US20160373120A1

US20160373120A1 - Calibration of digital-to-time converter

Info

Publication number: US20160373120A1
Application number: US14/745,545
Authority: US
Inventors: Aaron J. Caffee; Brian G. Drost
Original assignee: Silicon Laboratories Inc
Current assignee: Silicon Laboratories Inc
Priority date: 2015-06-22
Filing date: 2015-06-22
Publication date: 2016-12-22
Anticipated expiration: 2035-06-22
Also published as: US9531394B1

Abstract

A technique for calibrating a digital-to-time converter includes an apparatus including a digital-to-time converter configured to generate an output signal based on a digital code, an input signal, and a gain calibration signal. The output signal has edges linearly delayed from corresponding edges of the input signal based on the digital code. The digital code vacillates between an evaluation code and a calibration code. The apparatus includes a reference signal generator configured to provide a delayed version of the input signal. The delay of the reference signal generator is matched to a delay of the digital-to-time converter. The apparatus includes a calibration circuit configured to generate the gain calibration signal based on the output signal and the delayed version of the input signal. The calibration code may alternate between a first calibration delay code and a second calibration delay code.

Description

BACKGROUND

Field of the Invention
The present invention is related to data converters and more particularly to digital-to-analog converters.
Description of the Related Art
In general, a digital-to-time converter is a digital-to-analog data converter that generates an output signal having a period or edge delay based on a digital control word and a reference clock signal. For example, a clock divider circuit is a digital-to-time period converter that generates an output clock signal based on a reference clock signal and a digital control word indicating a divide ratio (e.g. an integer divide ratio or a fractional divide ratio). The analog output is the period of the output clock signal. The output clock period is based on the digital control word and the clock period of the reference clock signal. The output clock signal has a digitally-defined, full-scale range and a least-significant bit clock period.
A typical integer clock frequency divider is a digital-to-time period converter that generates an output clock signal having a period based on an integer multiplication of the input clock period and has a static and bounded quantization error and random jitter based on thermal noise. The fractional clock frequency divider is a dithering modulus clock frequency divider generating output clock signals having an average output clock period that is based on a fractional multiplication of the input clock period. A particular digital control word has an integer portion and a fractional portion. The dithering modulus clock divider provides a noise-shaped integer control signal to an integer clock divider that receives a reference clock signal. The noise-shaped integer control signal dithers between integer clock divider values causing the integer divider to generate an output clock signal having a period that, on average, is the target fractional clock period. The analog quantization error of the dithering noise-shaped integer control signal causes substantial deterministic jitter that dominates the jitter performance.
Digital-to-time period converters are used in clock synthesizer applications. Integer-based PLLs have been used extensively in clock multiplication applications that demand superior spurious performance. In order to generate fractional output frequencies, a prescaler digital-to-time converter (DivP) is used in the reference path of the PLL while a separate digital-to-time converter is used in the feedback path (DivN). In this way, the reference clock frequency can be multiplied by N/P. While this technique has proved useful in industry, the frequency resolution supported by the clock multiplier directly constrains the maximum achievable bandwidth to about f_ref/(10×P). As a result, the system is more dependent on voltage-controlled oscillator phase noise performance which can often translate into higher power consumption. In addition, the low loop bandwidth of the clock synthesizer reduces its usefulness in data communication applications to supporting significantly lower data rates than might otherwise be desired.
Referring to FIG. 1, an exemplary clock synthesizer utilizes phase-locked loop (PLL) 100, which includes a phase/frequency detector 102, a charge pump loop filter 104, and a voltage-controlled oscillator (VCO) 106. Voltage controlled oscillator 106 may be implemented as a ring oscillator, an LC oscillator, or other suitable oscillator structure. Phase/frequency detector 102 receives reference clock signal ck_in, which may be provided by a fixed source, such as a crystal oscillator, a microelectromechanical structure (MEMS) oscillator, or other suitable source. Fractional divider 108 introduces a digital quantization error that causes phase noise (i.e., jitter) in the feedback clock signal, ck_fb. For example, ck_fbmay have jitter of up to one cycle of the VCO output ck_vco.
A reference signal having a frequency f_insupplied to PLL 100 is multiplied based on a divider value to generate a synthesized clock signal ck_vco. Frequency f_outof the output clock signal, ck_out, can be determined by a divider value D_fbof the feedback divider in the PLL, and in embodiments including output divider 110, a divider value D_outof output divider 110:
$f_{out} = f_{in} \frac{D_{fb}}{D_{out}} .$
The feedback divider includes a digital-to-time converter that may be a fractional-N feedback divider 108 that receives clock signal ck_vcoas the reference clock signal and digital control word D_fb. Digital control word D_fbis a divide value sequence of integers corresponding to a target fractional divider ratio. The synthesized clock signal ck_vcomay be further divided by an output divider, which may be an integer clock divider including a digital-to-time period converter that generates the output clock signal ck_outbased on clock signal ck_vcoas the reference clock signal and integer divide value D_out. In applications with demanding spur specifications, another integer clock divider (not shown) receives an input clock and a divide value D_inand generates reference clock signal ck_inprovided to phase frequency detector 102. Note that in such embodiments an integer clock divider may be used instead of fractional divider 108 and D_in, D_fb, and D_outhave integer values to generate output clock signal ck_outa signal having a frequency:
$f_{out} = f_{in} \frac{D_{fb}}{D_{in} D_{out}} .$
However, an input integer divider provides frequency resolution at the expense of phase noise. Using a fractional-based digital word for D_fband fractional divider 108 instead of an integer input divider and an integer feedback divider, supports higher output frequency resolution. The PLL behaves like a digital-to-analog converter reconstruction filter and reduces or eliminates high-frequency quantization noise. The higher PLL bandwidths improve VCO noise suppression but introduce deterministic jitter.
As the frequencies of ck_inand ck_outincrease, the target clock period decreases and jitter introduced by the digital-to-time converter circuits become a larger fraction of the target clock period, thereby causing output clock signals from conventional clock synthesizer designs to fail to meet target performance specifications. Accordingly, improved techniques for digital-to-time conversion are desired.

