Decision feedback equalizer with dynamic feedback control Number:7,394,849 from the United States Patent and Trademark Office (PTO) owispatent

Title: Decision feedback equalizer with dynamic feedback control

Abstract: A decision feedback equalizer with dynamic feedback control for use in an adaptive signal equalizer. Timing within the decision feedback loop is dynamically controlled to optimize recovery of the data signal by the output signal slicer. The dynamic timing is controlled by a signal formed as a combination of feedback and feedforward signals. The feedback signal is an error signal related to a difference between pre-slicer and post-slicer signals. The feedforward signal is formed by differentiating and delaying the incoming data signal.

Patent Number: 7,394,849 Issued on 07/01/2008 to Ibragimov,   et al.

---

Inventors:	Ibragimov; Edem (Santa Clara, CA), Yu; Qian (Santa Clara, CA), Choudhary; Prashant (San Jose, CA)
Assignee:	Scintera Networks Inc. (Sunnyvale, CA)
Appl. No.:	11/539,810
Filed:	October 9, 2006

---

Related U.S. Patent Documents

---

	Application Number	Filing Date	Patent Number	Issue Date
	11070485	Mar., 2005	7120193
	10768408	Jan., 2004	7035330
	10322024	Dec., 2002	7039104
	10290571	Nov., 2002	6940898

---

Current U.S. Class:	375/233 ; 375/229; 375/230; 375/231; 375/232; 375/234; 375/346; 375/350
Current International Class:	H03H 7/30 (20060101); H03H 7/40 (20060101); H03K 5/159 (20060101)
Field of Search:	375/229-236,346,350

---

References Cited [Referenced By]

---

U.S. Patent Documents

5031194	July 1991	Crespo et al.
5119401	June 1992	Tsujimoto
5191462	March 1993	Gitlin et al.
5561687	October 1996	Turner
5646958	July 1997	Tsujimoto
6141378	October 2000	d'Oreye de Lantramange
6144697	November 2000	Gelfand et al.
6240133	May 2001	Sommer et al.
6829297	December 2004	Xia et al.
2001/0007479	July 2001	Kim
2001/0026197	October 2001	Tomisato et al.
2004/0091040	May 2004	Shanbhag et al.
2004/0091041	May 2004	Shanbhag et al.
2004/0240539	December 2004	Shanbhag et al.

Other References

John G. Proakis, Digital Communications Fourth Edition, 2001, pp. 598-659, McGraw-Hill Companies, Inc., New York, NY. cited by other . John G. Proakis, Digital Communications Fourth Edition, 2001, pp. 660-708, McGraw-Hill Companies, Inc., New York, NY. cited by other . Heinrich Meyr et al., Digital Communication Receivers, Synchronization, Channel Estimation, and Signal Processing. 1998, pp. 270-323, John Wiley & Sons, Inc. New York, NY. cited by other.

Primary Examiner: Odom; Curtis
Attorney, Agent or Firm: Vedder Price P.C.

---

Parent Case Text

---

RELATED APPLICATIONS

This is a division of U.S. patent application Ser. No. 11/070,485, filed Mar. 2, 2005, which is a continuation-in-part of U.S. patent application Ser. No. 10/768,408, filed Jan. 30, 2004, which is a continuation-in-part of U.S. patent application Ser. No. 10/322,024, filed Dec. 17, 2002, which is a continuation-in-part of U.S. patent application Ser. No. 10/290,571, filed Nov. 8, 2002.

---

Claims

---

What is claimed is:

1. An apparatus including a decision feedback equalizer with dynamic feedback control for adaptively controlling a pre-slicer data signal that is sliced to provide a post-slicer data signal, comprising: first signal combining circuitry that combines a feedback signal and an input signal representing a plurality of data to provide a pre-slicer signal; controllable delay circuitry, coupled to said first signal combining circuitry, that delays said pre-slicer signal in response to a delay control signal to produce a corresponding delayed pre-slicer signal; signal slicing circuitry, coupled to said first signal combining circuitry, that slices said pre-slicer signal to produce a post-slicer signal indicative of said plurality of data; decision feedback circuitry including input signal timing control, coupled to said controllable delay circuitry and said signal slicing circuitry, that feeds back said post-slicer signal in response to a timing control signal to produce said feedback signal; second signal combining circuitry, coupled to said controllable delay circuitry and said signal slicing circuitry, that combines said delayed pre-slicer signal and said post-slicer signal to produce a difference signal indicative of a difference between said delayed pre-slicer signal and post-slicer signal; signal differentiation circuitry with a selected signal delay that differentiates and delays said input signal to produce a resultant signal, wherein respective portions of said resultant signal are delayed relative to corresponding portions of said input signal by said selected signal delay; and third signal combining circuitry, coupled to said second signal combining circuitry and said signal differentiation circuitry, that combines said difference signal and said resultant signal to produce said delay control signal, wherein said selected signal delay is selected and said delayed pre-slicer signal is delayed relative to said pre-slicer signal such that said timing control signal has a substantially zero alternating current (AC) signal component.

2. The apparatus of claim 1, wherein said first signal combining circuitry comprises a signal summing circuit.

3. The apparatus of claim 1, wherein said signal slicing circuitry comprises a voltage comparison circuit.

4. The apparatus of claim 1, wherein said decision feedback circuitry comprises: signal timing control circuitry, coupled to said controllable delay circuitry and said signal slicing circuitry, that selectively delays said post-slicer signal in response to said timing control signal to produce a delayed post-slicer signal; and feedback filter circuitry, coupled to said signal timing control circuitry, that filters said delayed post-slicer signal to produce said feedback signal.

5. The apparatus of claim 1, wherein said decision feedback circuitry comprises: signal timing control circuitry, coupled to said controllable delay circuitry and said signal slicing circuitry, that selectively delays said post-slicer signal in response to said timing control signal to produce a delayed post-slicer signal; tapped delay circuitry, coupled to said signal timing control circuitry, that successively delays said delayed post-slicer signal to produce a plurality of further delayed post-slicer signals; mixer circuitry, coupled to said tapped delay circuitry, that mixes each one of said plurality of further delayed post-slicer signals with a respective one of a plurality of coefficient signals to produce a plurality of mixed signals; and fourth signal combining circuitry, coupled to said mixer circuitry, that combines said plurality of mixed signals to produce said feedback signal.

