CA2205686A1 - Adaptive multiple sub-band common-mode rfi suppression - Google Patents

Abstract

A noise suppression circuit for a two-wire communications channel comprises a hybrid device, for example a hybrid transformer or circuit, for providing a differential output signal corresponding to a received signal in the two-wire channel. A summing device extracts from the two-wires of the channel a common mode signal and supplies it to a noise estimation unit which derives from the common mode signal an estimate of a noise level in at least one sub-band having a bandwidth considerably narrower than an operating bandwidth for the channel. An amplifier adjusts the amplitude of the noise estimate to correspond to the residual noise in the differential output signal. It is then subtracted from the differential output signal. A noise detection and control unit scans the operating band, identifies a sub-band having an instant highest noise level, and sets the noise estimation unit to the detected sub-band. The noise estimation unit suppresses the noise in that sub-band.
In preferred embodiments, the noise estimation unit comprises several channels, each comprising a tunable filter, a phase shifter and an amplifier and the noise detection and control unit sets the channels, in succession, to different sub-bands in descending order of noise level.

Description

CA 0220~686 1997-0~-16 ADAPTIVE MULTIPLE SUB-BAND COMMON-MODE RFI SUPPRESSION

FIELD OF THE INVENTION
This invention relates to a method and apparatus for reducing intelrerellce in 5 communications channels and is especially, but not exclusively, applicable to the suppression of common mode noise, including radio frequency inlelreLence, in digital subscriber loops of telephone systems.

BACKGROUND OF THE INVENTION
As illustrated in Figure l of the accompanying drawings, and labelled PRIOR ART,a balanced digital subscriber loop comprising a twisted wire pair carries both differential and common mode cullenls induced by the signal and noise sources, respectively. In a perfectly balanced loop, the common mode cullell~s will not inlelrele with the differential current (information signal). However, due to the circuit unbalance caused by bridge taps and poorly twisted cable etc., longitudinal current injected by external noise sources will be converted into differential current at the receiver and detected as noise. Such noise can lead to errors by introducing jitters in timing extraction circuits or by causing false pulse detection. In digital loops, common mode noise can be conveniently categorized into the following: Impulse noise, Radio Frequency Intelrerellce (RFI) and Crosstalk Noise.
With the trend towards higher bit rates in the loops, radio inte,rerellce, typically caused by radio stations in the vicinity and hence tr~n~mitting on certain frequencies with a relatively narrow band width, is ~suming greater ~i~nific~nce.
When telephone subscriber loops operated at relatively low frequencies, perhaps 3,000 Hz. or 4,000 Hz., inle,rerence could be dealt with adequately by using twisted wire cable 2s which helps to cancel out any induced interference and by means of hybrid transformers.
With the introduction of VDSL (Very high speed Digital Subscriber Loops) and ADSL
(Asymmetric Digital Subscriber Loops), the frequency of operation is approaching the radio frequency bands and the afore-mentioned balancing of the cable is no longer sufficient to reduce the inlelrelcllce. As a result, common mode noise increases.
Various techniques, other than such balancing, are known for reducing intelrelellce or noise in a communication ch~nnel For example, U.S. patent number 5,555,277, discloses a technique for cancelling common mode switching noise to achieve reduced error rates in CA 0220~686 1997-0~-16 a local area network. This technique involves gain controllers at both tr~n~mitt~r and receiver ends to m~int~in signal integrity during tr~nsmission. In addition, noise cancellation at the receiver is performed by generating the inverted copy of the received signal, amplifying both signals, and then summing them to cancel the induced noise. This is not entirely satisfactory because noise p~ttern~ are random and could change in the interval between the tr~n~mi~ion of the primary and inverted signals.
To mitig~te common mode noise on a pair of signal conductors, the system disclosed in US patent number 3,705,365, Dec. 5, 1972 uses a two conductor shielded cable, a three-winding transformer, and a bipolar differential amplifier. The common mode signal from the cable shield is used to cancel the common noise using the third winding in the transformer.
U.S. patent number 4,283,604 discloses a current source circuit with common modenoise rejection for a two-wire tr~n~mi~ion system. The circuit provides an interface for coupling signals with a telephone pair and operates to cancel or negate the effect of ullw~-led impedance on the pair improving common mode impedance across the pair, thus enhancing the common mode rejection to noise ratio in the two-wire system.
European patent application number 0 453 213 A2 discloses a radio receiver in which an adaptive notch filtPrin~ approach is used to reduce inlelrelel ce in a radio frequency received signal carrying digital data at 2.4 kilobits per second using 3 to 30 MHz r.f.
carrier. The adaptive notch filter is implemented using frequency domain analysis of quantized data to detect inle rel~nce by comparing the received signal frequency spectrum with a known spectrum template. Any frequency band with higher power than the reference template is considered to have an intelrerence. A programmable notch filter is then tuned to nullify the signal power in the respective frequency band, i.e. the received signal in the selected band is cancelled along with the inlelrelence, resulting in an undesirable loss of signal information. A further disadvantage is that the technique also requires the inl~lrerence rejection filter to be trained, which entails the tr~n~mis~ion of a training sequence periodically between tr~n~mi~ions of the data stream.
Despite the existence of these various techniques, there is still a need for a satisfactory way of reduçing common mode noise in subscriber loops operating at the proposed very high speed levels of VDSL or ADSL. At a VDSL workshop at IEEE Globecom, November18, 1996 in London, England, John Cioffi and John Bingham proposed doing so by CA 0220~686 1997-0~-16 inserting an impedance between ground and the centre tap of the usual hybrid transformer at the end of the subscriber loop and extracting across the impedance a signal representing common mode noise. Cioffi et al proposed to filter this common mode noise signal using an adaptive wide band filter to provide a radio frequency noise estim~te and subtracted it from a differential signal obtained from the secondary of the hybrid transformer to produce an error signal for tuning the adaptive filter to reduce that error signal to zero. The circuit can only tune the filter when there are quiet periods in the received signal. This is not entirely satisfactory because it involves timing to ensure that the quiet periods are detected.
It also increases overhead, reducing the efficiency of the tr~n~mi~ion. Moreover, the insertion of an impedance in the ground line is undesirable where the ground impedance is relatively high, perhaps because of the ground is rocky. Finally, the adaptive filter has to have a bandwidth at least as wide as the bandwidth of the received signal which, in the case of VDSL, might be from zero to about 10Mz.
An object of the present invention is to elimin~te or at least miti~te the disadvantages of the known techniques and provide a noise suppression arrangement that is better adapted to the reduction of common mode noise in two-conductor communications channels, such as twisted wire subscriber loops.