SUMMARY OF EMBODIMENTS OF THE INVENTION

In at least one embodiment of the invention, an apparatus includes a digital-to-time converter configured to generate an output signal based on a digital code, an input signal, and a gain calibration signal. The output signal has edges linearly delayed from corresponding edges of the input signal based on the digital code. The digital code vacillates between an evaluation code and a calibration code. The apparatus includes a reference signal generator configured to provide a delayed version of the input signal. The delay of the reference signal generator is matched to a delay of the digital-to-time converter. The apparatus includes a calibration circuit configured to generate the gain calibration signal based on the output signal and the delayed version of the input signal. The calibration code may alternate between a first calibration delay code and a second calibration delay code. The apparatus may include an integer frequency divider configured to generate the input signal and the evaluation code based on a reference clock signal and a digital divider code. The digital-to-time converter may be configured as a subrange data converter with respect to the integer frequency divider. The first delay code of the calibration code may correspond to a minimum delay and the second delay code of the calibration code may correspond to the minimum delay plus a delay of the integer frequency divider corresponding to the least-significant bit of the digital divider code. The calibration circuit may generate a skew correction signal based on the output of the charge pump circuit. The skew correction signal may be indicative of skew between the output signal and the delayed version of the input signal. The apparatus may include a select circuit configured to provide the evaluation code based on digital quantization error of the input signal in response to a first value of a control signal and to provide the calibration code in response to a second value of the control signal. The digital-to-time converter may include a circuit configured to generate an output signal having a pulse width based on a digital code. The pulse width may be based on the digital code. The circuit may include a first node selectively coupled to a first reference signal, a second node configured to receive a second reference signal, and a capacitive digital-to-analog converter configured to generate a voltage on the first node in response to the digital code. The circuit may include a comparator configured to generate the output signal based on a first signal on the first node and the second reference signal on the second node and a current source selectively coupled to the first node and responsive to the gain control signal.
In at least one embodiment of the invention, a method includes generating an output signal based on a digital code, an input signal, and a gain calibration signal. The output signal has an edge linearly delayed from a corresponding edge of the input signal based on the digital code. The digital code vacillates between an evaluation code and a calibration code. The method includes generating a first phase error between the delayed signal having a first calibration delay based on the digital code having a first calibration value of the calibration code and the first signal during a first calibration interval. The method includes generating a second phase error between the delayed signal having a second calibration delay based on the digital code having a second calibration value of the calibration code and the first signal during a second calibration interval. The method includes generating the gain calibration signal based on the first phase error and the second phase error. The method may include generating a skew correction signal based on a common mode of the first phase error and the second phase error. The method may include adjusting a gain of a digital-to-time converter based on the gain calibration signal. Generating the output may include interleaving transmitting a first edge of a first evaluate operation with a calibration operation, interleaving transmitting a second edge of a second evaluate operation with a second calibration operation, and combining the first and second edges to generate the output having a 50% duty cycle. The first delay code may correspond to zero delay and the second delay code may correspond to a delay equivalent to delay defined by a least-significant bit delay of an integer range of the digital divider code.
A technique for improving PLL bandwidth for fractional clock multiplier values utilizes fractional-N PLL clock synthesizers. This technique uses a programmable feedback divider that, by dynamically dithering between different divide values, implements an average effective feedback divide value that supports fractional clock multiplication without the need for a prescaler. Accordingly, the technique improves PLL bandwidth as compared to integer implementations by a factor of the prescaler value P. However, since the divider dynamically switches between different divide values, large phase errors at the phase detector input result in excessive control voltage ripple and, as a result, substantially degrade spurious performance.
A technique for reducing the control voltage ripple includes reducing the bandwidth of the PLL. However, that solution detracts from the advantage of a clock synthesizer using a fractional-N PLL over an integer-based clock synthesizer. Another technique for reducing control voltage ripple includes using current-mode phase interpolation at the output of the charge pump of the PLL. Still referring to FIG. 1, a technique for reducing deterministic jitter and supporting higher PLL bandwidths for a particular deterministic jitter specification compensates for noise at the output of charge pump 104. Fractional divider 108 provides digital signal D_Q,fbto an auxiliary charge pump-based digital-to-analog converter (not shown) and charge pump loop filter 104 subtracts the error in the charge domain. The control signal that dithers the feedback divider to achieve a particular fractional multiplication may be used as the most-significant bit of the multiplier and, by tracking the instantaneous divider error, the error residue in the form of a digital word may be provided to a higher resolution current-mode digital-to-analog converter that perfectly cancels the deterministic ripple on the control voltage. This implementation may be complex and impracticable (e.g. requires a linear phase frequency detector and charge pump loop filter that is matched to the fractional divider 108 for effective cancellation), does not cancel the quantization noise at the source, causes excessive charge pump output thermal noise due to greater turn-on times, and is not easily adapted for reducing deterministic jitter in the output divider (e.g., integer divider 110).
Digital-based quantization noise cancellation techniques have similar drawbacks: those techniques cancel quantization noise at a digital loop filter input, not at the fractional divider source, require linear element matching to the fractional divider gain for effective cancellation, cause excessive thermal noise due to greater turn-on times, and are not easily adapted for output dividers. In addition, digital loop elements (e.g., phase detector, loop filter) introduce additional quantization noise. Accordingly, improved techniques for clock synthesis are desired.

BRIEF DESCRIPTION OF THE DRAWINGS

The present invention may be better understood, and its numerous objects, features, and advantages made apparent to those skilled in the art by referencing the accompanying drawings.

FIG. 1 illustrates a functional block diagram of an exemplary system for generating a clock signal using fractional frequency synthesis.

FIG. 2 illustrates a functional block diagram of an exemplary system for generating a clock signal using an interpolative divider.

FIG. 3 illustrates a functional block diagram of an exemplary interpolative divider of FIG. 2.

FIG. 4 illustrates timing waveforms for the exemplary interpolative divider of FIG. 3.

FIG. 5 illustrates a circuit diagram of a digital-to-time converter cell of an exemplary phase interpolator of FIG. 3.

FIG. 6 illustrates a circuit diagram of a digital-to-time converter cell of an exemplary phase interpolator of FIG. 3.

FIG. 7 illustrates timing waveforms for the exemplary digital-to-time converter cell of FIG. 5.

FIG. 8 illustrates a circuit diagram of an exemplary slew rate based delay control digital-to-time converter cell.

FIG. 9 illustrates timing waveforms for the exemplary slew rate based delay control digital-to-time converter cell of FIG. 8.

FIG. 10 illustrates a circuit diagram of an exemplary reset voltage based delay control digital-to-time converter cell consistent with at least one embodiment of the invention.

FIG. 11 illustrates timing waveforms for the exemplary reset voltage based delay control digital-to-time converter of FIG. 10.

FIG. 12 illustrates a circuit diagram of an exemplary pedestal voltage based delay control digital-to-time converter consistent with at least one embodiment of the invention.

FIG. 13 illustrates a circuit diagram of an exemplary capacitive based voltage digital-to-analog converter circuit of FIG. 12.

FIG. 14 illustrates timing waveforms for the exemplary pedestal voltage based delay control digital-to-time converter of FIG. 12.

FIG. 15 illustrates an exemplary voltage-to-current generator circuit including a switched-capacitor resistor of the exemplary pedestal voltage based delay control digital-to-time converter of FIG. 12.

FIG. 16 illustrates an exemplary voltage-to-current generator circuit of the exemplary pedestal voltage based delay control digital-to-time converter of FIG. 12.

FIG. 17 illustrates a functional block diagram of an exemplary digital-to-time converter and gain calibration system consistent with at least one embodiment of the invention.

FIG. 18 illustrates an exemplary edge generation circuit of FIG. 17 consistent with at least one embodiment of the invention.

FIG. 19 illustrates a functional block diagram of an exemplary time-to-digital converter and analog gain calibration system consistent with at least one embodiment of the invention.

FIG. 20 illustrates a functional block diagram of an exemplary time-to-digital converter and digital gain calibration system consistent with at least one embodiment of the invention.

FIG. 21 illustrates a functional block diagram of an exemplary phase interpolator and phase interpolator gain calibration circuit consistent with at least one embodiment of the invention.

FIG. 22 illustrates a functional block diagram of exemplary time reference generator circuits of FIG. 21 consistent with at least one embodiment of the invention.

FIG. 23 illustrates timing waveforms for the exemplary phase interpolator and phase interpolator gain calibration circuit and time reference generator circuits of FIGS. 21 and 22.

FIG. 24 illustrates a functional block diagram of an exemplary phase interpolator and phase interpolator gain calibration circuit consistent with at least one embodiment of the invention.

FIG. 25 illustrates timing waveforms for the exemplary phase interpolator and phase interpolator gain calibration circuits of FIGS. 21 and 24.

FIG. 26 illustrates a functional block diagram of an exemplary phase interpolator and phase interpolator gain calibration circuit consistent with at least one embodiment of the invention.

FIG. 27 illustrates a functional block diagram of an exemplary gain error detection and correction circuit consistent with at least one embodiment of the invention.

FIG. 28 illustrates a functional block diagram of an exemplary gain error and skew detection and correction circuit consistent with at least one embodiment of the invention.

FIG. 29 illustrates a functional block diagram of an exemplary gain error detection and correction circuit consistent with at least one embodiment of the invention.

FIG. 30 illustrates a functional block diagram of an exemplary gain error detection and correction circuit including a bang-bang phase detector consistent with at least one embodiment of the invention.

FIG. 31 illustrates a functional block diagram of an exemplary gain error detection and correction circuit including pre-charge voltage buffers consistent with at least one embodiment of the invention.

FIG. 32 illustrates a functional block diagram of an exemplary gain error detection and correction circuit having reduced charge sharing consistent with at least one embodiment of the invention.

FIG. 33 illustrates a functional block diagram of an exemplary time interleaving interpolative divider consistent with at least one embodiment of the invention.

FIG. 34 illustrates a functional block diagram of an exemplary time-to-digital converter consistent with at least one embodiment of the invention.

FIG. 35 illustrates a functional block diagram of an exemplary successive approximation time-to-digital converter consistent with at least one embodiment of the invention.

FIG. 36 illustrates a functional block diagram of an exemplary high-resolution time-to-digital converter consistent with at least one embodiment of the invention.

The use of the same reference symbols in different drawings indicates similar or identical items.

