The public switched telephone network (PSTN) provides the most widely available form of electronic communication for most individuals and businesses. Structured originally to provide voice communication with its consequent narrow bandwidth requirements, the PSTN increasingly relies on digital systems to meet the service demand. However, a major limiting factor in the ability to implement high rate digital transmission has been the subscriber loop between the telephone central office (CO) and the remote subscriber terminal (RT). This loop most commonly comprises a single pair of twisted wires which are well suited to carrying low- frequency voice communications for which a bandwidth of 0-4 kHz is quite adequate, but which do not really accommodate broadband communications (i. e. bandwidths on the

order of hundreds of kilohertz or more) without adopting new techniques for communication.

One approach to this problem has been the development of discrete multitone digital subscriber line technology (DMT DSL). This and other forms of discrete multitone digital subscriber line technology (such as ADSL, HDSL, etc. ) will commonly be referred to hereinafter generically as"DSL technology"or simply as"DSL".

In DSL technology, communication over the local subscriber loop between the central office and the remote subscriber terminal is accomplished by modulating the data to be transmitted onto a multiplicity of discrete frequency carriers which are summed together and then transmitted over the subscriber loop. Individually, the carriers form discrete, non-overlapping communication subchannels of limited bandwidth ; collectively, they form what is effectively a broadband communications channel. At the receiver end, the carriers are demodulated and the data is recovered from them.

The data symbols that are transmitted over each subchannel carry a number of bits that may vary from subchannel to subchannel, dependent on the signal-to-noise ratio (which is simply referred to as"SNR"in the following) of the subchannel. The number of bits that can be accommodated under specific communication conditions is known as the"bit allocation"of the subchannel, and is calculated for each subchannel in a known manner as a function of the measured SNR of the subchannel and the bit error rate associated with it. The bit allocation information is stored at both ends of the communication pair link for use in defining the number of bits to be used on the respective subchannels in transmitting data to a particular receiver. Other subchannel parameters such as subchannel gains, time and frequency domain equalizer

coefficients, and other characteristics may also be stored to aid in defining the subchannel.

Although DSL communication systems do in fact provide a significantly increased bandwidth for dated communications, special precautions are required to avoid interference with ordinary voice communications and associated signaling that may be taking place over the subscriber line at the same time that the broadband data is being carried. These voice communications and their associated signaling, commonly referred to as"plain old telephone service"or"POTS", presently are isolated from the data communications by modulating the data communications onto frequencies that are higher than those used for POTS. The data communications and POTS signals are thereafter separately retrieved by appropriate demodulation and filtering. The filters which separate the data communications and the POTS are commonly referred to as"splitters". The voice and data communications must be separated at both the central office and the remote subscriber terminal, and thus splitters must be installed at both locations.

U. S. patent application no. 20030026282, on which the above explanation regarding the DSL technology is mainly based, discloses a splitterless multicarrier modem for use in DSL communications, which transmits and receives data over the local subscriber loop in common with voice information over the loop, while avoiding the need for voice/data splitters.

The modem responds to disruptions associated with "disturbance events"such as on-hook to off-hook transitions and the like by rapidly switching between pre-stored channel parameter control sets defining communications over the loop under varying conditions. This U. S. patent application has a filing and publication date later than the priority application underlying the present patent application.

However, this publication does not deal with the problem of noise being induced by voice communications and affecting the

SNR when retrieving the information from the data communication signal.

For example, ADSL data communication services can be deployed with an ISDN ("integrated services digital network") basic access (BA) on the same subscriber line using frequency- division duplexing. The presence of the ISDN signal introduces in-band noise for the respective ADSL system implemented in an ADSL modem which can cause performance degradation. The in-band noise (which is referred to as"ISDN noise"in the following) depends on the characteristics of the respective splitter and the receive-path filtering of the ADSL system. In an ideal situation, a particular training connection of an ADSL modem is established with the ISDN noise being present throughout the training phase. Such a training connection is established during the set-up procedure of the ADSL modem connection and helps to check the connection properties and, in dependence thereon, to adjust the transmission settings both at the central office side and at the subscriber terminal side. In particular, the training phase allows to obtain information on the ISDN noise which may be present on the subscriber line, and thereafter the data information contained in the ADSL data signal can be retrieved by appropriate bit loading with the channel estimation being based on the complete information on the ISDN noise.