6. The apparatus of claim 1, wherein said second signal combining circuitry comprises a signal summing circuit.

7. The apparatus of claim 1, wherein said signal differentiation circuitry comprises high pass filter circuitry that high pass filters and delays said input signal to produce said resultant signal.

8. The apparatus of claim 1, wherein said signal differentiation circuitry comprises: a high pass filter circuit that high pass filters said input signal to produce a high pass filtered signal; and signal delay circuitry, coupled to said high pass filter circuit, delays said high pass filtered signal to produce said resultant signal.

9. The apparatus of claim 1, wherein said third signal combining circuitry comprises: a signal multiplication circuit that multiplies said difference signal and said resultant signal to produce a product signal; and low pass filter circuitry, coupled to said signal multiplication circuit, that low pass filters said product signal to produce said delay control signal.

10. The apparatus of claim 1, wherein said third signal combining circuitry comprises: a signal multiplication circuit that multiplies said difference signal and said resultant signal to produce a product signal; and signal integration circuitry, coupled to said signal multiplication circuit, that integrates said product signal to produce said delay control signal.

11. An apparatus including a decision feedback equalizer with dynamic feedback control for adaptively controlling a pre-slicer data signal that is sliced to provide a post-slicer data signal, comprising: first signal combiner means for combining a feedback signal and an input signal representing a plurality of data and generating a pre-slicer signal; controllable delay means for delaying said pre-slicer signal in response to a delay control signal and generating a corresponding delayed pre-slicer signal; signal slicer means for slicing said pre-slicer signal and generating a post-slicer signal indicative of said plurality of data; decision feedback means for controlling signal timing by feeding back said post-slicer signal in response to a timing control signal and generating said feedback signal; second signal combiner means for combining said delayed pre-slicer signal and said post-slicer signal and generating a difference signal indicative of a difference between said delayed pre-slicer signal and said post-slicer signal; signal differentiator means with a selected signal delay for differentiating and delaying said input signal and generating a resultant signal, wherein respective portions of said resultant signal are delayed relative to corresponding portions of said input signal by said selected signal delay; and third signal combiner means for combining said difference signal and said resultant signal and generating said delay control signal, wherein said selected signal delay is selected and said delayed pre-slicer signal is delayed relative to said pre-slicer signal such that said timing control signal has a substantially zero alternating current (AC) signal component.

12. A method for providing decision feedback equalization with dynamic feedback control for adaptively controlling a pre-slicer data signal that is sliced to provide a post-slicer data signal, comprising: combining a feedback signal and an input signal representing a plurality of data and generating a pre-slicer signal; delaying said pre-slicer signal in response to a delay control signal and generating a corresponding delayed pre-slicer signal; slicing said pre-slicer signal and generating a post-slicer signal indicative of said plurality of data; feeding back said post-slicer signal with controlled signal timing in response to a timing control signal and generating said feedback signal; combining said delayed pre-slicer signal and said post-slicer signal and generating a difference signal indicative of a difference between said delayed pre-slicer signal and said post-slicer signal; differentiating and delaying said input signal and generating a resultant signal, wherein respective portions of said resultant signal are delayed relative to corresponding portions of said input signal by a selected signal delay; and combining said difference signal and said resultant signal and generating said delay control signal, wherein said selected signal delay is selected and said delayed pre-slicer signal is delayed relative to said pre-slicer signal such that said timing control signal has a substantially zero alternating current (AC) signal component.

13. The method of claim 12, wherein said feeding back said post-slicer signal with controlled signal timing in response to a timing control signal and generating said feedback signal comprises: selectively delaying said post-slicer signal in response to said timing control signal and generating a delayed post-slicer signal; and filtering said delayed post-slicer signal and generating said feedback signal.

14. The method of claim 12, wherein said feeding back said post-slicer signal with controlled signal timing in response to a timing control signal and generating said feedback signal comprises: selectively delaying said post-slicer signal in response to said timing control signal and generating a delayed post-slicer signal; successively delaying said delayed post-slicer signal and generating a plurality of further delayed post-slicer signals; mixing each one of said plurality of further delayed post-slicer signals with a respective one of a plurality of coefficient signals and generating a plurality of mixed signals; and combining said plurality of mixed signals and generating said feedback signal.

15. The method of claim 12, wherein said differentiating and delaying said input signal and generating a resultant signal comprises: high pass filtering said input signal and generating a high pass filtered signal; and delaying said high pass filtered signal and generating said resultant signal.

16. The method of claim 12, wherein said combining said difference signal and said resultant signal and generating said timing control signal comprises: multiplying said difference signal and said resultant signal and generating a product signal; and low pass filtering said product signal and generating said timing control signal.

17. The method of claim 12, wherein said combining said difference signal and said resultant signal and generating said timing control signal comprises: multiplying said difference signal and said resultant signal and generating a product signal; and integrating said product signal and generating said timing control signal.

---

Description

---

BACKGROUND OF THE INVENTION

1. Field of the Invention

The present invention relates to signal transmission and detection, and in particular, to techniques for compensating for signal distortions caused by signal dispersion and nonlinearities within the signal transmission media.

2. Description of the Related Art

Signal processing architectures for intersymbol interference (ISI) equalization as used for communications transmission and/or storage systems may be divided into two categories: discrete-time architecture and continuous-time architecture. Discrete-time architectures, commonly used in current systems, use a sampled approach to render the input continuous-time, analog waveform in discrete form. Typically, a high resolution A/D converter, which follows the analog anti-aliasing filter, is used as the sampler at the analog front end. Continuous-time architectures use an analog continuous-time approach which directly processes and equalizes the incoming analog waveform while remaining in the continuous time domain until the final data bit stream is generated.

At present, those signal processing architectures having a feedforward transversal filter and a feedback filter as their basic components are considered, and in particular, the following scenarios: discrete-time/continuous-time architectures with fractionally-spaced (i.e., tap spacing less than symbol-spaced) feedforward taps; continuous-time architecture with feedback that is nominally symbol-spaced; continuous-time/discrete-time architectures with fractionally-spaced feedback. For purposes of the presently claimed invention, the following discussion concerns "fat tap" adaptation to cover the continuous-time architecture with fractionally-spaced feedback; however, such discussion may be readily extended to cover the other scenarios as well.