DETAILED DESCRIPTION

A technique that reduces control voltage ripple without narrowing the bandwidth of the PLL in a clock synthesizer uses a time-domain digital-to-analog converter (i.e., digital-to-time converter) in series with a dynamic feedback divider. Similar to the current-mode phase interpolation technique, the time-domain digital-to-analog converter receives an error residue that attempts to adjust for phase errors introduced by the dynamic feedback divider. This approach addresses the phase error early in the signal path, causes the downstream phase detector and associated charge pump loop filter circuitry of the PLL to see a much smaller signal, and as a result, operate more linearly than other implementations.
Referring to FIGS. 1 and 2, a technique for reducing jitter in PLL 100 includes delay-based quantization noise cancellation. The technique includes adjusting the phase of the frequency-divided signal at the output of a fractional divider, according to an error control signal D_Q, generated by fractional divider 108. The technique applies the error control signal D_Qto phase interpolator 204 and performs the subtraction in the time domain. Accordingly, any linearity requirement imposed by the quantization noise cancellation is limited to phase interpolator 204. This technique reduces phase errors seen by the front-end of PLL 200 (e.g., phase detector 102) as compared to the phase errors seen by the front-end of PLL 100. In addition, ratio-metric techniques may be used to reduce phase interpolator gain error. Moreover, the delay-based quantization noise compensation is easily adapted to output divider configurations. Interpolative divider techniques for reducing noise associated with a fractional-N divider are described in U.S. Pat. No. 7,417,510, filed Oct. 17, 2006, entitled “Direct Digital Interpolative Synthesis,” naming Yunteng Huang as inventor, which application is hereby incorporated by reference.
Referring to FIG. 2, interpolative divider 202 includes fractional divider 108 and phase interpolator 204. Fractional divider 108 may include a dithering modulus divider (e.g., a typical first order sigma-delta modulator or higher-order sigma-delta modulator) that generates noise-shaped clock signal ck_DMDand digital quantization error signal D_Q. Phase interpolator 204 adjusts the phase of clock signal ck_DMDaccording to the digital quantization error signal to reduce error in the feedback clock signal ck_fb, thereby introducing a phase adjustment prior to phase/frequency detector 102 of PLL 200. Interpolative divider 202 is a sub-ranging digital-to-time period converter that generates an output signal with an average fractional period resolution. Phase interpolator 204 is a digital-to-time converter that uses digital quantization error signal D_Qto divide output quantization noise by up to 2^N ^fracdepending on N_PIwith respect to N_frac, where N_PIis the number of bits by which phase interpolator 204 reduces the quantization noise, N_fracis the number of bits used to represent the fractional portion of the divide ratio, i.e., the fractional portion bit-width and N_PIis less than or equal to N_frac. Error of phase interpolator 204 limits the resolution of the digital-to-analog conversion of interpolative divider 202.
Referring to FIG. 3, in at least one embodiment, fractional divider 108 of interpolative divider 202 includes finite-state machine 302 that generates digital code D_N, which dithers between different divide values, and an associated digital quantization error signal D_Q. Counter 304 generates a gating signal that is used to select and stretch pulses of clock signal ck_ref(e.g., by digitally gating clock signal ck_refwith the gating signal using AND-gate 306 or other suitable logic circuit) to generate clock signal ck_DMD.
Ideally, the phase interpolator transfer function, i.e., the output time delay as a function of digital quantization error signal D_Q, is:
T _PI(D _Q)=T _PI,DAC +T _PI,OS
where
$T_{PI, DAC} = \frac{T_{ref}}{2^{N} PI} D_{Q},$
and T_PI,OSis an output delay offset. The output delay offset is not a function of digital quantization error signal D_Q, but rather is affected by environmental factors, and may be assumed to be zero with respect to deterministic jitter. The full-scale range of the phase interpolator delay is ideally one dithering modulus divider bit, i.e., one least-significant bit of digital code D_N, which corresponds to reference clock period T_ref, which is one period of reference clock ck_ref. For example, for a two-bit phase interpolator, digital quantization error signal D_Qmay have values 00, 01, 10, and 11, which correspond to discrete T_PI,DACvalues of 0, T_ref/4, 2×T_ref/4, and 3×T_ref/4. A delay of 4×T_ref/4 corresponds to an entire period of reference clock signal ck_ref, i.e., reference clock period T_ref. The phase interpolator error includes difference in phase interpolator output from the expected discrete T_PI,DACvalues. The component of phase interpolator error that is linearly related to digital quantization error signal D_Qis referred to as gain error and a component of phase interpolator error that is randomly related to digital quantization error signal D_Qis referred to as mismatch error. In addition, higher order gain error may be present (e.g., systematic non-linearities). Both gain error and mismatch error affect peak-to-peak deterministic jitter.
Referring to FIGS. 3 and 4, for an exemplary fractional divide ratio of 3.5, D_period=(11.10)₂, the noise-shaped integer divider value, digital code D_Ndithers between 3 and 4 and the digital quantization error signal D_Qdithers between 2^N ^frac ^-1and 0. Integer divider 110 generates clock signal ck_DMDthat has a period of three times the period of reference clock signal ck_refwhen digital code D_N=3 (e.g., between time t₁and t₂) and has a period of four times the period of reference clock signal ck_refwhen digital code D_N=4 (e.g., between times t₂and t₃). Then, phase interpolator 204 delays the edge of clock signal ck_DMDaccording to the digital quantization error signal to generate output clock signal ck_out, having a period that is 3.5 times the period of reference clock signal ck_ref. Note that the period between times t₁, and t′₂, the period between times t′₂and t₃, and the period between times t₃and t′₄is 3.5 times the period of reference clock signal ck_ref. However, the output duty cycle of output clock signal ck_outis not 50%. In at least one embodiment, a divide by two may be used to generate an output signal having a 50% output duty cycle. The deterministic jitter of clock signal ck_DMDat t₂and t₄is half of reference clock period, T_ref/2 and is cancelled by phase interpolator 204.
As discussed above, a typical phase interpolator circuit is a digital-to-time converter that delays an input edge signal to generate an output edge signal based on an analog reference signal (e.g., a voltage, current) and digital control word D_Qindicating the amount of delay to be applied. The full-scale range and least significant bit delay tend to be determined by analog components. Referring to FIGS. 5 and 7, in typical unit cell 500 of a digital-to-time converter, in response to a reset control signal, capacitor C_sis coupled to reference voltage V_rstto store an initial charge on capacitor C_s, which resets the voltage on node V_sto the level of reference voltage V_rst. In response to the reset signal being deasserted and a slew control signal being asserted, the charge stored by capacitor C_sincreases by charge delivered from current source I_s. When the voltage on node V_sreaches trip voltage V_tof comparator 508, at time t_trip, the comparator output signal transitions, e.g., from low to high, generating a rising edge of ck_out. The time that it takes for node V_sto charge from reference voltage V_rstto trip voltage V_tis time delay T_delay. Note that comparator delay T_compideally has no dependence on the control code and introduces negligible deterministic jitter. Time delay T_delayof delay cell 500 is typically controlled by adjusting the slew rate of the voltage on node V_s. Referring to FIG. 6, in at least one embodiment, delay cell 500 includes an inverter-based comparator and the reference voltage is derived from the trip-point of inverter 510 and no additional trip voltage is needed. Note that since low frequency noise affects charge stored on capacitor C_s, an auto-zeroing technique (e.g., correlated double sampling) is used to effectively reduce or eliminate flicker noise.
Referring to FIGS. 8 and 9, the slew rate of the voltage on node V_sof unit cell 700 of a digital-to-time converter may be changed by selectively adjusting one or more of the current output by current source 702 or the capacitance C_sof capacitor 704. When controlling the current provided by current source 702 based on digital control word D_Q, individual current sources I_LSBmay be enabled to obtain:
$T_{PI} (D_{Q}) = \frac{C_{S}}{I_{S}} (V_{t} - V_{rst}) + T_{comp} = \frac{C_{S}}{D_{Q} I_{LSB}} (V_{t} - V_{rst}) + T_{comp} .$
Note that in this case, time delay T_PIis a non-linear function of digital control word D_Q. For the greatest value of digital control word D_Q, the slew rate is the fastest, the trip voltage is reached soonest, at time t_trip1, and the delay of the rising clock edge of ck_outis the smallest, as illustrated by V_s(t) curve 802. For the smallest value of digital control word D_Q, the slew rate is the slowest, the trip voltage is reached latest at time t_trip3and the delay of the rising clock edge of output clock signal ck_outis the greatest, as illustrated by V_s(t) curve 806. The code-dependent slew rate results in a complex gain variability that may be difficult to calibrate. Controlling the slew rate using selectively enabled current sources, in particular, is susceptible to substantial random mismatch due to use of active devices, and trades off deterministic jitter with random jitter.
When controlling slew rate by varying target capacitance C_s, individual capacitors C_LSBmay be selectively coupled via switches to obtain the target capacitance C_sand thus a target delay period:
$T_{PI} (D_{Q}) = \frac{C_{S}}{I_{S}} (V_{t} - V_{rst}) + T_{comp} = \frac{D_{Q} C_{LSB}}{I_{S}} (V_{t} - V_{rst}) + T_{comp} .$
Note that T_PI(D_Q) is a linear function of digital control word D_Q.
However, variable slew rate affects the delay of the voltage comparator that receives signals from this node and, ultimately, causes nonlinearity in the delay control. While nonlinearity of a digital-to-time converter may be acceptable in some applications, other applications, such as frequency synthesis, are not as tolerant. Therefore, it is desirable to eliminate variable slew rate as a source of error. Although trim at production test can reduce or eliminate gain error at a particular temperature and age, variation over temperature and age will ultimately degrade performance. Accordingly, high performance applications require active gain error calibration, which substantially increases complexity.
