Input jitter attenuation in a phase-locked loop

[0002] This invention relates to phase-locked loops (PLLs), and more particularly to improving input jitter attenuation in PLLs.

[0003] PLLs are used in many communications systems because of their remarkable versatility. For example, a PLL may be used to perform frequency synthesis, tone decoding, signal modulation and demodulation, clock generation, and pulse synchronization. In addition, PLLs may be used in analog, digital, and hybrid analog/digital systems.

[0004] A conventional PLL includes a phase detector and a voltage-controlled oscillator (VCO). The phase detector compares the phase of a reference frequency and a feedback frequency (e.g., the output of the VCO), and generates an output that is a function of the phase difference (e.g., a phase-error signal). The phase-error signal is used to adjust the VCO's output frequency in the direction of the reference frequency. If conditions are right, the VCO locks to the reference frequency and maintains a fixed phase relation with the reference frequency.

[0005] In the simplest PLL, the phase detector may be connected directly to the VCO to form a first order loop (i.e., a loop that has a single pole in the closed loop transfer function). First order loops provide large phase margins; however, a first order loop's bandwidth and steady-state phase-error are undesirably coupled. Therefore, most PLLs include an integrator circuit, for example, a loop filter, that is connected between the phase detector and the VCO to form, for example, a second order PLL (i.e., a loop that has two poles).

[0006] A second order PLL provides a loop that has a high loop gain at low frequencies. The loop filter typically includes a capacitor that stores a voltage that is used to control the VCO. However, the second pole provided by the additional integrator circuit of the loop filter generates a 90° negative phase shift. The negative phase shift must be offset by a corresponding positive phase shift of a zero (i.e., a frequency that causes the loop transfer function to be zero) for the loop to remain stable. To provide an acceptable phase margin (i.e., the difference between 180° and the phase shift around the loop at the unity gain frequency), a resistor, for example, may be placed in series with the capacitor of the loop filter to introduce the zero and cause the loop filter to be a low pass filter.

[0007] The low pass loop filter is able to attenuate some high frequency noise in the loop. However, any input jitter or phase noise introduced in the reference frequency (e.g., by surrounding circuits and/or the coupling of the reference frequency source to the PLL) may be amplified by the loop filter. As a result, the amplified input jitter may appear in the output of the VCO, and, if large enough, may cause significant interference from the adjacent channel in a transceiver. Typically, a sufficiently large damping ratio is chosen to reduce the effect of input jitter in the PLL.

[0008] FIG. 1 is an exemplary block diagram of a PLL with zero in feedback path.

[0009] FIG. 2 shows a relation of the phases of the exemplary waveforms in the feedback path of the PLL of FIG. 1.

[0010] FIG. 3 is a comparison of the loop noise magnitude of a conventional PLL and the PLL of FIG. 1.

[0011] FIG. 4 is an exemplary block diagram of a PLL.

[0012] FIG. 5 is an exemplary block diagram of a discriminator block for the PLL of FIG. 4.

[0013] Like reference symbols in the various drawings indicate like elements.

[0014] As shown in FIG. 1, an exemplary PLL 100 includes a forward path 101 and a feedback path 102. The forward path 101 may include a frequency-phase detector 110, a loop filter 120, and a VCO 130. The feedback path 102 may include a frequency divider 140, a discriminator block 145, and a frequency divider 147.

[0015] The frequency-phase detector 110 determines the relative phase difference between the edges of a reference frequency f_ref(e.g., an input clock) and a feedback frequency f_FB, and generates a current I_PDthat is a function of the relative phase difference. The frequency-phase detector 110 may include a phase comparator 150 and a charge pump 151. The phase comparator 150 receives two inputs (i.e., the reference frequency f_refand the feedback frequency f_FB), and generates an output pulse on one of the lines 157 and 158 based on any difference between their phases. The width of the pulse is equal to the time between the respective edges of the reference frequency f_refand the feedback frequency f_FB. The phase comparator 150 generates pulses only if there is a phase-error. The pulses stop when the phase of the reference frequency f_refand the feedback frequency f_FBare locked.

