Sampled current-integrating decision feedback equalizer and method

1. Technical Field

The present invention relates generally to equalization techniques for high-speed data communications and more specifically to implementations of decision feedback equalizer (DFE) circuits with improved performance.

2. Description of the Related Art

With advances in digital computing capabilities fueled by scaling of semiconductor technologies, demands for high-bandwidth transmission of data in systems such as servers and data communication routers continue to increase. However, the limited bandwidth of electrical channels makes it difficult to increase data rates beyond a few gigabits per second. Channel bandwidth degradation is the result of many physical effects, including skin effect, dielectric loss, and reflections due to impedance discontinuities. Consequently, high data rate pulses transmitted through these channels will broaden to greater than a unit interval (UI), thus creating intersymbol interference (ISI) with preceding bits (precursors) and succeeding bits (postcursors).

One method of compensating signal distortion due to ISI is to add equalization to the transmitting or receiving circuitry. In particular, a nonlinear decision feedback equalizer (DFE) is preferred over linear equalization techniques for equalizing high-loss channels. Unlike linear equalizers, a DFE reduces signal distortion without amplifying noise or crosstalk from adjacent channels, making it well-suited for equalizing channels where the loss exceeds, e.g., 20-30 dB.

In a DFE, previous bit decisions are fed back with weighted tap coefficients and added to the received input signal. The tap coefficients are adjusted, either manually or by means of an adaptive algorithm, to match the inverse of the channel characteristics. Consequently, ISI is removed from the received input signal such that a decision can be made on the received bit with a low bit error rate (BER). It is common for the addition function to be implemented using a current-mode logic (CML) analog summer. In such an implementation, current from multiple differential pairs is steered into a resistive load. The voltage at the output of the summer is proportional to a linear combination of the input signal plus the weighted feedback taps.

One drawback of this resistive approach is that as more feedback taps are added to compensate for multiple postcursors, the capacitive loading at the summation node increases hence degrading its settling time. While the load resistance can be decreased to improve settling time, this requires higher current levels and thus higher power consumption to achieve a desired differential output voltage.

To alleviate the settling time requirements of the resistively loaded summer, a current-integrating summer can be employed to sum the input signal with the weighted decisions of previous bits. In this approach, the resistive loads of a CML-based summer are replaced by capacitive loads. A differential voltage is developed at the output based on the amount of current steered into the load over a fixed amount of time. At the end of the integration period (e.g., one UI long), a decision element makes the bit decision and the differential output voltage is reset.

By integrating current onto a capacitor, settling time requirements are eliminated. Moreover, it becomes feasible to increase the number of taps without significant power penalty, particularly if the capacitance associated with an additional tap is small compared to the integrator load capacitor.

One consequence of implementing a DFE with a current-integrating summer is the additional loss introduced by this block. Ideally, a received binary signal should be equal to the voltage level associated with a 1 or 0 bit over the entire unit interval. However, rise and fall times associated with the transition from a 1 to a 0 (or vice versa) can consume a non-negligible fraction of a UI. Unfortunately integration will continue during these transitions, resulting in a lower output voltage than in the ideal case where the input levels remain constant during the integration period.

It can be shown that the loss of a resettable integrator has a sin(x)/x frequency dependence. For a frequency equal to one-half of the symbol (or baud) rate and assuming a one UI integration period, this corresponds to an integrator loss of 3.9-dB in addition to channel and packaging losses. As a result, the signal-to-noise ratio of the equalized output is decreased making it even more difficult to compensate high-loss channels.

A decision feedback equalizer (DFE) and method including a branch coupled to an input and including a sample-and-hold element are configured to receive and sample a received input signal from the input and a current-integrating summer. The current-integrating summer is coupled to an output of the sample-and-hold element. The summer is configured to receive and sum currents representing at least one previous decision and an input sample. The at least one previous decision and the input sample are integrated onto a node, wherein the input sample is held constant during an integration period, thereby mitigating the effects of input transitions on an output of the summer.