Rather than use a variable slew rate delay cell, a digital-to-time converter includes a fixed slew rate delay cell. It is desirable to maintain a constant slew rate as an input node passes through a trip point of the voltage comparator. By changing the initial voltage, V_initial, on a fixed capacitor, C_slew, and then initiating slewing with a constant current, Islew, a variable delay element whose delay depends on
$\frac{V_{initial} \times C_{slew}}{I_{slew}}$
is formed. If this initial voltage level is sufficiently far away from the comparator trip point voltage level, then the digital-to-time converter should have minimal modulation of comparator delay and, therefore, a more linear delay versus control characteristic. Various different techniques may be used to realize the initial capacitor voltage. An exemplary digital approach uses a capacitor digital-to-analog converter. In addition, by deriving the slewing current from a voltage-to-current generator loaded by a switched-capacitor resistor, a difference in delay between digital-to-analog converter codes will be ratio-metric to capacitance. Furthermore, a current generator that uses a voltage that is proportional to the voltage reference used by the capacitor digital-to-analog converter causes the difference in delay between digital-to-analog converter codes to be ratio-metric in voltage and proportional to the frequency used to drive the switched-capacitor resistor. Thus, the difference in delay between any two digital-to-analog converter codes will, using ideal elements, be directly proportional to the difference in codes used. Note that conversion gain from voltage-to-time delay may also be trimmed by adjusting the capacitance and/or voltage ratio. Other noise-shaping techniques such as chopping, auto-zeroing, delta-sigma modulation, dynamic element matching (either white or high-pass noise-shaped), and current-copier mirroring techniques may be used to further improve upon the accuracy of the design depending on the application.
Referring to FIGS. 10 and 11, digital-to-time converter 900 includes constant current source 502, which provides constant current I_S, and predetermined capacitor 506, which provides constant capacitance C_S. Digital-to-time converter 900 is configured to generate a selectively controllable reset voltage, V_rstbased on a digital control word D_Qand reference voltage V_ref. During a first time interval, e.g., the time interval during which the reset control signal closes switch 504 and evaluate control signal ck_DMDopens switch 507, digital-to-time converter 900 applies reset voltage V_rstto charge fixed capacitor C_sto a level determined according to a particular reset voltage level selected based on digital control word D_Q. Note that the digital control word D_Qmay include multiple bits having digital voltage levels that are thermometer coded, binary coded, or coded using other suitable techniques. During a second time interval, switch 504 is open according to the reset control signal and switch 507 is closed according to clock signal ck_DMD. Digital-to-time converter 900 applies the constant current generated by current source 502 to node V_S, causing the voltage level on node V_Sto increase from the selected reset voltage level to a second level, V_C, with linear delay control, as illustrated by V_S(t) curves 1002, 1004, and 1006 for increasing digital control word D_Qand decreasing V_rst. The resulting phase interpolator delay is:
$T_{PI} (D_{Q}) = \frac{C_{S}}{I_{S}} (V_{t} - V_{rst}) + T_{comp} = \frac{C_{S}}{I_{S}} V_{t} + T_{comp} - \frac{C_{S}}{I_{S}} \frac{V_{rst}}{2^{N} PI} D_{Q}, where T_{PI, OS} = \frac{C_{S}}{I_{S}} V_{t} + T_{comp} and T_{PI, DAC} = - \frac{C_{S}}{I_{S}} \frac{V_{rst}}{2^{N} PI} D_{Q} .$
Digital-to-analog converter 902 of digital-to-time converter 900 may include passive elements (e.g., resistors and capacitors) that are easily matched by manufacturing techniques. In at least one embodiment, digital-to-time converter 900 includes only one current source and one slew control switch, which reduces or eliminates code-dependent switching errors. Digital-to-time converter 900 includes comparator 508 having a simple, fixed, code-independent delay. However, the code-dependent reset charge injection of digital-to-time converter 900 may cause complex gain and non-linear current variations.
To address the reset charge injection of digital-to-time converter 900, digital-to-time converter 1100 has a code-independent charge injection topology as illustrated in FIGS. 12-14. Digital-to-time converter 1100 includes comparator 508 having a simple, fixed, code-independent delay. After a reset interval charges node V_sto a fixed reset voltage V_rst, digital-to-time converter 1100 subtracts a fixed pedestal voltage V_pedfrom that reset voltage on node V_sto provide code-independent charge injection. As a result, digital-to-time converter 1100 provides a ratio-metric topology having linear delay control and code-independent charge injection. Accordingly:
$T_{PI} (D_{Q}) = \frac{C_{S}}{I_{S}} (V_{t} - V_{rst}) + T_{comp} + \frac{C_{S}}{I_{S}} \frac{V_{ref}}{2^{N} PI} D_{Q},$
where
$T_{PI, OS} = \frac{C_{S}}{I_{S}} (V_{t} - V_{rst}) + T_{comp} and T_{PI, DAC} = \frac{C_{S}}{I_{S}} \frac{V_{ref}}{2^{N} PI} D_{Q} .$
Digital-to-analog converter 1102 may include capacitor-based units that provide improved matching over current source-based units of other phase interpolator topologies. Since digital-to-time converter 1100 uses code-independent reset charge injection, phase interpolator errors due to charge injection are reduced or eliminated. For all values of digital control word D_Q, digital-to-analog converter 1102 charges the voltage on node V_sto the same, predetermined reset voltage level. In response to deassertion of the reset control signal, digital-to-analog converter 1102 applies a pedestal voltage to the sense node of comparator 508 according to digital control word D_Q. In at least one embodiment, digital-to-analog converter 1102 logically combines digital control word D_Qwith the reset control signal to generate a gated version of digital control word D_Q, D_cap, individual bits of which are coupled to corresponding bit cells 1204, 1206, and 1208 for proper pedestal voltage generation following the reset phase. In at least one embodiment of digital-to-analog converter 1102, each of the bit cells 1204 includes corresponding inverter 1210 and corresponding pedestal capacitor 1212 that separately charges or discharges the sense node to realize a combined voltage on the sense node of V_rst-V_ped(D_Q). As D_Qincreases, V_ped(D_Q) increases, thereby increasing the time between the slew start time t_startto the trip point t_tripof comparator 508 and thus increasing the delay of ck_out. For example, D_Qassociated with V_ped1and t_trip1is less than the value of D_Qassociated with V_ped2and t_trip2and the value of digital control word D_Qassociated with V_ped3and t_trip3, as illustrated by corresponding V_s(t) curves 1302, 1304, and 1306, respectively.
Referring to FIG. 15, in at least one embodiment of a digital-to-time converter, current source 502 is a ratio-metric voltage-to-current converter. Current source 502 generates a fixed current using switched-capacitor resistor 1410 controlled by reference clock signal ck_R. Switched-capacitor resistor 1410 is used instead of an integrated circuit resistor for ratio-metric operation of the digital-to-time converter (e.g., phase interpolator). The resistance is a function of the reference clock, which is provided by a high precision frequency source (e.g., accurate within a few parts per million). Current source 502 provides a constant current I_Susing precision current mirror techniques:
I _S f _R C _R V _ref,
where C_Ris the capacitance of a capacitor in switched-capacitor resistor 1410 and f_Ris the frequency of reference clock signal ck_R. Ratio-metric operation is achieved by using the same reference voltage V_refused by digital-to-analog converter 1102. If the period of the reference clock T_Ris set to be equal to N_ref×T_ref, and the capacitor is trimmed to have C_R=N_refC_S(assuming a current mirror ratio of one) then:
$T_{PI} (D_{Q}) = T_{ref} \frac{V_{t} - V_{rst}}{V_{ref}} + T_{comp} + \frac{T_{ref}}{2^{N} PI} D_{Q},$
Trimming may reduce or eliminate gain error at production test although variation due to environmental factors remains. Current-copier techniques may also be used to reduce or eliminate current mirror errors. Increasingly ratio-metric behavior that reduces deterministic jitter and random jitter may be achieved by deriving V_tand V_rstfrom V_ref. In at least one embodiment of current source 502, the current I_smay be digitally controlled according to the value of digital current control signal D_mirror, which may be a binary coded-signal or a thermometer-coded signal that selectively enables devices within current mirror 1405 to generate a current I_Shaving a corresponding level. The resistance of resistor 1407 may be trimmed to reduce or eliminate gain error of an associated digital-to-time converter with respect to temperature variation. For example, resistor 1407 includes resistances having different temperature coefficients (e.g., p-type or n-type polysilicon resistors, p-type or n-type diffusion resistors, metal resistors, or other suitable resistors). The resistance of resistor 1407 may be selectable according to digital temperature sensitivity control signal D_temp. Selectively controlling current and temperature sensitivity may be used to reduce or eliminate gain errors of the digital-to-time converter as a function of temperature alone or in conjunction with other gain calibration techniques.
The techniques described above may result in a residual phase error primarily due to the quantization error resulting from the digital-to-time converter and the noise of the system. The quantization error may be due to data converter non-idealities such as offset, gain error, and integral nonlinearity. While offset is not detrimental to clock synthesizer applications, addressing gain error and integral nonlinearity are critical to spurious performance. Background calibration of the digital-to-time converter, described below, addresses gain error with the assumption that the native integral nonlinearity of the converter is acceptable.
Techniques for background calibration of the digital-to-time converter gain error may be used. Note that, while the calibration may be performed digitally, hardware requirements and associated power consumption may be impracticable for low cost and/or low power applications. As described above, the digital-to-time converter and dithering modulo divider form a true fractional divider. In typical usage, the fractional divider is always driving edges into the phase detector of the PLL. However, since signal edges only occur every PLL reference clock period (e.g. 20 ns for a 50 MHz clock), calibration circuitry may perform operations between edges. Therefore, by monitoring the gain of the digital-to-time converter of the phase interpolator in the background between PLL reference clock edges, error detection and correction circuitry may reduce or eliminate gain error using analog techniques.