Fractional-spaced feedforward filters have commonly been used either as stand-alone linear equalizers or in combination with Decision Feedback. Advantages of fractional-spaced versus symbol-spaced feedforward filters include: added robustness to constant or slowly varying sampling phase offset or sampling jitter; and improved signal-to-noise ratio (SNR) sensitivity, particularly in the absence of complete channel information, due to the role of the fractional-spaced filter as a combined adaptive matched filter and equalizer.

The adaptation technique for the tap coefficients have always implicitly assumed independence in the adaptation of the successive tap coefficients, which has been based on minimizing the mean squared error (MSE) as computed using the difference between the slicer input and output. This adaptation technique is referred to as LMSE (least mean squared error) or MMSE (minimized mean squared error) adaptation. It can be shown that the LMSE adaptation for both fractional feedforward or symbol spaced feedback at iteration k+1 reduces to the following coefficient update equations: c.sup.(k+1)=c.sup.(k)+.mu.e.sup.(k)s (discrete-time adaptation case) where c.sup.(k) is the tap coefficient vector and e.sup.(k) the corresponding error at the k.sup.th iteration, s is the vector with components as the input waveform to the corresponding tap mixer and .mu. is a constant and is an adaptation parameter; and

.intg..times..mu..function..times..function..times.d.times..times..times..- times..times..times. ##EQU00001## with similar terminology as above.

When continuous-time feedback is considered, a number of difficulties are encountered. For example, it is difficult and sometimes unfeasible to design precisely symbol-spaced, flat group delay filters. If the total group delay on the feedback path for canceling successive past symbols is even slightly different from the corresponding symbol period, the performance loss can be substantial. This may necessitate the need for using fractionally-spaced feedback filters. Fractionally-spaced feedback filters may also be needed to account for the slicer-induced jitter and/or the data bits pattern-specific group delays due to frequency dependent group delays of the slicer, mixer and any other analog/digital component on the feedback data path. Further when an equalizer with fractionally-spaced feedback taps is used, independent LMS adaptation of the successive feedback taps fails because of the strongly correlated nature of the fractional tap-spaced feedback data.

SUMMARY OF THE INVENTION

A decision feedback equalizer is provided with dynamic feedback control for use in an adaptive signal equalizer. Timing within the decision feedback loop is dynamically controlled to optimize recovery of the data signal by the output signal slicer. The dynamic timing is controlled by a signal formed as a combination of feedback and feedforward signals. The feedback signal is an error signal related to a difference between pre-slicer and post-slicer signals. The feedforward signal is formed by differentiating and delaying the incoming data signal.

In accordance with one embodiment of the presently claimed invention, a decision feedback equalizer with dynamic feedback control for adaptively controlling a pre-slicer data signal that is sliced to provide a post-slicer data signal includes signal combining circuitry, controllable delay circuitry, signal slicing circuitry, decision feedback circuitry and signal differentiation circuitry. First signal combining circuitry combines a feedback signal and an input signal representing a plurality of data to provide a pre-slicer signal. The controllable delay circuitry is coupled to the first signal combining circuitry and delays the pre-slicer signal in response to a delay control signal to produce a corresponding delayed pre-slicer signal. The signal slicing circuitry is coupled to the first signal combining circuitry and slices the pre-slicer signal to produce a post-slicer signal indicative of the plurality of data. The decision feedback circuitry includes input signal timing control, is coupled to the controllable delay circuitry and the signal slicing circuitry, and feeds back the post-slicer signal in response to a timing control signal to produce the feedback signal. Second signal combining circuitry is coupled to the controllable delay circuitry and the signal slicing circuitry, and combines the delayed pre-slicer signal and the post-slicer signal to produce a difference signal indicative of a difference between the delayed pre-slicer signal and post-slicer signal. The signal differentiation circuitry, with a selected signal delay, differentiates and delays the input signal to produce a resultant signal, wherein respective portions of the resultant signal are delayed relative to corresponding portions of the input signal by the selected signal delay. Third signal combining circuitry is coupled to the second signal combining circuitry and the signal differentiation circuitry, and combines the difference signal and the resultant signal to produce the delay control signal, wherein the selected signal delay is selected and the delayed pre-slicer signal is delayed relative to the pre-slicer signal such that the timing control signal has a substantially zero alternating current (AC) signal component.

In accordance with another embodiment of the presently claimed invention, a decision feedback equalizer with dynamic feedback control for adaptively controlling a pre-slicer data signal that is sliced to provide a post-slicer data signal includes signal combiner means, controllable delay means, signal slicer means, decision feedback means and signal differentiator means. A first signal combiner means is for combining a feedback signal and an input signal representing a plurality of data and generating a pre-slicer signal. The controllable delay means is for delaying the pre-slicer signal in response to a delay control signal and generating a corresponding delayed pre-slicer signal. The signal slicer means is for slicing the pre-slicer signal and generating a post-slicer signal indicative of the plurality of data. The decision feedback means is for controlling signal timing by feeding back the post-slicer signal in response to a timing control signal and generating the feedback signal. A second signal combiner means is for combining the delayed pre-slicer signal and the post-slicer signal and generating a difference signal indicative of a difference between the delayed pre-slicer signal and the post-slicer signal. The signal differentiator means, with a selected signal delay, is for differentiating and delaying the input signal and generating a resultant signal, wherein respective portions of the resultant signal are delayed relative to corresponding portions of the input signal by the selected signal delay. A third signal combiner means is for combining the difference signal and the resultant signal and generating the delay control signal, wherein the selected signal delay is selected and the delayed pre-slicer signal is delayed relative to the pre-slicer signal such that the timing control signal has a substantially zero alternating current (AC) signal component.