A method for decision feedback equalization includes sampling an input signal to provide a sampled portion, holding the sampled portion at an input of a current-integrating summer and summing currents representing the sampled portion of the input signal with currents representing at least one previous decision tap during an integration period wherein the sampled portion is held constant during the integration period, thereby mitigating the effects of input transitions on an output of the summer.

These and other features and advantages will become apparent from the following detailed description of illustrative embodiments thereof, which is to be read in connection with the accompanying drawings.

The disclosure will provide details in the following description of preferred embodiments with reference to the following figures wherein:

FIG. 1 is a schematic diagram of a prior art implementation of a decision-feedback equalizer with a current-integrating summation amplifier;

FIG. 2 is a schematic diagram of a CMOS implementation of the current-integrating summation amplifier employed in the DFE architecture of FIG. 1 ;

FIG. 3 is a plot of a sinusoidal input to a current-integrating amplifier with a frequency equal to half the baud rate, and a response of the resettable current-integrating amplifier over a 1 UT integration window illustrating a signal loss of 3.9 dB as compared with a value of the input waveform at the optimal sampling point;

FIG. 4 is a half-rate DFE architecture with sample-and-hold stages and current-integrating summing stages representing one illustrative embodiment;

FIG. 5 is a timing diagram of the DFE architecture shown in FIG. 4 in response to a sinusoidal input with a frequency equal to half the baud rate;

FIG. 6 is a timing diagram of the DFE architecture shown in FIG. 4 in response to binary data;

FIG. 7 is a block diagram of a speculative half-rate DFE architecture employing sample-and-hold and current-integrating summer stages in accordance with another embodiment;

FIGS. 8A-8D are schematic diagrams showing possible implementations of sample-and-hold stages in various semiconductor technologies such as CMOS transmission gate ( FIG. 8A ), CMOS switched source follower ( FIG. 8B ), bipolar switched emitter follower ( FIG. 8C ) and a diode bridge ( FIG. 8D );

FIGS. 9A-9B show schematic diagrams for possible implementations of current-integrating summing amplifiers in bipolar technology ( FIG. 9A ) and BiCMOS technology ( FIG. 9B );

FIG. 10 is a full-rate DFE architecture with a sample-and-hold stage and current-integrating summing stage representing one embodiment; and

FIG. 11 is a timing diagram of a full-rate DFE architecture shown in FIG. 10 in response to a sinusoidal input with a frequency equal to half the baud rate.

The present principles provide systems, circuits and methods for implementing a decision feedback equalizer (DFE) with a low-power current-integrating summer without incurring additional losses associated with an integrator as described above. The DFE architecture in accordance with useful embodiments includes a sampler followed by a current-integrating summer. A received input data signal is sampled and held constant during the integration period, thereby mitigating the effects of input transitions on the integrator output.

Embodiments of the present invention can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment including both hardware and software elements. In one embodiment, the present invention may be implemented in software, which includes but is not limited to firmware, resident software, microcode, etc.

Furthermore, a circuit in accordance with the present invention can take the form of a computer program product accessible from a computer-usable or computer-readable medium providing program code for use by or in connection with a computer or any instruction execution system. For the purposes of this description, a computer-usable or computer readable medium can be any apparatus that may include, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The medium can be an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system (or apparatus or device) or a propagation medium. Examples of a computer-readable medium include a semiconductor or solid state memory, magnetic tape, a removable computer diskette, a random access memory (RAM), a read-only memory (ROM), a rigid magnetic disk and an optical disk. Current examples of optical disks include compact disk-read only memory (CD-ROM), compact disk-read/write (CD-R/W) and DVD.

A data processing system suitable for storing and/or executing program code may include at least one processor coupled directly or indirectly to memory elements through a system bus. The memory elements can include local memory employed during actual execution of the program code, bulk storage, and cache memories which provide temporary storage of at least some program code to reduce the number of times code is retrieved from bulk storage during execution. Input/output or I/O devices (including but not limited to keyboards, displays, pointing devices, etc.) may be coupled to the system either directly or through intervening I/O controllers.

