Title of Invention  A METHOD AND APPARATUS FOR GENERATING A CLOCK ERROR ESTIMATION 

Abstract  A method and apparatus is provided that computes an optimal estimate of known clock frequency error between the transmitter and receiver using a known pilot signal and the statistics of the noise process known. The estimate is computed such that the residual clock error is below the least count (the smallest frequency correction that can be imparted) of the VCXO that controls the receiver sample clock. A tracking technique based on a measure of drift in taps of frequency domain equalizers of different subcarriers is disclosed. This tracking ensures that the residual mean square error is within a predefined bound. Finally, the least count effects in digitally controlled oscillators (DAC controlled VCXOs and Numerically Controlled Oscillators (NCXO)) are addressed by a dithering mechanism. The dithering mechanism involves imparting positive and negative clock corrections for different lengths of time in such a manner that the residual clock error becomes zero mean. In data mode, a tracking scheme makes uses of variations in frequency domain equalizer taps for determination of clock error estimates, computes a residual clock error estimate different from the clock error estimate generated from pilot channel using training mode scheme, and SNR based combination of errors is computed to obtain clock correction, and a dithering mechanism computes the actual correction to be given to the VCXO such that the residual phase error is maintained at an acceptably low value. 
Full Text  ( BACKGROUND OF THE TNVENTTON 1. Field of the Invention. This application pertains generally to data communication systems and more specifically to the field of clock recovery in multicarrier systems. 2. Description of the Background Art. A multi carrier traogmission system is one that employs firequency division multiplexed (FDM) subcarriers for transmission. A comprehensive description of multicarrier systems is given by John A. C. Bingham in "Multicarrier Modulation for Data Transmission: An idea whose time has come", EEEE Communication Magazine, Vol. 28, No. 5, pp. 514, "May 1990. In conventional communication systems a receiver has to perform clock and carrier recovery. However, typical multicarrier systems use lookup tables for generating individual carriers through a sample clock. As a result carrier firequencies are dependent on the clock firequency. Therefore independent carrier firequency recovery is not required. Clock recovery is the problem of synchronizing a receiver"s clock to that of the transmitter. The problem of clock recovery is made complicated, however, by the presence of channel noise, frequency offset, clock drift and phase jitter. Channel noise comprises thermal noise and noise induced due to other similar systems existing in that environment. Frequency offset can arise due to other communication equipment present in the channel. These equipment = wmodui&tion and subsequent demodulation may shift the signal spectrum, as described by John A. C. Bingham in "Method and Apparatus for Correcting for Clock and Carrier Frequency Offset and Phase Jitter in Multicarrier Modems", United States Patent No. 5,228,062, herein incorporated by reference, and issued July 13,1993. Clock drift is caused by crystal imperfections and temperature variations. Phase jitter is mainly attributed to power line coupling, however phase jitter can also arise from crystal imperfections and other circuit limitations. Any clock recovery circuit has to therefore estimate these parameters and then track them in the presence of frequency offset, clock drift and phase jitter. In the past many schemes have been suggested for addressing the above problems. Almost all ofthese techniques are based on the use of phase locked loops (PLLs). These schemes work well in general, however the loop filters need to be designed carefiilly to provide the required performance. In multicarrier systems the symbol periods are typically large and corresponding PLL delays would also be large, making the fracking problem difficult. The clock recovery problem in multicarrier systems has been addressed by U.S. Patent No. 5,228,062. In this method, a combination of block and serial processing techniques have been proposed for compensating offset and jitter. These schemes use a plurality of pilot tones during an initial training mode for estimation of clock error, frequency offset and phase jitter components. The training mode is followed by a data mode, dining which the actual data is encoded into symbols (referred to as data mode symbols hereinafter) and transmitted through the channel. In data mode, the clock error due to clock drift, frequency offset and phase jitter estimates are updated using data mode symbols. This scheme and its limitations are described below. The clock error estimation is done by first fitting a straight line to the phase error profile of the received symbols. The straight line fit is typically carried out in a linear least square sense, where the parameters of the straight line are found in such a way that the square of the error between the resulting straight line and the phase error profile is minimum. The slope of the linear fit is then passed through a low pass filter designed using conventional PLL theory to obtain a correction for the clock circuit. This scheme however is limited in performance due to two main reasons: a) The linear least squares estimation carried out is suboptimal when the noise in the phase samples is coloured. Noise colouring is relevant in some applications of multicarrier