In at least one embodiment, a gain calibration technique exploits linearity of the digital-to-analog converter of the phase interpolator described above. Systematic nonlinearity may be reduced or eliminated with disciplined unit cell design. Use of simple capacitor-based units have reduced random mismatch as compared to units including transistors or other non-linear devices. Use of active gain error calibration reduces overhead hardware and design complexity. In addition, a target refresh rate of the calibration is independent of a divide ratio. In at least one embodiment, since the digital-to-analog converter is inactive following phase interpolator edge transmission, phase interpolator calibration is performed between edge transmissions. In at least one embodiment, the calibration technique is addressed at the block level, e.g., locally for each phase interpolator. Local calibration facilitates system integration, duplication, and design reuse.
Referring to FIG. 17, a technique for compensating for a gain error of a digital-to-time converter of a phase interpolator generates a gain control signal G_refthat is fed back to the digital-to-time converter 1650. Digital-to-time converter 1650 generates output analog signal t′_outbased on a digital input signal D′_in. During evaluation intervals (i.e., when calibration is not enabled, e.g., intervals during which cal=‘0’), selection module 1654 provides digital signal D_inas digital input signal D′_in. During background calibration intervals (e.g., intervals during which cal=‘1’) selection module 1654 provides a calibration digital signal D_calas digital input signal D′_in. Referring to FIGS. 17 and 18, edge generator 1520 provides clock signal t_refand calibration clock signal t_cal, both generated with the same delay to reference clock signal ck_ref. Clock signal t_refis used by digital-to-time converter 1650, e.g., as evaluate control signal ck_DMDof digital-to-time converter including unit cell 700 of FIG. 8, digital-to-time converter 900 of FIG. 10, or digital-to-time converter 1100 of FIG. 12. Referring back to FIGS. 17 and 18, calibration clock signal t_calvacillates between a low value that corresponds to a minimum delay and a high value that corresponds to a minimum delay plus one full reference clock signal ck_refperiod delay (e.g., generated using delay element 1522, which may be a flip-flop or other element providing a suitable delay) according to the value of delay control signal D_delay
In at least one embodiment, edge generator 1520 provides calibration clock signal t_refbased on reference clock signal ck_refand gating clock signal gate, which is a frequency-divided version of reference clock signal ck_ref(e.g., frequency-divided by N). Accordingly, calibration clock signal t_calincludes a stream of pulses occurring when calibration is enabled. Individual pulses of the stream alternate between pulses having edges that nominally correspond to edges of pulses of clock signal t_refand pulses having edges that nominally correspond to edges of t_refdelayed by an entire period of reference clock signal ck_ref, which will be described further below with reference to FIGS. 20-24, where t_ref, t_cal, and t′_outof FIGS. 17 and 18 correspond to clock signals ck_DMD, ck_D, and ck_PI, respectively, of FIGS. 21-25.
Referring back to FIGS. 17 and 18, gain adjustment generator 1652 may provide output clock signal t′_outas output clock signal t_outduring evaluation intervals. When performing background calibration, gain adjustment generator 1652 generates an analog gain adjustment that is fed back to digital-to-time converter 1650 to compensate for gain error of digital-to-time converter 1650.
Referring to FIG. 19, the gain calibration techniques described herein with regard to digital-to-time converters may be applied to time-to-digital converters that include a digital-to-time converter. For example, the techniques may be adapted to generate an analog gain adjustment G_refthat is fed back to the time-to-digital converter 1750 to compensate for gain error of time-to-digital converter 1750. Time-to-digital converter 1650 generates output digital signal D′_outbased on an analog input signal t′_in. During evaluation intervals (i.e., when calibration is not enabled, e.g., intervals during which cal=‘0’), selection module 1754 provides analog signal t_inas t′_in. During background calibration intervals (e.g., intervals during which cal=‘1’), selection module 1754 provides a calibration analog signal t_cal. In at least one embodiment of the gain calibration technique, gain adjustment generator 1752 generates a digital gain adjustment that is internally applied to the output digital signal D′_outto generate adjusted output digital signal D_out(FIG. 20).
In at least one embodiment of the calibration technique, the digital-to-time converter is a phase interpolator configured to generate phase interpolator output ck_PI, based on clock signal ck_DMD, which is the output of a fractional divider (e.g., the output of feedback clock signal ck_fbor output clock signal ck_outoutput by respective phase interpolators 204 responsive to respective fractional dividers 108 of corresponding interpolative dividers 202 of FIGS. 2 and 3). Referring to FIG. 21, error detection and correction module 1508 generates gain control signal g_ctrlbased on phase interpolator output ck_PIand feeds back gain control signal g_ctrlto phase interpolator 1506 to compensate for gain error of phase interpolator 1506. Delay module 1504 is a delay matching module that matches the delay of the phase interpolator 1506 such that delayed reference signal ck_Dis delayed by the same amount as phase interpolator output ck_PI(e.g., delay introduced by comparator 508 of digital-to-time converter 1100 of FIG. 12) in the minimum delay condition (e.g., delay control signal D_PIequals zero).
Referring back to FIG. 21, error detection and correction module 1508 compares edges of the output of phase interpolator 1506 (i.e., phase interpolator output ck_PIgenerated during calibration intervals), and the output of delay module 1504, delayed reference signal ck_D, generated during calibration intervals to determine the gain error. Delay module 1504 receives pulses for error detection as calibration timing reference signal ck_CDTC. During evaluation intervals (e.g., intervals during which cal=‘0’) during which phase interpolator 1506 transmits delayed edges, calibration timing reference signal ck_CDTCmay be static and does not transition. Calibration digital-to-time converter 1502 generates the calibration timing reference signal ck_CDTCbased on a reference clock signal ck_refand a delay control signal D_delay. When in calibration intervals (e.g., intervals during which cal=‘1’), calibration digital-to-time converter 1502 delays the same, or related, gate signal as is used to derive clock signal ck_DMDbased on delay control signal D_delayto generate timing reference signal ck_CDTC.
In at least one embodiment of phase interpolator 204, delay control signal D_delayvacillates between a low value that corresponds to a minimum delay and a high value that corresponds to a minimum delay plus one full ck_refperiod delay (e.g., a delay generated using delay element 1612 for a delay equivalent to D_PI=2^NPI) or an entire least-significant bit of FDIV. Accordingly, the output of calibration digital-to-time converter 1502 is a timing reference signal ck_CDTCthat is a stream of pulses occurring when calibration is enabled. Individual pulses of the stream alternate between pulses having edges that nominally correspond to edges of pulses of clock signal ck_DMD, and pulses having edges that nominally correspond to edges of ck_DMDdelayed by an entire period of ck_ref. Although illustrated using rising edges of ck_ref, other embodiments generate calibration timing reference signal ck_CDTCbased on falling edges of ck_ref.
Referring to FIGS. 21 and 22, divider 110 generates a clock signal ck_DMD, having pulses with periods that vacillate between different clock periods to generate the reference timing signal for both calibration and evaluation intervals having a period based on the integer value D_N. For example, during the interval between time t₁and time t₂, clock signal ck_DMDhas a period according to integer value D_Nin an evaluation mode (i.e., a transmit mode) corresponding to a clock signal being generated by the target application (e.g., the value of D_Ncorresponds to three periods of ck_ref). During a calibration interval, e.g., the interval between time t₂and time t₃, clock signal ck_DMDincludes a pulse (e.g., rising edge) with a period according to integer value D_Ncorresponding to a clock signal being generated for use during the calibration interval (e.g., integer value D_Ncorresponds to four periods of ck_ref). In at least one embodiment of phase interpolator 204, multiplexer 1503 spreads a binary or otherwise coded delay control signal D_delayto a number of bits corresponding to the number of unit delay cells included in phase interpolator 1506. For example, when D_delay=‘0,’ and NPI=2, then phase interpolator 1506 has four unit cells (one of which is in addition to unit cells corresponding to bits output by multiplexer 1503 and which is driven by the output of a logical AND of D_delayand cal), then D_PI=‘0000,’ and when delay control signal D_delay=‘1,’ then D_PI=2^NPI=‘1111,’ which corresponds to delaying by an entire period of ck_ref.
FIGS. 22 and 23, illustrate at least one embodiment of timing reference signal generation. In at least one embodiment, calibration digital-to-time converter 1502 is a single bit digital-to-time converter with a least-significant bit of T_ref. When delay control signal D_delay=0, ck_CDTCis delayed from ck_DMDby T_CDTC(0), which is ideally zero. When delay control signal D_delay=1, ck_CDTCis delayed from ck_DMDby T_CDTC(l), which is ideally T_ref. The calibration codes are time interleaved with the phase interpolator evaluation values as signal D_N. Note that in at least one embodiment, corresponding pulse edges of the timing reference signal ck_Dare generated with a fixed delay amount T_Dthat matches the minimum delay (e.g., D_PI=0) of phase interpolator 1506.
In at least one embodiment of a gain calibration technique, a first phase of the gain calibration, delay control signal D_delay=‘0,’ and error detection and correction module 1508 generates an indication of a time difference between an edge of fixed delay signal ck_Dand a corresponding edge of phase interpolator output ck_PI:
ΔT _cal(0)=T _PI(0)−[T _CDTC(0)+T _D].