Network adapters may also be coupled to the system to enable the data processing system to become coupled to other data processing systems or remote printers or storage devices through intervening private or public networks. Modems, cable modem and Ethernet cards are just a few of the currently available types of network adapters.

The circuit as described herein may be part of the design for an integrated circuit chip. The chip design may be created in a graphical computer programming language, and stored in a computer storage medium (such as a disk, tape, physical hard drive, or virtual hard drive such as in a storage access network). If the designer does not fabricate chips or the photolithographic masks used to fabricate chips, the designer transmits the resulting design by physical means (e.g., by providing a copy of the storage medium storing the design) or electronically (e.g., through the Internet) to such entities, directly or indirectly. The stored design is then converted into the appropriate format (e.g., Graphic Data System II (GDSII)) for the fabrication of photolithographic masks, which typically include multiple copies of the chip design in question that are to be formed on a wafer. The photolithographic masks are utilized to define areas of the wafer (and/or the layers thereon) to be etched or otherwise processed.

The resulting integrated circuit chips can be distributed by the fabricator in raw wafer form (that is, as a single wafer that has multiple unpackaged chips), as a bare die, or in a packaged form. In the latter case the chip is mounted in a single chip package (such as a plastic carrier, with leads that are affixed to a motherboard or other higher level carrier) or in a multichip package (such as a ceramic carrier that has either or both surface interconnections or buried interconnections). In any case the chip is then integrated with other chips, discrete circuit elements, and/or other signal processing devices as part of either (a) an intermediate product, such as a motherboard, or (b) an end product. The end product can be any product that includes integrated circuit chips, ranging from toys and other low-end applications to advanced computer products having a display, a keyboard or other input device, and a central processor. The present embodiment may be applied to optical fiber transmission systems.

Referring now to the drawings in which like numerals represent the same or similar elements and initially to FIGS. 1 and 2 , an implementation of a prior art half-rate decision feedback equalizer 10 with current-integrating summation amplifiers (S 1 and S 2 ) is shown. One possible CMOS implementation of a current-integrating summer stage is presented in FIG. 2 . While a two-tap half-rate DFE architecture 10 is presented in FIG. 1 , one skilled in the art can readily extend the use of the current-integrating summing amplifier to other DFE architectures, including speculative and non-speculative, with any number of feedback taps. The DFE in FIG. 1 includes decision element represented as latches L 1 , L 2 , L 3 and L 4 , which have outputs to generate taps H 1 and H 2 based on clock signals CLK and CLKB (clock bar). The differential output of Latch L 1 (or L 2 ) in FIG. 1 is applied to the gates of transistors Q 5 and Q 6 in FIG. 2 , while the differential output of Latch L 3 (or L 4 ) in FIG. 1 is applied to transistors Q 7 , Q 8 , Q 9 , and Q 10 in FIG. 2 .

To illustrate the basic operation of the summer, consider the case when H 1 and H 2 tap weight coefficients set by currents I H1 , I H2P , and I H2N in FIG. 2 are all zero. A differential input signal is applied to the gates of the differential pair formed by transistors Q 1 and Q 2 , which may or may not include resistive degeneration (R DEG as seen in FIG. 2 ) depending on linearity requirements. A clock timing signal (CLK) is applied to the gates of the PMOS loads Q 3 and Q 4 connected to the drains of Q 1 and Q 2 .

When CLK is high, the PMOS devices Q 3 and Q 4 are turned off. In this case, the differential pair Q 1 and Q 2 produces differential currents proportional to the differential input signal (V in ). These currents are integrated onto capacitance at drain nodes of Q 1 and Q 2 , and a differential voltage is produced at the output of the summer (V out ) in response to the applied input signal (V in ).

At the end of the integration, a decision circuit such as latch L 1 in FIG. 1 makes a decision on the polarity of the received bit. CLK then becomes low, and PMOS transistors Q 3 and Q 4 turn on to reset the output signal by pulling the drains of Q 1 and Q 2 to the supply voltage V DD .

