GB2440589A

GB2440589A - Phase difference calculator

Info

Publication number: GB2440589A
Application number: GB0511161A
Authority: GB
Inventors: Bahador Bakpaz
Original assignee: TECTEON PLC
Current assignee: TECTEON PLC
Priority date: 2005-06-01
Filing date: 2005-06-01
Publication date: 2008-02-06
Also published as: WO2006129061A3; CA2611302A1; GB0511161D0; WO2006129061A2

Abstract

A phase reversal detector converts an input tone from time domain into frequency domain, determines a phase difference over a time interval, and detects a phase reversal, based on the phase difference. The converter derives a Fourier transform coefficient having a centre frequency corresponding approximately to a principal frequency being sought in the incoming signal. It uses a very short window of samples, the window length being selected to be shorter than that required for an exact correspondence. By using an approximate correspondence a much shorter window can be used, which enables a significant reduction in computational load and a shorter detection time. The converter output is a complex number. The phase difference is obtained using a dot product of the outputs of the converter obtained at different times. An average of consecutive phase differences can be used to reduce the effect of noise.

Description

1 2440589

TECTEON PATENT APPLICATION

PHASE DIFFERENCE CALCULATOR

FIELD OF THE INVENTION

The invention relates to methods and apparatus for detecting phase differences for detecting phase reversals, for applications including echo cancellers, to software for carrying out such methods and to systems incorporating the above, and methods of using such systems, including methods of providing telecommunications services to subscribers, over a telecommunications network having such an echo canceller, and to converters, and phase difference calculators for these and for other applications.

BACKGROUND

It is known to detect phase reversals in tones. One of the main applications for detecting phase reversals is in echo cancellers in telecommunications networks.

In order to avoid interaction between an internal echo canceller of a fax or modem, and a network voice echo canceller, such faxes or modems are required to transmit an echo canceller disabling tone. This tone instructs the network to disable its voice echo cancellers to leave a clear channel for data transmission. This disabling tone is transmitted at the start of each modem call. ITU-T has specified the requirements in G.164/G.165 standards, which specify the disabling tone as a 2100 Hz tone. Modems designed in accordance with the G. 165 standard reverse the phase of their tone every 450 ms +1-25 ms. Hence echo cancellers need tone detectors which can detect phase reversals.

A further standard ITU-T V.8 also specifies some amplitude modulation of the 2100Hz tone. There is a therefore need for a tone detector which can reliably detect a tone which is subject to amplitude, frequency and phase variations. There is also a need for the tone detector to detect quickly the presence of the tone to enable a successful modem call.

The detector should detect the phase changing in a range of 155 & 205 degrees (180 25) degrees and should not detect the phase changing in a range -1 10,+1 10 degrees.

The problem can be formulated in mathematical terms in the following way: The tone can be expressed as: x[n] = Asin(2izfT.n + ) (1) A phase reversal must be detected when 0 changes from ç to 00 + p where,LE [(180-25) ,(180+ 25) ] and not be detected if 1iE [-110 ,1 1001 Where: T. 1 s fE [2100-4f,2100+Af] A: Signal amplitude 4: Signal phase Known methods of implementing phase reversal detection include: a) detecting a dip in the output of a bandpass filter b) correlating the input signal with a locally generated tone, and detecting a sudden change in correlation level, c) measuring the instantaneous frequency, and detecting a sudden change, d) measuring a group delay and looking for a sudden change in this, e) demodulating the signal to baseband (DC) and looking for a change in polarity of the DC signal, and f) looking for a change of sign in the phase error signal of a second order phase locked loop (PLL) arrangement for detecting a tone.

It is known from US patent 5 815 568 (Trump) that an optimal solution for the problem would be to implement a matched filter for the disabling signal. However it is recognized that this would lead to an algorithm with unacceptably high computational complexity and long processing delay. This patent also indicates that known PLL type solutions are susceptible to out of band noise signals which lower the probability of a correct detection. If band pass filtering were included in the scheme, it would remove, in addition to the out of band noise, the sharpness of the phase reversal and the phase locked loop would be able to track the resulting smooth phase change. This would result in less difference between the input and output signals and hence worsen the detection.

Trump proposes using a feed forward receiver structure with in-phase and quadrature channels rather than a phase locked loop and phase reversal detection. First, a first power estimate is determined from an input signal. In addition, the input signal is divided into in-phase and quadrature components. The in-phase and quadrature components are then sub-sampled and used to determine a second power estimate. The first and second power estimates are compared to determine whether a predetermined tone is present. When the predetermined tone is detected, a phase reversal in the predetermined. tone is detected by using the in-phase and quadrature components.

However this will be sensitive to noise, will have a heavy computational burden, and be sensitive to frequency changes.