[0016] The pulses on lines 157 and 158 may be input to a charge pump 151. The charge pump 151 generates the current I_PDto correct any phase-error resulting from the difference between the phase of the reference frequency f_refand the phase of the feedback frequency f_FB. The current I_PDis input to the loop filter 120. In the frequency domain, the current input to the loop filter 120 may be expressed as I_PD(s)=K_PDθ_PD(s), where K_PDis the gain constant of the phase frequency detector and θ_PDis the difference between the phase of the reference frequency f_refand the phase of the feedback frequency f_FB(i.e., the phase-error).

[0017] The loop filter 120 outputs a VCO control voltage V_filthat controls the frequency generated by the VCO 130. The loop filter 120 may include one or more capacitors (not shown) that store the VCO control voltage V_fil. In the Frequency domain, the VCO control voltage may be expressed as V_fil(s)=I_PD(s)F(s) where F(s) is the transfer function of the loop filter 120. In one implementation, the loop filter 120 does not contain any resistors in series with the one or more capacitors, and, therefore, does not introduce a zero in the forward path transfer function. As a result, the one or more capacitors may act as pure integrators.

[0018] If a phase-error occurs in the PLL 100, then the phase comparator 150 generates a short train of pulses that cause the charge pump 151 to charge or discharge the one or more capacitors of the loop filter 120. If the phase of the feedback frequency f_FBlags behind the reference frequency f_ref, the phase comparator 150 outputs an up-pulse on the line 157. The up-pulse causes the charge pump 151 to charge the one or more capacitors of loop filter 120. If the phase of the feedback frequency f_FBleads the reference frequency f_ref, then the phase comparator 150 generates a down-pulse on the line 158. The down-pulse causes the charge pump 151 to partially discharge current from the one or more capacitors of the loop filter 120. The changes to the voltage stored by the one or more capacitors adjust the VCO control voltage V_fil.

[0019] The VCO 130 outputs a frequency f_vcothat is a function of the VCO control voltage V_filreceived from the loop filter 120. In the frequency domain, the VCO output phase θ_vcomay be expressed as ${\theta }_{\mathrm{vco}}\left(s\right)=K_{\mathrm{vco}}\frac{V_{\mathrm{fil}}\left(s\right)}{s},$

[0020] where K_vcois the gain constant of the VCO 130.

[0021] The output frequency f_VCOis input to the feedback path 102. In particular, the VCO output frequency f_VCOmay be input to the frequency divider 140. The frequency divider 140 may be used as a pre-scaler to divide down the VCO output frequency f_VCOby an integer R. The frequency divider 140 outputs the frequency f_div1, which may be input to the discriminator block 145 and the frequency divider 147. In the Frequency domain the frequency f_div1may be expressed as $f_{\mathrm{div1}}\left(s\right)=\frac{f_{\mathrm{vco}}\left(s\right)}{R}.$

[0022] The block 145 may include a frequency discriminator 172, a delay unit 173, an adder 175, and an amplifier 177. The block 145 receives the frequency f_div1and a reference frequency f_x, for example, a low frequency clock signal, as inputs. The block 145 outputs ΔN_out, which is a component of the divider control signal N for divider 147.

[0023] The clock frequency f_xis input to the frequency discriminator 172 and the delay unit 173. The frequency discriminator 172 (e.g., a frequency-to-voltage converter) compares the clock frequency f_xand the frequency f_div1to determine any difference between the frequencies. The frequency discriminator 172 generates an error voltage that is a function of the determined frequency difference. The error voltage is input to the delay circuit 173 and to the adder 175. The adder 175 determines the difference of the output of the frequency discriminator 172 and the delay circuit 173. The output of the adder 175 is input to inverter 177 to generate ΔN_out. In the frequency domain, ΔN_outmay be expressed as $\Delta N_{\mathrm{out}}\left(s\right)=\frac{K_{d}s{\theta }_{\mathrm{div1}}\left(s\right)}{f_x},$

[0024] where K_dis the gain constant of the block 145.

[0025] The output ΔN_outfrom the block 145 is input to an adder 180 along with input N_o. The output N of the adder 180 is used to control the divider 147. N may be an integer if N_ois set to a constant, or N may be a fraction if N_ois the output of, for example, a delta sigma modulator (not shown).