SUMMARY OF THE INVENTION
According to one aspect of the present invention, a noise ~upplession circuit for a two-conductor communications channel comprises a hybrid device, for example a hybridtransformer or circuit, for providing a differential output signal corresponding to a received signal in the channel, a device for extracting from the channel a common mode signal, a noise estim~tion unit for deriving from the common mode signal an estimate of a noise level 2 5 in a sub-band having a bandwidth considerably narrower than an operating bandwidth for a received signal in the channel, amplifier means for adjusting the amplitude of the noise estimate to correspond to residual noise in the differential output signal, means for subtracting the noise estim~te from the differential output signal, and noise detection and control means for moniloling the received signal, detecting a sub-band having an instant highest noise level, and setting the noise estim~tion unit to the detected sub-band.

CA 0220~686 1997-0~-16 BRIEF DESCRIPTION OF THE DRAWINGS
Embodiments of the invention will now be described by way of example only and with reference to the accompanying drawings in which:-Figure l, labelled PRIOR ART, is a schematic diagram of a two-wire communications 5 channel showing the paths of common mode and differential mode currents and the introduction of radio frequency intelrerence or noise from an RF radiation source;
Figure 2 is a simplified schematic block diagram of a noise suppression circuit according to the present invention and comprising a noise detection and control unit and a noise estim~tion unit;
Figure 3 is a block schematic diagram of the circuit of Figure 2 but showing the noise estim~tion unit in more detail; and Figure 4 is a schematic block diagram of the circuit of Figure 2 but showing the noise detection and control unit in more detail.

DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS
Referring to Figure 2, the differential signal coupled with the common mode signal is presented at line interface 1 (hybrid). The common mode signal is extracted from the twisted wire pair by summer 2, filtered by adaptive bandpass filter 3, and then phase-inverted by programmable phase inverter 4. The resulting signal is scaled by adaptive gain 20 block 5 and is combined with the differential mode output of line interface 6 to miti~;~te common mode noise present at the output. The improved differential mode output 7 with higher Signal to Noise Ratio (SNR) is then presented to the receiver.
The centre frequencies of bandpass filter 3 are adjusted to coincide with the centre frequencies of narrowband noises detected by pelrorn~ g spectral analysis of the common 2 5 mode signal in adaptive controller 8. The amount of nonlinear phase inversion is controlled by the adaptive controller so that the nonlinear phase delays inherent in hybrid 6 can be colllpen~t~d. The gains in adaptive gain block S are also controlled by the adaptive controller 8. Typically, the adaptive controller 8 can be implemented by a low speed microcontroller because the real time requirement to pelrorlll the above adaptive 30 computations is low.
Dominant common mode noise such as RFI ingress and Impulse noise will be reducedsignificantly. The noise reduction in a twisted-pair cable will increase the reach of digital CA 0220~686 1997-0~-16 subscriber 1~QP modems and improve the SNR which allows higher ~i~n~lling rate in the co~res~ollding cable.
Good measurement of the common mode signal that exists in an unbalanced twisted pair cable is crucial. In embodiments of the present invention, common mode noise is estimated 5 by adding the in-phase TIP signal and anti-phase RING signal with respect to ground reference in a twisted pair cable. It should be noted that the common mode signals in both TIP and RING portion of the cable are in phase with each other. Therefore, the common mode signal is extracted while the differential mode signal is cancelled out in the adding process of TIP and RING signals at summer, as illustrated in Figure 3.
The extracted common mode signal is then processed by a noise çstim~tor in the form of a multi-channel signal suppresser 10. The signal suppresser comprises a group of proces~ing channels 11, each of which comprises a programmable (adaptive) narrowband bandpass filter 12. The bandwidth and centre fre~uency of the bandpass filter can be programmed to match the bandwidth and centre frequency of narrowband noise detected by ~lrol,l,ing spectral analysis of the common mode signal at the adaptive controller (to be discussed later). As a result, the output of the bandpass filter will be a narrowband component of the common mode noise. It should be noted that circuits according to the invention can have any number of proces~ing channels depending on the number of narrowband common mode noise sub-bands that need to be suppressed.
2 o Differential mode and common mode signals propagate differently on twisted pair and the hybrid. There exists a phase relationship between the two propagation modes.Furthermore, this relationship is not constant with respect to frequency. The difference in phase between the two propagation mode increases with fre~uency. Beyond a few mçg~hertz, multiple periods of phase delay are exhibited between the common mode and differential mode. Therefore, the programmable phase shifter 13 must be capable of r~lignin~ the common mode signal to the differential mode signal such that there is a 180 phase difference between the signals in the fre~uency bandwidth of the noise suppression filter channel.
The amplitude of the differential mode noise is dependent on the loop balance.
Therefore, the common mode signal must be scaled to match the amplitude of the differential noise by adaptively controlling the gain of the adaptive gain block 14. It should be noted that better cable, such as data grade cable, exhibits better noise immunity;