It is pointed out that in the implementation of FIG. 2 , the current I MAIN will still flow through Q 3 and Q 4 during this reset state and hence the outputs will not exactly equal V DD . To alleviate this problem, transistors Q 3 and Q 4 have large gate widths such that their “on” resistance in the triode region of operation is low.

Alternatively, note that in a half-rate architecture one summing amplifier (S 1 or S 2 ) integrates current while the other (S 2 or S 1 ) is reset. This makes it possible to share the bias current in the main linear amplifier between the summers S 1 and S 2 such that no current flows through the PMOS transistors Q 3 and Q 4 during the reset operation.

In the more general case where x(t) in FIG. 1 is an input signal including ISI from previous bits, previous decisions can be fed back to steer tap weight currents I H1 , I H2P , and I H2N in FIG. 2 such that the ISI is partially or completely removed in the signal at the output of the integrating summer (S 1 and/or S 2 ). In the summer implementation of FIG. 2 , the sign of the H 2 tap coefficient is set to positive by turning off I H2N and setting current source I H2P to the appropriate tap weight. Alternatively, I H2P can be set to zero and I H2N set to the appropriate current level to produce a negative tap coefficient. H 1 sign selection is not implemented, as it is assumed that the coefficient is always negative; however, sign selection for this tap could be implemented in the same manner described for H 2 . For all feedback taps, the magnitude of the tap coefficient can be determined by some adaptation method.

In the prior art DFE with current-integrating summer shown in FIG. 1 , changes to the input signal V in during the integration period of 1 UI will result in a change in the amount of differential current produced by transistors Q 1 and Q 2 of FIG. 2 . If the differential input signal V in decreases during the integration period, the rate of change in the output voltage V out will also decrease and the differential output signal will be lower at the end of the integration period.

Consequently, the current-integrating summer exhibits frequency-dependent loss. Consider the case when the input signal is a sinusoid with a peak amplitude of A and a frequency equal to half the data symbol rate. If this input waveform were to be applied directly to a slicer, the optimal time to make a bit decision would be when the waveform reaches its maximum value. However, if this input waveform is first integrated before being applied to the slicer, the ideal integration window coincides with the time between the zero crossings of the half-baud sinusoid.

Referring to FIG. 3 , a plot of normalized amplitude versus unit interval is shown for a sinusoidal input 302 to a current-integrating amplifier with a frequency equal to half the baud rate. A response curve 304 of a resettable current-integrating amplifier over the 1 UI integration window illustrating a signal loss of 3.9 dB at point 306 as compared with the value of the input waveform at the optimal sampling point 308 is also shown. The magnitude of the output signal at the end of a 1-UI integration window 306 is reduced by 2/π as compared to the peak magnitude 308 of the input signal level, corresponding to a loss of 3.9 dB. As mentioned above, this loss assumes a 1-UI integration window. In general, it can be shown that the resettable integrator exhibits a sin(x)/x frequency response.

For a given slicer sensitivity, the additional loss will result in higher bit error ratio (BER) and degrade system performance when equalizing high-loss channels. One may attempt to eliminate this frequency-dependent integrator loss by inserting a peaking amplifier before the integrator to compensate for high-frequency signal attenuation. However, exact compensation of the sin(x)/x integrator response cannot be achieved through this method. Moreover, the addition of a peaking amplifier will lead to higher noise, reduced linearity, and most likely higher power dissipation in the receiver.

The present principles described herein eliminate integrator loss by integrating a sampled waveform rather than the continuous-time input. A modified half-rate DFE architecture with input sampling circuits is depicted in FIG. 4 . To illustrate the operation of this architecture, relevant waveforms are shown in FIG. 5 for an input sinusoid with a frequency equal to half the baud rate. These waveforms do not account for finite delays through the various circuit blocks of FIG. 4 . For simplicity, FIG. 5 illustrates the operation when feedback tap coefficients are turned off.

Referring to FIG. 4 , a DFE 400 includes two branches 402 and 404 . In this case, branch 402 outputs data associated with even intervals or bits (y even ), and branch 404 outputs data associated with odd intervals or bits (y odd ). An input function x(t) represents information received by the DFE 400 which may include ISI information or other signal in need of equalization. In the present example, input x(t) includes sinusoidal wave ( FIG. 5 ).