Another attempt is known from US patent 6055310 (Zhang et al). Here a phase change is detected by delaying an input sampled tone by a variable amount to obtain an output sampled tone until a phase of said output sampled tone is an integer multiple of 27t of said input sampled tone; fixing said variable delay regardless of any phase difference thereafter arising between said input sampled tone and said output sampled tone; and thereafter monitoring said input sampled tone and said output sampled tone for a phase difference arising between said input sampled tone and said output sampled tone, to detect the phase change. This aims at increased accuracy of detection, but imposes a heavy computational burden because it uses an adaptive algorithm.

Yet another attempt is known from US patent 6259750 (Miller et al). This involves generating signals representing the quadrature components of the input signal, and looking for a migration in the quadrature plane of the position of said quadrature components by an amount greater than a predetermined threshold to indicate the phase reversal. The detector may also include means for bandpassing the incoming signal, which may be hard limited to eliminate the need for an automatic gain control circuit while ensuring the functionality of the detector over a large dynamic range of the input signal. It is based on the idea that a migration of the coordinates in the quadrature plane occurs at each phase reversal. Owing to filtering effects in the telephone network and the analog-to-digital converter, etc., the migration of the quadrature point when a phase jump occurs does not generally happen instantaneously. It can take at least 5 ms before the position of the quadrature point stabilizes. Accordingly, the detection of the phase reversal is based on the average position in the quadrature plane over a predetermined period, in this case 4 ms. Again this is a computationally intensive method.

SUMMARY OF THE INVENTION

It is an object of the invention to provide improved apparatus or methods. According to a first aspect of the invention, there is provided a phase reversal detector having: a converter for deriving phase information by applying a transform to an input signal, an output of the transform being a complex number whose phase is approximately linearly related to a phase of the input signal, a phase difference calculator for determining a difference in phase of the input signal over a time interval, using the phases of the complex numbers derived for that time interval, and a phase reversal detector for detecting a reversal in the phase, based on the phase difference.

An advantage of such a converter is that the amount of processing can be dramatically reduced by using such a transform. An advantage of having the converter output having phase linearly related to the phase of the input signal is that it can enable the subsequent phase difference calculation to be relatively simple. This can lead to a radical reduction in the amount of processing. Particularly for applications such as telecommunications signal processing, this can be very valuable, as the processing capacity is limited and costly to increase. Therefore, a reduction in processing load will often directly affect the amount of telecommunications traffic, such as calls, that can be handled, and therefore directly increase the revenue for the service operator.

Another advantage of the above-mentioned features is a reduced processing delay, before the result of the phase reversal detection is output. Overall, a better balance can be achieved between processing load, delay, and accuracy of detection.

As a preferred additional feature, the transform can be a Fourier transform for which a principal frequency f of the input signal, a sampling rate T, and a window length N of the transform are arranged such that f*T*N is approximately an integer. An advantage of this is that for a Fourier transform, it can enable the output to be simplified compared to a conventional, more complete Fourier transform.

As another preferred feature, the transform can involve deriving a transform coefficient corresponding approximately to the principal frequency of the input signal, the window length being selected to be shorter than that required for an exact correspondence.

An advantage of using an approximate correspondence is that it can enable a much shorter window to be used. The use of shorter windows enables a significant reduction in computational load and a shorter detection time. The length of window determines the frequency separation between the coefficients. Hence a larger window will result in more coefficients and one will be close to or on the desired frequency. The shorter detection time can arise partly from the reduction in amount of calculation, and partly from the need to wait until the end of a window, for the result, such wait being shorter for a shorter window. Any loss in accuracy from using a short window can be tolerated or compensated in other ways, since in this case only phase information is used, not frequency information. This can enable a much improved balance between detection accuracy, detection time, and computational load. It represents a move away from the most promising of the known techniques, back to transforms which were considered too complex or computationally heavy until now. The amount of reduction in computational load can be dramatic. A window length normally set at 80 samples can be reduced to 8 samples in one example.

A preferred additional feature is the output being based on a transform coefficient chosen as having a centre frequency which corresponds to within 20% of the principal frequency. The closeness of the coefficient can be controlled by choosing the window size and which coefficient is chosen.

Another preferred additional feature is the window length being less than 20 samples of the input signal. This more specific approximation gives particularly good results in terms of sufficient accuracy with very low computational load. Some benefit is still achieved with 30 or 40 samples for example.

Another preferred additional feature is the window length being 8 samples, to give a Fourier coefficient at 2000Hz. If the desired frequency is 2 100Hz, this can give sufficiently accurate results with very low computational load.

As another preferred feature, the phase difference calculator is arranged to carry out a dot product of the complex numbers. An advantage of this is that it can be more computationally efficient since it enables a reduction in the number of conversions from complex to angle representation. Such conversions are computationally heavy. (A dot product here is defined as a multiplication of two complex numbers where one of the complex numbers is complex conjugated, otherwise a phase addition would result.) As another preferred feature, the phase difference calculator is arranged to carry out two consecutive dot product operations. An advantage of the second dot product is that it is a computationally very efficient way of cancelling the constant c which is related to the sinusoid's frequency part (c = Z7zJTN, where f= sinusoid's frequency).