systems, such as those that work in an Asymmetric Digital Subscriber Line (ADSL) environment. In this environment noise resulting fiom other loops in the same binder is coloured and moreover is a dominant component of the total received noise. Further description of the noise in an ADSL environment can be found in ANSI Tl.4131998, American National Standard for Telecommunications Network and Customer Installation Interfaces  Asymmetric Digital Subscriber Line (ADSL) Metallic Interface, 1998. b) The clock recovery and tracking proposed in many conventional systems involve sending a fiequency correction to the receiver clock. The correction can be incorporated through a change in the receiver clock or by carrying out interpolation of received samples. Typically a voltage controlled oscillator (VCXO) is controlled through this correction to change the receiver clock. It is customary to provide the VCXO input through a digital to analog converter (DAC) the input to which is the correction value provided as a discrete number. These systems are limited in their ability to provide a continuous range of clock frequencies (since the frequency correction resolution is determined by the smallest nonzero quantity the DAC can output), resulting in a nonzero residual error within the least count of the system. Such an error unless explicitly randomized will manifest as a biased phase error and eventually lead to loss in synchronization and failure. Typical implementations do not address this problem as estimation errors due to system noise and clock jitter provides some degree of randomization. The measurement of clock error can also be carried out using the scheme proposed by G. Ungerboeck, "Fractional TapSpacing Equalizer and Consequences for Clock Recovery in Data Modems", IEEE Tr. Communications, COM24, No. 8, pp. 856864, Aug., 1976. This scheme uses a measure of the drift in the taps of a time domain equalizer to get an estimate of the frequency correction. It is particularly suited for communication receivers that utilize adaptive equalizers. However, these schemes cannot be directly applied to high data rate multicarrier systems, as the updation of time domain equalizer taps is computationally expensive. This invention proposes a novel scheme for clock synchronization in multicarrier systems which addresses the limitations described above, including the effect of channel noise and clock drift on the synchronization problem. The effect of frequency offset and phase jitter is not explored in this invention. However, those skilled in the art can combine conventional frequency offset and phase jitter solutions with the novel scheme proposed in this invention for a complete clock recovery. The proposed scheme also provides two additional features. The first is an optimal method for determining the acquisition time, which is used for reducing the training time. The second is a procedure for optimal tracking of nonstationarity arising out of clock drift. SUMMARY OF THR TNVENTTON In accordance with the present invention, an optimal estimate of clock frequency error between the transmitter and receiver using a pilot signal and the statistics of the noise process is computed during a training mode. In a preferred embodiment, the estimate is such that the residual clock error is below the least count (the smallest frequency correction that can be imparted) of the VCXO that controls the receiver sample clock. The present invention also provides for optimal estimates for systems in which colored noise is introduced, for example, in an ADSL environment. Fiulher, a tracking technique based on a measure of drift in taps of frequency domain equalizers of different subcarriers is disclosed. This fracking ensures that the residual mean square error is within a predefined bound. Finally, the least count effects in digitally confroUed oscillators (DAC confrolled VCXOs and Numerically Controlled Oscillators (NCXO)) are addressed by a dithering mechanism. The dithering mechanism preferably involves imparting positive and negative clock corrections for different lengths of time in such a manner that the residual clock error becomes zero mean. A multicarrier system which consists of a training phase preceding a data transmission phase is provided. The training phase includes a transmission of a known signal in a particular subcarrier referred to as a pilot channel. This is used to obtain an optimal estimate of the error between the transmitter and receiver clocks in the following way: M) Determination of the discrete Fourier transform coefficients of the pilot channel which are used to a) obtain the signal to noise ratio (SNR) in the channel and b) obtain the autocorrelation statistics of the noise process. 2) Design of a minimum variance unbiased estimator filter using the SNR, noise autocorrelation statistics and predefined estimation error which could, for example, be based on the least count of the system that controls the sample clock. This ensiffes that the filter designed is optimal in the number of samples required for a given noise statistics and estimation error. Alternately, it provides an estimate with minimum variance for a given number of samples and noise statistics. This filter is used for determination of the clock error estimate that is sent as a correction to the receiver clock. The training phase may also involve determination of the following: a) parameters of the phase jitter and fiequency offset components in the system which can be done using techniques described in by U. S. Pat. no. 5,228,062, b) taps of a fiequency domain equalizer for compensating the chaimel characteristics. The initial estimate is then refined during data mode in the following way: 1) The firequency domain equalizer taps are adapted using a least mean squares (LMS) algorithm using the decoded output as a reference signal. A measure of the drift in these taps is used to obtain an estimate of the residual clock error. 