[0026] The frequency divider 147 receives the output of the frequency divider 140 and divides the frequency f_div1by N to generate the feedback frequency f_FB. The feedback frequency f_FBis input to the frequency-phase discriminator 110 to determine any phase-error generated by the PLL 100 and to adjust the output frequency of the PLL 100 (if necessary). In the frequency domain, the feedback phase θ_FBmay be expressed as ${\theta }_{\mathrm{FB}}\left(s\right)=\frac{{\theta }_{\mathrm{div1}}\left(s\right)}{N_o}+\frac{2\pi \Delta N_{\mathrm{out}}\left(s\right)}{N_o}.$

[0027] The effect on the feedback frequency f_FBby adjusting ΔN_outis shown in FIG. 2. FIG. 2 shows an example of the relation of the wave diagrams for the frequency f_div1and the feedback frequency f_FBwhere N_o=4. As shown in FIG. 2, if the output of the discriminator block ΔN_out=0, the divider 147 acts as an integer N divider that divides the frequency f_div1by 4. However, if ΔN_out=1, then the frequency divider 147 divides the frequency f_div1by 5 but also produces a change in phase Δθ_div2equal to $\frac{2\pi \times 1}{4}.$

[0028] The change of phase Δθ_div2from frequency f_div1to the feedback frequency f_FBintroduces a zero in the feedback path 102.

[0029] A comparison of the loop gains for a conventional PLL and the PLL 100 shows that the PLL 100 attenuates input jitter and preserves closed loop stability. For example, the loop gain G_PLLof a conventional PLL may be expresses as follows: $G_{{\mathrm{PLL}}^{2\mathrm{nd}}}=K_{\mathrm{PLL}}\frac{\left(1+s/{\omega }_z\right)}{s^2}\left(2^{\mathrm{nd}}\mathrm{order}\mathrm{loop}\right);$ $G_{{\mathrm{PLL}}^{3\mathrm{rd}}}=K_{\mathrm{PLL}}\frac{\left(1+s/{\omega }_z\right)}{s^2\left(1+s/{\omega }_{\mathrm{p1}}\right)}\left(3^{\mathrm{rd}}\mathrm{order}\mathrm{loop}\right);$ $G_{{\mathrm{PLL}}^{4\mathrm{th}}}=K_{\mathrm{PLL}}\frac{\left(1+s/{\omega }_z\right)}{s^2\left(1+s/{\omega }_{\mathrm{p1}}\right)\left(1+s/{\omega }_{\mathrm{p2}}\right)}\left(4^{\mathrm{th}}\mathrm{order}\mathrm{loop}\right),$

[0030] where K_PLLis the loop gain constant, ω_zis the frequency of a loop gain zero, and ω_pis the frequency of a loop gain pole.

[0031] The loop gain G_PLLof the PLL 100, shown in FIG. 1 may be determined as the product of the individual gains of each element in the loop. Therefore, the loop gain may be expressed as G_PLLFIG. 1=G_pdG_filG_vcoG_div1G_div2. Substituting the values for each of the loop elements described above gives: $G_{{\mathrm{PLL}}^{\mathrm{Fig}\mathrm{.1}}}=K_{\mathrm{PD}}F\left(s\right)\frac{K_{\mathrm{vco}}}{s}\frac{1}{R}\left(\frac{1}{N_o}+\frac{2\pi \Delta N_{\mathrm{out}}}{N_o}\right).$

[0032] Substituting for ΔN_outgives: $G_{{\mathrm{PLL}}^{\mathrm{Fig}\mathrm{.1}}}=\frac{K_{\mathrm{PD}}{K}_{\mathrm{vco}}}{{\mathrm{RN}}_o}\frac{F\left(s\right)}{s}\left(1+\frac{2\pi K_{d}s}{f_x}\right).$

[0033] Simplifying the equation gives: $G_{{\mathrm{PLL}}^{\mathrm{Fig}\mathrm{.1}}}=\frac{K_{\mathrm{PD}}{K}_{\mathrm{vco}}}{{\mathrm{RN}}_o}\frac{F\left(s\right)}{s}\left(1+\frac{s}{{\omega }_z}\right)$

[0034] where the frequency of the zero is expressed as ${\omega }_z=\frac{f_x}{2\pi K_d}.$