As another preferred additional feature, the phase difference calculator is arranged to determine an average of a series of phase differences detected at different times. This can enable greater immunity to noise, without unduly adding to the computational load or the time of detection. It can also provide some compensation for any loss of accuracy caused by using the approximation, or the short window.

As another preferred additional feature, each of the differences is a change in phase of the input signal over a time period, the time periods being arranged to overlap. This can ensure a phase reversal will always be detected, and in at least some cases, can be detected more than once, to give greater immunity to noise.

The detector additionally can have the time periods arranged to overlap such that there is always coverage by two or more of the time periods. This can give further immunity from noise. Increasing the length of the time period, or increasing the frequency of the phase difference calculations, will increase the amount of overlap, and therefore increase the number of phase difference values that can contribute to the average. The more values that are averaged, the better will be the immunity to noise, though at the cost of increased computational load.

The detector additionally can be arranged to determine the phase difference in complex values, then convert to a phase angle. This enables the computational load to be reduced.

The detector may be in the form of software. This recognises the value of software as a significant component of working systems and a component which can be independently traded or upgraded.

An echo canceller can have the detector with any of the features set out above. This is one of the most significant applications. The detector can make a significant difference to the commercial value of such cancellers, which themselves can be critical to the performance of telecommunications networks.

Another application is in a method of providing telecommunications services to subscribers, over a route in a telecommunications network using the echo canceller.

This aspect recognises the added value to the end user of a telephone service, of improved echo cancellation performance. It also recognises the value of the activity or operation, once the apparatus is installed, in generating more revenue from telephone calls, which may be much greater than the cost of the apparatus.

A second aspect of the invention provides a phase difference calculator for processing complex numbers representing phase information, each of the complex numbers having a phase which has a linear relationship to the phase information, the calculator having: a first dot product generator for generating a dot product of two of the complex numbers, to derive a phase difference, and a second dot product generator for removing a constant from the phase difference.

An advantage of the use of two stages of dot product operations is that step changes in phase such as phase reversals, can be extracted.

As a preferred feature, the complex numbers represent phase information of a single input signal at different times.

As another preferred feature, a series of phase differences are calculated over different time intervals, the time intervals being arranged to overlap, and an average phase difference is determined from the series.

As another preferred feature, the phase difference is used to detect a phase reversal.

A third aspect of the invention provides a phase difference calculator having: a phase difference arranged to determine a series of phase differences, each difference being over a time interval of more than one sample, the series being such that the time intervals overlap, and an averager for determining an average phase difference from the series of phase differences.

An advantage of taking an average is improved immunity to noise. Having the overlap ensures that a discontinuity such as a phase reversal will be captured by more than one of the differences in the series. Thus detection of the discontinuity can be improved by the combination of improved noise immunity and multiple capture.

Another aspect of the invention provides a phase reversal detector having: a converter for converting an input signal from time domain representation into a frequency domain representation, a phase difference calculator for detecting a difference in phase of the input signal over a time interval, using the frequency domain representation, and a phase reversal detector for detecting a reversal in the phase, using an average of a series of phase differences detected by the detector at different times.

The use of an average phase difference enables more immunity from noise, and means there is less need for each phase difference to be calculated so accurately. Reduced accuracy requirements can reduce computational load or time of calculation, or both.

Thus the balance between accuracy, time of calculation and computational load, can be improved.

As an additional feature the time periods are arranged to overlap. This can ensure a phase reversal will always be detected, and in at least some cases, can be detected more than once, to give greater immunity to noise.

Another possible additional feature is the time periods being arranged to overlap such that there is always coverage by two or more of the time periods. This can give further immunity from noise. Increasing the length of the time period, or increasing the frequency of the phase difference calculations, will increase the amount of overlap, and therefore increase the number of phase difference values that can contribute to the average. The more values that are averaged, the better will be the immunity to noise, though at the cost of increased computational load.

The detector may be arranged to determine the phase difference in complex values, then convert to a phase angle. This enables the computational load to be reduced.

The detector may be in the form of software. An echo canceller can have a detector with any of these features. A method of providing telecommunications services to subscribers, over a route in a telecommunications network can make use of such an echo canceller. This aspect recognises the added value to the end user of a telephone service, of improved echo cancellation performance. It also recognises the value of the activity or operation, once the apparatus is installed, in generating more revenue from telephone calls, which may be much greater than the cost of the apparatus.

A detector can be used in a method of using a telecommunication service as a subscriber, provided over a route in a telecommunications network using such as echo canceller.

Other aspects provide for methods or software corresponding to any of the apparatus or system aspects, or combinations or components of the above aspects. Other advantages than those set out above may be apparent to those skilled in the art, particularly over other prior art of which the inventor is not yet aware. The features of dependent claims within each aspect can be combined with each other or with other aspects of the invention as would be apparent to those skilled in the art.

