lines. In practice, the engine(s) may be located at a VDSL street cabinet VTU-C, at a VTU-R in a building or flat.

The inventive technology is especially advantageous for xDSL lines with bitrates higher than 10 Mbps, for example for those between 10 and 100 Mbps, where the proposed compression allows achieving essential economy.

Process splitting (A)

In a typical DSM implementation, the data before a so-called IFFT block (performing mathematical processing for signal conversion from the frequency domain to the time domain) is transmitted out of the xDSL modem(s) chip and to the vectoring engine (block 2 of the DSM engine) for DSM manipulations/precoding. Then the manipulated data goes back to the same chip of the xDSL modem(s) in order to be further processed in the IFFT block.

Splitting the above processing chain into two parts and implementing them by separated stmctures on the chip or even by different chips allows making the chip(s) more specific and thus could lead to much higher efficiency. Pre-coder coefficients rate transfer reduction (B)

As mentioned earlier in the description, when implementing DSM

L3 precoding, a huge coefficient data file has to be retrieved from memory to the actual pre-coder in order to perform the pre-coding. We mentioned one example of a 48 users' pre-coding matrix, when the speed at which the memory must be read is 1 180Gbpsb (48*2* 16*4000*4000*48), which is too high and non- implementable by currently available DDR memory technology. The proposed idea comprises using a compression mechanism that would compress the data to be stored in the memory block (block 1). In this way, when the data is being sent to the pre-coding processing block (block 2), it is compressed for using a narrower bus and, upon receiving it in the pre-coding block (block 2) is restored. In other words, the idea here is to compress the data, that is stored in the main memory, by a dedicated compressing algorithm and transfer it to the processing engine, so as to allow further expanding the data back (a.k. close to the original values of the data) in a simple manner.

In the proposed technology of compressing the cross-talk coefficients stored in the memory, the idea is to bring only some of the initial coefficients from block 1 to block 2 and use one coefficient for several bins of the xDSL signal. Such a solution, however, would cause degradation in the pre-coding performance (as can be seen in the simulation results shown in Figs 3a and 3b). It has been shown that using such a method of decimation (grouping of as little as 4 coefficients per group which become equal) may reduce the pre-coding performance by 15 dB (where the precoder overall gain is approximately 35 dB). In order to overcome the above problem while maintaining reasonable size of the memory and interface, the Inventors suggest implanting, in the chip/scheme of the vector engine (block 2), an interpolation mechanism being an expansion/decompression part. The modem(s) chip will actually read only part of the coefficients (since it receives the compressed data) but then will interpolate the rest of the coefficients. Such a solution will of course still cause some performance degradation due to inaccuracy of the coefficients, but since this inaccuracy is somehow reduced, the overall degradation will be less severe.

Dynamic resource allocation ( C ):

We will understand a cancellation resource unit of a DSM engine as a single basic unit of the engine, used for performing a single data multiplication for adding one cancellation coefficient to one incoming data constellation (for a number of bins of one line).

For example, a cancellation resource may be a MAC (multiply accumulate) unit in an FPGA implementation of the vectoring engine or a specialized block used for that purpose in an ASIC implementation of the vectoring engine. The number of such single resources (say, such "FPGA" element blocks) may be, say, one order of magnitude greater than the number of "victim" (disturbed) lines M in the xDSL system.

The Inventors has recognized the following problem: if the DSM cross-talk cancellation engine is able to cancel D disturbers (disturbing lines), and a specific deployment has D+1 disturbers, the impact of such a discrepancy becomes veiy sever. Namely, cancellation of D disturbers by the DSM engine may not help at all, since the remaining disturber (even a sole line!) may easily ruin operation of the victim line in the xDSL system.

The Inventors propose providing such a DSM cross-talk cancellation engine technique, which would be not critical or at least would have low sensitivity to the exact number of disturbing and disturbed (victim) lines in an xDSL system, and/or would be adaptable to an xDSL system where the number of interfering/interfered lines could change with time.

The above object may be achieved by the following general approach:

- to allocate a basic number C being close or equal to k*N (0<k<l) out of the total number of N cancellation resource units, wherein N reflecting the total cancellation power of a DSM cross-talk cancellation engine, and wherein N being greater (preferably at least one order of magnitude greater) than the total number M of victim lines,

- to apply C cancellation resource units to all M disturbed (victim) lines,

- to distribute the remaining (l-k)*N resource units (i.e., the remaining portion of the total cancellation power) between the remaining disturbed (victim) lines; the distribution may be performed, for example, to new victim lines which occur suddenly and/or to those existing victim lines who have more than a preselected number of disturbers to begin with.

