For the past few years there has been interest in transmitting and receiving high¬

speed data over telephone wires to individual subscribers without having to provide

higher-quality cables. High speed in this context means rates such as 1.5Mb/s (USA), 2 Mb/s (Europe). Two systems have emerged:

1) High-speed digital subscribers line (HDSL). The equipments at each end of the line are nominally indentical. They transmit and receive data simultaneously at the same rates, ie full duplex. It is expected that two or three lines will be needed to achieve the desired capacity, each signalling at 50% or 33% respectively relative to a single-line implementation. 2) Asymmetric digital subscribers line (ADSL). On the basis that most subscribers do not have the means or the need to generate such large amounts of data, the channel from subscriber to exchange is reduced to a low-speed signalling circuit, typically 16kb/s, while retaining a high-speed channel from exchange to subscriber. An obvious application for this is the transmission of a selectable broadcast-quality video signal. An ADSL system has a much longer

transmission range than an HDSL system of gthe same baud rate as will be

discussed later.

Transmission standards are still under discussion. For HDSL the main contenders are pulse-amplitude modulation, probably 2B1Q as used on ISDN, or Quadrature Amplitude Modulation (QAM) (and a recent variant termed carrierless amplitude/phase modulation (CAP). For ADSL it is most likely that the American

National Standards Institute (ANSI) will specify discrete multiple tone (DMT).

DMT is fully described in tutorial paper T1 E1.4/91 -157 "A Multicarrier Primer"

J.M. Cioffi, Amati Communications Corporation presented to Standards Committee Tϊ-

Telecommunications, of ANSI and imported herein by reference, but its basic principle

is as follows:

Successive blocks of the high-speed data stream are demultiplexed into a large

number of channels (typically 256). Each channel may be several bits wide and

different channels may be of different widths. Each channel is then QAM-modulated on

to one of 256 carters. The carrier frequencies are all harmonics of the lowest one. This

composite spectrum is transmitted for the duration of the data block length. Viewed another way, each subset of bits within the data block describes the amplitude and phase of a tone burst at a frequency specific to that group. The tone burst lasts for a data-block duration, and the tones are a simple harmonic series (Figures 1 & 2).

The claimed advantages of DMT lie not so much in its transmission performance or distance under normal conditions but in its implementation and flexibility: (a) the composite spectrum is generated and decoded by a fast Fourier transform (FFT) and

its inverse, without the need for generating individual carriers; (b) since each carrier occupies a small bandwidth it suffers little frequency distortion so

removing the need for complicated line equalisers; (c) by monitoring the error rate of each carrier, the receiver can instruct the transmitter to preferentially use those parts of the spectrum with the best performance.

HDSL is limited by near-end crosstalk (NEXT). To the attenuated signal from the distant transmitter is added an unwanted signal that originates from transmitters physically close to the receiver and which is coupled via inter-wire capacitance. With

conventional coding, the wanted signal and the NEXT occupy the same frequencies.

The problem is greater at the exchange end where lines are concentrated than it is at

the subscriber's end.

With ADSL the subscriber's receiver only receives NEXT from low-speed sources

so the limiting factor is noise from unwanted high-speed transmitters at the exchange

(far-end crosstalk). The noise source is thus the same as for HDSL but the noise is

attenuated by the line giving an ADSL system better Signal/Noise ratio (SNR).

In full-duplex operation the received signal may contain an element of the signal transmitted on the same wire-pair that is caused by impedance discontinuities in the transmission medium or imbalance in a hybrid circuit. In many data systems it is necessary to include an echo canceller to remove this interfering signal.

According to the present invention there is provided a communications system employing a high-speed digital link with discrete multiple tone (DMT) wherein the outward paths from a central transmitter/receiver to further transmitter/receivers receiving signals therefrom use one set of carrier frequency channels and the return signal paths use a different set of carrier frequency channels.

The communciations system may be a telecommunications system, wherein the central transmitter/receiver is an exchange and the further transmitter/receivers are

subscribers connected to the exchange. The present invention will now be described by way of example, with reference

to the accompanying drawings, in which:-

Figure 1 shows a diagrammatic representation of a DMT transmitter coder;

Figure 2 shows a conventional arrangement of DMT 'go' and 'return' channels;

Figures 3 and 4 show possible arrangements of DMT 'go' and 'return' channels

according to the present invention;

Figure 5 shows diagrammatically the forms of crosstalk in a telecommunications

system according to the present invention;

Figure 6 shows diagrammatically an n-sample post-cursor attached to a 256-

sample data frame;

Figure 7 illustrates the alignment of data and post-cursor with crosstalk;

Figures 8 and 9 illustrate the alignment of transmitted data according to two

alternative methods.