BRIEF DESCRIPTION OF THE FIGURES

Embodiments of the invention will now be described with reference to the figures as follows: Figure 1 shows a conventional network showing an echo canceller having a phase reversal detector, Figure 2 shows in schematic form a phase reversal detector according to a first embodiment of the invention, Figure 3 shows an example of the fourier coefficient determining part of the embodiment of figure 2, Figure 4 shows an example of how to implement the phase difference calculation element of Figure 2, Figure 5 shows an example of how to implement the instantaneous phase angle difference calculation of figure 4, Figure 6 shows an example of how to implement the angular velocity cancellation element of figure 4, Figure 7 shows an example of how to implement the phase reversal decision element of figure 2, Figure 8 shows another embodiment having a moving average window element, Figure 9 shows a schematic view of samples of three frames to show how they may be overlapped, and Figure 10 shows a schematic view of a sequence of frame timings for the transform and phase difference calculation elements.

DETAILED DESCRIPTION

Figure 1, Telephone Network.

To show a high level view of an application of the detector, Figure 1 shows an example of a telecommunications network in the form of a prior art telephone network having an echo canceller which can be adapted to incorporate embodiments of the invention. In this figure, a long-distance telephone network 50 is shown, for making a telephone call from one subscriber to another. This can be used for voice or data transmission as is well known. For convenience, one side of the network is denoted the near end, and the other side is denoted the far end. A subscriber's handset 90 is coupled to a private branch exchange (P B X) by a 2-wire subscriber line 45. In the P B X, a hybrid coil 60 is used to convert between the two wire subscriber line and a 4 -wire line to the Central Office or local exchange 10. The conversion to 4-wire enables the voice signals in two directions to be a separated, which is useful for digitising and further processing. Each P B X may support tens or hundreds of subscribers, and will have sufficient hybrid coils according to how many calls are to be supported simultaneously.

Connections from many PBXs and many subscriber lines may be concentrated at a Central Office 10, which maybe many miles away from the subscriber. The central office contains the echo canceller 70, and a switch 80. For the sake of clarity, many other functions of the Central Office are not illustrated. There may be many echo cancellers provided, according to how many calls are to be handled simultaneously.

Conventionally, each Central Office concentrates many calls on to one or more or trunk routes 130 which make up the long distance telephone network 50. At the far end, similar elements and functions are provided. A far end Central Office 20 contains an echo canceller 110 and a switch 100. 4-wire lines 150 are provide to connect the Central Office to one or more P B Xs 30. Each will contain a hybrid 120. Two-wire subscriber lines 160 couple handsets 165 to the hybrid.

As the echo cancellers are intended to cancel echoes arising from the hybrids at each end of the circuit, in principle, they can be located anywhere in between the hybrids.

They are in practice usually located in a central office where many lines are switched and concentrated. This is convenient to enable them to be shared to make more efficient use of limited processing resource, and for ease of access.

Figure 2, Phase Reversal Detector Overview As explained above, a phase reversal detector may be included in a tone detector for use by the echo canceller. Figure 2 shows how the phase reversal detector may be implemented according to a first embodiment of the invention. It includes a transform section 210, a phase difference calculator section 220, and a phase reversal decision section 230. Each of these will be described in more detail below. Other sections may be added without departing from the invention. Individual sections may be used in applications other than phase reversal detection.

The transform section receives as an input a tone signal from a filter or a tone detector (not illustrated) which isolates the tone from other components of an incoming signal.

The transform section serves to convert the time domain representation of the incoming tone into a frequency domain representation which makes phase information in the signal readily accessible.

In the example described below, the transform is a fourier type transform. Other transforms can be used if they can extract the phase information. If other transforms are used, other blocks need to be adapted accordingly, especially the Phase Difference calculator block.

The phase difference calculator section takes the frequency domain representation which contains the phase information, and determines a phase difference over a period of time. In other words, a difference between the current phase and the phase a short time earlier, is determined. Based on this, a decision as to whether a phase reversal has occurred can be taken. Optionally this decision block may be arranged to determine the decision using criteria following the standards mentioned above. (According to the above standard, phase reversal is a phase difference greater than 155 degrees and less than 205 degrees). The decision is made with a simple threshold (relay block) 400 comparing the phase difference with minimum and maximum phase differences.

Of these blocks, the transform block is notable for enabling a very significant reduction in computational load, usually expressed in MIPs (Millions of Instructions per Second).

Before describing how each of these sections may be implemented, an explanation of the mathematics underlying a best mode implementation of a suitable transform will be presented.