In view of the above, there are proposed a method for performing cross-talk cancellation by a L3 DSM engine, and the specific DSM engine for implementing the method, preferably the method and engine adapted to high bit rate xDSL lines of tens of Mbps, and/or the method and engine capable of dynamic allocation of it cross-talk- - cancellation resources.

There is further proposed a software product comprising computer implementable instructions and/or data, suitable for being stored on a computer readable medium and adapted to enable and support the described methods when run on a computer.

Details of the invention will be explained as the description proceeds.

Brief description of the drawings

The invention will be further described with reference to the following non-limiting drawings, in which:

Fig. 1 is an exemplary block-diagram of the proposed split architecture of a DSM cross-talk cancellation engine for xDSL lines. Fig. 2a illustrates a schematic block diagram of the DSM engine between a DSP circuit and a xDSL modems chip, with a brief illustration of decimation and interpolation operations.

Fig. 2b is an exemplary implementation of the interpolation module. Figs 3a, 3b are simulations illustrating how performance of crosstalk cancellation in an xDSL system degrades when one type of compression of cross-talk cancellation coefficients is not followed by decompression.

Figs 4a, 4b are simulations illustrating how performance of crosstalk cancellation in an xDSL system degrades when one type of compression of cross-talk cancellation coefficients is followed by decompression thereof.

Fig. 5 is a simulation showing how other, more advanced types of the coefficients' decompression can affect the discussed performance.

Fig. 6 schematically illustrates how the idea of dynamic cross-talk resource allocation in a DSM cross-talk cancellation engine can be implemented for D victim lines in an xDSL system.

Detailed description of the preferred embodiments

Fig. 1 illustrates a schematic block-diagram of the proposed DSM Cross-talk cancellation engine 10 for high bit rate multi channel xDSL systems.

The DSM engine includes the following main parts:

1. Block 1, (marked 12) containing: A compression unit 1 1 for compressing the crosstalk cancellation coefficients received from a DSP block, before storing them in a coefficient memory 13. The compression unit may be further decomposed into (i.e., presented by) a lossy compression sub-unit and a lossless compression sub- unit.

2. A communication unit 14 (a bus with interfaces) which is responsible for sending the compressed coefficients from memory 13 to block 2 marked as a cross-talk cancellation processor 16. The bus may be located in the same card or even in the same chip between block 1 and block 2

3. Block 2 (separate from Block 1 , while not mandatory external/ remote) comprises a decompression unit 18 which processes the compressed cancellation coefficients to generate a replica of the original coefficient with as small distortion as possible, so that the application of the restored coefficient will give nearly the same performance as applying the original ones. Block 2 also contains a vectoring engine 20 (i.e., the pre-coding engine or the cross-talk cancellation processor) for applying the decompressed coefficients to the digital signals in the xDSL lines. To perform that, block 2 (more specifically, the processor 20) receives signals from xDSL modems.

Actually, the vectoring engine 20 performs processing of initial digital signals of the xDSL modems by pre-coding them, i.e. by applying to them the obtained resulting coefficients. At the output, the modem signals in the xDSL system become modified and thus cross-talk in the system is minimized. Compression and decompression of the crosstalk coefficients serve to address the following problems:

- Limited memory capacity to store the full array of crosstalk cancellation coefficients;

- Limited communication bandwidth between the coefficient memory to cancellation processor.

Fig. 2a illustrates a schematic diagram explaining the proposed process of fomiing pre-coding coefficients for cross-talk cancellation in xDSL lines. The scheme shows that the DSM engine 10 receives cross-talk cancellation coefficients (Cn...) from a DSP block 30, while they are compressed in block 1 1 of the engine 10 (in this example, they are decimated - i.e., only one of Kcoefficients is stored in the memory 13 of the engine, K=4, i.e. one coefficient for 4 frequency bins-. As can be seen, DSP block 30 performs channel estimation and produces four times more coefficients than stored in the DSM engine 10. ( In the frame of this description, we do not discuss how the initial coefficients are formed in the DSP block.) The memory 13 of the engine 10 also stores the step value and the step sign for further restoration of the coefficients. The the interpolation module 18 of the DSM engine lOexpands the coefficients with possible accuracy, to obtain their restored values (marked as waved symbols Cn...). Then a signal, modified in the vectoring block 20 using the restored coefficients, is applied to xDSL modems block 40.

Fig. 2b shows in more details, how the above-mentioend interpolation can be performed. The latch 22 captures the previous coefficient C[n] read from memory 13. It always stores the current coefficient, and when the next coefficient arrives (in our example, C[n+4]), it replaces the previous one. The subtraction unit 24 computes the difference between the previous and the current coefficients. The divider 26 computes the step size per frequency bin,

The coefficient generator 18 finds the linear interpolated values of the missing coefficients, i.e. c[n+l], c[n+2] and c[n+3] ( restored values, marked as waved symbols). The restored value at bin n is the same as c[n] itself.