2) If the pilot signal is continued in the data mode, then the scheme deployed during training mode is used to obtain another estimate of the residual clock error. The error estimates obtained in the above steps are combined to obtain a possible clock correction. 4) The correction obtained is input to a dithering algorithm to obtain the actual correction. The dithering algorithm ensures that the phase error does not build up beyond manageable proportions whenever the clock error is below the least count. The filter is optimal in the number of samples it uses for determination of estimate. This can be used to terminate clock recovery faster than existing schemes, allowing for shorter training times. Conventional phase jitter and frequency offset tracking mechanisms can also be easily combined with the above optimal clock recovery mechanism. In data mode, a tracking scheme that makes uses of variations in firequency domain equalizer taps for determination of residual clock error estimate, different fro"m the error estimate generation method used in the training mode is disclosed. A SNR based combination of errors is computed to obtain an estimated residual clock correction value. A dithering mechanism computes the actual coirection to be given to the VCXO such that the residual phase error is maintained at an acceptably low \alue. The invention provides for a method for generating a clock error estimate in order to synchronize a receiver clock with a transmission clock in a multicarrier transmission system in which possibly colored noise is introduced during the transmission of a data signal, comprising receiving a pilot signal in a predetermined subcarrier frequency; determining the signaltonoise ratio of the pilot signal; determining noise characteristics of the pilot signal; and filtering the pilot signal to generate a clock error estimate responsive to the signaltonoise ratio, the noise characteristics and a predefined estimation error. A method of correcting residual clock error in a multicarrier transmission system in which data signals are transmitted to a receiving system and are sampled in accordance with the output of a receiving clock system, comprising generating an initial clock error estimate; adjusting the output of the receiving clock system responsive to the generated initial clock error estimate; receiving data signals from the transmission system; measuring drift in frequency domain taps of the receiving system to obtain an estimate of the residual clock error; adjusting the receiving clock output responsive to the estimated residual clock error; and sampling the data signals in accordance with the adjusted receiving clock system output. An apparatus for generating a clock error estimate in order to synchronize a receiver clock with a transmission clock in a multicarrier transmission system in which possibly colored noise is introduced during the transmission of a data signal, comprising a data sampler, for sampling an input pilot signal of a predetermined carrier frequency in accordance with a clock signal received from a receiver clock system; a phase rotation analyzer, coupled to the data sampler and a test pilot signal input, for comparing a phase of the input pilot signal to a phase of the test pilot signal to determine a phase rotation of the input pilot signal, wherein the test pilot signal has substantially the same phase as the input pilot signal prior to transmission of the input pilot signal; a signaltonoise ratio estimator, coupled to the phase rotation analyzer, for computing a signaltonoise ratio for the input pilot signal responsive to the phase rotation of the input pilot signal; a noise characteristic estimator, coupled to the phase rotation analyzer, for computing noise characteristics of the input pilot signal responsive to the phase rotation of the input pilot signal; a linear filter, coupled to the noise characteristic estimator, the signaltonoise ratio estimator, the phase rotation analyzer, and to which a least count value of the clock recovery system is input, for computing a number of taps and weights of the taps responsive to the signaltonoise ratio estimate, the noise characteristic estimate, the least count value, and the samples received from the phase rotation analyzer to generate a linear estimate of the clock error; the receiver clock system, having a clock control input coupled to the output of the linear filter, for adjusting an output clock signal responsive to the clock error estimate generated by the linear filter. DETAILED DESCRIPTION OF THE TNVENTTON The transmitter 32 and receiver 28 units of a multicairier system connected through a subscriber loop are shown in Fig. 1. The input data bits are passed through a serial to parallel buffer and encoder 5 and modulated using a multicarrier modulator 6. In multicarrier systems, the preferred method of modulation is using the inverse discrete Fourier transform (IDFT). The modulated data may be further processed using other signal processing modules and then conveted to serial data 7. This data is transmitted onto the subscriber loop or channel 11 in the analog from using a digital to analog converter (DAC) 8. The noise in the system is shown additive 9 at the receiver end and is usually coloured, comprising crosstalk and thermal noise. At