2.5De beschrijving van EP 456 bevat onder meer de volgende passages:
BACKGROUND OF THE INVENTION
[0001] This invention relates generally to methods, systems and apparatus for managing digital communications systems. More specifically, this invention relates to dynamically controlling system parameters that affect performance in communication systems such as DSL systems.
Description of Related Art
[0002] The present invention refers to digital communication systems where the transmission medium typically is copper wiring. Most commonly the copper wiring consists of twisted pairs (also referred to as "lines" or "loops") categorized according to several manufacturing specifications (for example AWG-26, AWG-24, CAT-3, CAT-5, CAT-6). Typical communication systems making use of copper wiring include Digital Subscriber Line (DSL) systems, such as ISDN, HDSL, ADSL and VDSL, and Local Area Networks (LAN), such as Ethernet. A transceiver (for example, a user modem) is situated at each end of the communications line that incorporates the copper wiring.
[0003] Existing phone lines typically are "bundled" in some way. "Bundling" several pairs (in a binder or otherwise) can improve service to a single user or permit service for multiple users. (…)
[0005] The bundling of twisted pairs arises either out of necessity (for example, the existing telephone loop infrastructure) or because of the benefits of improved performance (for example, 1000-BaseT Ethernet). In either case however, communications in these settings suffer from interference arising from electromagnetic coupling between neighboring pairs, referred to as "crosstalk" interference. This means that any signal received by a modem at the end of a twisted pair generally contains not only the transmitted signal of the specific pair (which itself is likely distorted to some extent), but also distorted signals transmitted on neighboring pairs. It is apparent, therefore, that the transmission characteristics of a specific pair (for example, the pair's transmitted power) can materially influence communication on a neighboring pair due to the induced crosstalk. Therefore, transmissions on neighboring pairs (especially those belonging to a bundle or sharing the same binder) are coupled in certain ways. The interfering signals are commonly treated as noise. However, crosstalk can be identified in some situations. (See United States Serial No. 09/788,267 ). If crosstalk coupling functions can be identified, it may be possible to remove the crosstalk interference. (….)
[0009] One of the shortcomings of current multi-user communication systems is power control. In typical communication systems, which are interference-limited, each user's performance depends not only on its own power allocation, but also on the power allocation of all other users. Consequently, the system design generally involves important performance trade-offs among different users. The DSL environment can be considered a multi-user system, which would benefit from an advanced power allocation scheme that maximizes or allows selection from most or all of the achievable data rates for multiple DSL modems in the presence of mutual interference.
[0010] As mentioned above, DSL technology provides high speed data services via ordinary telephone copper pairs. The DSL environment is considered a multi-user environment because telephone lines from different users are bundled together on the way from the central office, and different lines in the bundle frequently create crosstalk into each other. Such crosstalk can be the dominant noise source in a loop. However, early DSL systems such as ADSL and HDSL are designed as single-user systems. Although single-user systems are considerably easier to design, an actual multi-user system design can realize much higher data rates than those of single-user system designs.
[0011] As the demand for higher data rates increases and communication systems move toward higher frequency bands, where the crosstalk problem is more pronounced, spectral compatibility and power control are central issues. This is especially true for VDSL, where frequencies up to 20MHz can be used.
[0012] Power control in DSL systems differs from power control in wireless systems because, although the DSL environment varies from loop to loop, it does not vary over time. Since fading and mobility are not issues, the assumption of perfect channel knowledge is reasonable. This allows the implementation of sophisticated centralized power control schemes. On the other hand, unlike the wireless situation where flat fading can often be assumed, the DSL loops are severely frequency selective. Thus, any advanced power allocation scheme needs to consider not only the total amount of power allocated for each user, but also the allocation of power in each frequency. In particular, VDSL systems suffer from a near-far problem when two transmitters located at different distances from the central offices both attempt to communicate with the central office. When one transmitter is much closer to the central office than the other, the interference due to the closer transmitter often overwhelms the signal from the farther transmitter.
[0013] DSL modems use frequencies above the traditional voice band to carry highspeed data. To combat intersymbol interference in the severely frequency selective telephone channel, DSL transmission uses Discrete Multitone (DMT) modulation, which divides the frequency band into a large number of sub-channels and lets each sub-channel carry a separate data stream. The use of DMT modulation allows implementation of arbitrary power allocation in each frequency tone, allowing spectral shaping.