Each branch 402 and 404 respectively includes: a sample and hold circuit SH 1 and SH 2 , a current-integrating summer S 1 and S 2 and latches (L 1 , L 3 ) and (L 2 , L 4 ). DFE 400 illustratively depicts a half-rate architecture. This architecture may be extended to full rate architectures, quarter rate architectures, and so on.

Referring to FIG. 5 with continued reference to FIG. 4 , at a time t 1 , CLK goes high and the sample-and-hold circuit SH 1 samples the input sinusoid x(t) at its maximum value. In practice, a clock-and-data recovery circuit (not illustrated) would align the phase of the clock such that all samples are taken at the optimal sampling points in the center of the bit interval. After t 1 , the sample-and-hold circuit SH 1 holds the sampled value x[n]. A time interval t 1 to t 2 represents the integration window or period during which the integrating summer S 1 responds to the sampled input x[n]. During this time, one advantage in accordance with the present principles is realized. This includes that the integrator S 1 is not affected by changes to x(t) during the integration window. Also, between times t 1 and t 2 , latch L 1 is transparent (e.g., acts as an amplifier) and the signal at node ( 2 ) tracks that at node ( 1 ), preferably with some gain. At time t 2 , CLK goes low and a bit decision is made as L 1 regenerates and latches its data. Additionally, the current-integrating summer S 1 resets its output and the sample-and-hold SHI begins tracking x(t).

Although not depicted in the timing diagram of FIG. 5 , after time t 2 sample-and-hold SH 2 will operate in hold mode and the integration window for current-integrating summer S 2 will begin. A generalized timing diagram representative of the response at various nodes of the DFE 400 to random non-return-to-zero (NRZ) data is depicted in FIG. 6 .

Referring to FIG. 6 with continued reference to FIG. 4 , while the basic operation depicted in FIG. 6 is the same as FIG. 5 , FIG. 6 shows the response of integrating summer S 1 in situations where x[n] is a logic ‘1’ or a logic ‘0’ responsive to a digital input signal for x(t) as opposed to the analog sinusoid for input x(t). Note that the waveforms in FIGS. 5 and 6 represent the differential signals at each of the respective nodes.

As mentioned above, the timing diagrams of FIGS. 5 and 6 do not account for finite delays through the various circuit blocks depicted in FIG. 4 . In situations where the CLK-to-Q delay of latches L 1 or L 2 are significant fractions of a UI, it is often desirable to use a speculative architecture (sometimes referred to as loop unrolling in the art).

Referring to FIG. 7 , a speculative half-rate DFE architecture 500 employing sample-and-holds and current-integrating summers in the branches is illustratively depicted in accordance with another embodiment. Since sample-and-hold blocks SH 1 and SH 3 (or SH 2 and SH 4 ) operate on a same clock phase, they may be combined into a single sample-and-hold element. Further, the output of latches L 1 and L 5 and latches L 2 and L 6 are respectively provided to multiplexers 502 and 504 .

Multiplexer 502 provides an output to a latch L 3 , and multiplexer 504 provides an output to a latch L 4 . The outputs of latches L 3 and L 4 are cross-coupled to multiplexers 504 and 502 to provide a select signal for controlling the outputs of the multiplexers. The outputs of latches L 3 and L 4 also provide feedback taps H 2 to current-integrating summers S 1 and S 3 and current-integrating summers S 2 and S 4 , respectively. +H 1 and −H 1 represent static offsets used in a speculative architecture. In the speculative architecture of FIG. 7 , H 1 is a static tap or static speculative tap. In this look-ahead architecture, one path assumes the previous bit was a ‘1’ (−H 1 ), and the other assumes the previous bit was a ‘0’ (+H 1 ). The multiplexer 502 (or 504 ) following latches L 1 and L 5 (or L 2 and L 6 ) then selects the correct path once the previous bit in question has been determined.