the receiver, the data is converted into digital form using an analog to digital converter (ADC) 12 and processed further before going through a serial to parallel converter 13. This data is demodulated using a multicarrier demodulator 14, decoded and converted to serial mode to obtain the transmit data bits 15. The clock recovery problem pertains to matching the sample clocks at the transmit and receive side. The oscillators at the transmitter and receiver for controlling the sampling rate though nominally equal have a frequency difference within a known bound (typically 50 parts per million). The differences arise because of physical characteristics of crystals and temperature variations. In the following, a software implementation of the clock recovery scheme for training and data mode is disclosed. Those skilled in the art would recognize that the schemes could be implemented in hardware, or any convenient combination of hardware and software. Clock recovery mechanisms at the receiver modify the receiver clock so as to synchronize the transmitter and receiver sampling rates. If the sampling rate at the transmitter is indicated by fs, and at the receiver by fs", then the clock recovery mechanism has to estimate delta_f=fs"fs, which is then sent as a correction to the receiver clock. Typically, fs and fs" are large compared to deha_f The mechanism for computation of delta_f is divided into two phases, the first phase, training, is described below. • Training phase: A pilot signal is transmitted as shown in Fig. 2. The pilot is generated using the pilot signal generator 31 and transmitted onto the channel 11 using the transmitter 32. The signal passes through the channel 11 as indicated in Fig. 1. The noise corraption during transmission through the channel is shown as a noise source 33 whose output is added to the channel 11 output. The ADC 12 at the rate fs" samples the received signal corrupted by noise as shown in Fig. 2. Samples for a symbol are collected in a buffer 20 and then input to a phase rotation analyzer 25. In a preferred embodiment, the phase rotation analyzer 25 comprises a DFT module 30 which receives the input signal and computes the DFT coefficients. The phase rotation of the received signal from the transmittal signal, produced by pilot signal generator 35, is computed by the rotator block 40 and output into the arctan computation block 50, after multiplying with the complex conjugate (block 45) of the previous output (block 34) of the phase rotator 40. The arctan block 50 computes the difference between consecutive phase values and inputs the difference into two blocks, the SNR estimator 60 and the autocorrelation estimator 70 for analyzing noise characteristics of the signal. The SNR estimator 60 computes the variance of a given block of input samples and outputs the SNR value. The autocorrelation estimator 70 computes autocorrelations at various lags. Based on the SNR, autocorrelation values and least count of the control system, denoted by LC, (obtained from system parameters) the minimum variance imbiased estimator (MVUE) block 80 computes the number of taps and their weights for an optimal hnear estimate (this method is described subsequently). The samples used for SNR autocorrelation estimation can be reused for computing the final estimate by passing them through a delay unit 55. Using this . ^Stimat^ the clock correction for training mode is derived. The theory for the blocks above is described below. Note that the individual estimation strategies used below are only representative; other estimators can be used without changing the overall nature of the scheme, however these will be suboptimal. The frequency estimator used in the above scheme is a linear estimate operating on the phase of the received sinusoid. The theoretical lower bound of an unbiased estimator is given by CramerRao lower bound (CRLB). Linear estimators are computationally less expensive and it is possible to design a linear MVUE for frequency estimation problem which can achieve CRLB whenever the SNR is reasonably good (> 6 dB), which is so in the case of typical DMT systems. Let the received signal in a symbol be x(i) = s(i) + w(i) (1) where s(i) are the samples of a symbol and w(.) is the additive Gaussian noise. The following are reasonable assumptions for the ensuing analysis: A) E[s(i)*w(i)] = 0 for all i, i.e. noise is independent of signal. B) w(.) is wide sense stationary (WSS); in particular, w(.) could be identically distributed with finite variance. C) E[w(i)] = 0 for all i, i.e., w(.) is zero mean. The procedxire for frequency estimation is as follows: 1. DFT coefficient corresponding to pilot channel fc is computed using N samples of the input signal x(.) (use shifting window). y"(k)r= sum(x(N*k+i)*exp(j*wc*i)) k=0,l,2,...., M1 ... (2) i=0 where wc = 2 * pi * fc / fs, fs is the sampling rate. This corresponds to the block 30. 1.5. The samples y"(k) are rotated by the phase of the transmitted symbol to obtain y(k). This is done in block 40. 2. The variables z(.) are then computed from y(.) as follows: z(k) = angle( y(k) conj(y(kl))) k = 1,3,..., M1. This is done in block 50. 