However, in practice, ISDN noise might not be present for the training, but may be turned on only later during the normal operation of the ADSL system. If the training is performed without considering the ISDN noise, the resulting bit loading operation might be too optimistic for the connection. The disturbances from the ISDN noise can cause CRCs ("cycle redundancy check") and link failure.

Therefore, the object underlying the present invention is to provide a method and a corresponding apparatus for retrieving

information, in particular data information, transmitted with a signal, in particular a data communication signal, over a signal channel, in particular a communication channel of a communication network, by taking into account noise induced by a further signal, in particular a voice communication signal, transmitted over the same signal channel at the same time, even if no such noise is present for measuring it during a training phase so as to allow a reliable retrieval or recovery of the aforesaid information if the noise may be turned on only after the respective connection is established.

According to the present invention, this object is achieved by a method for retrieving information transmitted with a signal over a signal channel as defined in independent claim 1 and a signal processing apparatus as defined in independent claim 27. The dependent claims define preferred and advantageous embodiments of the present invention.

According to the present invention, it is proposed that noise affecting a first signal and caused by a second signal being transmitted over the same signal channel or signal line and at the same time as the first signal, but in a different frequency range, is estimated so as to allow to retrieve the information transmitted with the first signal by taking into account the estimated noise caused by the second signal.

The present invention is in particular applicable to fixed- line or wired communication networks, and the first signal is preferably a data communication signal, for example an ADSL signal or a DSL signal in general. The second signal is preferably a voice communication signal, for example an ISDN signal. As a matter of course, the present invention is not limited to DSL-ISDN-systems, but may be applied to communication networks and even to signal transmission in general, where a plurality of signals are transmitted at the same time over one and the same physical signal channel using

frequency-division multiplexing and where the recovery of information transmitted with one of these signals is disturbed by noise caused by another one of the signals.

However, in the following, the invention will be explained with reference to the preferred field of application, namely the use in a communication network, the aforesaid first and second signals being communication signals transmitted over a communication channel.

Before estimating the noise, it is preferably detected whether noise caused by the second communication signal is present and can be measured so as to be able to conduct a respective training phase. If such noise is present, the regular training procedure may be performed. However, if such noise is not present, then in accordance with the present invention the effect of the induced noise on the first communication signal, in particular on the measurement of the SNR thereof, is estimated. This estimation is then used for adjusting the respective communication apparatus, which is for example a DSL modem, for retrieving the information from the first communication signal by taking into account the estimated noise.

According to a preferred embodiment, a pre-measured and pre- stored noise spectrum is used and shaped with the filter components of the respective communication apparatus (modem) before it is added to the total noise. The total noise is then used to compute the SNR, in particular the SNR for each tone.

It is required to conduct the pre-measurement of this noise spectrum only once, with noise being present on the communication channel, whereby this pre-measured or pre- estimated noise spectrum may be determined using averaging periodograms, and the pre-estimated noise spectrum may then be recorded and saved along with the corresponding settings

of the respective communication apparatus used for this measurement.

During the normal operation of the communication apparatus, the average input signal power may be determined when transmitting a test signal, for example the periodic REVERB signal defined for ADSL systems. Then, the settings of the respective communication apparatus are preferably adjusted in dependence on this average input signal power. Furthermore, by using a further test signal, for example the random MEDLEY signal defined for ADSL systems, the noise spectrum for this further test signal may be estimated, and a shaped noise spectrum may be computed. In an ADSL modem, this can be done by shaping the noise spectrum with the combined frequency response of the time domain equalizer (TDQ) and frequency domain equalizer (FDQ), which are used in every regular ADSL modem. Finally, according to the preferred embodiment, the Medley SNR is calculated for every tone using the shaped noise spectrum so as to be able to retrieve very reliably the transmitted information.