In a second phase of the gain calibration, D_delay=‘1,’ and error detection and correction module 1508 generates an indication of a time difference between an edge of delay signal ck_Dand an edge of phase interpolator output ck_PI:
ΔT _cal(1)=T _PI(2^NPI)−[T _CDTC(1)+T _D].
Error detection and correction module 1508 determines the difference between the two measurements to generate indicator of linear gain error T_γ:
$\begin{matrix} T_{γ} = Δ T_{cal} (1) - Δ T_{cal} (0) \\ = [T_{PI} (2^{NPI}) - T_{PI} (0)] - [T_{CDTC} (1) - T_{CDTC} (0)] \\ = T_{PI, FS} - T_{ref} . \end{matrix}$
where T_PI,FS[T_PI(2^NPI)−T_PI(0)], and
T _ref =[T _CDTC(1)−T _CDTC(0)].
Error detection and correction module 1508 generates gain control signal g_ctrlbased on gain error T_γ. Compensating for linear gain error T_γ, matches the gain of phase interpolator 1506, which is a subrange data converter, to the gain of the fractional divider 108, to obtain a linear transfer function.
Referring to FIGS. 24 and 25, in each calibration phase, phase interpolator 1506 introduces a different delay into clock signal ck_DMDto generate phase interpolator output ck_PI. When D_delay=0, clock signal ck_DMDis delayed relative to ck_CDTCby T_PI(0)−T_D, which is ideally zero. When D_delay=1, ck_DMDis delayed relative to ck_CDTCby T_PI(2^NPI)−T_D, which is ideally T_ref. The logical AND of D_delayand cal control signal introduce an extra least-significant bit to D_PIso that the full-scale range may be generated between the two calibration measurements (without this extra bit, the maximum code is 2^NPI−1). Note that T_skew=T_CDTC(0)+T_D−T_PI(0). The gain error and correction signal generation of error detection and correction module 1508 may be implemented using digital circuits, analog circuits, or a combination thereof. In at least one embodiment, an analog implementation includes phase detector 1802 and a loop filter 1804, which are configured to generate phase interpolator gain control signal g_ctrl. In at least one embodiment, digital control within error detection and correction module 1508 includes a time-to-digital converter coupled to a digital loop filter and a gain control signal provided to an auxiliary digital-to-analog converter. In at least one embodiment, an analog phase detector provides signals to an analog loop filter that generates signals provided to phase interpolator auxiliary digital-to-analog converter. Note that during calibration phases, gating circuit 1806 may block propagation of ck_PIto ck_out.
Rather than delaying both ck_Dand ck_PIin the calibration mode, by applying a delay that vacillates between a low value that corresponds to a minimum delay and a high value that corresponds to a minimum delay plus one full ck_refperiod delay in parallel, in separate paths as described above with reference to FIGS. 21-25, other embodiments of the calibration technique apply the two vacillating delays in series, in the same path, but out of phase with each other. For example, referring to FIG. 26, an embodiment of the calibration technique generates timing reference signal ck′_Dhaving a fixed delay. Delay element 1509 and logic gate 1511 generate timing reference signal ck′_D. Timing reference signal ck′_Dlags clock signal ck_DMDby one period of reference clock signal ck_ref. In calibration mode, the technique applies a first vacillating delay to clock signal ck′D_DMDto generate an input to the phase interpolator. In calibration mode, the phase interpolator applies a second vacillating delay serially to generate the output signal ck_PI. The first vacillating delay vacillates between the low value that corresponds to the minimum delay and the high value that corresponds to the minimum delay plus one full period of reference clock signal ck_ref, and the second vacillating delay vacillates between the high value that corresponds to the minimum delay plus one full period of reference clock signal ck_ref, and the low value that corresponds to the minimum delay. Accordingly, the first and second vacillating delays introduce a net delay of one full period of reference clock signal ck_ref. Error detection and correction circuit 1508 generates an indication of a time difference between an edge of fixed delay signal ck_Dand a corresponding edge of phase interpolator output ck_PI:
ΔT _cal(0)=[T _CDTC(1)+T _PI(0)]−[T _ref +T _D].
In a second phase of the gain calibration, D_delay=‘1,’ and error detection and correction module 1508 generates an indication of a time difference between an edge of delay signal ck_D, and an edge of phase interpolator output ck_PI:
ΔT _cal(i)=[T _CDTC(0)+T _PI(2^NPI)]−[T _ref +T _D].
Error detection and correction module 1508 determines the difference between the two measurements to generate indicator of linear gain error T_γ:
$\begin{matrix} T_{γ} = Δ T_{cal} (1) - Δ T_{cal} (0) \\ = [T_{CDTC} (0) + T_{PI} (2^{NPI})] - [T_{CDTC} (1) + T_{PI} (0)] \\ = T_{PI, FS} - T_{ref} . \end{matrix}$
where T_PI,FS[T_PI(2^NPI)−T_PI(0)], and
T _ref =[T _CDTC(0)−T _CDTC(1)].
When phase interpolator 1506 has the proper gain, T_PI,FS=T_refand no additional gain adjustment is needed. Although the fixed delay of T_refintroduced into ck′_Dis not necessary, it may reduce the systematic phase error between fixed delay signal ck_Dand phase interpolator output ck_PIin the calibration mode, which reduces the dynamic range requirements of error detection and correction module 1508.
Referring to FIG. 27, analog error detection and correction module 1508 generates a gain control compensation signal ΔI_s,γthat is provided to the current source of the digital-to-time converter of the phase interpolator. During a first calibration interval, delay control signal D_delayhas a first value. Phase detector 2002 generates up and down signals based on a first time difference in phases of reference signal ck_Dand phase interpolator output ck_PI. Charge pump 2004 converts the first time difference into differential charge that is stored on integrating capacitors 2012 and 2014, each having capacitance C_int. The charge pump output is based on the difference between the digital-to-time converter delay at a minimum delay (i.e., D_Q=0) and the delay through calibration digital-to-time converter 1502 and delay module 1504. During a second calibration interval, delay control signal D_delayhas a second value. The charge pump output is based on the difference between the digital-to-time converter delay at a maximum delay (i.e., D_Q=2^NPI) and the delay through calibration digital-to-time converter 1502 and delay module 1504. Phase detector 2002 generates up and down signals based on a second time difference in phases of reference signal ck_Dand phase interpolator output ck_PI. Charge pump 2004 converts the second time difference into a second differential charge, switch 2006 inverts the second differential charge, and stores that inverted difference in charge on integrating capacitors 2012 and 2014. When in the second calibration interval, the track signal causes charge transfer to hold capacitors 2016 and 2018, which updates V_hold. Transconductance amplifier 2008 generates the gain control signal ΔI_s,γ, that is used to adjust the slew current to correct for the gain error. Note that switch 2006 facilitates a subtraction function to isolate gain error T_γfrom the two measurements. In addition, note that the critical paths are identical between the two measurements. However, note that charge-sharing due to T_skewmay cause systematic gain error.
Referring to FIG. 28, in at least one embodiment, in addition to generating gain control signal ΔI_s,γ, analog error detection and correction module 1508 also generates a differential skew correction signal, ΔI_s,skew, which may be combined with a current source control signal to compensate for the skew between outputs of those two modules. One of the complementary signals (e.g., −ΔI_s,skew) of the differential signal is combined with a current source control signal of the phase interpolator 1506 and the other of the complementary signals (e.g., ΔI_s,skew) is combined with a current source control signal of delay module 1504. In operation, during a first calibration interval, D_delayhas a first value. Phase detector 2002 generates up and down signals based on a first time difference in phases of reference signal ck_Dand phase interpolator output ck_PI. Charge pump 2004 converts the first time difference into a first differential charge, the track₀signal causes charge transfer of that charge to integrating capacitor 2106. During a second calibration interval, D_delayhas a second value. Phase detector 2002 generates up and down signals based on a second time difference in phases of ck_Dand ck_PI. Charge pump 2004 converts the second time difference into a second differential charge, and the track₁signal causes charge transfer of that charge to integrating capacitor 2108. Differential-to-common mode transconductance amplifier 2008 applies gain control signal ΔI_s,γto adjust the slew current of the phase interpolator. A common-mode-to-differential-mode transconductance amplifier 2014 applies skew correction by differentially applying slew currents of phase interpolator 1506 and delay module 1504 although in other embodiments, all of the skew correction may be applied to the phase interpolator 1506 delay elements or all of the skew correction may be applied to the delay elements of delay module 1504. Note that charge-sharing due to T_γmay cause systematic gain error, charge pump 2004 has a finite output resistance, and supporting a differential voltage, V_int1−V_int0may cause gain errors. In addition, note that controlling slew current I_saffects the control loop for both calibration and evaluation:
$T_{PI} (D_{Q}) = \frac{C_{S}}{I_{S}} (V_{t} - V_{rst}) + T_{comp} + \frac{C_{S}}{I_{S}} \frac{V_{ref}}{2^{N} frac} D_{Q},$
Changing the slew current I_scontrols the gain of the phase interpolator, which is desired, but also affects the skew measurement in a calibration phase (e.g., calibration phase zero, T_skew+T_PI(0), which is undesirable.
Referring to FIG. 29, in at least one embodiment, an analog implementation of error detection and correction module 1508 decouples the gain and skew control loops. The gain is corrected by controlling a voltage reference in digital-to-analog converter 2204 of phase interpolator 1506. Error detection and correction module 1508 senses skew only from V_int0. The two integrator loop forces V_int1−V_int0to within a transconductor offset voltage, which reduces charge sharing error. The charge-sharing error may be further reduced by auto-zeroing the gain control transconductor. In other embodiments, correction of the delay module 1504 may also be implemented using voltage control of a reference voltage. Note that V_gctrlV_reffor trim at production test and upon startup.