[0014] As shown in Figure 4 , a DSL bundle 410 can consist of a number of subscriber lines 412 bundled together which, due to their close proximity, generate crosstalk. Near-end crosstalk (NEXT) 414 refers to crosstalk created by transmitters located on the same side as the receiver. Far-end crosstalk (FEXT) 416 refers to crosstalk created by transmitters located on the opposite side. NEXT typically is much larger than FEXT. The examples of the present invention presented herein use frequency duplexed systems for illustrative purposes.
(...)
[0027] Crosstalk coupling is strongest among the twisted-pairs in a binder group. Therefore, eliminating or mitigating self-FEXT within a binder group has the biggest performance benefit. "Unbundled" lines of different service providers may share a binder group which can result in the absence of collocation of the CO transceiver equipment.(…)
[0028] The crosstalk problem has been addressed before with some shortcomings. For example, in some systems, MIMO Minimum-Mean-Square-Error (MMSE) linear equalizers were derived. Another prior method employs the singular value decomposition to achieve crosstalk cancellation assuming co-location of both transmitters and receivers. Other earlier methods include "wider than Nyquist" transmitters which were shown to provide performance advantages compared to "Nyquist-limited" ones, and crosstalk cyclostationarity (induced by transmitter synchronization) combined with oversampling which were shown to result in higher SNR values.
[0029] The paper "A Multi-user Precoding Scheme achieving Crosstalk Cancellation with Application to DSL systems", by [uitvinder 1] and [uitvinder 2] discloses a precoding method for DMT modulation combined with an optimal distribution of energy for a system with multiple users.
[0030] None of the earlier methods or systems provided a relatively simple and effective reduction in crosstalk interference in wireline communication systems. However, vectored transmission (as defined in this invention) can achieve a high degree of crosstalk reduction without unreasonable complexity. Moreover, the use of vectored transmission can accommodate the approaching architectural changes coming to DSL service as well as providing an opportunity for dynamic system management which can overcome the shortcomings of prior systems and methods.
BRIEF SUMMARY OF THE INVENTION
[0031] The present invention relates to a method and apparatus for dynamically controlling a digital communication system, such as a DSL system, according to the accompanying claims.
[0032] Further details and advantages of the invention are provided in the following Detailed Description and the associated Figures.
(…)
[0074] In the following example, vectored transmission for DSL systems is explained. (…)
Channel Model and DMT Transmission
(…)
[0076] Two fundamental assumptions are used in connection with this discussion of a preferred embodiment. First, all users employ block transmission with a cyclic prefix (CP) of at least length ν. Also, block transmission and reception at the CO/ONU are synchronized as illustrated in the timing diagram of Figure 9.
(…)
[0080] Applying Discrete Fourier Transform (DFT) modulation, which is known in the art, an Inverse Discrete Fourier Transform (IDFT) operation is performed on each transmitted data block (prior to appending the CP), and a DFT operation is performed on each received data block (after discarding the CP), thus yielding a channel description where the samples are stacked in groups corresponding to users, and each of the groups contains samples corresponding to tones. It is desirable to reorganize these samples for further processing by stacking in groups corresponding to tones, where each group contains samples corresponding to different users. To this end, a permutation matrix P having NL rows and NL columns is defined, which is composed of the N x N blocks P
ijwhere
i,j=1,...,L. The block P
ijcontains all zeros, except for a one at position
(j,i). When matrix P is right-multiplied with a vector of size NL, the elements of P are re-ordered from L groups of N components into N groups of L components. Also, note that P-1=P* =P. Applying this reordering operation to both the transmitter and the receiver samples, yields:
[0081] Therefore,
Zi,
Uiand
Nicontain the received samples, transmitted symbols and noise samples of all users corresponding to tone i, and T
ifully characterizes MIMO transmission within tone
i. In the following, a distinction between upstream and downstream will be made by adopting the notation
Ti,upand
Ti,down.
[0082] Equation (22) shows that crosstalk cancellation can be performed independently in each tone. Therefore, as explained in more detail below, an array of canceller blocks can be employed at the CO/ONU to remove crosstalk within each tone for upstream communication. Similarly, precoder blocks can be used at the CO/ONU to pre-distort the transmitted signals within each tone, so that signals received at the CPE are crosstalk-free. Determining the parameters of the canceller/precoder blocks relies on perfect knowledge of the channel matrix and noise covariance matrix at the CO/ONU. This assumption is reasonable for DSL, since the twisted pair channels are stationary, and systems can afford training-based channel identification during initialization.