In accordance with another embodiment of the presently claimed invention, a method for providing decision feedback equalization with dynamic feedback control for adaptively controlling a pre-slicer data signal that is sliced to provide a post-slicer data signal includes:

combining a feedback signal and an input signal representing a plurality of data and generating a pre-slicer signal;

delaying the pre-slicer signal in response to a delay control signal and generating a corresponding delayed pre-slicer signal;

slicing the pre-slicer signal and generating a post-slicer signal indicative of the plurality of data;

feeding back the post-slicer signal with controlled signal timing in response to a timing control signal and generating the feedback signal;

combining the delayed pre-slicer signal and the post-slicer signal and generating a difference signal indicative of a difference between the delayed pre-slicer signal and the post-slicer signal;

differentiating and delaying the input signal and generating a resultant signal, wherein respective portions of the resultant signal are delayed relative to corresponding portions of the input signal by a selected signal delay; and

combining the difference signal and the resultant signal and generating the delay control signal, wherein the selected signal delay is selected and the delayed pre-slicer signal is delayed relative to the pre-slicer signal such that the timing control signal has a substantially zero alternating current (AC) signal component.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a block diagram of decision feedback equalization (DFE) circuitry in which an adaptive coefficient signal generator in accordance with the presently claimed invention can advantageously provide for improved data signal equalization.

FIG. 2 is a block diagram of the feedback portion of a DFE circuit in which an adaptive coefficient signal generator in accordance with the presently claimed invention is used to provide the adaptive coefficient signals.

FIG. 3 is a block diagram of one embodiment of an adaptive coefficient signal generator in accordance with the presently claimed invention.

FIG. 4 is a block diagram of an exemplary implementation of the adaptive coefficient signal generator of FIG. 3.

FIG. 5 is a block diagram of another embodiment of an adaptive coefficient signal generator in accordance with the presently claimed invention.

FIG. 6 is a block diagram of an exemplary implementation of the adaptive coefficient signal generator of FIG. 5.

FIG. 7 is a block diagram of an alternative implementation of the adaptive coefficient signal generator of FIG. 4.

FIG. 8 is a block diagram of an alternative implementation of the adaptive coefficient signal generator of FIG. 6.

FIG. 9 is a block diagram depicting signal adaptation using correlated taps in accordance with another embodiment of the presently claimed invention.

FIG. 10 is a block diagram of another embodiment of an adaptive coefficient signal generator in accordance with the presently claimed invention.

FIG. 11 is a block diagram of feedforward equalization (FFE) circuitry with which the adaptive coefficient signal generator of FIG. 10 can share adaptive filter coefficient signals.

FIG. 12 is a block diagram of an exemplary implementation of an adaptive filter coefficient signal generator for providing adaptive filter coefficient signals for the adaptive coefficient signal generator of FIG. 10.

FIG. 13 is a block diagram of an exemplary implementation of a portion of the adaptive coefficient signal generator of FIG. 10.

FIG. 14 is a block diagram of an alternative implementation of a portion of the adaptive coefficient signal generator of FIG. 10.

FIG. 15 is a block diagram of an exemplary implementation of the adaptive coefficient signal generator of FIG. 10.

FIGS. 16A and 16B illustrate expected performance improvement with use of an analog continuous-time feedforward filter in conjunction with "fat tap" adaptation in accordance with the presently claimed invention.

FIG. 17 is a block diagram of one example of an implementation of decision feedback equalization (DFE) circuitry providing improved data signal equalization in accordance with another embodiment of the presently claimed invention.

FIG. 18 is a block diagram of one example of an implementation of decision feedback equalization (DFE) circuitry providing improved data signal equalization in accordance with another embodiment of the presently claimed invention.

FIG. 19 is a block diagram of one example of an implementation of DFE circuitry providing improved data signal equalization in accordance with another embodiment of the presently claimed invention.

DETAILED DESCRIPTION

The following detailed description is of example embodiments of the presently claimed invention with references to the accompanying drawings. Such description is intended to be illustrative and not limiting with respect to the scope of the present invention. Such embodiments are described in sufficient detail to enable one of ordinary skill in the art to practice the subject invention, and it will be understood that other embodiments may be practiced with some variations without departing from the spirit or scope of the subject invention.

Throughout the present disclosure, absent a clear indication to the contrary from the context, it will be understood that individual circuit elements as described may be singular or plural in number. For example, the terms "circuit" and "circuitry" may include either a single component or a plurality of components, which are either active and/or passive and are connected or otherwise coupled together to provide the described function. Additionally, the term "signal" may refer to one or more currents, one or more voltages, or a data signal. Within the drawings, like or related elements will have like or related alpha, numeric or alphanumeric designators. Further, while the present invention has been discussed in the context of implementations using discrete electronic circuitry (preferably in the form of one or more integrated circuit chips), the functions of any part of such circuitry may alternatively be implemented using one or more appropriately programmed processors, depending upon the signal frequencies or data rates to be processed.

The subject matter discussed herein, including the presently claimed invention, is compatible and suitable for use with the subject matter disclosed in the following copending, commonly assigned patent applications (the disclosures of which are incorporated herein by reference): U.S. patent application Ser. No. 10/117,293, filed Apr. 5, 2002, and entitled "Compensation Circuit For Reducing Intersymbol Interference Products Caused By Signal Transmission Via Dispersive Media"; U.S. patent application Ser. No. 10/179,689, filed Jun. 24, 2002, and entitled "Crosstalk Compensation Engine For Reducing Signal Crosstalk Effects Within A Data Signal"; U.S. patent application Ser. No. 10/244,500, filed Sep. 16, 2002, and entitled "Compensation Method For Reducing Intersymbol Interference Products Caused By Signal Transmission Via Dispersive Media"; U.S. patent application Ser. No. 10/290,674, filed Nov. 8, 2002, and entitled "Compensation Circuit And Method For Reducing Intersymbol Interference Products Caused By Signal Transmission Via Dispersive Media"; U.S. patent application Ser. No. 10/290,993, filed Nov. 8, 2002, and entitled "Adaptive Signal Equalizer With Adaptive Error Timing And Precursor/Postcursor Configuration Control"; U.S. patent application Ser. No. 10/321,893, filed Dec. 17, 2002, and entitled "Adaptive Signal Latency Control For Communications Systems Signals"; U.S. patent application Ser. No. 10/321,876, filed Dec. 17, 2002, and entitled "Adaptive Signal Equalizer With Adaptive Error Timing And Precursor/Postcursor Configuration Control"; and U.S. patent application Ser. No. 10/179,996, filed Jun. 24, 2002, and entitled "Programmable Decoding of Codes of Varying Error-Correction Capability".