[0035] In addition, if the transfer function F(s) of the loop filter 120 is expressed as: $F\left(s\right)=\frac{1}{s\left({\tau }_1\right)}\left(2^{\mathrm{nd}}\mathrm{order}\mathrm{loop}\right);$ $F\left(s\right)=\frac{1}{s\left({\tau }_1\right)\left(1+s{\tau }_2\right)}\left(3^{rd}\mathrm{order}\mathrm{loop}\right);$ $F\left(s\right)=\frac{1}{s\left({\tau }_1\right)\left(1+s{\tau }_2\right)\left(1+s{\tau }_3\right)}\left(4^{\mathrm{th}}\mathrm{order}\mathrm{loop}\right);$

[0036] then substituting for F(s) in the loop gain G_PLLFIG. 1 equation for the PLL 100 in FIG. 1 gives: $G_{{\mathrm{PLL}}^{\mathrm{Fig}\mathrm{.1}}}=\frac{K_{\mathrm{PD}}{K}_{\mathrm{vco}}}{{\mathrm{RN}}_o{\tau }_1}\frac{\left(1+s/{\omega }_z\right)}{s^2}=K_{\mathrm{PLL}}\frac{\left(1+s/{\omega }_z\right)}{s^2}\left(2^{\mathrm{nd}}\mathrm{order}\mathrm{loop}\right);$ $G_{{\mathrm{PLL}}^{\mathrm{Fig}\mathrm{.1}}}=\frac{K_{\mathrm{PD}}{K}_{\mathrm{vco}}}{{\mathrm{RN}}_o{\tau }_1}\frac{\left(1+s/{\omega }_z\right)}{s^2\left(1+s/{\omega }_p\right)}=K_{\mathrm{PLL}}\frac{\left(1+s/{\omega }_z\right)}{s^2\left(1+s/{\omega }_p\right)}\left(3^{rd}\mathrm{order}\mathrm{loop}\right);$ $G_{{\mathrm{PLL}}^{\mathrm{Fig}\mathrm{.1}}}=\frac{K_{\mathrm{PD}}{K}_{\mathrm{vco}}}{{\mathrm{RN}}_o{\tau }_1}\frac{\left(1+s/{\omega }_z\right)}{s^2\left(1+s/{\omega }_{\mathrm{p1}}\right)\left(1+s/{\omega }_{\mathrm{p2}}\right)}=K_{\mathrm{PLL}}\frac{\left(1+s/{\omega }_z\right)}{s^2\left(1+s/{\omega }_{\mathrm{p1}}\right)\left(1+s/{\omega }_{\mathrm{p2}}\right)}\left(4^{\mathrm{th}}\mathrm{order}\mathrm{loop}\right);$

[0038] As can be seen, the PLL 100 of FIG. 1 includes the same loop zero in the loop gain function as that of a conventional PLL, and therefore provides the same loop stability. However, the gain of the forward path is the product of the gains of the elements of the forward loop and may be expressed as: G_FowardFIG. 1=G_pdG_filG_vco.

[0039] Substituting the values for the forward loop gives: $G_{{\mathrm{Forward}}^{\mathrm{Fig}\mathrm{.1}}}=K_{\mathrm{PD}}F\left(s\right)\frac{K_{\mathrm{vco}}}{s}.$

[0040] Substituting for F(s) gives: $G_{{\mathrm{Forward}}^{\mathrm{Fig}\mathrm{.1}}}=\frac{K_{\mathrm{PD}}{K}_{\mathrm{vco}}}{{\tau }_1{s}^2}\left(2^{\mathrm{nd}}\mathrm{order}\mathrm{loop}\right);$ $G_{{\mathrm{Forward}}^{\mathrm{Fig}\mathrm{.1}}}=\frac{K_{\mathrm{PD}}{K}_{\mathrm{vco}}}{{\tau }_1}\frac{1}{s^2\left(1+s/{\omega }_p\right)}\left(3^{\mathrm{rd}}\mathrm{order}\mathrm{loop}\right);\mathrm{and}$ $G_{{\mathrm{Forward}}^{\mathrm{Fig}\mathrm{.1}}}=\frac{K_{\mathrm{PD}}{K}_{\mathrm{vco}}}{{\tau }_1}\frac{1}{s^2\left(1+s/{\omega }_{\mathrm{p1}}\right)\left(1+s/{\omega }_{\mathrm{p2}}\right)}\left(4^{\mathrm{th}}\mathrm{order}\mathrm{loop}\right).$

[0041] As can be seen from the forward path gain equations, the numerator of each equation is a constant and does not contain a zero. As a result, the effect of the input jitter is attenuated as shown in FIG. 3.