The disclosed sampled current-integrating DFE 500 (and 400 ) is not dependent on the implementation of the various stages presented in the architectures of FIGS. 4 and 7 . A plurality of different technologies may be employed to realize advantageous effects in accordance with the present principles as depicted in FIGS. 8A-8D . For example, the sample-and-hold (SH) may be provided in CMOS technology and can be implemented as a transmission gate 602 as shown in FIG. 8A . In another embodiment, a higher-speed sample-and-hold (SH) may be implemented using a switched source follower circuit 610 shown in FIG. 8B . In yet another implementation, a common bipolar sample-and-hold circuit 620 including switched emitter followers (similar to circuit 610 ) may be employed for the sample-and hold circuit (SH) in FIG. 8C . R C and R D are resistors. In still another embodiment, a diode bridge circuit 630 may be employed as shown in FIG. 8D , where diodes D 1 -D 4 are employed. In all cases, a hold capacitance C H of the sample-and-hold circuits ( 602 , 610 , 620 , 630 ) can be realized using an on-chip capacitor or the parasitic input capacitance of a summing amplifier (S).

A current-integrating summing amplifier (S) may also be realized in a plurality of different technologies, for example, in bipolar and BiCMOS technologies. FIGS. 9A and 9B show illustrative implementations of the current-integrating summing amplifier (S). Referring to FIG. 9A , a bipolar implementation for a current-integrating summing amplifier 702 (S) using pnp loads 703 and bipolar npn differential stages 706 is illustratively shown. Referring to FIG. 9B , a BiCMOS implementation for a current-integrating summing amplifier 704 (S) using bipolar npn differential stages 706 with PMOS loads 708 is shown.

In accordance with the present principles, a full-rate architecture may also be employed. Referring to FIG. 10 , a DFE 800 with a sample-and-hold (SH 1 ) front-end and current-integrating summer (S 1 ) in a full-rate architecture is illustratively shown. In one possible embodiment, the full-rate one-tap DFE 800 includes a sample-and-hold element SH 1 , a current-integrating summer S 1 , and a D-type flip-flop (DFF). When the clock signal CLK is low (and hence CLKB is high), it is assumed that the sample-and-hold circuit SH 1 tracks the input x(t) and the output voltage of the integrator or summer S 1 is reset to a differential zero. The DFF is preferably triggered on falling edges of CLK. A timing diagram is shown in FIG. 11 to illustrate operation of the full-rate architecture 800 in response to a sinusoidal input with a frequency equal to half the baud rate. For simplicity, these waveforms do not account for the finite delays through each of the blocks in FIG. 10 . Furthermore, the H 1 tap coefficient is assumed to be zero. At time t 1 , the input waveform is sampled and integrating summer S 1 integrates in response to the sampled voltage. At time t 2 , the sample-and-hold circuit SH 1 tracks the input signal, the integrator S 1 resets to a differential zero, and the DFF makes a decision based on the value of the integrated voltage at the instant when CLK goes low.

Note H 1 , H 2 , H 3 represent taps from a previous decision 1 bit away, 2 bits away, 3 bits away, etc., and tap coefficients are weights that are employed to determine a tap's strength.

Other modifications and variations of the disclosed embodiments are also within the scope of the present principles. Such variations may include, for example, a quarter-rate instead of half-rate architecture, or other sample-and-hold or integrating summer implementations. The DFE in accordance with the present principles can have one or more dynamic feedback taps, and/or one or more static speculative taps. Such modifications and variations will be obvious to those skilled in the art in view of the teachings of the present disclosure.

Having described preferred embodiments for sampled current-integrating decision feedback equalizers and methods (which are intended to be illustrative and not limiting), it is noted that modifications and variations can be made by persons skilled in the art in light of the above teachings. It is therefore to be understood that changes may be made in the particular embodiments disclosed which are within the scope and spirit of the invention as outlined by the appended claims. Having thus described aspects of the invention, with the details and particularity required by the patent laws, what is claimed and desired protected by Letters Patent is set forth in the appended claims.