3. A weighting window h of length M1 is determined. M is obtained from the SNR and noise statistics of z(.), as follows: In the presence of clock error, the SNR estimated using DFT coefficients does not correspond to the actual SNR. In order to obtain a proper estimate, the variance of the phase noise of DFT coefficients can be computed as follows. Let s(.) in Eq. (1) be a sinusoid of amplitude A and frequency wc. SNR of s(.) is 20*log(A*N/(2*a)), where a^ = E[Sk w(k)*exp(j*wc*k)*2, w"(l)*exp(j*wc*l)] / (N*N) Phase samples of z(.) are obtained at the output of the block 50. Variance of phase noise of DFT coefficients is cj^p^ = C7V(2* A*A). Noise in z(.) corresponds to difference of a white Gaussian process as DFT operation whitens . . the additive noise. Clearly, variance of the coloured noise corrupting z(.) is 2*(j pi,. Hence SNR can be computed from the variance estimator of z(.) as: SNR = 10*log(2*(T"pJ. Let fee be the least count of clock correction. Further, let f be the frequency being estimated and fs be the sampling rate. If the estimation error e has to be below LC with 95% confidence, it is required that 3e = 2 71 f N LC / fs" For the phase samples corrupted by white noise, the estimation error is given by e" = 6/(SNR*M(M"l)) The value of M can be computed using the above two expressions. In case the noise is coloured, the error expression becomes an indirect function of M and the required value may be computed iteratively. 4. The weight vector for MVUE is given by h = R"^ (1) ones(Ml) / (ones(Ml)" R"" ones(Ml)) where R is the autocorrelation matrix of the noise corrupting z(.). ones(Ml) is the column vector of size M1, whose entries are unity. For example, when the noise corrupting y(.) is white, the weight vector h can be derived to be: h(i) = 1.5 M (M*M1)" (1 (2*iM+2)" /M") i=l,2,...,Ml. The weights obtained are used to filter the output of block 50 as shown in block 90. The outputs used for determining the filter parameters can also be used in this step. The filtering is carried out as follows: deha_f_est = sumi.n„Mi (z(i)*h(i)) ( fs*fs / (2*pi*fc*N)) where z(.) is the output of block 50. The output of 90 is a clock correction which is used to modify fs". The modification may be carried out in either the clock input to the ADC through a VCXO or NCXO. It may also be used to modify the interpolator which resamples the output from the ADC. The initial estimation of phase jitter and frequency offset parameters are also carried out during training mode. One of the ways of estimating the jitter parameters is detailed in by U.S. Pat. no. 5,228,062, issued July 13,1993. Tracking clock errors during data mode In typical implementations of multicarrier receivers, fiequency domain equalization is used for channel equalization. In the present embodiment of the invention, the presence of a trained firequency domain equalizer, whose taps are adapted during data mode, is assumed. The clock recovery mechanism in operation during data mode has to find a suitable interval at which clock corrections are provided. The tracking interval is decided based on the SNR, residual jitter, clock drift and other relevant parameters. Fig. 3 gives the block schematic for estimation of clock error using the firequency domain equalizer (FDE) taps. In typical implementations of multicarrier receivers, the FDE taps are updated during data mode, using an algorithm such as LMS. In such a case, a measure of the clock error is obtained using the variation in the equalizer taps themselves. It is customary to adapt the FDE taps using the difference between the FDE outputs and decoder outputs. The demodulated symbol in the ith subband 100 is multiplied by the FDE tap cj(n), at instant n, to get the output d_i(n). The symbol is then passed through the decoder 110 (which, in the simplest case, could be a quantizer) to get the decoded symbol r_i(n). The error between the demodulated and decoded symbol is used to update the tap c_i(n) 115 using the LMS block 120, with step size mu, to yield the tap for the instance n+1, c_i(n +1). The delay block 116, complex conjugation block 117, the multiplier 126 and the arctan block 118 compute z_i(n) = angle(c_i(n+l) conj(c_i(n))) to obtain the phase difference delta_phi(i) between the taps at successive instances of time. The phase difference, for small mu (which is typical during data mode adaptation), is given by deltaj3hi(i) = 2 * pi * f_0 * i * mu *delta_f / fs""2 = c * i * delta_f where f_0 is the frequency separation between successive subbands, c = 2 * pi * f_0 * mu / fs"^2 and delta__phi(i) is the incremental phase due to clock error delta_f The estimate of delta_f(i), the clock error as measured in the ith subband, is computed over the estimation length N which is computed using SNR and CRLB as described in data mode step 4 above. The estimation is carried out using the delay block 119, SNR estimator 123, MVUE weight vector generator 122 and the simimation unit 121, as in Fig. 2. While the sequel shows the derivation of the estimator for the white noise case, coloured noise can be handled as specified in case of the training mode. We use z_i(.) to compute an estimate of delta_phi(i). The expression for determining delta_phi(i) is delta_phi(i) = sumj., ,,,^1 hCj) * z_i(j) The clock error estimate from subband i is then obtained as delta_f(i) = delta__phi(i) / ( c * i) j^aving obtained an estimate of clock error in various subbands, we now provide a scheme for combining these estimates depending on the estimation error in each subband. This scheme results in the estimation error achieving CRLB as shown below: Let e_i denote the error in the estimation of delta_f(i) in subband i. Then E[e_i] = 0 and E[e_i^2] = 6 / (c^2 i"^2 SNRJ N^3 ) = b_i^2 where SNRJ is the SNR of c_i(.). Usually the estimation errors in different subbands are uncorrelated; hence the estimate of timing correction, deha_f, is given by delta_f = sum;.! „, ^ d_i delta_f(i) where M is the number of subbands over which the tapbased estimator is computed; d_i = (l/b_i^2) / sumi„, ^M (l/b_i"2 ), If the SNR in subband i is low, d_i can be set