CA 0220~686 1997-0~-16 therefore a smaller portion of the common mode noise will couple into the differential mode path.
Signals from each sub-channel 11 are combined at summer 15. Common mode noise ~u~ ion occurs at summer 16 when the phase-inverted common mode noise is added to 5 the differential mode signal co~ g the residual noise.
The noise detection and control unit 17 detects the residual noise at 18 to adaptively modify the parameters of each sub-channel 11 of the multichannel signal supplcssel 10 to minimi~e the noise at 18. Noise detection involves computing the average of the cross-correlation between the differential and common mode signals, and will be discussed later.
10Although spectral analysis of the common mode signal to detect residual noise in the differential mode signal could be pe ro med simply by computing a Fourier transform of the signal, this approach requires intensive computations and thus a high speed processor.
To avoid the need for a high speed processor, spectral analysis is accomplished by sweeping the entire frequency band of the digital subscriber loop incrementally using narrow band 15bandpass filters 19 and 20, as shown in Figure 4. The sweeping of the signal channel is controlled by a microcontroller 21 and the interactions between the analog components and the microcontroller are provided by way of the ADC 22 (analog to digital converter) and DAC 23 (digital to analog converter) . Radio frequency inte. rer~llce (RFI) could be detected by sweeping the entire frequency band of the common mode signal while differential mode 2 o noise could be detected by sweeping the cross-correlation between differential and common mode signals. It should be noted that common mode noise comprises p~sb~nd signals which need to be converted into baseband signals to fit into the frequency range of the 22 (ADC) by democl~ ting the noise with carrier signals. This is achieved by carrier signals typically generated by a voltage controlled oscillator 24 (VCO) whose frequency could be tuned by 25 adjusting the input voltage of the VCO 24. Such tuning is carried out by microcontroller 21 by way of D-to-A converter 23, multiplexer 32 and sample-and-hold circuit 35.Operation of the circuit will now be described. Referring to Figure 4, when the system is first switched on, the microcontroller 21 will cause the common mode signal (SCM) passband filter 19 and the differential mode signal passband filter 20 to scan the 30 entire frequency range. Amplifiers 25 and 26 each amplify a respective one of the two filtered signals from the two bandpass filters 19 and 20, respectively. The gain of each of the amplifiers 25 and 26 is adjustable by microcontroller 21 via the D-to-A converter 23 and CA 0220~686 1997-0~-16 sample-and-hold circuits 34. The amplified SCM signal is passed to a multiplier 27 which, under the control of voltage controlled oscillator 24 converts the passband common mode signal to a baseband signal. Low pass filter 28 removes from the baseband common mode signal high frequency components, such as harmonics, and applies the res~llting signal to the microcontroller 21 by way of its A-to-D converter 22. Meanwhile, following amplification by amplifier 26, the SDM signal from the differential mode signal bandpass filter 20 is passed to a multiplier 29 which multiplies it by the same sub-band of the common mode signal SCM and applies the resulting signal to a second low pass filter 30 for suppression of harmonics. Finally, sliding average integrator 31 extracts from the output of low pass filter 30 the DC component signal and applies it to the A-to-D converter 22. The operation of the differential mode signal path will be described later.
By way of D-to-A converter 23, multiplexer 32, and sample-and-hold circuits 34, the microcontroller 21 increments the frequency of tunable or programmable filters 19 and 20 in steps which are equal to the passband of the tunable filters 12A, 13A, and 14A in the noise estim~tion unit 10, i.e. in steps of 100 Khz. At each 100 Khz. interval, the microcontroller 21 measures and stores in memory the power level of the received signal.
When the microcontroller 21 has completely scanned the entire frequency range, it selects the highest power value recorded and then, by way of D-to-A converter 22, multiplexer 32 and one of three sample-and-hold circuits 33, sets the first tunable bandpass filter 12A in the noise estimation unit 10 to the centre frequency of that sub-band.
While this has been happening, the differential mode signal path has been free running. Once the first tunable filter 12A has been set or locked to its sub-band, and begins to suppress the noise in that sub-band, the noise component in the differential mode signal is reduced. As this dirr~fe.ltial mode signal is passed through the first passband filter 20, amplifier 26 and multiplier 29 in the SDM path (Figure 4), the multiplier 29 will pelrol---auto correlation by multiplying the common mode signal (SCM) by the differential mode signal (SDM), producing a residual signal which is rather noisy. Low pass filter 30 removes high frequency components or harmonics from the residual signal. The sliding average integrator 31 then extracts the DC component from the residual signal and applies it to the A-to-D converter 22. The microcontroller 21 determines whether or not the residue is positive or negative, indicating that the amplitude of the differential modes signal noise component is smaller or greater than the noise component extracted from Tip and CA 0220~686 1997-0~-16 Ring, and, by way of D-to-A converter 23, multiplexer 32 and sample-and-hold circuits 33, adjusts the phase shifter 13A and amplifier 14A to adjust the phase and gain in the first tunable filter channel A in the multi-channel noise estim~tor 10. The adjustment of phase and gain will continue until the residual signal is substantially zero. At this point, the first filter channel is correctly set to the first noisy sub-band.
The microcontroller 21 then repeats the process, looking for the next noisiest sub-band and sets the second tunable filter channel 12B, 13B and 14B to the second noisiest sub-band. It should be noted that, at this time, the first tunable filter channel A will have suppressed the first noisiest sub-band, so the microcontroller will again look for the noisiest sub-band in the operating range.
The process will be repeated again to set the third channel C to the third-noisiest sub-band, at which point all of the tunable filter channels have been set to respective sub-bands selected in descending order of noise power.
Assuming an operating fre~uency range from zero to lOMz, and individual filters of 100 kHz. bandwidth, it is expected that pelhaps three tunable filters will be sufficient for most applications to subscriber loops using twisted wire pairs. This recognises that the nature of noise or common mode noise in these systems tends to be concentrated in certain bands, perhaps because it is inlelrelellce from a neighbouring radio station. Nevertheless, it will be appreciated that a greater number could be used if desired.
In conclusion, the suppression of noise in the tr~n~mi~ion loop permits extension of loop lengths and higher data tr~n~mi~ion rate. It extends usable loop bandwidth in the presence of RFI and thus optimizes the use of available cable bandwidth.

Claims (3)

1. A noise suppression circuit for a two-conductor communications channel comprises a hybrid device for providing a differential output signal corresponding to a received signal in the channel, a device for extracting from the channel a common mode signal, a noise estimation unit for deriving from the common mode signal an estimate of a noise level in a sub-band having a bandwidth considerably narrower than an operating bandwidth for a received signal in the channel, amplifier means for adjusting the amplitude of the noise estimate to correspond to residual noise in the differential output signal, means for subtracting the noise estimate from the differential output signal, and noise detection and control means for monitoring the received signal, detecting a sub-band having an instant highest noise level, and setting the noise estimation unit to the detected sub-band.

2. A circuit as claimed in claim 1, wherein the noise estimation unit comprises a plurality of tunable narrowband filter units and the noise detection and control means is arranged to scan the operating band a plurality of times and, after each scan, adjust a corresponding one of the narrowband filter units to the noisiest sub-band detected during that scan.

3. A circuit as claimed in claim 1 or 2, wherein the noise detection and control means comprises a tunable bandpass filter having a bandwidth similar to that of a said sub-band, and means for controlling the tunable bandpass filter to scan the operating range in steps, recording the noise signal level at each step, identifying the highest recorded noise level, and setting the noise estimation means to the frequency corresponding to the highest recorded noise level.