Mathematical Explanation of an example of the Transform With reference to equation (1) set out above, consider an 8 point time frame on the first 8 point of x[n]: n=N-I -.2dc,z=N--.2IC A n=N-l -.227* A n=N-I X1[k]= x[n]e N = Asin(2,ltTn+ 1)e i" =- -- e2'" n=O n=O j,, 2j n=O A n=N-1.2,r(l-k) n=N-1.2,r(-l-k) e' N e' N ")= -e if k=l, lEN 3 n=o n=o 3 -----e' if k=-l, lEN 2j 0 Otherwise (2) Where: l=JT.N (3) if f=2lOOHz, T. = 1 s, N=8 then 1=2.1 and it can be seen that lEN. It follows that: A J2'21N S _________.27r('-2.I-k) X1[k] N)/(1-e N)e.(1-e N)/(1-e N)) 2j A..r(2.12)N 2r(2.I2) J2'2'2N 2)7i'2.12) X1[2]---(e'.(1-e N)/(1-e N)-e'.(1-e)/(1-e N))_. 2j

_.4_(7.87e.027e -.3e'29e' ) 2j 2j Thus: X1[2] 2j (4) Now, consider another 8 point time frame on the (D)th frame after the first frame n=N-1.2)7* X+1[kI= Asin(2,T(n+DN)+ 2)e'N D-l is the number of intermediate frames separating the frames used for determining the phase difference (PD), called camera frames. In this case the first and the (D)th frames are the camera frames. (5)

By the same calculation it can be shown: XD+l[2] __Nej 2eJ2 2j (6) Define: c = 2nJT.DN (7) PD1 = phase(XD+l[l]) -phase(X1[fl) = c +02 - 01 (8) by the same way it can be written for nth camera frames(In this case the nth and the (n+D)th frames are the camera frames: PD = phase(Xfl+D[l]) -phase(X[1]) = c + 02 -01 (9) If a phase reversal (PR) occurs in one of the intermediate frames between two camera frames: 02 -01 PD = phase(Xfl[l]) -phase(X,1[1]) = c + tp (10) Define: v, = PD -(11) According to equation (9) it can be written assuming that the (n+1)th frame contains the signal with phase reversal (where dots represent preceding and succeeding elements of the series: PD_,) = phase(X,1[1}) -phase(Xfl_D[l]) = c PDfl_D+l = phase(X1[1]) -phase(Xfl_D+l[l]) = c + p1 PDfl_D+2 = phase(X, 2[l]) -phase(XflD 2[l]) = c + p PDfl_D 3 = phase(X3[l]) -phase(Xfl_D 3[l]) = c + p PD = phase(Xfl D[l]) -phase(X[l]) = c + y PD1 = phase(Xfl+D l[l]) -phase(X1[l]) = c + PD7 phase(Xfl+D 2[l]) -phase(X2[l]) c (12) In the ideal case (PR occurs exactly between two neighbouring frames): /1 = =0 (13) By calculating the values of v it can be written: VflD = -PDflDD = C -C =0 V_/)1 = PD,Z_D+l -= 1/' VflD+2 = PDfl_D 2 -= V_13 = PDfl_D+3 -PDfl_2D+l = v, = PD -= v,1 = PD1 -PDfl_J) 1 = -v,2 = PD2 -PD,D+2 = v3 = PD,13 --PDfl_D+3 = Vfl+D = PDfl+D -PD = -p Vn0 1 = PDfl+D+ l -PD1 = 1/12 Vfl+D 2 = PD1,2 -PD+2 = c -c = 0 (14) Define: n+D Avrg1 = 1/2D * fl_(D)+lk1l (15) By the previous assumption ((n+l)th frame contains the PR): Avrg n+1 2D (16) In the ideal case when p1 = = 0 the above equation is exact.

Figure 3 Transform Figure 3 shows one way to implement a discrete Fourier transform (DFT) using the approximation described above. This block fundamentally transforms the input signal into a domain where thephase can be extracted. In this case it uses the Fourier domain.

It computes the nearest Fourier coefficient to the theoretical tone frequency (here 2100Hz) of the Discrete Fourier Transform. This approximation allows this very low MIPS algorithm. A conventional exact calculation of the fourier coefficients for a frequency of 2 100Hz with a sampling frequency of 8000Hz would normally require calculation of an 80 point fourier transform as minimum. Hence the reduction in computation load using this approximation, is enormous, typically several orders of magnitude.

A buffer 250 converts the input tone signal from a serial stream of samples into a parallel stream using a shift register type function. In this example eight samples are fed to a matrix multiply function 260. These samples are multiplied by a DSP constant 270. In this block the buffer is an 8 point buffer that is updated, 8 points by 8 points.

The fourier transform of the buffer is calculated around 2000Hz. (the resolution of the 8 points FT being 1000Hz, the nearest point to 2 100Hz will be 2000Hz). This approximation is good enough as explained above.

According to equation (2) above, the following calculation should be done for k=2: n=N-1 X1[k]= x[n]e N The equation can be written in another way: _J27r(2)(0) _J2lr(2)(1) _J27r(2)(2) J22t(2)(3) _J27r(2)(4) _327r(2)(5) X1[2]=x[O]e 8 +x[l]e 8 +x[2]e 8 +x[3]e 8 +x[4]e 8 +x[5Je 8 _127T(2)(6) _J21r(2)(7) +x[6]e 8 +x[7]e 8 = x[O] + x[1](-j) + x[2](-1) + x[3J(j) + x[4] + x[5](-j) + x[6](-1) + x[7J(j) = x[O] x[l I x[2] x[3] -J -1 i -J -1 x[4] x[5] x[6] x[7] In other words, this represents the dot product of the constant vector 270 stored within the DSP in a typical example, and a signal frame output by the buffer 250.