While the present invention is described with reference to systems using copper wire-pairs, the invention has general applicability to any arrangement where the signals from co-sited transmitters can interfere with one another, for example, infra-red links within a room or in close proximity to one another.

In an HDSL system using DMT wherein all routes from exchange to subscribers use one set of carrier frequencies and the routes from subscribers to exchange use a different set, NEXT and echos occupy different frequencies from the desired received signal (Figures 3, 5). At first sight it would appear that a comb filter would be needed to remove the NEXT before the signal is applied to the FFT, but it is simpler to omit the filter; the noise then appears as small signals at those outputs from the FFT that

correspond to the interfering carriers. Since it is known that the transmitter does not

use those channels, the noise may simply be ignored.

With NEXT eliminated the limiting factor becomes far-end crosstalk (FEXT) as with ASDL. A penalty of at least 3dB exists relative to ADSL since the data rate

applied to each carrier must be doubled, but the performance should greatly exceed

HDSL using Phase Amplitude Modulation (PAM) or QAM.

For a signal to be properly detected the receiver FFT function must be carried out on the correct block of data. If the FFT window is not aligned with the data block

then data from an adjacent frame will be included while data from the correct frame will

be omitted. Misalignment of the frame in this way causes broadband distortion of the

received signal and hence a transfer of signal from one carrier frequency to another. The crosstalk component of a received signal will therefore only be fully suppressed if the FFT operation is synchronised with the crosstalk. In general this will be inconsistent with synchronising with the received data, and repeating part of the frame makes alignment unnecessary.

The phase of a data frame can be adjusted provided ail or part of the frame is repeated. If the first M samples of an N-sample frame are appended to the frame as a post-cursor, then the last N samples of the new frame are a rotated form of the original and can be demodulated without adding distortion (Figures 6, 7); the only effect is to add a time-delay. If the duration of the M samples equals or exceeds twice the propagation delay of the cable and the frames are correctly aligned then NEXT will be properly suppressed; note that all co-sited transmitters must be synchronised for their crosstalk to be cancelled. Referring to Figure 8, frame b is transmitted to a subscriber

with a delay T _d . The subscriber's frame c is transmitted in synchronism

with the locally-received data and arrives at the exchange with a total deay of 2T _d . To c is added the crosstalk from frame a on a nearby wire-pair. The window for the FFT function performed on received frame c starts at the beginning of frame c and includes

includes the last part of crosstalk frame a and the beginning of its post-cursor. When

an FFT is performed on the noisy signal there is no transference of crosstalk energy

from one carrier frequency to another and the crosstalk is rejected. Note that the length

M of the post-cursor must be the same for all transmitters. Regarding echoes, a

reflection from the hybrid at the transmitter will be suppressed, but echoes originating further down the line may not be aligned with crosstalk frame a and will leak through.

This method of synchronisation is only possible at one end of the line. It should

therefore be used at the exchange, where the crosstalk is greater ( ie alignment of the

transmitted and received frames should be carried out at the subscribr's end).

NEXT at the subscriber's end will arise where the cable converges with those from other subscribers; it will be subject to a certain amount of cable attenuation and it will be advanced or retarded relative to the received signal depending on the various cable lengths. Since the post-cursor is designed for the longest possible delay, advanced crosstalk can always be removed but retarded crosstalk requires the use of a precursor in a similar fashion to that described previously. However, DMT may be implemented using a precursor so further coding may not be needed.

The frames transmitted from and received by the exchange may be synchronised by a second method not requiring a post-cursor. Instead of the subscribr's equipment

maintaining its transmitted and received frames in complete alignment it may advance

the transmissions by time 2T _d thereby compensating for the

cable delay (Figure 9). It is necessary for the exchange to inform the subscriber's

equipment of the amount of slip required, which might be achieved by either.

1 ) the subscriber sending a repeated alignment frame. The exchange computes

the required slip and informs the subscriber accordingly.

2) the exchange detecting that the subscriber's data cannot be satisfactorily

demodulated and sending a command 'slip' until the correct delay is reached.

Each time the subscriber's equipment receives 'slip' it adjusts the alignment of

its transmissions by a suitable increment.

The measure of the cable delay, together with other information such as the state

of an adaptive equaliser and the optimum assignment of data among the various

channels, may be stored in the receiver between calls to reduce the set-up time.

Regarding channel selection, the most obvious implementation is to use the even-numbered channels for one path and the odd channels for the other. However, they may be grouped in any way that gives adequate capacity (eg Figure 4). In the limit, one path uses all the high channels and the other all the low channels which is then similar to V22bis standard for data modem operation. Further, the system may be made asymmetric by allocating more channels to one path than the other.