(…)
Crosstalk Cancellation via QR Decomposition
[0085] Starting with Equation (22), the methods to remove crosstalk within each tone are described first for upstream and then for downstream communication. In the following, the matrices
Ti,upand
Ti,downare assumed to be non-singular (the justification for this assumption and the consequences of ill-conditioning are discussed below).
[0086] For upstream transmission, the co-location of the CO/ONU transceiver equipment gives the opportunity to perform joint signal processing of the received samples. The computation of the QR decomposition of matrix
Ti,upyields:
where Q
iis a unitary matrix and R
iis an upper triangular matrix. If the received samples are
"rotated/reflected" by
then Equation (22) becomes:
where
has an identity covariance matrix. Since R
iis upper triangular and
has
uncorrelated components, the input U
ican be recovered by back-substitution combined with
symbol-by-symbol detection. Thus, as seen in Figure 11, a decision feedback structure 1100 is created with the feedforward matrix module 1110 using
and the feedback matrix module 1120 using
I – Ri. The detection of the
kthelement of U
iis expressed as:
where
is the (
k, j) element of R
i. Assuming that the previous decisions are correct, crosstalk is completely cancelled, and L "parallel" channels are created within each tone. The operations described above can be used to define a preferred canceller block corresponding to a single tone, which is shown in Figure 11. Combining the canceller blocks of all tones, and taking into account DMT transmission, a system 1200 for upstream vectored DMT transmission is shown in Figure 12. The transmitters 1210-1 through 1210-L send their respective signals through channel 1220. The receivers 1230-1 through 1230-L receive the signals from channel 1220 and process the received signals using canceller blocks 1240-1 through 1240-L which, in the preferred embodiment, resemble the block of Figure 11.
[0087] For downstream transmission in the preferred embodiment, joint signal processing of the transmitted symbols is used. The QR decomposition of
results in:
where again Q
iis a unitary matrix and R
iis an upper triangular matrix. Assuming that the symbols are "rotated/reflected" by
prior to being transmitted:
crosstalk-free reception is achieved, where the transmitted symbols in tone i are the elements of Ũ
i. (…)
And M
i,kis the constellation size of user
kon tone
i, while
dis the constellation point spacing. (…).
These operations result in:
which implies crosstalk-free reception. The preferred MIMO precoder described above corresponds to a single tone and is shown in Figure 13. Combining the precoders of all tones and including the DMT transmitters and receivers, the vectored DMT system for downstream transmission is shown in Figure 14. This system resembles the system of Figure 12, except that signals are "preprocessed" with precoders 1420-1 through 1420-L before being sent by the system transmitters 1410-1 through 1410-L, respectively.
[0088] Assuming that transmit and receive filtering at the CO/ONU and at the CPE is identical, and noise within a tone has the same statistics for all users, the reciprocity property for twisted pair transmission implies that
Ti,up=
Ti,down. In that case, Equations (23) and (27) give the QR decomposition of the same matrix.
[0089] For the upstream channel, regardless of the loop topology, the diagonal element of a column of Ti is larger in magnitude than the off-diagonal elements of the same column. This occurs because, in upstream transmission, the crosstalk coupled signal originating from a specific transmitter can never exceed the "directly" received signal of the same transmitter, and typically the magnitude difference is more than 20 dB. The insertion loss of a signal is always smaller than the coupling loss that it experiences when it propagates into a neighboring pair.
[0090] Visualizing the columns of T
iin vector space, it is seen that the columns are almost orthogonal to each other, which implies that Q
iis close to being an identity matrix. Thus, the magnitudes of the diagonal elements of R
ido not differ significantly from those of the diagonal elements of T
i, which indicates that QR cancellation performs almost as good as crosstalk removal. (…)
[0091] The preceding discussion concerning upstream transmission can be readily extended to downstream transmission by starting with the observation that the crosstalk signals at a specific receiver can never exceed the magnitude of a "directly" received signal. Alternatively, the same conclusions can be reached by using the transpose relationship between the upstream and downstream channel matrices.