Referring to FIG. 30, in at least one embodiment, error detection and correction module 1508 includes bang-bang phase detector 2302 coupled to digital accumulator 2308, and current digital-to-analog converter 2310, which reduces or eliminates T_skewand issues associated with an analog control loop implementation (e.g., charge sharing). Charge pump 2304 and capacitor 2305 are configured as an integrating switched current digital-to-analog converter. The charge pump output may include small up/down charge packets to reduce V_gctrlripple. However, note that this configuration dithers around on average, causing ripple in the control signal and resulting in quantization noise. In other embodiments, correction of the delay module 1504 may be implemented using a charge pump and integrating switched-current digital-to-analog converter control.
Referring back to FIG. 28, charge-sharing at the output of charge pump 2004 behaves like a switched-capacitor resistor between nodes V_int0and V_int1, which results in a reduction in DC charge pump gain. Sustaining a differential voltage across nodes V_int0and V_int1may cause a substantial gain error. A technique for reducing or eliminating that contribution to gain error is illustrated in FIG. 31. Error detection and correction module 1508 includes pre-charge voltage buffer 2406 and pre-charge voltage buffer 2408 for each of capacitor nodes C_int0and C_int1. Pre-charge voltage buffer 2406 and pre-charge voltage buffer 2408 are configured to force the respective nodes coupled to the output of the charge pump 2004 to V_int0and V_int1, respectively, prior to connecting those nodes to C_int0and C_int1, respectively. Accordingly, the charge-sharing is reduced to the buffer offset voltage. In other embodiments, the switched- capacitor buffers 2406 and 2408 of error are modified to be an integrator with noise cancellation, as illustrated in FIG. 32 for one calibration phase (i.e., for the D_delay=0 calibration phase). Although not shown, the integrator with noise cancellation circuitry is replicated for the other calibration phase (i.e., for the D_delay=1 calibration phase). Such technique substantially reduces charge-sharing in analog error detection and correction module 1508.
Interpolative divider techniques that interleave evaluate or transmit operations with calibration operations, as described above, use a reference clock signal that is at least twice the frequency of the transmit operation. Referring to FIG. 33, by interleaving multiple interpolative dividers, an output clock signal with approximately 50% duty cycle may be achieved. The interleaved interpolative divider implementation may double the output clock frequency or relax calibration timing requirements. Dithering-modulo-divider and controller 2602 alternates between phase interpolator 2604 and phase interpolator 2606. While phase interpolator 2604 transmits a rising edge, which causes a rising edge on ck_out, phase interpolator 2606 makes a calibration measurement. Phase interpolator 2606 transmits a rising edge, which causes a falling edge on ck_out, while phase interpolator 2604 performs a calibration measurement. Flip-flop 2608 latches the rising edge ck_PIrfrom phase interpolator 2604 and the falling edge ck_PIffrom phase interpolator 2606 and generates output clock ck_outhaving approximately 50% duty cycle. Note that more interleaving can support higher frequency operation at e.g., radio frequencies.
The gain calibration interleaving with digital-to-time signal evaluation techniques described herein may be applied to time-to-digital signal conversion applications. For example, referring to FIG. 34, a gain measurement is interleaved between edges of clock ck_in. Use of the calibration digital-to-time converter facilitates a two-point measurement with a precise time base having reference period T_ref. Similar to the digital-to-time calibrations discussed above, the control loop may change reference voltage V_refto force accurate time-to-digital gain. A difference between code D_Tgenerated by analog-to-digital converter 2704 determines a gain of the time-to-digital converter. Skew between the calibration mode and the input path introduces a fixed offset between time stamps. As a result, additional range greater than reference period T_ref(as defined by the period of clock signal ck_ref) may be required for the analog-to-digital conversion in order to reduce or eliminate over-ranging (i.e., saturation of the analog-to-digital converter).
Referring to FIG. 35, exemplary time-to-digital converter 3200 includes capacitance C_Sdistributed across capacitors 3208 and 3210 of feedback digital-to-analog converter units (e.g., digital-to- analog converter units 3214 and 3216 similar to digital-to-analog converter 1102 of digital-to-time converter 1100 of FIGS. 12-14) of a digital-to-analog converter configured to sample-and-hold input signal ck_in. Comparator 3212 uses the comparator threshold voltage V_Tto compare against the stored input time signal V_Sgenerated by sample-and-hold of input signal ck_in. Comparator 3212 provides the results of the comparison to successive approximation register 3202. Successive approximation register 3202 provides a digital approximation code to the internal capacitor-based digital-to-analog converter that supplies comparators 3212 with an analog voltage representation of the digital approximation code. Thus, analog-to-digital converter 3200 generates output digital bits b₀, and b₁, which collectively form an output digital signal D_outand are based on input analog signal V_S, originally derived from an input time interval. The time domain information may be mapped into voltages and digital-signal processing techniques may be used to determine distances, velocities, etc. based on that time domain information in radar or other applications.
An embodiment of the digital-to-time converter and calibration technique described above may be included as a feedback digital-to-analog converter in a high resolution time-to-digital converter application. Referring to FIG. 36, an exemplary high-resolution time-to-digital converter receives input signal ck_in, and digital control words D_f,intand D_f,fracwhich correspond to coarse and fine frequency divider control codes, respectively. Accumulator 3612 integrates the fine frequency divider control code to generate a fine delay control code (i.e., fine phase domain control code D_φ,frac) and summing node 3618 combines any carry from the integration with the coarse frequency divider control code and provides the sum to digital-to-frequency converter 3616. Phase detector 3602 generates analog phase error signal φ_εindicative of the phase difference between input signal ck_inand feedback signal ck_fb. Loop filter 3604 integrates φ_εand quantizer 3606 generates digital phase output D_φoutbased on the loop filter output.
Summing node 3610 combines digital phase output D_φoutwith fine phase domain control code D_φ,fracto generate digital phase control code D_φDTCfor conversion to an analog signal by digital-to-time converter and calibration module 3608, which generates feedback signal ck_fb. Feedback signal ck_fbis an analog representation of the digital output D_φoutDigital-to-frequency converter 3616 and edge gating circuit 3614 form a digitally controlled integer divider and generate clock signal ck_DMDbased on reference clock signal ck_refand digital control word D_f, which indicates a coarse (or integer portion) of a frequency divider value. Since the gain of the digital-to-time conversion performed by digital-to-time converter and calibration module 3608 is calibrated consistent with techniques described above, feedback clock signal ck_fbhas reduced deterministic jitter, which allows for digital output D_φoutto provide a higher resolution digital representation of the time information of input signal ck_inthan other time-to-digital converters. Note that in other high-resolution time-to-digital converters consistent with the teachings herein, phase-detector 3602, loop filter 3604, and quantizer 3608 may be replaced with a bang-bang phase detector or other suitable implementations.
Thus, improved techniques for digital-to-time conversion and phase interpolation for clock synthesis have been disclosed. Systematic nonlinearities in digital-to-analog conversion are reduced by using capacitor-based unit delay cells and ratio-metric design techniques. In addition, an active gain error calibration technique for block level implementation is disclosed for high-performance applications. Techniques described herein reduce or eliminate gain error of a digital-to-time converter. In applications such as fractional clock synthesis, the reduction or elimination of that gain error reduces spurious outputs due to phase wrapping errors caused by the gain error of the digital-to-time converter. Accordingly, fractional clock synthesizers using those techniques may produce clocks with spurious performance similar to integer clock synthesizer counterparts but with increased power efficiency due to increased bandwidth (e.g. suppressed VCO noise) of the fractional synthesizer. In addition, the digital-to-time conversion techniques described herein may have reduced area and power consumption as compared to other techniques.
While circuits and physical structures have been generally presumed in describing embodiments of the invention, it is well recognized that in modern semiconductor design and fabrication, physical structures and circuits may be embodied in computer-readable descriptive form suitable for use in subsequent design, simulation, test or fabrication stages. Structures and functionality presented as discrete components in the exemplary configurations may be implemented as a combined structure or component. Various embodiments of the invention are contemplated to include circuits, systems of circuits, related methods, and tangible computer-readable medium having encodings thereon (e.g., VHSIC Hardware Description Language (VHDL), Verilog, GDSII data, Electronic Design Interchange Format (EDIF), and/or Gerber file) of such circuits, systems, and methods, all as described herein, and as defined in the appended claims. In addition, the computer-readable media may store instructions as well as data that can be used to implement the invention. The instructions/data may be related to hardware, software, firmware or combinations thereof.
The description of the invention set forth herein is illustrative, and is not intended to limit the scope of the invention as set forth in the following claims. For example, while the invention has been described in an embodiment in which a digital-to-time converter of a phase interpolator of a clock synthesizer is compensated, one of skill in the art will appreciate that the teachings herein can be utilized with other digital-to-time and time-to-digital converter applications. Variations and modifications of the embodiments disclosed herein, may be made based on the description set forth herein, without departing from the scope and spirit of the invention as set forth in the following claims.