The result is used in the phase difference calculation block described below to calculate the phase difference between (X2[2],X1[2]) and between (X3[2], X2[2]) and between (X[2], X_1[2]). In the averaging method described below, using a delay D between the phases, the phase difference between ((Xfl[2],X,D[2J)) is calculated.

Figure 4 Phase Difference Calculation This block use the information computed before to extract the phase information. The two main tasks of this block are: 1) Compute the instantaneous angle difference as illustrated by block 300, and 2) cancel the angular velocity as illustrated by block 310.

For an example input signal: s(t) = cos(w*t + p(t)) (22) S(t) is the input signal, w is the angular velocity, p is the phase, t is the time and (w*t + p(t)) is the instantaneous angle.

Instantaneous angle diff = Phase (s(T1)) -Phase (s(T2)) = w*(T1T2) + p(T1) -p(T2) (23) w*(T1T2) is a constant to be removed to get the desired value p(T1) -p(T2) which is the phase difference.

The last block (complex to angle) 320 is an implementation choice, made to save MIPS.

Rather than computing the phase before (as in the equation above), the phase is computed afterwards.

To detect a phase reversal, the phases at two instants are subtracted, represented by subtracting 02 + c and. If the result is c it is clear that no PR has occurred (02 = 01). But if the result is c+180 then a PR has occurred (02 = + 180).

There are two principal problems in implementing the above calculations: a) There is no easy access to the phases (02 + c and ) directly because the output of the converter is a complex number in form of a + bj. To convert this form to the single exponent expression (re0) is very computationally heavy.

b) The second problem is that the value of the constant( c) is unknown.

Solving the problems: a) It became apparent that it is not necessary to access to 01 or 02 + C independently, to be able to subtract them. According to complex numbers theory the result of a dot product of two complex numbers (ie ) and (r1eJ 2)is rr2e(O2 1) Thus a dot product of the two complex numbers (output of the converter in two camera frames) gives a complex number in the form of (a + bj) which hasaphaseof(2+c)- 1.

b) Constant cancellation: For the case that a PR has occurred, the result of part a) above is a complex number whose phase is 1 80+c. The previous sample of result of part a) is a complex number whose phase is c because the PR had not yet occurred. If the phases of the new sample of result of part a' and the previous sample of result of part a' are subtracted, this removes the constant and leaves the PR (for example 180 or 0). As mentioned, the result of part a) is in the form of a + bj. Therefore another dot product can be used to compute a complex number in form of a + bj whose phase is PR (180 or 0). This second dot product is the dot product of a new sample and a previous sample of the result of part a). To threshold the phase angle, it should be converted from a complex number to the form of real angle.

The instantaneous phase angle difference block 300 will now be described in more detail with reference to figure 5.

Figure 5 Instantaneous Phase Difference As shown in figure 5, a dot product generator 350 is provided to generate a dot product of the input and a delayed version of the same input, generated using a delay element 360. This is basically the implementation of equation (23) above. It could be done in various different ways. This block calculates the product of the second input and the conjugated value of the first input. The phase of the result equals the phase of the second input subtracted by the phase of the first. The first input to this block is shifted by D samples of the"8 points fourier transform" (D*8=8D samples of the input signal).

If the phase reversal occurs in that interval (8D points) the phase of the dot product output changes. In this example, a short window or frame contains 8 samples of input signal. A sample of Fourier transform is X[2] as mentioned above. For calculating a sample of Fourier transform, a frame that contains 8 samples of input signal is used.

The value of the delay D is important. Increasing D (the delay) will improve the robustness of the algorithm to noise. Increasing D enables more instances of the phase reversal to be computed each having different values of noise. This gives more chances of detecting the phase reversal correctly. If an average of many phase differences is used, as described below with reference to figures 8 and 9 below, the noise effect can be reduced further.

A compromise has to be found. A value of D between 4 and 8 can give good results, where D-l is the number of disjunctive frames between the frames used for calculating a phase (termed "camera frames" below). In other words, the second camera frame is the Dth frame after the first camera frame.

Figure 6 Angular Velocity Canceller Figure 6 shows a second dot product generator 380 for generating a dot product of the output of the instantaneous phase angle difference block and a delayed version of the same signal. Again a delay element 360 is provided. The delay value D should be the same for the instantaneous phase angle difference calculation and the angular velocity canceller. As explained above, the angular velocity canceller had the effect of removing a constant term, to leave the phase difference as set out in equation (9) or (23) above.