to zero in order to obtain a good correction. The estimator block 150 computes the estimate in accordance with the just mentioned equations. The estimation error is given by E[ (delta_f  delta_f)^2 ] = 1 / sum,., ,„^, ( l/b_i"^2) which is the same as the CRLB for this estimation problem. In order to obtain small estimation errors, it is required that the estimation is carried over only those subbands that have high SNR. A preferred implementation of frequency correction during data mode is shown in Fig. 4. The FDE taps in subbands 1,2,.. .,M1 are denoted by cl, c2,... Each of these taps form the input to the subband frequency estimators 151, 152, 153,154, respectively. These estimators measure the frequency of the each subband using the estimator described for the training mode. The correction derived from each of the subbands 151, 152, 153 and 154 is combined in the estimator 150. The value estimated in each subband is weighted by the weights di and added in the block 150 to get the FDE based clock error estimate. The procedure for computation of d_i is as described earlier. Note that those subbands that have high SNR receive more weight in this scheme. The final clock error estimate is computed using the estimate derived using the FDE taps and the one obtained using the pilot frequency. In the particular realization depicted in Fig. 4, the estimate is derived as a linear combination of the two individual estimates. The weights h and h" are chosen based on the confidence in the corresponding estimates. This could be done, for instance, by setting each weight inversely proportional to the corresponding estimation error and normalizing in such a way that the sum of h and h" is imity. The scheme for computing the clock error estimate using the pilot channel is the same as the one described during the training mode, except that the estimation length is equal to the tracking length N. In case the pilot channel is not available during data mode the same scheme as used for any other subchannel can be applied to this channel too. Block 160 computes the actual ^timate of clock correction as a weighted linear combination of the frequency domain equalizer based estimate and pilot based estimate. Once the difference in transmitter DAC and receiver ADC clock frequency has been reduced to a value below LC, it is impossible to further reduce the difference by sending corrections to the VCXO. The aim then is to see to it that the residual frequency error does not result in the accumulation of a biased phase that in the end would lead to a loss in synchronization. To alleviate this problem, a ditherer 170 is applied which ensures that once the frequency error has achieved an uncorrectably small value, the same does not lead to a persistent oversampling or undersampling as time goes on. The frequency error in the scheme is randomized about zero in such a way that on the whole the system maintains a small zeromean frequency error. The ditherer 170 comprises of two cyclic buffers 180,190 and a dither analyzer 165. The error estimate buffer 180 stores the incoming required correction values, while the actual correction buffer 190 stores the actual corrections imparted for each of the input corrections. If the actual corrections over some instances happen to be zero (i.e., all the corrections computed over this instances have absolute values below LC), the residual frequency error will result in a phase error, which if allowed to build up over time, could cause loss of synchronization. In order to avoid this problem, the dither analyzer 165 sets counter 200 to zero every time the new value written into the actual correction buffer 190 is nonzero. The counter is incremented every time a zero is written into buffer 190. Whenever the count exceeds a predeteraiined threshold Nc, a threshold detector 157 is activated and the dither analyzer sends a nonzero correction whose magnitude is equal to LC, and sign the same as that of the most recent corrections is given to the clock correction module 210. The correction is written into the actual correction buffer and the  counters reset to zero. The correction is fed to the clock corrector block 210, which implements the correction mechanism. The numbers N and Nc are chosen in such a manner that the net buildup of phase error during (N * Nc) symbol intervals does not affect the decoding process within the prespecified error tolerance. The above described embodiment is a preferred embodiment of the invention. The individual estimation strategies used in training and data mode are only representative; other schemes that perform the tasks indicated could replace these without changing the overall nature of the scheme however they may lead to suboptimal results. The ways in which the schemes are combined depends upon the actual multicarrier system which may or may not have all the characteristics described above. The procedures will have to be modified depending on these characteristics, and should be obvious to those skilled in the art. Finally, all the above described procedures could either be implemented in hardware or software. While the present invention has been described with reference to certain preferred "mbodiments, those skilled in the art will recognize that various modifications may be provided, hese and other variations upon and modifications to the preferred embodiments are provided for y the present invention. We Claim: 1. A method for generating a clock error estimate in order to synchronize a receiver clock with a transmission clock In a multicarrier transmission system in which possibly colored noise is introduced during the transmission of a data signal, comprising: Receiving a pilot signal in a predetermined subcarrier frequency; determining the signaltonoise ratio of the pilot signal; determining noise characteristics of the pilot signal; and filtering the pilot signal to generate a clock error estimate responsive to the signaltonoise ratio, the noise characteristics and a predefined estimation error. 