The present invention prevents CRCs and link failure when no noise signal is present after the connection is established.

It achieves as good data rates as those with noise being turned on during a training phase and is robust for different line conditions.

In the following, with reference to the accompanying drawings, the present invention will be explained and described in more detail referring to preferred embodiments of the present invention.

Figure 1 shows schematically the transmission of an ISDN communication signal and an ADSL communication signal between a central office and a subscriber terminal using a communication apparatus in the form of an ADSL modem according to a preferred embodiment of the present invention,

Figure 2 shows a flow chart illustrating a method for ISDN noise estimation according to a preferred embodiment of the present invention, and Figures 3-5 show charts for illustrating the results and advantages which can be achieved with the aid of the present invention.

Figure 1 schematically shows a communication between a central office (CO) 1 and a remote subscriber terminal (RT) of a fixed-line communication network. The remote subscriber terminal is associated with a splitter 2 being connected to an ISDN device 3, for example an ISDN telephone, and an ADSL modem 4 being connected to an input/output device 7 for inputting and outputting respective data or information.

Between the central office 1 and the splitter 2, the ISDN communication signal and the ADSL communication signal are transmitted in both directions at the same time on the same communication channel, that is to say on the same subscriber line, using frequency-division duplexing. The splitter 2 extracts the ISDN communication signal from the composite signal received from the central office 1 and forwards the same to the ISDN device 3. Furthermore, the splitter 2 extracts the ADSL communication signal from the composite signal and forwards the same to the ADSL modem 4. Vice versa the splitter 2 receives the ISDN communication signal from the ISDN device 3 and the ADSL communication signal from the ADSL modem 4 and combines both communication signals to the composite communication signal so as to transmit it to the central office 1. As the ISDN communication signal and the ADSL communication signal are transmitted over the same subscriber line using frequency-division duplexing, both communication signals are transmitted in frequency ranges which do not overlap one another. In particular,. the ADSL communication signal is transmitted in a frequency range

which is a multiple of the frequency range of the ISDN communication signal.

As can be taken from Figure 1, the ADSL modem 4 comprises means 5 for estimating and detecting ISDN noise present in the ADSL communication signal and for adjusting settings of the ADSL modem, in particular of means 6 which are provided for retrieving and further processing data contained in the ADSL communication signal, depending on the ISDN noise estimated by the means 5. In the following, the means 5 are referred to as ISDN noise estimation means, and the means 6 are referred to as data retrieval means. Of course, both means 5,6 may also be provided in the form of a single component in the ADSL modem 4.

As a matter of course, an ADSL modem 4 of the type shown in Figure 1 may also be provided in the central office 1.

The ISDN noise estimation means 5 are designed to predict or estimate the influence of the ISDN noise on the training process described above, even if ISDN noise is not present.

In particular, the ISDN noise estimation means 5 perform an algorithm which conducts an ISDN noise detection. If no ISDN is detected, the algorithm estimates the effect of the ISDN induced noise on the measurement of the SNR. This estimation is then used to adjust the SNR of the data retrieval means 6 for bit loading. In particular, the goal is to match the adjusted SNR with the SNR trained during the training process with ISDN noise being present in order to avoid overloading.

It should be noted that the ISDN noise estimation means 5 estimate the effect of the ISDN noise on the measurement of the SNR for every tone of the ADSL communication signal, and this estimation is then used to adjust the SNR for every tone of the ADSL communication signal for bit loading.

A preferred embodiment of the algorithm performed by the ISDN noise estimation means 5 is shown in Figure 2 in the form of

a flow chart. The algorithm can be subdivided into three main phases, namely a phase comprising steps 100,110 for pre- estimation of an input ISDN noise spectrum, a phase comprising step 200 for ISDN noise detection, and a phase comprising steps 300-340 for conducting ISDN noise estimation and adjusting the settings of the data retrieval means 6 of Figure 1. Furthermore, as shown in Figure 2, the individual steps of the algorithm can also be subdivided in terms of the respective ADSL signal conditions, that is to say depending on the kind of signals being present during carrying out the respective steps, which will be described in more detail in the following.

The steps 100,110 need not be performed continuously during the normal operation of the splitter 2 and the ADSL modem 4.

Instead, it is only required to conduct these steps once for a particular combination of the splitter 2 and the ADSL modem 4, and the results obtained during these steps can be saved in the ADSL modem 4 or in the ISDN noise estimation means 5 so as to be able to access this information during the normal operation of the ADSL modem 4. For example, the steps 100, 110 can be performed at the beginning of a start-up or initialization phase of the ADSL modem 4.

In step 100, the ISDN noise estimation means 5 of the ADSL modem 4 estimate. the ISDN noise spectrum using averaging periodograms Nk2 with ISDN noise being present at the input of the ADSL modem 4, where k designates the k-th tone. In particular, the step 100 (as well as the step 110) can be performed during the so-called CQUIET2 and QUIET2 mode of the central office and the remote subscriber terminal, respectively, which is defined for ADSL communication. The pre-estimated noise spectrum Nk2 is recorded and saved along with the respective settings, in particular the settings of a programmable gain amplifier (PGA) included in the ISDN noise estimation means 5 and/or the data retrieval means (6), used for this measurement. It should be noted that post-processing

of the pre-estimated noise spectrum may be necessary as follows so as to achieve optimal performance: <BR> <BR> <BR> <BR> <BR> 2<BR> <BR> (1) Nk = Nk gk The value ek has to be determined experimentally for every tone.

In step 110, the average power PISDN of the ISDN input signal is computed and recorded, and a threshold for ISDN noise detection is set to be one half of PISDN- In the same signal condition, i. e. in the C QUIET2 and RQUIET2 mode of the central office 1 and the remote subscriber terminal 2,4, respectively, it is then detected whether ISDN noise is present. This is done by computing the average power PICN of the input signal, that is to say by measuring the idle channel noise (ICN) power, and comparing it with the threshold set in step 110. In step 200, it is decided that ISDN noise is present if PICN is higher than the threshold. If ISDN noise can be detected in step 200, the following steps 300-340 are skipped and the regular training procedure known from the prior art is performed in step 400.

If however, no ISDN noise is detected in step 200, the algorithm proceeds with steps 300-340 for estimating the influence of ISDN noise on the ADSL communication signal.

In the next phase of the algorithm shown in Figure 2, a periodic or cyclic test signal, which is called REVERB signal under ADSL, is transmitted between the central office and the remote subscriber terminal, and under this signal condition the ADSL modem 4 or the ISDN noise estimation means 5 thereof compute the average input signal power Preverb. Then the settings of the ADSL modem 4, in particular the PGA settings,

are determined based on PISDN f Preverb, where PISDN is the scaled version of PISDN according to the difference of the PGA settings used for the measurement of PISDN and Preverbr respectively (see steps 300 and 310 of Figure 2).

During the next phase of the algorithm of Figure 2, another test signal is transmitted between the central office 1 and the remote subscriber terminal 2,4, which under ADSL is known as the so-called MEDLEY signal comprising a random signal sequence. With this random test signal being present at the input of the ADSL modem 4, first the ISDN noise power spectrum is estimated for the Medley test signal on the basis of the pre-estimated ISDN noise power spectrum, which was recorded and saved in the ADSL modem 4 during step 100.

Then, in step 330, a shaped ISDN noise power spectrum is computed by shaping the input ISDN noise power spectrum Nk with the combined frequency response of the time domain equalizer (TDQ) and the frequency domain equalizer (FDQ) of the ADSL modem 4 as follows : Gk and Hkl are the frequency response of the time domain equalizer and the frequency domain equalizer, respectively, at the k-th tone. L is the number of symbols used for periodogram averaging. Xk, l is the frequency component of the input ISDN noise at the k-th tone of the 1-th symbol. The input ISDN noise power spectrum N 2 is defined as follows:

This noise shaping and compensation method is adapted to different line conditions by estimating the ISDN induced noise power with the trained time domain equalizer and frequency domain equalizer of the ADSL modem 4.