Claims

1. An apparatus comprising:

a digital-to-time converter configured to generate an output signal based on a digital code, an input signal, and a gain calibration signal, the output signal having edges linearly delayed from corresponding edges of the input signal based on the digital code, the digital code vacillating between an evaluation code and a calibration code;

a reference signal generator configured to provide a delayed version of the input signal, the delay of the reference signal generator being matched to a delay of the digital-to-time converter; and

a calibration circuit configured to generate the gain calibration signal based on the output signal and the delayed version of the input signal.

2. The apparatus, as recited in claim 1, wherein the calibration code alternates between a first calibration delay code and a second calibration delay code.

3. The apparatus, as recited in claim 1, further comprising:

an integer frequency divider configured to generate the input signal and the evaluation code based on a reference clock signal and a digital divider code.

4. The apparatus, as recited in claim 3, wherein the digital-to-time converter is configured as a subrange data converter with respect to the integer frequency divider, a first delay code of the calibration code corresponding to a minimum delay and the second delay code of the calibration code corresponding to the minimum delay plus a delay of the integer frequency divider corresponding to the least-significant bit of the digital divider code.

5. The apparatus, as recited in claim 1, wherein the calibration circuit comprises:

a phase detector configured to generate a phase difference signal based on the output signal and the delayed version of the input signal; and

a loop filter configured to generate the gain calibration signal based on the phase difference signal.

6. The apparatus, as recited in claim 1, wherein the calibration circuit comprises:

a phase detector configured to generate an up down output signal based on the output signal and the delayed version of the input signal;

a charge pump circuit configured to generate an output based on the phase difference signal; and

a switched capacitor circuit configured to generate the gain calibration signal based on the output of the charge pump circuit and a control signal.

7. The apparatus, as recited in claim 6, wherein the calibration circuit generates a skew correction signal based on the output of the charge pump circuit, the skew correction signal being indicative of skew between the output signal and the delayed version of the input signal.

8. The apparatus, as recited in claim 1, further comprising:

a select circuit configured to provide the evaluation code based on digital quantization error of the input signal in response to a first value of a control signal and to provide the calibration code in response to a second value of the control signal.

9. The apparatus, as recited in claim 1, wherein the calibration code has a first calibration value during a first calibration interval and a second calibration value during a second calibration interval.

10. The apparatus, as recited in claim 1, further comprising:

a control circuit configured to generate a control signal to enable calibration time interleaved with signal generation based on digital quantization error.

11. The apparatus, as recited in claim 1, wherein the reference signal generator comprises:

a calibration digital-to-time converter configured to generate a first signal based on the input signal and the digital code; and

a delay element configured to generate the delayed version of the input signal based on the first signal, the delay element having a delay matched to a delay of the digital-to-time converter.

12. The apparatus, as recited in claim 1, wherein the digital-to-time converter comprises:

a circuit configured to generate an output signal having a pulse width based on a digital code, the pulse width being based on the digital code, comprising:

a first node selectively coupled to a first reference signal;

a second node configured to receive a second reference signal;

a capacitive digital-to-analog converter configured to generate a voltage on the first node in response to the digital code;

a comparator configured to generate the output signal based on a first signal on the first node and the second reference signal on the second node; and

a current source selectively coupled to the first node and responsive to the gain control signal.

13. A method performed by an apparatus, the method comprising:

generating an output signal based on a digital code, an input signal, and a gain calibration signal, the output signal having an edge linearly delayed from a corresponding edge of the input signal based on the digital code, the digital code vacillating between an evaluation code and a calibration code;

generating a first phase error between the delayed signal having a first calibration delay based on the digital code having a first calibration value of the calibration code and the first signal during a first calibration interval;

generating a second phase error between the delayed signal having a second calibration delay based on the digital code having a second calibration value of the calibration code and the first signal during a second calibration interval; and

generating the gain calibration signal based on the first phase error and the second phase error.

14. The method, as recited in claim 13, further comprising:

generating a skew correction signal based on a common mode of the first phase error and the second phase error.

15. The method, as recited in claim 13, further comprising:

adjusting a gain of a digital-to-time converter based on the gain calibration signal.

16. The method, as recited in claim 13, wherein generating the output comprises:

interleaving transmitting a first edge of a first evaluate operation with a calibration operation;

interleaving transmitting a second edge of a second evaluate operation with a second calibration operation; and

combining the first and second edges to generate the output having a 50% duty cycle.

17. The method, as recited in claim 13, further comprising:

generating the input signal and the digital code based on a reference clock signal and a digital divider code.

18. The method, as recited in claim 17, wherein the first delay code corresponds to zero delay and the second delay code corresponds to a delay equivalent to delay defined by a least-significant bit delay of an integer range of the digital divider code.

19. An apparatus comprising:

means for generating an output signal based on a digital code, an input signal, and a gain calibration signal, the output signal having edges delayed from corresponding edges of the input signal based on the digital code, the digital code vacillating between an evaluation code and a calibration code;

means for generating a delayed version of the input signal based on the calibration code having a first calibration value and the calibration code having a second calibration value; and

means for generating a gain calibration signal based on phase errors between the output signal and the delayed version of the input signal, generating based on the calibration code having a first calibration value and the calibration code having a second calibration value.

20. The apparatus, as recited in claim 19, further comprising:

means for generating the input signal and the evaluation value of the digital code by frequency dividing a reference clock signal based on a digital divider code, the evaluation value of the digital code corresponding to an error value of the means for generating,

wherein the first calibration value corresponds to zero delay and the second calibration value corresponds to a delay equivalent to delay defined by a least-significant bit delay of an integer range of the digital divider code.