2. The method as claimed in claim 1 wherein computing the signaltonoise ratio comprises: determining discrete Fourier transform coefficients of the pilot signal. 3. The method as claimed in claim 1 wherein determining noise characteristics comprises: determining discrete Fourier transform coefficients of the pilot signal; and determining autocorrelation statistics of the noise responsive to the determined discrete Fourier transform coefficients. 4. The method as claimed in claim 1 in which the receiver clock Is generated by a control system comprising: generating a predefined estimation error responsive to a least count of the receiver clock control system 5. The method as claimed in claim 1 In which receiving a pilot test signal comprises: sampling the received pilot signal and accompanying noise; receiving a test signal having signal characteristics of the predetermined pilot signal prior to transmission of the predetermined pilot signal to the receiving system, wherein said test signal is generated locally to the receiving system; determining a phase rotation of the received pilot signal responsive to comparing a phase of the received pilot signal with a phase of the test signal; and wherein determining a signaltonoise ratio of the pilot signal further comprises determining the signaltonoise ratio of the received pilot signal responsive to the phase rotation of the received pilot signal. 6. The method as claimed in claim 5 wherein determining noise characteristics of the pilot signa[ further comprises determining the noise characteristics of the pilot signal responsive to the phase rotation of the received pilot signal. 7. The method as claimed in claim 1 comprising: sampling the pilot test signal; determining a least count value for the receiving clock system; and wherein filtering the pilot signal to generate a clock error estimate responsive to the signaltonoise ratio, the noise characteristics and a predefined estimation error further comprises: " Computing a number of taps and weights of the taps for a linear estimate filter responsive to the least count value, the signaltonoise ratio, the noise characteristics, and samples of the pilot signal. 8. The method as claimed in claim 7 wherein the samples are delayed prior to computing the taps and weights of the filter. 9. A method of correcting residual clock error in a multicarrier transmission system in which data signals are transmitted to a receiving system and are sampled in accordance with the output of a receiving clock system, comprising: generating an initial clock error estimate; adjusting the output of the receiving clock system responsive to the generated initial clock error estimate; receiving data signals from the transmission system; measuring drift in frequency domain taps of the receiving system to obtain an estimate of the residual clock error; adjusting the receiving clock output responsive to the estimated residual clock error; and sampling the data signals in accordance with the adjusted receiving clock system output. 10. The method as claimed in claim 9 comprising: dithering the estimate of the residual clock error to minimize phase error buildup; and adjusting comprises: adjusting the receiving clock system output responsive t the dithered residual clock error estimate. 11. The method as claimed in claim 10 wherein dithering the estimate of the residual clock error further comprises: determining a least count of the receiving clock system; dithering the estimate of the residual clock error responsive to the residual clock error estimate being below the least count. 12. The method as claimed in claim 11 wherein dithering comprises: receiving the estimated clock correction value; storing the estimated clock correction value; storing the estimated clock correction values to be transmitted to the receiving clock system; resetting a counter responsive to a clock correction value to be transmitted being nonzero; Incrementing the counter responsive to a clock correction value to be transmitted being equal to zero; transmitting a nonzero clock correction to the receiving clock system responsive to a value of the counter exceeding a predetermined threshold. 13. The method as claimed in claim 12 wherein the nonzero correction transmitted to the receiving clock system has a magnitude equal to the least count value of the receiving clock system and a sign equal to a most recent estimated clock correction transmitted to the receiving clock system. 14. The method as claimed In claim 9 comprising: receiving a pilot signal; computing a clock error estimate responsive to the pilot signal; combining tlie clock error estimate computed from the pilot signal with the estimated residual clock error; and wherein adjusting comprises: Adjusting the receiving clock system output responsive to the combined clock error estimate and the residual clock error estimate. 