[0042] FIG. 3 shows the noise magnitude |H_noise(f)| (e.g., including input jitter) of the PLL as a function of frequency. For example, a conventional third order PLL that has a zero produced by the loop filter has a noise transfer function expressed as: $H_{\mathrm{noise}}\left(f\right)=\frac{K_{\mathrm{PLL}}\left(1+s/{\omega }_z\right)}{s^3/{\omega }_{\mathrm{p1}}+s^2+K_{\mathrm{PLL}}s/{\omega }_z+K_{\mathrm{PLL}}}$

[0043] The magnitude of the noise |H_noise(f)| versus frequency for a conventional PLL is represented in FIG. 3 by curve 301. The power of the noise is represented by the area under the curve 301. As shown in FIG. 3, the magnitude for noise of a conventional PLL has a peaking response, which is caused by the zero introduced by the loop filter in the forward path.

[0044] In contrast, the function for noise H_noise(f) of a third order PLL 100 of FIG. 1 with a zero produced in the feedback path (e.g., by the block 145) may be expressed as: $H_{\mathrm{noise}}\left(f\right)=\frac{K_{\mathrm{PLL}}}{s^3/{\omega }_{\mathrm{p1}}+s^2+K_{\mathrm{PLL}}s/{\omega }_z+K_{\mathrm{PLL}}}$

[0045] The magnitude of noise |H_noise(f)| versus frequency for the PLL 100 of FIG. 1 is represented in FIG. 3 by curve 302. Comparing the noise transfer function equations, the denominators of the equations for a conventional PLL and the PLL 100 of FIG. 1 are the same. However, with the elimination of the zero from the forward path, the numerator of the noise transfer function for the PLL 100 of FIG. 1 is a constant. As a result, the peaking response of curve 301 may be reduced or eliminated, and the area under curve 302 is smaller than that under curve 301. Therefore, the power of the noise in the forward path of PLL 100 is attenuated.

[0046] The elements used to introduce the zero in the feedback path (e.g., the block 145 and the divider 147) may be implemented digitally. Therefore, the PLL 100 may be implemented using, for example, a single supply, low-voltage, digital complementary metal oxide semiconductor (CMOS) process. As a result, the PLL 100 may be manufactured using process invariant procedures. In addition, the PLL 100 incurs no significant penalty for power dissipation associated with the additional circuitry (e.g., the discriminator block) that corresponds to placing a zero in the feedback path 102. The PLL 100 also provides the loop designer with excellent control over the placement of the loop zero in the closed loop transfer function. For example, the placement of the zero may be easily set or adjusted by varying the reference frequency f_xand/or the discriminator block gain constant K_d.

[0047] FIG. 4 shows another exemplary PLL 400 that attenuates input jitter and has a zero in the feedback path. As shown in FIG. 4, the PLL 400 may include a frequency-phase detector 110, a loop filter 120, and a VCO 130 in the forward path 401. The feedback path 402 may include a divider 440, a pulse swallow unit 441, a block 445, and a divider 447.

[0048] The block 445 includes a frequency-to-voltage discriminator 451, a delta-sigma modulator 453, an inverter 455, and a dual modulus divider 457. The frequency f_div1output from the first divider 440 is input to a phase detector block 451 that compares the phase of the signal f_div1with a reference frequency, for example, an input clock f_x, to generate an error voltage that is a function of the phase-error between f_div1and f_x. The error voltage from the discriminator 451 is input to the delta-sigma modulator 453. The delta-sigma modulator 453 generates an output (i.e., a ±1) based on the error voltage. The output of the delta-sigma modulator 453 is input to an inverter 455 to generate the control input M_i.

[0049] The control input M_iis used to control the dual modulus divider 457. The dual modulus divider 457 outputs a control pulse n_ionce every M or M+1 cycles of the signal; f_div1where M is an integer based on the control input M_i. The control pulse n_iis used to control the pulse swallow unit 441.