Figures 7, 8 Phase Reversal Detector with Averaging Figure 7 shows an alternative embodiment similar to that of figure 2 and additionally having a moving averaging window 450. This takes the output of the phase difference calculator and determines an average phase difference from a series of phase differences determined at different times. The moving window refers to the time span of the series of values and should not be confused with the short window used in the earlier mentioned transform. The output of the moving average window element is fed to the phase reversal decision block 230 which can remain the same as that described above with reference to figure 2. The other blocks, the approximated Fourier coefficient and the phase difference calculator, also can be implemented as described above, though the transform to the frequency domain 410 may be implemented with or without the approximation described above. Corresponding reference numerals have been used where appropriate.

One way of implementing the moving averaging window is shown in figure 8. This block performs an averaging of the phase differences. Depending on D (the delay value used in the phase difference calculator as described above), the phase difference containing a phase reversal will be computed at several different times. Because of noise in the incoming tone, the quality of phase difference computed is degraded. By computing it several times and averaging the values, the noise effect will be reduced.

As negative and positive phase differences are equivalent, an absolute value is taken first using element 480. This is fed into delay lines 490 (in the form of a buffer with overlap). The average is generated by the matrix sum element 500 fed by the buffer. In other words the elements at the output of the absolute function can be averaged.

The size of the window in the Moving Averaging Window can be related to the time period of the delay D between the two Fourier coefficients used to calculate the phase difference. This is because having a delay enables more than one successive phase difference to span a phase reversal. Hence ideally, the length of the window should correspond to D. If D is longer, then fewer than the maximum phase differences are being used in the average, so less than optimum noise immunity is obtained. If D is shorter than the window, then some phase differences which do not span the phase reversal will be included in the average so the average will be diluted and so less accurate.

Figure 9 Overlapping short windows Figure 9 shows three short windows each having 8 samples or points arranged to overlap so that the second window overlaps the first and third. All frames are effectively camera frames in this example, with no intermediate frames, though it is possible to add intermediate frames to this example. The use of overlapping windows or frames can be seen as an alternative way of enabling a phase reversal to be covered by more than one phase difference calculation. It represents an alternative to the use of camera frames and moving averages, described with reference to Figures 4-8 and 10. In principle, it is also possible to apply this frame overlap to the method using camera frames and moving averages. This could give some improvement in accuracy of the average phase difference, for the worst case of a PR occurring in the middle of a frame, though at the expense of more calculation.

"m" is the distance (in samples) between the position of a PR and the nearest start point of a frame. Clearly if a PR occurs in the sample shown as m=0 or m=1 or m=2 or m=3 or m=4 the nearest start point of a frame is the start point of their own frame. But if the PR occurs in sample m=5 or m=6 or m=7 the nearest start point of a frame is the start point of the next frame. Thus m=5 for one frame is m=3 for next frame and so on. To improve accuracy for the worst case of a PR occurring in the middle of a frame, whichever of the two overlapping calculated phase differences has a greater value, can be chosen, rather than taking an average. It is possible to have more than 50 % overlap, for more accuracy at the expense of calculation load.

One advantage of overlapping as shown in figure 9 is that it can enable the speed of detection to be increased, as will now be explained. By overlapping, it is possible to reduce D and still have a PR covered by multiple phase difference calculations. By reducing D, the speed of detection is increased, because the detection calculation usually takes at least 2D frames after the PR. If D is reduced without overlapping, the number of phase difference calculations covering the PR will also reduce, causing reduced accuracy.

Figure 10 OverlapDing time periods This figure relates to the phase difference calculation using camera frames, as described above with reference to figures 7 and 8. It shows several rows each representing a successive time. A phase difference calculation is made at each of the times represented by the rows. The boxes in each row represent a short window or frame of samples of the incoming tone signal. The columns at each end marked "first camera frame" and "second camera frame" are those frames used for determining fourier coefficients. The phase difference is taken between the first and second camera frames in each row. The shaded part indicates the part after a phase reversal. As time advances, the time period of the capture window between the camera frames effectively moves over the phase reversal point in the incoming signal. As can be seen the phase reversal point is captured by a number of rows, and therefore by a number of phase difference calculations. Any number of the rows can be included in the moving averaging window to derive the average phase difference. Preferably as explained above, the averaging window should match the delay D between the camera frames or fourier coefficients used for the instantaneous phase difference calculations. In the example illustrated there are 3 rows for which the camera frames are wholly on different sides of the phase reversal.

Other Variatiüns And Remarks Other variations and implementations within the scope of the claims will be apparent to those skilled in the art, and are not intended to be excluded. Although described in relation to telecommunications applications, such as echo cancellation, clearly there are many other applications, and so many claims are intended to encompass all applications. Typically the tone processing sections including the phase reversal detector and other modules in the central office are implemented as software modules run on one or more DSPs (Digital Signal Processor). Accordingly, the phase reversal detector features can be implemented in well known programming languages such as C or Ada, or others, as would be well known to those skilled in the art. The resulting code can be cross-compiled into a lower level language appropriate to run on a DSP, such as the fixed or floating point types made by TI or Motorola or others, or on a general purpose microprocessor, or any type of firmware, or programmable or fixed hardware, or any combination. The software can in principle be implemented as instructions or as combinations of data, instructions, rules, objects and so on. Some features can in principle be implemented in dedicated hardware for greater speed of operation.