[0092] The computational cost incurred by the QR cancellation is decomposed into the cost of the QR decompositions and the cost associated with signal processing. DSL channels are stationary, so the QR decompositions need to be computed infrequently (preferably during initialization). (…) On the other hand, the real-time computational burden due to the canceller and precoder blocks cannot be reduced. In a straightforward implementation, the operations dominating the total cost are those of Equations (24) and (28).
[0093] Although the assumption of perfect channel matrix knowledge is reasonable in the given environment, it is still worth briefly considering the effects of channel estimation errors. (…) A similar analysis can be applied for downstream communication, but modulo arithmetic complicates the expressions. Ignoring the modulo operations, the effects of the estimation errors can be separated into a detection bias term and a residual crosstalk term. [0094] The results of this analysis reveal that the impact of channel estimation errors is aggravated when any of the diagonal elements of R̂
iare small. Although channel matrix singularity is almost impossible in the DSL environment, an ill-conditioned channel (implying small diagonal elements) cannot be ruled out, thus increasing the impact of channel estimation errors and posing several computational problems. Such cases arise in high frequencies (for example, in loop topologies that have extreme loop length differences) or in the presence of bridged taps. Nevertheless, the energy allocation algorithms discussed below prevent the occurrence of such phenomena by not allowing transmission in frequencies where the diagonal elements of Ri are small.
[0095] As seen above, the elimination of crosstalk in the signals of a system will substantially improve performance of the system. Optimizing energy allocations in the system, when taken in conjunction with the crosstalk elimination likewise improves system performance. Also, as noted above, appropriate energy allocation can help avoid problems resulting from the impact of estimation errors in ill-conditioned channels.
Transmission Optimization
[0096] "Transmission optimization" as used in the following example will refer to maximization of a weighted data rate sum. However, in the broadest sense of the present invention, the term "optimization" is not so limited. Optimization may also mean determining the maximum rates available and allocating or provisioning available resources (including data rates for various users) within a digital communication system.
[0097] The methods disclosed in the following discussion concern energy allocation in frequency generally, energy allocation in frequency while observing constraints on induced crosstalk, and energy allocation combined with upstream/downstream frequency selection.
Energy Allocation in General
[0098] The optimization objective is the maximization of the weighted sum of the data rates of all users:
Where ak ≥ 0 is the weight assigned to the
kthuser, and R
kis the achievable data rate of the
kthuser, which may refer to either the upstream or the downstream direction. In order to compute the data rate, an appropriate known gap approximation is employed. Taking into account the fact that vectoring essentially "diagonalizes" the channel (and assuming no error propagation in the upstream direction), the upstream and downstream achievable rates are obtained:
where r [lees ‘Γ’, zie WO 008, p. 36, hof] is defined as the transmission gap, and depends on
the probability of error requirement, the coding gain and the required martin [lees: ‘margin’, zie WO 008, p. 36, hof]. Also,
Nupand
Ndownare the sets of upstream and downstream tone
indices correspondingly (…). Error propagation effects are generally limited since DSL systems operate at very small probabilities of error.
[0099] The parameters with respect to which optimization takes place are
for upstream and
for downstream transmission. These parameters are constrained by limits on the transmitted energy. In upstream transmission, the total transmit energy is constrained by:
Where
is the energy of (U
i)
kin Equation (25), and ε
k,upis the maximum allowed upstream transmitted energy of user
k. Since
it is deducted that:
In downstream transmission, the total transmit energy constraint is expressed as:
Where
is the energy of (U
i)
kin Equation (21) [lees: ‘(28)’, hof], and ε
k,downis the
maximum allowed downstream transmitted energy of user
k. Unfortunately, this constraint does not translate directly to a constraint for
due to non-linear precoding.
[0100] However, simulation results for extreme loop topologies indicate that use of the preferred precoder described above does not result in considerable correlation between the transmitted signals of different users. It is reasonable to assume that this result holds generally, since the simulated loops correspond to a worst-case situation with regard to the crosstalk coupling.
[0101] Therefore, the approximation
is made and Equation (38) for downstream becomes:
[0102] With this in mind, it is seen that the energy allocation problem of Equation (33) becomes independent for each user, and thus the ak weights are irrelevant in this scenario. The optimization problem for each transmission direction is broken into
kwaterfilling problems (…)
Solutions to these problems can be derived using known techniques. The resulting transmission spectra are optimal in the context of vectored DMT.