[0050] The pulse swallow unit may be used to adjust the phase of the signal f_div1. If the control pulse n_iis high, the transition of input signal f_div1(i.e., the triggering of an edge of the frequency waveform) is delayed. As a result, the phase of the signal f_div1is adjusted before it is input to the divider 447.

[0051] As a result of the phase adjustment to f_div1, the block 445 in combination with the pulse swallow unit introduces a zero in the feedback path 402 to provide stability in the closed loop transfer function of the PLL. For example, the pulse swallow unit 441 multiplies the loop transfer function by a factor of $\left(1-\frac{1}{n_i}\right),$

[0052] where n_i(i.e., the control pulse) is the division index output from the dual modulus divider 457. Since the index n_iis derived from the block 445, its gain may be expressed as $\frac{K_d}{s}$

[0053] where K_dis the gain constant of the block 445. As a result, the loop gain transfer function is multiplied by $\left(1+\frac{s}{K_d}\right)$

[0054] to provide a loop stabilizing zero. However, the gain of the forward path of PLL 400 remains: $G_{{\mathrm{Forward}}^{\mathrm{Fig}\mathrm{.4}}}=\frac{K_{\mathrm{PD}}{K}_{\mathrm{vco}}}{{\tau }_1{s}^2}\left(2^{\mathrm{nd}}\mathrm{order}\mathrm{loop}\right);$ $G_{{\mathrm{Forward}}^{\mathrm{Fig}\mathrm{.4}}}=\frac{K_{\mathrm{PD}}{K}_{\mathrm{vco}}}{{\tau }_1}\frac{1}{s^2\left(1+s/{\omega }_p\right)}\left(3^{\mathrm{rd}}\mathrm{order}\mathrm{loop}\right);$ $G_{{\mathrm{Forward}}^{\mathrm{Fig}\mathrm{.4}}}=\frac{K_{\mathrm{PD}}{K}_{\mathrm{vco}}}{{\tau }_1}\frac{1}{s^2\left(1+s/{\omega }_{\mathrm{p1}}\right)\left(1+s/{\omega }_{\mathrm{p2}}\right)}\left(4^{\mathrm{th}}\mathrm{order}\mathrm{loop}\right).$

[0055] As a result, the numerator is a constant and the phase noise of the forward path is not amplified. In addition, the closed loop transfer function is stable due to the zero introduced to the feedback path 402.

[0056] As shown in FIG. 5, another exemplary block 445 may combine frequency to voltage conversion and delta sigma modulation. The block 445 may be used in the PLL 400 of FIG. 4 may include a phase detector 500 (e.g., a frequency-to-voltage converter) and two dual modulus dividers 551 and 457. The frequency f_div1is input to a dual modulus divider 551. The dual modulus divider 551 divides the frequency down by one of two moduli (N or N+1). The divided frequency is input to the discriminator 500, and is compared with a reference frequency f_x, for example, a clock signal. According to one implementation, the reference frequency f_refmay be used as the reference frequency f_x. The output of the discriminator 500 is used as the control input or division index of the two dual modulus dividers 457 and 551. The dual modulus divider 457 outputs a control pulse n_ionce every M or M+1 cycles of the frequency f_div1, as described above.

[0057] In addition to attenuating noise in the forward path by placing the loop-stabilizing zero in the feedback path, the circuits of FIGS. 4 and 5 may be implemented using a single supply, low-voltage, digital CMOS process. As a result, the PLL 400 may be manufactured using a process invariant procedure, and the PLL 400 incurs no significant penalty for power dissipation associated with the additional circuitry of the block 445. The PLL 400 also provides the loop designer with excellent control over the placement of the loop zero.

[0058] A number of exemplary implementations have been described. Nevertheless, it will be understood that various modifications may be made. For example, different types of phase comparators, loop filters, VCOs, and dividers may be used in the PLL. In addition, suitable results still could be achieved if the steps of the disclosed techniques were performed in a different order and/or if components in a disclosed architecture, device, or circuit were combined in a different manner and/or replaced or supplemented by other components. Accordingly, other implementations are within the scope of the following claims.