In accordance with the presently claimed invention, a group of two or more adjacent fractionally-spaced feedback taps are viewed to be correlated and corresponding to essentially the same symbol so that this group of fractionally-spaced taps essentially emulates a symbol-spaced feedback tap. Such a group of fractionally-spaced taps that emulates a symbol-spaced feedback tap is referred herein as a "fat tap." Thus, successive fat taps emulate different symbol-spaced feedback taps. The adaptation of the taps within the fat tap should have some kind of adaptive timing interpolation between them so that the effective timing of the fat tap corresponds to the corresponding symbol timing. Further, both taps in the fat tap need to adapt together using LMSE adaptation, for instance, since their inputs are correlated.

The emphasized case will be that in which successive fat taps consist of disjoint taps and, thus, represent independent symbols (e.g., referred to as Independent Fat Tap Adaptation (IFTA)). Also emphasized will be the case with the constraint of only two taps within the fat tap.

Let the feedback tap spacing be denoted by .tau. which is sufficiently less than the symbol period. It is assumed that the feedback symbol waveform is approximately piece-wise linear within the span of time interval .tau.. Thus, if the feedback signal is denoted as s(t), the parametric equation of the line passing through s(t) and s(t-.tau.) may be denoted via: rs(t)+(1-r)s(t-.tau.) for the parameter r (in general, -.infin.<r<.infin.). With the piecewise linear assumption of s(t), we have: s(t-.tau..sub.r).apprxeq.rs(t)+(1-r)s(t-.tau.) for a suitable choice of the delay .tau..sub.r. Note that with the piecewise linear approximation: .tau..sub.r.apprxeq.(1-r).tau..

With a linear approximation of the feedback signal waveform, we represent the two tap coefficients of the fat tap as cr and c(1-r) where the quantity r is the timing interpolation parameter which ideally should achieve effective symbol spaced timing for the fat tap, and the quantity c is used for LMSE adaptation for estimating the past ISI that needs to be cancelled. Thus, with these two parameters, the fat tap tries to emulate the corresponding symbol-spaced feedback tap with LMS tap coefficient c. The correlation ratio between the two adapting taps of the fat tap is then

##EQU00002## . It is then required to adapt the parameters c and r for the fat tap.

For the continuous-time case, the LMSE adaptation of the parameters c and r can then be shown to be:

.function..mu..intg..times..function..function..function..tau..times.d ##EQU00003## .function..mu..intg..times..function..function..function..function..tau..- times.d ##EQU00003.2## If, as may be typical

.times..times..times..times..times..times. ##EQU00004## .function..function..tau..apprxeq..times.dd.times..function..tau. ##EQU00004.2## for a constant k, the update equation for r(t) may also be simplified to:

.function..mu..intg..times..function.dd.times..function..tau..times.d ##EQU00005## Note that

dd.times..function..tau. ##EQU00006## may easily be implemented by passing the output of the fat tap through a C-R differentiator block, which is a high-pass filter.

For the discrete-time case, the LMSE adaptation of the parameters c and r can then be shown to be: c.sup.(k+1)=c.sup.(k)+.mu..sub.ce.sup.(k)[r.sup.(k)s(t)+(1-r.sup.(k)s(t-.- tau.)] r.sup.(k+1)=r.sup.(k)+.mu..sub.rc.sup.(k)e.sup.(k)[s(t)-s(t-.tau.)] The update equation for r can also be simplified to (if its known that sign(c)=-1): r.sup.(k+1)=r.sup.(k)-.mu..sub.re.sup.(k)[s(t)-s(t-.tau.)]

The next step is determining which pair of two (or more) taps belong to the same fat tap. The specific two adjacent taps which form a fat tap corresponding to a specific past symbol may vary depending on the group delay variations of different analog or digital components within the feedback path such as the slicer, delay elements, summer, mixer, etc. One approach is to hypothesize that different (disjoint) pairs of taps correspond to different fat taps corresponding to different past symbols. Then, depending on the values of the feedback tap coefficients after convergence, specifically the timing interpolation parameter which should ideally be within the range (0,1) and/or the hypothesis which corresponds to the minimum mean squared error, the winning hypothesis may be selected as part of the start-up procedures of the circuit. (As noted above, ideally 0<r<1, although in general, -.infin.<r<.infin.. However, it should be understood that maximum advantages of timing interpolation as provided by the presently claimed invention are realized when 0<r<1 (e.g., where the complement 1-r of r when r=0.2 is 1-r=1-0.2=0.8). When r lies outside of the range (0,1), i.e., -.infin.<r<0 or 1<r<.infin.(e.g., where the complement 1-r of r when r=1.2 is 1-r=1-1.2=-0.2), extrapolation takes place instead of interpolation and performance degradation may result.

Referring to FIG. 1, a decision feedback equalizer (DFE) 100 typically includes a feed forward filter 102 and a feedback filter 104. The feedback filter 104 processes the decisions d.sub.k from the output of the final signal slicer 106. The coefficients, or gains, F*.sub.i can be adjusted to cancel ISI on the current symbol based upon past detected symbols. The feed forward filter 102 has N.sub.1+N.sub.2+1 taps while the feedback filter 104 has N.sub.3 taps, and the output {circumflex over (d)}.sub.k 109 of the final signal summer 108 can be expressed as follows:

.times..times..times..times. ##EQU00007##

where

c*.sub.n=tap gains for feed forward filter 102

y.sub.n=input signals to feed forward filter 102

F*.sub.i=tap gains for feedback filter 104

d.sub.i(i<k)=previous decision made upon detected signal d.sub.k

Accordingly, once the output d.sub.k is provided by the output summer 108 of the feed forward filter 102, the final output d.sub.k is decided. Then, the final output d.sub.k and the previous decisions d.sub.k-1, d.sub.k-2, . . . are fed back through the feedback filter 104, thereby providing the solution for the next decision {circumflex over (d)}.sub.k+1 at the output 109 of the final signal summer 108 in the feed forward filter 102 in accordance with the foregoing equation.

More specifically, with respect to the feedback filter section 104, the output signal 107 is successively delayed through a sequence of time delay elements 110a, 110b, . . . 110n (e.g., with each successive time delay element 110 imparting a time delay equal to one symbol). Each of the successively time-delayed signals 111a, 111b, . . . 111n are processed in their respective adaptive gain stages 112a, 112b, . . . 112n to provide corresponding adaptive feedback signals 113a, 113b, . . . 113n which are summed together in a signal summing circuit 114 to produce the equalization feedback signal 115.

The adaptive gain stages 112a, 112b, . . . 112n process their respective time-delayed signals 111a, 111b . . . 111n based upon an error signal 117 representing the error, if any, between the post-slicer data signal 107 and the pre-slicer data signal 109. This error signal 117 is typically generated by subtracting the pre-slicer data signal 109 from the post-slicer data signal 107 in a signal summing circuit 116.

Referring to FIG. 2, a feedback filter 104a for use in a DFE in a continuous time signal application includes the time delay elements 110 as well as the adaptive gain stages 112. In accordance with the presently claimed invention, each of the time delay elements 110 is a fractional delay element and, therefore, imparts to its data signal a time delay of a fraction of the data symbol, and each of the adaptive gain stages 112 is a multiplier that multiplies its respective time-delayed input signal 111 by one of two adaptive coefficient signals 205b, 205c (discussed in more detail below). Additionally, a fat tap adaptation (FTA) stage 200 is included. As discussed in more detail below, this FTA stage 200 provides the adaptive coefficient signals 205b, 205c. In accordance with one embodiment, this FTA stage 200 provides these adaptive coefficient signals 205b, 205c in accordance with the error signal 117 and the corresponding time-delayed signal 111a, 111b processed by the related adaptive gain stages 112aa, 112ba. In accordance with another embodiment, the FTA stage 200 also uses the equalization feedback signal 115.

It should be understood that, in conformance with the discussion herein, additional pairs of adaptive gain stages 112 can also have corresponding FTA stages 200; however, only one such FTA stage 200 is shown in FIG. 2 for purposes of simplified illustration.

By using the adaptive coefficient signals 205b, 205c provided by the FTA stage 200, adaptive gain stages 112aa and 112ba operate as correlated taps since adaptive coefficient signals 205b and 205c are correlated (discussed in more detail below). On the other hand, those among the remaining adaptive gain stages 112ca, . . . , 112na-1-112na that do not use a corresponding FTA stage 200 are independent taps to the extent that their respective adaptive coefficient signals Cca, . . . , Cna-1, Can are independent, i.e., uncorrelated.

Referring to FIG. 3, one example 200a of the FTA stage 200 (FIG. 2) includes multiplier circuits 202a, 202b and control signal generator circuitry implemented as a coefficient signal generator 204a and a timing interpolation parameter signal generator 206a, all interconnected substantially as shown. The adjacent time-delayed feedback signals 111a, 111b are multiplied in their respective multiplier circuits 202a, 202b with the error signal 117. The resulting product signals 203a, 203b are processed by the control signal generator circuits 204a, 206a. As discussed in more detail below, the first control signal generator circuit 204a provides an adaptation control signal 205a to the second control signal generator circuit 206a which, in return, provides two other adaptation control signals 207a, 207b to the first control signal generator 204a. As a result of processing these input signals 203a, 203b, 207a, 207b, the first control signal generator circuit 204a provides the adaptive coefficient signals c(t)r(t) 205b, c(t)(1-r(t)) 205c, where signal element c(t) is the weighting factor and signal element r(t) is the factor indicative of the degree of correlation between the adjacent time-delayed feedback signals 111a, 111b.

Referring to FIG. 4, one example 200aa of the FTA circuit 200a of FIG. 3 implements the control signal generators 204a, 206a substantially as shown. In the first control signal generator 204aa, the initial product signals 203a, 203b are further multiplied in further multiplier circuits 212aa, 212ba with the adaptation control signals 207aa, 207ba from the other control signal generator 206aa. An additional constant signal 211 can also be multiplied as part of the product operations, or can be implemented as a constant scaling factor within the multiplier circuits 212aa, 212ba.

The resulting product signal 213aa, 213ba are summed in a signal summing circuit 214. The resulting sum signal 215 is integrated in an integration circuit 216 (e.g., a low pass filter) to produce the first adaptation control signal 205aa. This adaptation control signal 205aa, in addition to being provided to the other control signal generator 206a, is multiplied within further multiplication circuits 218a, 218b with the other adaptation control signals 207aa, 207ba provided by the other control signal generator 206aa. The product signals resulting from these multiplication operations are the adaptive coefficient signals 205ba, 205ca.

In the second control signal generator 206aa, the initial product signals 203a, 203b are differentially summed in a signal summing circuit, where the second product signal 203b is subtracted from the first product signal 203a. The resulting difference signal 223 is multiplied in a multiplier circuit 224a with the adaptation control signal 205aa provided by the first control signal generator 204aa. As with the input multiplier circuits 212aa, 212ba of the first control signal generator 204aa, an additional constant signal 221 can also be used in this multiplication operation, or, alternatively, be implemented as a constant scaling factor within the multiplication circuit 224a operation.

The resulting product signal 225 is integrated by another signal integration circuit 226 (e.g., a low pass filter) to produce one of the adaptation control signals 207aa used by the first control signal generator 204aa.

This adaptation control signal 207aa is further processed by a signal complement circuit 228 in which the input signal 207aa is subtracted from a reference signal having a normalized value, with the resulting difference signal 207ba serving as the other adaptation control signal used by the first control signal generator 204aa. For example, if the value of the incoming signal 207aa were considered to have a normalized signal value range bounded by the values of zero and unity, the signal complement circuit 228 subtracts the incoming signal 207aa from the value of unity to produce the output signal 207ba.

Referring to FIG. 5, another embodiment 200b of the FTA circuitry 200 (FIG. 2) uses alternative implementations 204b, 206b of the control signal generators to process the incoming data signals 111a, 111b, the error signal 117 and the equalization feedback signal 115.

Referring to FIG. 6, one implementation 200ab of the FTA circuit 200b of FIG. 5 can be implemented substantially as shown. In the first control signal generator 204ba, the time-delayed data signals 111a, 111b and error signal 117 are multiplied in the multiplier circuits 212ab, 212bb (with the multiplication, or scaling, constant 211 included as part of the operation as discussed above) along with the adaptation control signals 207ab, 207bb from the other control signal generator 206ba. The resulting product signals 213ab, 213bb are summed in the summing circuit 214. The resulting sum signal 215 is integrated in the signal integration circuit 216 (e.g., a low pass filter) to produce an adaptation control signal 205ab (which, in this implementation, is used internally and is not provided to the other control signal generator 206ba). This signal 205ab is multiplied in output multiplier circuits 218a, 218b with the adaptation control signals 207ab, 207bb provided by the other control signal generator 206ba to produce the adaptive coefficient signals 205bb, 205cb.

In the second control signal generator 206ba, the equalization feedback signal 115 is differentiated in a signal differentiation circuit 232 (e.g., a high pass filter). The resulting differentiated signal 233 is multiplied in a multiplier circuit 224b with the error signal 117 (with the multiplication, or scaling, constant 221 included as part of the operation as discussed above). The resulting product signal 207ab forms one of the adaptation control signals provided to the first control signal generator 204ba. This signal 207ab is also complemented by the signal complement circuit 228 (as discussed above) to produce the other adaptation control signal 207bb provided to the first control signal generator 204ba.

Referring to FIG. 7, an adaptive coefficient signal generator in accordance with another embodiment 300 of the presently claimed invention includes error mixer stages 302, 304, a complement integrating mixer stage 306, an integrating mixer stage 308 and a complement coefficient mixer stage 312, all interconnected substantially as shown. The time-delayed data signals 111a, 111b are received and processed by the error mixer stages 302, 304. Each error mixer stage 302, 304 multiplies these signals 111a, 111b with the error signal 117 in signal multipliers 352, 354. The resulting product signals 353, 355 are summed in a signal combiner 356 and the sum signal 357 is integrated by a signal integration circuit 358 (e.g., a low pass filter).

The signals 303, 305 produced by these mixer stages 302, 304 are further processed by the complement integrating mixer stage 306 and integrating mixer stage 308. The complement integrating mixer 306 multiplies signals 303 and 305 in signal multipliers 360, 362 with the signal 309 produced by the integrating mixer stage 308. The resulting product signals 361, 363 are summed in a signal combiner 364. The sum signal 365 is integrated by a signal integration circuit 366 (e.g., a low pass filter).

The integrating mixer stage 308 multiplies signal 303 and signal 305 (which is inverted by a signal inversion circuit 310) in signal multipliers 370, 372 with the signal 307 produced by the complement integrating mixer stage 306. The resulting product signals 371, 373 are summed in a signal combiner 374. The sum signal 375 is integrated by a signal integration circuit 376 (e.g., a low pass filter).

The signals 307, 309 produced by the complement integrating mixer stage 306 and integrating mixer stage 308 are multiplied in the signal multipliers 380, 382 of the complement coefficient mixer stage 312 with the signal 309 produced by the integrating mixer stage 308. The resulting product signals 381, 383 are integrated in respective signal integration circuits 384, 386 (e.g., low pass filters), thereby producing the adaptive coefficient signals 205bc, 205cc.

Referring to FIG. 8, an adaptive coefficient signal generator in accordance with another embodiment 240 of the presently claimed invention shares some similarities with the implementation shown in FIG. 6. The data signals 111a, 111b are multiplied in multiplier circuits 212ac, 212bc (with a multiplication, or scaling, constant 211 as desired). The resulting product signals 213ac, 213bc are summed in a signal combiner 214a and the sum signal 215a is buffered by a gain stage 214b. The resulting buffered signal 215b is multiplied in a signal multiplier 118 with an integrated signal 245 (discussed in more detail below) to produce the equalization feedback signal 115a, and is also processed by two signal delay stages 246, 248 (discussed in more detail below).

The buffered signal 215b is processed by the first signal delay stage 246, which compensates for signal delays introduced by the signal slicer 106 (FIG. 1), the feedback signal multiplication circuit 118 and the feedback signal summing circuitry 108 (FIG. 1) while subtracting out a delay corresponding to that introduced by the signal inversion circuitry 250 and signal differentiation circuit 232a. The delayed signal 247 is inverted in a signal inversion circuit 250. The inverted signal 251 is differentiated in a signal differentiation circuit 232a (e.g., a high pass filter).

The differentiated signal 233a is multiplied in a signal multiplier 224c with the error signal 117 (along with a multiplication, or scaling, constant 221 as desired). The resulting product signal 225c is integrated in a signal integration circuit 226a (e.g., a low pass filter) to produce the adaptation control signal 227a for signal multiplier 212bc. This signal 227a is also complemented by a signal complement circuit 228a (as discussed above) to produce the other adaptation control signal 229a for signal multiplier 212ac.

The second signal delay stage 248 compensates for signal delays introduced by the signal slicer 106 (FIG. 1), the feedback multiplier 118 and feedback signal summer 108 (FIG. 1). The resulting delayed signal 249 is multiplied in a signal multiplier 242 with the error signal 117. The resulting product signal 243 is integrated in a signal integration circuit 244 (e.g., a low pass filter). The integrated signal 245 is multiplied with the buffered signal 215b in the feedback signal multiplier 118 to produce the equalization feedback signal 115a.

Referring to FIG. 9, the adaptive coefficient signals as discussed above can be used in an adaptive signal equalizer with fractionally-spaced feedback in a number of ways. As discussed above, the time-delayed data signals 111a, 111b are multiplied in signal multipliers 112aa, 112ba with the adaptive coefficient signals 205b, 205c, with the resulting product signals 113aa, 113ba combined in a signal combiner 114a to produce a signal 115a which may or may not be the final equalization feedback signal (discussed in more detail below). For example, in scenario A, the first adaptation control signal 205b corresponds to a product of a weighting factor c(t) and the correlation factor r(t), while the second adaptation control signal 205c corresponds to a product of the weighting factor c(t) and the complement [1-r(t)] of the correlation factor r(t). The sum signal 115a can be used directly as the final equalization feedback signal or can be further scaled (e.g., multiplied by a value of unity) in the output multiplier 114b to produce the final equalization feedback signal 115b.

In scenario B, the first adaptation control signal 205b corresponds to a product of a partial weighting factor c.sub.1(t) and the correlation factor r(t), while the second adaptation control signal 205c corresponds to a product of the partial weighting factor c.sub.1(t) and the complement [1-r(t)] of the correlation factor r(t). The sum signal 115a is multiplied in the multiplier 114b by a final weighting factor c.sub.2(t) to produce the final equalization feedback signal 115b.

In scenario C, the first adaptation control signal 205b corresponds to the correlation factor r(t), while the second adaptation control signal 205c corresponds to the complement [1-r(t)] of the correlation factor r(t). The sum signal 115a is multiplied in the output multiplier 114b by the weighting factor c(t) to produce the final equalization feedback signal 115b.

A number of enhancements or modifications may be used to improve the performance over the IFTA with two taps within the fat tap.

Fixed Ratio Fat Tap with Hypothesis Testing

In this modification, adaptation of r may not occur in a continuous-time basis. One tap in the fat tap may be set to be at c with LMS adaptation (discrete-time or continuous-time), while the coefficient of the other tap within the fat tap is related to the first tap coefficient as a multiple by a correlation parameter (of the form

##EQU00008## as described above). A discrete set of such hypotheses corresponding to different values of the correlation parameter

##EQU00009## may be assumed. Each hypothesis is tested and the different taps, which may be LMS-adaptable, are adapted and, after convergence, the steady-state mean square error and/or the adapted filter coefficients may be used to decide on the right hypothesis (the right hypothesis may be selected to be the one with minimum MSE and/or acceptable patterns within the tap coefficients).

Multi-tap FAT Tap with Linear Interpolation

Multiple taps (more than two) and/or a variable number of taps may be used within a fat tap. A simple but effective approach here is to do multiple stages of linear interpolation, each stage consisting of a linear interpolation between some two points obtained from the earlier stage to give one new point which may be used in the next stage. The multi-tap fat tap will then have more than two parameters to adapt.

As an example, consider three feedback taps within a fat tap with input signals s(t),s(t-.tau.), s(t-2.tau.). Then, s(t-.tau..sub.r.sub.1)=r.sub.1s(t)+(1-r.sub.1)s(t-.tau.) may first be formed as a linear interpolation of s(t), s(t-.tau.), and then s(t-.tau..sub.r.sub.2)=r.sub.2s(t-.tau..sub.r.sub.1)+(1-r.sub.2)s(t-2.tau- .) is expected to be the symbol-spaced feedback signal. The feedback tap coefficients for the fat tap with input signals s(t), s(t-.tau.), s(t-2.tau.) are then cr.sub.1r.sub.2, c(1-r.sub.1)r.sub.2, c(1-r.sub.2). The adaptation updates of the 3 parameters r.sub.1, r.sub.2, c in the continuous-time domain are as follows:

.function..mu..intg..times..function..function..times..tau..function..tau.- .times.d ##EQU00010## .function..mu..intg..times..function..function..function..function..funct- ion..tau..times.d ##EQU00010.2## .function..mu..intg..times..function..function..function..times..tau..fun- ction..tau..times.d ##EQU00010.3##

Knowing a priori the signs of c, r.sub.2 the above equations may be simplified. With an intermediate output of the fat tap defined as f.sub.i(t)=cr.sub.1r.sub.2s(t)+c(1-r.sub.1)r.sub.2s(t-.tau.), and the final output as f(t)=f.sub.i(t)+c(1-r.sub.2)s(t2.tau.), then the following simplified update equations result:

.function..mu..intg..times..function.d.function.d.times.d ##EQU00011## .function..mu..intg..times..function.d.function.d.times.d ##EQU00011.2## Note that

d.function.d ##EQU00012##

d.function.d ##EQU00013## may easily be implemented by passing the outputs of the fat tap f.sub.i(t), f(t) through the C-R differentiator block, which is a high-pass filter.

Multi-tap Fat Tap with Superlinear Interpolation

More general interpolation can also be employed, especially when more than two taps are included within the fat tap, such as quadratic interpolation. For example, with three feedback taps within a fat tap and with input signals s(t),s(t-.tau.),s(t-2.tau.), the corresponding tap coefficients may be given as cf.sub.0(r), cf.sub.1(r), cf.sub.2(r) for some appropriately selected functions f.sub.0(), f.sub.1((), which in general may also be functions of more than one parameter. The adaptation updates are then given as:

.function..mu..intg..times..function..function..function..function..times.- .tau..function..function..tau..times.d ##EQU00014## dd.times..function..function..function.'.function..function.'.function..t- imes..tau.'.function..function..tau. ##EQU00014.2## which may be approximated by the following explicit update equation for r(t):

.function..mu..intg..times..function..function.'.function..function.'.func- tion..times..tau.'.function..function..tau..times.d ##EQU00015##

Fat Tap with Gain Offset

To compensate for a residual but unknown gain offset between the taps in a fat tap or to control the linearity range, a fat tap with gain offset may be used. For example, consider two feedback taps within a fat tap with input signals s(t),s(t-.tau.). The tap coefficients for these two taps are then respectively cr, ca(1-r). The adaptation updates for the three parameters (c,a,r) are then given as:

.function..mu..intg..times..function..function..function..tau..times.d ##EQU00016## .function..mu..intg..times..function..function..function..function..tau..- times.d ##EQU00016.2## .function..mu..intg..times..function..function..function..tau..times.d ##EQU00016.3##

Correlated Fat Tap Adaptation (CFTA)

Multiple fat taps may share one or more taps such that each tap could correspond to more than one symbol (e.g., two symbols). It would then be expected that a set of fat taps together emulate multiple symbol-spaced feedback taps. For example, consider three feedback taps with input signals s(t), s(t-.tau.), s(t-2.tau.) such that these together could correspond to two symbol-spaced feedback taps. The first fat tap which corresponds to the first past symbol consists of the feedback taps with inputs s(t),s(t-.tau.), and the second fat tap corresponds to the second past symbol and consists of the feedback taps with inputs s(t-.tau.), s(t-2.tau.); thus, the fat taps have an overlapping feedback tap. The feedback tap coefficients may then be expressed as c.sub.1r.sub.1,c.sub.1(1-r.sub.1)+c.sub.2r.sub.2,c.sub.2(1-r.sub.2).