Specific examples of the principles of this invention the compression and the associated decompression units may utilize one or more of the following methods:

Lossless data compression algorithms. The LZW algorithm (Lempel-Ziv- Welch universal lossless data compression algorithm) can be applied on the crosstalk cancellation coefficients. For typical crosstalk cancellation scenarios in xDSL this method can achieve compression ratios between 0.3 and 0.6. Since this is a lossless compression no loss of performance whatsoever is caused by this method. (However, in our case we need compression factors of about 100, at least of about 10. Here LZW becomes inefficient.) A simpler form of lossless compression is a so-called sign compression method where repeated sign bits in a two's complement format are compressed by shifting towards the most significant bit. Another lossless compression method can be the Huffman encoding.

Lossy data compression algorithms. The JPEG algorithm is a lossy compression algorithm where the coefficients are mapped to a two dimensional array which is then compressed like an image. As an example the collection of the crosstalk cancellation coefficients of all the users in a single frequency bin can be viewed as a two dimensional bitmap image to be compressed by the JPEG algorithm; Typical compression factors achieved by JPEG are between 7 and 20, but the performance with application of coefficients restored after applying the JPEG decoder is in general very poor, and that performance is comparable to the performance without any cancellation of cross-talk.

2a. Decimation algorithms. Decimation by an integer factor of K is an algorithm where one out of K coefficients is stored in the coefficient memory. This is also a lossy compression method. Such methods will be further illustrated by figures 3a to 4b (with and without restoration).

3. Compression by a general lossy/lossless parametric model of the crosstalk cancellation coefficients. Such a parametric model describes the collection of crosstalk coefficients, e.g. across the frequency bins, by a function with some fixed parameters and an input vector as follows:

where k is the frequency bin index, Θ1 up to 0L are some fixed parameters and u(k) is an input vector. Compression is achievable when the fixed L parameters and the input vector u can be faithfully described by fewer bits than the bits required to represent the original crosstalk cancellation vector P m,n(k) . In this case the compression unit stores the parameters and the input vector u in the coefficient memory and the decompression unit restores the original coefficients by applying the functional F.

One specific example of the above principle, suitable for the proposed application is the following model:

P„,„ (^k) = ^A,„„ W exp(i fi>_{0 ra} k + φ_{0ιη η} + δ,_{η ιι} (h))

The model parameters are:

- A_{m n}(k) - which is a slowly varying complex envelope function for bin k; This function can be further parameterized by a polynomial in k with constant coefficients

- A sinusoidal function with constant frequency ω₀ and constant

phase p₀ for m and n modems

- A phase correction sequence 6_lllin(k) which can be further compressed by some factor K, e.g. by repeating each value K times

- k is index of bins, for example 1 -4096

- m, n - numbers of modems; "m" of a victim, "n" of a disturber

- P - value of a decompressed coefficient for cross-talk cancellation, being close to the original coefficient (P real); A decompression method associated with the decimation (compression) algorithm may be an interpolation algorithm which estimates the values of crosstalk coefficients in frequency bins outside the decimation set. Specific examples of the interpolation algorithms are, for example, linear interpolation where the values of coefficients at interior bins are obtained by a straight line drawn between two adjacent decimation points. It has been shown by the Inventors for DSM, that typical compression factors still allowable in case of further use of linear interpolation are up to a compression factor of 8. Such methods will be illustrated by exemplary Figs 4a, 4b.

A more general interpolation method is a higher order spline interpolation which fits a polynomial of order R between two adjacent decimation points. An example of using such interpolation will be shown in Fig. 5.

Figs. 3a, 3b respectively illustrate simulation results of performance degradation without interpolation, for 3 bin decimation and for 5 bin decimation (K=3, K=5).

Figs. 4a, 4b illustrate simulation results of performance degradation with linear interpolation, for 3 bin decimation and for 5 bin decimation respectively.

Fig. 5 illustrates simulation results of performance degradation in case of 64 and 128 decimation factors when cubic spline interpolation is used(for decompression of preliminarily compressed coefficients), in comparison with the case where no interpolation is applied. The graph of Fig. 5 compares the Signal to Noise Ratio (SNR) per bin on a specific line when applying a cubic spline interpolation (upon compressing the initial coefficients by decimation factors of 64 and 128), with SNR achieved when using the original crosstalk coefficients (i.e., without any decimation). In this example, the cubic spline interpolation was carried out on the magnitude (in units of dB) and phase (in units of radians) of the crosstalk cancellation coefficients. It can be seen that the performance is only marginally affected even with a compression factor of 128 (!).

Fig. 6 schematically illustrates M victim lines in an xDSL system (M=4), each being disturbed by a specific number of other disturbing lines. Victim 1 - by 10 lines, victim 2 - by 12 lines, victim 3 - by 7 lines, victim 4 - by 15 lines.

As mentioned above, N cancellation resource units are intended for cross-talk cancellation in the xDSL system. The problem is that the N resources are not sufficient for cancelling cross-talk of all disturbers in each victim line. In order to allocate the resourcesOptimaMy and dynamically, the Inventors propose :

- to allocate a number C = k*N, (say k=0.5) out of the total number of N cancellation resource units, which will be sufficient for cancellation, for example, of 7 disturbers (in this example, the minimal existing current number of disturbers is selected ) in each of the disturbers' lines;

- to apply C cancellation resource units to all M disturbed (victim) lines, (shown by the double hatching in M lines, up to the level 7)

- to distribute the remaining 0.5N resource units dynamically, between the remaining disturbed (victim) lines according to any criterion (shown by a simple hatching). For example, the remaining resources may be distributed to any newly appearing victim lines - see line 10, and to the existing victim lines which have more than a pre-selected number of disturbers to begin with.

This pre-selected number of disturbers may be, for example, the maximal number of disturbers still existing in the x-DSL system after applying the C resource units to the M disturbed lines.

Though the above description has referred only to a number of specific examples of the method and implementations of the system, it should be appreciated that other versions of the method and examples of the systems could be proposed and are to be considered part of the invention, whenever defined by the claims that follow.

Claims

Claims:

1. A method of DSM cross-talk cancellation in an xDSL system, comprising processing of an initial plurality of crosstalk cancellation coefficients; the method being characterized by

- distributing the processing of said initial plurality of crosstalk cancellation coefficients between at least two blocks, a first block comprising a memory of coefficients and a second block being separate from the first block and comprising a cross-talk cancellation processor,

- compressing the initial cross-talk coefficients before storing in the memory of coefficients,

- transferring the compressed coefficients to the second block,

- decompressing the received compressed coefficients before applying to the cross-talk cancellation processor for implementing cross-talk cancellation in the xDSL system.

2. The method according to Claim 1 , wherein the xDSL system is a VDSL system.

3. The system according to Claim 1 or 2, wherein the xDSL system is characterized by bit rates in the range between 10 and 100 Mbps. ¾

4. The method according to any one of the preceding claims, wherein the compression is performed according to one or both of the following models: lossy and lossless.

5. The method according to any one of the preceding claims, wherein the decompression is performed according to one or more of the following models: LZW compression/decompression, cubic spline interpolation.

6. A method for allocation of resources of a DSM cross-talk cancellation engine comprising N basic cross-talk cancellation resource units for handling M victim lines in an xDSL system, wherein M«N;

the method comprises:

- selecting a basic number C of said resource units, C= k*N (0<k<l),

- applying C cancellation resource units to all M victim lines,

- dynamically distributing the remaining (l -k)*N resource units between remaining and/or newly appearing victim lines.

7. The method according to any one of Claims 1 to 6, being implemented by a DSM cross-talk cancellation engine, and wherein allocation of resources of said DSM cross-talk cancellation engine is performed according to Claim 6.

8. A DSM cross-talk cancellation engine for cross-talk cancellation in xDSL lines, comprising

- a first block containing

a compression unit for compressing an initial plurality of crosstalk cancellation coefficients,

a coefficient memory for storing the compressed coefficients,

- a second block located separately from the first block and containing

a decompression unit for expanding the compressed cancellation coefficients;

a cross-talk cancellation processor for applying the decompressed coefficients to the xDSL system;

- a communication unit responsible for transferring the compressed coefficients from said memory to the second block.

9. The DSM cross-talk cancellation engine according to Claim 8, wherein the first block and the second block are remote from one another.

10. A DSM cross-talk cancellation engine comprising N basic cross-talk cancellation resource units for handling M victim lines in an xDSL system, wherein M«N;

the engine being capable of :

- selecting a basic number C of said resource units, C= k*N (0<k<l),

- applying C cancellation resource units to all M victim lines,

- dynamically distributing the remaining (l-k)*N resource units between remaining and/or newly appearing victim lines.

11. The DSM cross-talk cancellation engine according to Claim 8, comprising N basic cross-talk cancellation resource units for handling M victim lines in an xDSL system, wherein M«N;

the engine being capable of :

- selecting a basic number C of said resource units, C= k*N (0<k<l),

- applying C cancellation resource units to all M victim lines,

- dynamically distributing the remaining (l-k)*N resource units between remaining and/or newly appearing victim lines.