15. An apparatus for generating a clock error estimate in order to synchronize a receiver clock with a transmission clock in a multicarrier transmission system in which possibly colored noise is introduced during the transmission of a data signal, comprising: a data sampler, for sampling an input pilot signal of a predetermined carrier frequency in accordance with a clock signal received from a receiver clock system; a phase rotation analyzer, coupled to the data sampler and a test pilot signal input, for comparing a phase of the input pilot signal to a phase of the test pilot signal to determine a phase rotation of the input pilot signal, wherein the test pilot signal has substantially the same phase as the input pilot signal prior to transmission of the input pilot signal; a signaltonoise ratio estimator, coupled to the phase rotation analyzer, for computing a signaltonoise ratio for the input pilot signal responsive to the phase rotation of the input pilot signal; a noise characteristic estimator, coupled to the phase rotation analyzer, for computing noise characteristics of the input pilot signal responsive to the phase rotation of the input pilot signal; a linear filter, coupled to the noise characteristic estimator, the signaltonoise ratio estimator, the phase rotation analyzer, and to which a least count value of the clock recovery system is input, for computing a number of taps and weights of the taps responsive to the signaltonoise ratio estimate, the noise characteristic estimate, the least count value, and the samples received from the phase rotation analyzer to generate a linear estimate of the clock error; the receiver clock system, having a clock control input coupled to the output of the linear filter, for adjusting an output clock signal responsive to the clock error estimate generated by the linear filter. 16. The apparatus as claimed in claim 15, wherein the phase rotation analyzer comprises: a discrete fourier transform module, coupled to the output of the data sampler, for determining discrete fourier transform coefficients for the input pilot signal; a rotation comparator, coupled to the discrete fourier transform module and a test pilot signal, for computing a difference inphase between the test pilot signal and the input pilot signal; a modified phase rotation estimator, coupled to the rotation comparator, for multiplying the difference In phase between the test pilot signal and the input pilot signal with a complex conjugate of a pervious sample of the test pilot signal to generate a modified phase rotation value; and an arctan computation block, coupled to the modified phase rotation estimator, for computing a difference between consecutive phase values of the input signal responsive to modified phase rotation value. 17. The apparatus as claimed in claim 15 comprising: a ditherer, coupled to the linear filter, for receiving the clock error estimate, and generating a dithered clock estimate output responsive to a value of the clock error estimate. 18. The apparatus as claimed in claim 17 wherein the ditherer receives a least count value from the receiving clock system, and the dither analyzer comprises: an input cyclic buffer, for receiving the storing the estimated clock correction value; an actual correction buffer, coupled to the input cyclic buffer, for receiving the estimated clock correction value, transmitting a dithered clock correction value to the receiving clock system responsive to receiving a dithered clock correction value from a dither analyzer, transmitting the estimated clock correction value otherwise, and storing the transmitted clock correction value; the dither analyzer, coupled to the input cyclic buffer and a counter, for resetting a counter responsive to the estimated clock correction value being nonzero, incrementing the counter responsive to the estimated clock correction value being zero, comparing the counter value to a predetermined threshold value, and transmitting a dithered clock correction value to the actual correction buffer responsive the value of the counter exceeding the threshold. 19. The apparatus as claimed in claim 18 wherein the dithered clock correction value is a nonzero value. 20. A method for generating a clock error estimate in order to synchronize a receiver clock with a transmission clock in a multicarrier transmission system in which possibly colored noise is introduced during the transmission of a data signal substantially as herein described with reference to and as illustrated by the accompanying drawing. 21. A method of correcting residual clock error in a multicarrier transmission system in which data signals are transmitted to a receiving system and are sampled in accordance with the output of a receiving clock system substantially as herein described with reference to and as illustrated by the accompanying drawing. 22. An apparatus for generating a clock error estimate in order to synchronize a receiver clock with a transmission clock in a multicarrier transmission system in which possibly colored noise is introduced during the transmission of a data signal substantially as herein described with reference to and as illustrated by the accompanying drawing. Dated this 10th day of June, 1999 

0638mas1999 claimsduplicate.pdf
0638mas1999 correspondenceothers.pdf
0638mas1999 correspondencepo.pdf
0638mas1999 description (complete)duplicate.pdf
0638mas1999 description (complete).pdf
638mas1999 abstract duplicate.pdf
638mas1999 drawings duplicate.pdf
Patent Number  216216  

Indian Patent Application Number  638/MAS/1999  
PG Journal Number  13/2008  
Publication Date  31Mar2008  
Grant Date  10Mar2008  
Date of Filing  11Jun1999  
Name of Patentee  SASKEN COMMUNICATION TECHNOLOGIES LIMITED  
Applicant Address  139/25, DOMLUR LAYOUT, RING ROAD, DOMLUR POST, BANGALORE  560 071,  
Inventors:


PCT International Classification Number  H04L 007/00  
PCT International Application Number  N/A  
PCT International Filing date  
PCT Conventions:
