Predicted parallel branch slicer and slicing method thereof

Abstract

A predicted parallel branch slicer for use in an adaptive decision feedback equalizer includes Mk adders commonly receiving a signal to be processed and respectively receiving Mk preset values, and performing respective addition operations to generate Mk output signals; Mk slicers in communication with the Mk adders, receiving and processing the Mk output signals to obtain Mk signals of Mk levels, respectively; a multiplexer in communication with the Mk slicers, receiving the signals of the Mk levels; and k delay units interconnected with one another in series and being in communication with the multiplexer, and generating k selection signals of different delay time in response to an output of the multiplexer, the selection signals being provided for the multiplexer to select one of the signals of the Mk levels to be outputted.

Description

FIELD OF THE INVENTION

[0001] The present invention relates to a predicted parallel branch slicer, and more particularly to a predicted parallel branch slicer for use in an adaptive decision feedback equalizer (DFE) for properly slicing signals. The present invention also relates to a slicing method of a predicted parallel branch slicer for use in an adaptive DFE.

BACKGROUND OF THE INVENTION

[0002] Recently, the Institute of Electric and Electronic Engineers stipulated a transmission standard of Gigabit Ethernet. In this transmission standard, four unshield twisted pair-category 5 (UTP-CAT5) transmission lines are used to transmit data at a rate of giga bits per second. For complying with such high speed transmission, the transceiver of each node has to overcome the noise problems resulting from, for example, inter-symbol interference (ISI), echo, near-end cross talk (NEXT) and far-end cross talk (FEXT) phenomena.

[0003] Please refer to FIG. 1A which is a functional block diagram schematically showing a transceiver of a node in a Gigabit Ethernet. In the signal receiving path, an analog signal is processed by the UTP-CAT5 transmission lines 10, hybrid 11, analog front end (AFE) 12 and analog-to-digital converter (ADC) into a digital data signal x(n) essentially suffered from the ISI phenomena (The FEXT phenomena can be ignored). The digital data signal x(n) is transmitted to the subsequent adaptive decision feedback equalizer (ADFE) 16 to be further processed in order to remove the ISI effect, and then transmitted to be processed by the downstream decoder 17, packet and cell switch (PCS) 18 and medium access controller 19. Finally, the processed digital data is transmitted to the network node itself, e.g. a personal computer. The PCS 18 also outputs some signals which pass through an adaptive echo canceller 14 and an adaptive NEXT canceller 15 and then enter the ADFE 16. FIG. 1B shows the waveform diagram of the channel impulse response of a digital data signal x(n). The left portion from the dash line is so called as precursor ISI, and the right portion from the dash line is so called as postcursor ISI.

[0004] Please refer to FIG. 2A which is a schematic functional block diagram of a conventional adaptive decision feedback equalizer. The conventional ADFE uses a feed forward equalizer (FFE) 21 and a feed back equalizer (FBE) 22 to eliminate the precursor ISI and postcursor ISI, respectively. The coefficients of the FEE 21 and FBE 22 are determined and refreshed by a first and a second coefficient refresher 23 and 24 according to the error signal e(n) and the previous values thereof. The slicer 25 quantizes the signal y(n) to recover the digital data signal d(n). The operational principle of the ADFE shown in FIG. 2A is based on least-mean-square (LMS) algorithm, involving the following equations: 1$\begin{array}{c}y\left(n\right)=\sum _{k=0}^{N_f-1}x\left(n-k\right)w_k\left(n\right)-\sum _{k=1}^{N_b-1}d\left(n-k\right)f_k\left(n\right)\\ d\left(n\right)=Q\left[y\left(n\right)\right]\\ e\left(n\right)=d\left(n\right)-y\left(n\right)\\ w_k\left(n+1\right)=w_k\left(n\right)+\mu x\left(n-k\right)e\left(n\right)\\ f_k\left(n+1\right)=f_k\left(n\right)+\mu d\left(n-k\right)e\left(n\right).\end{array}$

[0005] The data-processing rate of this ADFE is confined by the bandwidth of the decision feedback loop (DLP), and thus is limited within a certain level. In order to solve this problem, a pipeline method is developed, which is referred to FIG. 2B. FIG. 2B shows another conventional adaptive decision feedback equalizer. The detailed description of the pipeline method is referred to Naresh R. Shanbhag, Keshab K. Parhi, “Pipelined adaptive DFE architectures using relaxed look-ahead,” IEEE Trans. Signal Processing, vol. 43, No. 6, pp. 1368-1385, June 1995, which is incorporated herein for reference. Different from the ADFE of FIG. 2A, k delay units are additionally provided for the decision feedback loop (DLP). It is to be noted that the additional k delay units are shown outside the FBE 22 in FIG. 2B for simplifying the drawing. In practice, however, the additional k delay units are generally arranged inside the FBE 22. Accordingly, the FBE 22 is divided into (k+1) groups of sub-circuits with a delay unit in each sub-circuit. The pipeline operation is performed with the (k+1) groups of sub-circuits, so as to improve the overall processing speed.

[0006] This pipeline method, although has relatively high processing rate, suffers from a low signal-to-noise ratio. Due to the increased delay time, the waveform response of the FBE 22 will become the one illustrated in FIG. 2C. Since the presence of the additional k delay units, the postcursor ISI relating to the delay time of preceding k delay units is limited to zero, as indicated by the arrow in FIG. 2C. Accordingly, the FBE cannot perform well, and so as to reduce the overall signal-to-noise ratio of the system and deteriorate the signal quality. Modification on the FFE 21 may alleviate the problem, but make the circuitry of the FFE 21 even more complicated. Moreover, the signal quality has not been significantly improved.

SUMMARY OF THE INVENTION

[0007] Therefore, an object of the present invention is to provide an adaptive decision feedback equalizer, which has a high signal-to-noise ratio and a processing speed complying with the requirement of Gigabit Ethernet.

[0008] According to a first aspect of the present invention, a predicted parallel branch slicer for use in an adaptive decision feedback equalizer comprises adders of a number of an expression Mk, where M is a base greater than one and k is an exponent, commonly receiving a signal to be processed and respectively receiving Mk preset values, and performing respective addition operations to generate Mk output signals; slicers of a number of the expression Mk, in communication with the Mk adders, receiving and processing the Mk output signals to obtain Mk signals of Mk levels, respectively; a multiplexer in communication with the Mk slicers, receiving the signals of the Mk levels; and delay units of a number k, interconnected with one another in series and being in communication with the multiplexer, and generating k selection signals of different delay time in response to an output of the multiplexer, the selection signals being provided for the multiplexer to select one of the signals of the Mk levels to be outputted.

[0009] Preferably, the Mk adders are Mk adders with constant coefficients.

[0010] Preferably, the signals of the Mk levels at least include a plurality of data signals of the Mk levels and a plurality of error signals of the Mk levels.

[0011] Preferably, the multiplexer is a combination of a first multiplexer and a second multiplexer, which are in communication with the Mk slicers and the k serially interconnected delay units. The first multiplexer receives the data signals of the Mk levels and the selection signals from the delay units, the second multiplexer receives the error signals of the Mk levels and the selections signal from the delay units, and the second multiplexer generates the at least one output according to an output of the first multiplexer.

[0012] In accordance with the present invention, the one of the signals of the Mk levels selected to be outputted by the multiplexer is the one including a data signal closest to the selection signals, which means the difference between the selected one and the corresponding selection signal is smallest.

[0013] In an embodiment, the one of the signals of the Mk levels selected to be outputted by the multiplexer is the one having an error signal closest to the selection signals, which means they are closet to each other than anyone of other signals of the Mk levels and anyone of other selection signals.

[0014] In another embodiment, the one of the signals of the Mk levels selected to be outputted by the multiplexer is the one which is closest to the selection signals, which means the differences between any other signals of the Mk levels and the selection signals is larger than the difference between the one and the selection signals.

[0015] Preferably, a value VeTT received by each of the Mk adders at a sampled point n is equal to the product of an optimal coefficient Ve=[C1, C2, . . . , Ck] and a value T of k preceding levels, where T=[a(n−1), a(n−2), . . . , a(n−k)]. More preferably, C1, C2, . . . , Ck are k constant coefficients realized according to a simulated waveform of a channel impulse response on a transmission line of Gigabit Ethernet stipulated by IEEE.

[0016] A second aspect of the present invention relates to a predicted parallel branch slicing method for use in an adaptive decision feedback equalizer. The method comprises steps of realizing the first k coefficients of a feed back equalizer according to a channel feature, and operating to obtain Mk preset values, where M is a base greater than one and k is an exponent; receiving a signal to be processed and the Mk preset values which are operated to obtain Mk output signals; respectively receiving and slicing the Mk output signals to obtain Mk signals of Mk levels; generating a sliced output signal according to the Mk signals of Mk levels; and generating a plurality of selection signals of k kinds of different delay time according to the sliced output signal and k different delay operations, and selecting one of the Mk signals of Mk levels to be outputted.

[0017] Compared to the prior art, the present invention performs a pipeline operation of the feed back equalizer by providing additional delay units so as to improve the processing speed. Further, a specially designed predicted parallel branch slicer is used to keep the signal-to-noise ratio at a satisfactory level so as to present from the possible deterioration of signal quality due to pipeline operation. Moreover, the complexity of the circuitry is acceptable (essentially only some slicer and multiplexers are additionally used).

BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The present invention may best be understood through the following description with reference to the accompanying drawings, in which:

[0019] FIG. 1A is a functional block diagram schematically showing a transceiver of a node in a Gigabit Ethernet;

[0020] FIG. 1B is a waveform diagram of the channel impulse response of a digital data signal x(n);

[0021] FIG. 2A is a functional block diagram schematically showing a conventional adaptive decision feedback equalizer;

[0022] FIG. 2B is a functional block diagram schematically showing another conventional adaptive decision feedback equalizer;

[0023] FIG. 2C is an exemplified waveform diagram showing the response variation resulting from unduly increased delay time;

[0024] FIG. 3 is a functional block diagram schematically showing an adaptive decision feedback equalizer according to a preferred embodiment of the present invention;

[0025] FIGS. 4A˜4D are waveform diagrams showing the channel impulse responses obtained by using different lengths of UTP-CAT5 transmission lines according to the Gigabits Ethernet stipulated by IEEE;

[0026] FIG. 5 is an exemplified waveform diagram showing the response variation by using the adaptive decision feedback equalizer of FIG. 3; and

[0027] FIGS. 6A and 6B are schematic functional block diagrams showing two embodiments of the predicted parallel branch slicers for use in the adaptive decision feedback equalizer of FIG. 3.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT

[0028] The present invention will now be described more specifically with reference to the following embodiments. It is to be noted that the following descriptions of preferred embodiments of this invention are presented herein for purpose of illustration and description only; it is not intended to be exhaustive or to be limited to the precise form disclosed.

[0029] Please refer to FIG. 3. The adaptive decision feedback equalizer according to a preferred embodiment of the present invention uses a predicted parallel branch slicer (PPBS) 30 to substitute for the simple slicers of the prior art in order to eliminate the effect of the additional k delay units on the signal-to-analog ratio of the system. In order to clarify the technique and principle of the present invention, please refer to FIGS. 4A˜4D first, which are waveform diagrams showing the channel impulse responses obtained by using different lengths of unshield twisted pair-category 5 (UTP-CAT5) transmission lines according to the Gigabits Ethernet stipulated by IEEE. The lengths of the transmission lines shown in FIGS. 4A˜4D are 25, 50, 75 and 100 meters, respectively. For FIGS. 4A˜4D, the unit of the horizontal axis is time (such as second) and the unit of the vertical axis is voltage (such as volt). It is observed from the figures that the waveforms vary with the lengths of the transmission lines, but the variations are not significant. Therefore, k non-zero constants are preset according to observed waveforms, which are used for simulating the first k coefficients of the response waveform of the combined channel and feed forward equalizer (FFE) 31. It is understood that the response waveform of the combined channel and FFE 31 is equivalent to the waveform entering the downstream feed back equalizer (FBE) 32. By keeping the first k coefficients of the waveform unchanged, the other coefficients of the downstream FBE can be obtained according to the lease mean square (LMS) algorithm. Accordingly, the first k terms of the postcursor ISI can be eliminated by the first k constant coefficients of the FBE (referring to the PPBS 30 described later), and the rest postcursor ISI can be eliminated by the adaptive coefficients of the latter coefficients of the FBE (referring to the FBE 32 described later). For example, when k is equal to 2 and the constant coefficients are C1 and C2, the waveform entering the FBE will be similar to that shown in FIG. 5. The first k coefficients will be very close to the optimal values. Reminded that the first k coefficients of the FBE 22 of FIG. 2B are fixed at zero, which is far way from the practice, the signal-to-noise ratio is thus adversely reduced. By the present invention, this problem can be solved. Next, the equalizer with the first k constant coefficients is transformed into the predicted parallel branch slicer 30 to introduce pipeline process into the FBE.

[0030] Please refer to FIG. 6A, which is a schematic functional block diagram showing the predicted parallel branch slicer 30 in a preferred embodiment, wherein the number k of delay units is equal to 2, and the signal x(n) is a pulse amplitude modulation (PAM) signal with five levels, i.e. −2, −1, 0, 1 and 2. The predicted parallel branch slicer 30 includes twenty-five, i.e. 52, adders 60 and twenty-five slicers 61 of a parallel branch structure. The value VeTT received by the input end of each adder 60 is equal to the product of an optimal coefficient Ve=[C1, C2] and a value T of two preceding levels, where T=[a(n−1), a(n−2)]. In a 5-level case, the value T has 25 possible combinations. Two delay units 62 and 63 are additionally provided in the decision feedback loop (DLP) for use. It is apparent from the figure, the signal b(n) obtained by adding the output of the feed forward equalizer (FFE) 31 with the output of the feed back equalizer (FBE) 32 are further processed by addition and quantization operations with the 25 possible combinations so as to obtain 25 possible quantized data signals d and corresponding error signals e. The signals d and e are inputted to two 25-to-1 multiplexers 64 and 65, respectively, to be selected. The multiplexers 64 and 65 make selections based on the outputs x(n−1) and x(n−2) of the delay units 62 and 63. That is, that x(n−1) and x(n−2) are substantially which of the 25 possible combinations is determined to realize the selections of the multiplexers 64 and 65. Ideally, x(n−1) or x(n−2) will be exactly the same as one of the 25 possible combinations. Further explanation, please refer to the paper of the inventors: IEEE workshop on Signal Processing Systems (SIPS'02). Sam Diego, Calif., USA, Oct. 16-18, 2002. In practice, however, x(n−1) or x(n−2) will be very close to rather than exactly identical to one of the 25 possible combinations due to certain noises and/or errors. Under this circumstance, one of the 25 possible combinations, which is closest to x(n−1) or x(n−2) in level, is determined, and that combination is selected to be outputted by the multiplexers 64 and 65. Herein, the term “closest to” could be viewed as the determined combination with a smallest difference among other combinations. In other words, the term “closest to” could be viewed as the following: the difference between the elements of the selected combination is smaller than the difference between the elements of any other combinations. The delay units 62 and 63 can be arranged in the FBE 32 so that the FBE 32 is divided into three groups of sub-circuits to perform pipeline operations, thereby speeding up the processing.

[0031] Please refer to FIG. 6B. In this embodiment of predicted parallel branch slicer 30, k is equal to 3, and the signal x(n) is a pulse amplitude modulation (PAM) signal with two levels, i.e. −1 and 1. The predicted parallel branch slicer 30 includes eight, i.e. 23, adders 70 and eight slicers 71 of a parallel branch structure. The value VeTT received by the input end of each adder 70 is equal to the product of an optimal coefficient Ve=[C1, C2, C3] and a value T of three preceding levels, where T=[a(n−1), a(n−2), a(n−3)]. In a 2-level case, the value T has 8 possible combinations. Three delay units 72, 73 and 74 are additionally provided in the decision feedback loop (DLP) for use. It is apparent from the figure, the signal b(n) outputted from the feed forward equalizer (FFE) 31 are processed by addition and quantization operations with the 8 possible combinations so as to obtain 8 possible quantized data signals d and corresponding error signals e. The signals d and e are inputted to two 8-to-1 multiplexers 75 and 76, respectively, to be selected. The multiplexers 64 and 65 make selections based on the outputs x(n−1), x(n−2) and x(n−3) of the delay units 72, 73 and 74. That is, that x(n−1), x(n−2) and x(n−3) are which of the 8 possible combinations is determined to realize the selections of the multiplexers 75 and 76. The delay units 72, 73 and 74 can be arranged in the FBE 32 so that the FBE 32 is divided into four groups of sub-circuits to perform pipeline operations, thereby speeding up the processing.

[0032] While the invention has been described in terms of what are presently considered to be the most practical and preferred embodiments, it is to be understood that the invention need not be limited to the disclosed embodiment. On the contrary, it is intended to cover various modifications and similar arrangements included within the spirit and scope of the appended claims which are to be accorded with the broadest interpretation so as to encompass all such modifications and similar structures.

Claims

1. A predicted parallel branch slicer for use in an adaptive decision feedback equalizer comprising: adders of a number of an expression Mk, where M is a base greater than one and k is an exponent, commonly receiving a signal to be processed and respectively receiving Mk preset values, and performing respective addition operations to generate Mk output signals; slicers of a number of said expression Mkin communication with said Mk adders, receiving and processing said Mk output signals to obtain Mk signals of Mk levels, respectively; a multiplexer in communication with said Mk slicers, receiving said signals of said Mk levels; and delay units of a number k, interconnected with one another in series and being in communication with said multiplexer, and generating k selection signals of different delay time in response to an output of said multiplexer, said selection signals being provided for said multiplexer to select one of said signals of said Mk levels to be outputted.

2. The predicted parallel branch slicer according to claim 1 wherein said Mk adders are Mk adders with constant coefficients.

3. The predicted parallel branch slicer according to claim 1 wherein said signals of said Mk levels at least include a plurality of data signals of said Mk levels and a plurality of error signals of said Mk levels.

4. The predicted parallel branch slicer according to claim 1 wherein said multiplexer is a combination of a first multiplexer and a second multiplexer, which are in communication with said Mk slicers and said k serially interconnected delay units.

5. The predicted parallel branch slicer according to claim 4 wherein said signals of said Mk levels at least include a plurality of data signals of said Mk levels and a plurality of error signals of said Mk levels.

6. The predicted parallel branch slicer according to claim 5 wherein said first multiplexer receives said data signals of said Mk levels and said selection signals from said delay units, said second multiplexer receives said error signals of said Mk levels and said selections signal from said delay units, and said second multiplexer generates said at least one output according to an output of said first multiplexer.

7. The predicted parallel branch slicer according to claim 5 wherein said one of said signals of said Mk levels selected to be outputted by said multiplexer is the one including a data signal closest to said selection signals.

8. The predicted parallel branch slicer according to claim 5 wherein said one of said signals of said Mk levels selected to be outputted by said multiplexer is the one having an error signal closest to said selection signals.

9. The predicted parallel branch slicer according to claim 1 wherein said one of said signals of said Mk levels selected to be outputted by said multiplexer is the one which is closest to said selection signals.

10. The predicted parallel branch slicer according to claim 1 wherein a value VeTT received by each of said Mk adders at a sampled point n is equal to the product of an optimal coefficient Ve=[C1, C2, . . . , Ck] and a value T of k preceding levels, where T=[a(n−1), a(n−2), . . . , a(n−k)].

11. The predicted parallel branch slicer according to claim 10 wherein C1, C2, . . . , Ck are k constant coefficients realized according to a simulated waveform of a channel impulse response on a transmission line of Gigabit Ethernet stipulated by IEEE.

12. A predicted parallel branch slicing method for use in an adaptive decision feedback equalizer, comprising steps of: realizing the first k coefficients of a feed back equalizer according to a channel feature, and operating to obtain Mk preset values, where M is a base greater than one and k is an exponent; receiving a signal to be processed and said Mk preset values which are operated to obtain Mk output signals; respectively receiving and slicing said Mk output signals to obtain Mk signals of Mk levels; generating a sliced output signal according to said Mk signals of Mk levels; and generating a plurality of selection signals of k kinds of different delay time according to said sliced output signal and k different delay operations, and selecting one of said Mk signals of Mk levels to be outputted.

13. The predicted parallel branch slicing method according to claim 12 wherein a certain one of said Mk signals of Mk levels and a certain one of said selection signals, which are closest to each other, are determined, and said certain one of said Mk signals of Mk levels is selected to be outputted.

14. The predicted parallel branch slicing method according to claim 13 wherein said signals of said Mk levels at least include a plurality of data signals of said Mk levels and a plurality of error signals of said Mk levels.

15. The predicted parallel branch slicing method according to claim 14 wherein a certain one of said data signals of Mk levels and a certain one of said selection signals, which are closest to each other, are determined, and said certain one of said data signals of Mk levels is selected to be outputted.

16. The predicted parallel branch slicing method according to claim 14 wherein a certain one of said error signals of Mk levels and a certain one of said selection signals, which are closest to each other, are determined, and said certain one of said error signals of Mk levels is selected to be outputted.

17. The predicted parallel branch slicing method according to claim 12 wherein any of said Mk preset values is equal to a product of k-level optimal coefficients and k preceding levels.

18. The predicted parallel branch slicing method according to claim 12 wherein said k-level optimal coefficients are realized according to a simulated waveform of a channel impulse response on a transmission line of Gigabit Ethernet stipulated by IEEE.

19. The predicted parallel branch slicing method according to claim 14 wherein said step of generating said sliced output signals is performed by slicing said data signals of said Mk levels and said selection signals first and then slicing said error signals of said Mk levels and said selection signals.

20. The predicted parallel branch slicing method according to claim 19 wherein a result generated by slicing said data signals of said Mk levels and said selection signals is referred by the slicing operation of said error signals of said Mk levels and said selection signals.