As has been described above, a phase reversal detector converts an input tone from time domain into frequency domain, determines a phase difference over a time interval, and detects a phase reversal, based on the phase difference. The converter derives a Fourier transform coefficient having a centre frequency corresponding approximately to a principal frequency being sought in the incoming signal. It uses a very short window of samples, the window length being selected to be shorter than that required for an exact correspondence. By using an approximate correspondence a much shorter window can be used, which enables a significant reduction in computational load and a shorter detection time. An 8 sample window can be used to give a coefficient at 2000Hz, close enough to a 2 100Hz tone. The converter output is a complex number.

The phase difference is obtained using a dot product of the outputs of the converter obtained at different times. An average of consecutive phase differences can be used to reduce the effect of noise.

Claims

Claims 1. A phase reversal detector having: a converter for deriving

phase information by applying a transform to an input signal, an output of the transform being a complex number whose phase is approximately linearly related to a phase of the input signal, a phase difference calculator for determining a difference in phase of the input signal over a time interval, using the phases of the complex numbers derived for that time interval, and a phase reversal detector for detecting a reversal in the phase, based on the phase difference.

2. The detector of claim 1, the transform being a Fourier type transform, for which a principal frequency f of the input signal, a sampling rate T, and a window length N of the transform are arranged such that f*T*N is approximately an integer.

3. The detector of claim 2, the transform involving deriving a transform coefficient corresponding approximately to the principal frequency of the input signal, the window length being selected to be shorter than that required for an exact correspondence.

4. The detector of claim 2, the output being based on a transform coefficient chosen as having a centre frequency which corresponds to within 20% of the principal frequency.

5. The detector of claim 2, the window length being less than 20 samples of the input signal.

6. The detector of claim 2, the window length being 8 samples, to give a Fourier coefficient at 2000Hz.

7. The detector of claim 1, the phase difference calculator being arranged to carry out a dot product of the complex numbers.

8. The detector of claim 1, the phase difference calculator being arranged to carry out two consecutive dot product operations.

9. The detector of claim 1, arranged to determine an average of a series of phase differences detected at different times.

10. The detector of claim 9, each of the differences being a change in phase of the input signal over a time period, the time periods being arranged to overlap.

11. The detector of claim 10, the time periods being arranged to overlap such any time is covered by two or more of the time periods.

12. The detector of claim 1, arranged to determine the phase difference in complex values, then convert to a phase angle.

13. The detector of claim 1, in the form of software.

14. An echo canceller having a detector as set out in claim 1.

15. A method of providing telecommunications services to subscribers, over a route in a telecommunications network using the echo canceller of claim 14.

16. A phase difference calculator for processing complex numbers representing phase information, each of the complex numbers having a phase which has a linear relationship to the phase information, the calculator having: a first dot product generator for generating a dot product of two of the complex numbers, to derive a phase difference, and a second dot product generator for removing a constant from the phase difference.

17. The phase difference calculator of claim 16, the complex numbers representing phase information of a single input signal at different times.

18. The phase difference calculator of claim 17, arranged such that the series of phase differences are calculated over different time intervals, the time intervals being arranged to overlap, and an average phase difference being determined from the series.

19. The phase difference calculator of claim 17, additionally having a phase reversal detector for detecting a phase reversal from the phase difference.

20. The phase difference calculator of claim 17 arranged to determine a series of phase differences, each difference being determined over a time interval of more than one sample, the series being such that the time intervals overlap, and the calculator having an averager for determining an average phase difference from the series of phase differences.

21. A phase reversal detector having: a converter for converting an input signal from time domain representation into a frequency domain representation, a phase difference calculator for detecting a difference in phase of the input signal over a time interval, using the frequency domain representation, and a phase reversal detector for detecting a reversal in the phase, using an average of a series of phase differences detected by the detector at different times.

22. The detector of claim 21, the time periods being arranged to overlap.

23. The detector of claim 22, the time periods being arranged to overlap such that there is always coverage by two or more of the time periods.

24. The detector of claim 21, arranged to determine the phase difference in complex values, then convert to a phase angle.

25. The detector of claim 21, in the form of software.

26. An echo canceller having the detector of claim 21.

27. A method of providing telecommunications services to subscribers, over a route in a telecommunications network using the echo canceller of claim 26.

28. A method of using a telecommunication service as a subscriber, provided over a route in a telecommunications network using the echo canceller of claim 26.

29. A converter for deriving phase information of an input signal by applying a transform to the input signal, an output of the transform being a complex number whose phase is approximately linearly related to a phase of the input signal, transform being a Fourier type transform, for which a principal frequency f of the input signal, a sampling rate T, and a window length N of the transform are arranged such that f*T*N is approximately an integer.