In step 340 of the algorithm shown in Figure 2, finally, the Medley SNR is calculated using the shaped noise power spectrum. In general, the Medley SNR for the k-th tone is computed using the signal power Sk2 and the noise power E of the k-th tone as follows: The noise power spectrum is estimated by averaging periodograms computed with discrete Fourier transformation (DFT). One way to adjust the Medley SNR is to directly subtract the pre-estimated SNR deduction (ASNRk) for each tone from the measured SNRk, as shown in equation 5: (5) S#Rk = SNRk - #SNRk However, this is a non-adaptive approach where ASNRkhas to be pre-measured or pre-estimated for different loop length and gross talk noises. Another way for the SNR adjustment is to combine the ISDN induced noise power (Ek°N) 2 with the measured noise power as follows:

As can be seen, according to the algorithm of Figure 2, a pre-measured or pre-estimated ISDN noise power spectrum is used to reduce the PGA settings of the ADSL modem 4 in order to avoid clipping. In addition, the Medley SNR for each tone is lowered or adjusted by using the shaped ISDN noise power spectrum so as to enable appropriate bit loading.

In general, in order to avoid overestimation of the ISDN power contribution for long loops, Nk should be measured or estimated with ISDN attached on one side only (either the central office 1 or the remote subscriber terminal 2, depending on where the detection scheme is to be implemented), but not both together. For long loops the remote ISDN signal will not contribute significantly, and it is for long loops where the ISDN detection and estimation approach is most important. It is also important to disable bitswapping at the local end in view of ISDN noise results in increased margins at the lower tones if ISDN noise is not present. These increased margins would be destroyed by a bitswapping mechanism.

The performance of the proposed algorithm was evaluated with the GEMINAX (Global Enhanced Multiport Integrated Transceiver) test platform of the applicant. The results are illustrated in Figures 3 to 5.

In Figure 3, the Medley SNRs for both 1000m and 3000m loops from the noise shaping are compared with those trained with ISDN on. Curves a, b relate to a 1000m loop, and curve a corresponds to the results obtained by training with ISDN on, while curve b corresponds to the results obtained with ISDN off by using the algorithm of Figure 2. Curves c, d relate to a 3000m loop, and curve c corresponds to the results obtained by training with ISDN on, while curve d corresponds to the results obtained with ISDN off by using the algorithm of

Figure 2. For both loops, it is observed that the adjusted SNRs using the predicted ISDN induced noise spectrum are very close to the SNRs measured when ISDN noise is present. The closely matched SNRs indicate that the proposed algorithm achieves as good data rates as those trained with ISDN on. It should be noted that Figure 3 shows the Medley SNR (dB) over the tone number with white noise at-140 dBm.

Figure 2 shows US loop reach test results by depicting the payload rate (kbps) over the loop length (m) with additive white noise (AWGN) being present. In addition, the test was conducted in the interleaved mode with TCM on and RS on with an ETSI 0.4 mm cable. Curve a of Figure 4 shows the US data rates obtainable by using the ISDN detection and estimation method of Figure 2 with ISDN off, while curve b shows the US data rates obtainable by training with ISDN on. These results demonstrate that the proposed algorithm performs well regardless of the loop length.

Finally, Figure 5 compares the US loop reach test results with ETSI B noise on. The test conditions interleaved mode, TCM on, RS on, ETSI 0.4 mm cable) were the same as for the test shown in Figure 4. Again, the payload rate (kbps) is depicted over the loop length (m). Curve a of Figure 5 shows the US data rates obtained by using the algorithm of Figure 2 with ISDN off, while curve b shows the US data rates obtained by training with ISDN on. The closely matched rates show the robustness of the proposed method to different noise conditions.