Application Note. PLL Jitter Measurements.: Purpose

PLL jitter measurements.
Application Note. PLL jitter

measurements.
The procedures described in this application note are deliberately broad and generic. Your
specific design might require procedures that are slightly different from those described here.
Purpose
This application note illustrates how to use the Spectre and SpectreRF simulators within the
Analog Design Environment (ADE) to measure jitter characteristics of the PLL circuits.
Audience
Designers of PLL circuits or their blocks who are interested to use Cadences tools for a noise
and jitter performance verification.
Overview
Phase Lock Loops are widely used in electronics and communication for such timing related
functions as clock generation, clock recovery, demodulation, and clock skew reduction. The
undesired variation of the PLL operation due to internal and external noise sources are hard
to predict and simulate. But it is essential to be able to predict and verify the timing
performance of PLL in the presence of internally and externally generated jitter and phase
noise.
We introduce the step by step methodology of jitter computation for the PLL using voltagedomain behavioral models for its blocks. The major steps of the process are the following:
i. SpectreRF simulation of the individual blocks to measure the jitter and operating
parameters of the models.
ii. Creation or modification of the existing behavioral models of the blocks with the jitter.
iii. Time domain simulation of the original PLL using behavioral models of the blocks.
June 2006
Product Version 5.1.41

Application Note. PLL jitter measurements.
iv. Post processing of the simulation results to find the jitter and noise characteristics of the
entire PLL.
We assume that the user is familiar with the general PLL functionality and his/her particular
design under investigation. This will be essential to properly setup the test benches for the
simulations. It is also assumed that there are test benches for each of the block to run at least
a simple transient simulation under typical operating conditions. User is also advised to
familiarized her/himself with the basic ideas of the Jitter measurements under SpectreRF, for
which we recommend [1]. The familiarity with the SpectreRF setup, Direct Plot and Calculator
is also preferred.
PLL jitter and its contributors

Simple Charge Pump PLL
A typical charge pump PLL is shown at Figure 1. It consists of phase/frequency detector
(PFD), charge pump (CP), loop filter (LPF), voltage controlled oscillator (VCO) and frequency
dividers (FD), also called counters. The phase and frequency of VCO is forced to follow the
input signal periodic signal using the negative feedback. The loop is locked when the
frequency of both PFD inputs is the same and the phase skew is constant or zero.
Figure 1 The block diagram of a charge pump PLL.
Vin +
Vfb -
Phase/Freq
Detector
(PFD)
Charge
Pump
(CP)
Icp
Low-Pass
Filter
(LPF)
Vcntl
Voltage Control
Oscillator
(VCO)
Output
Voltage
HLP(s)
Vosc
Frequency
divider
(FD)
June 2006

In case of the frequency multiplier, the input is a signal from a reference oscillator (not shown)
which could be passed through a frequency divider/pre-scaler (not shown) before it enters the
PLL. There also could be several outputs taken from the VCO using post-scalers (not shown).
Phase noise and timing jitter

The operation of the phase lock loop is mostly affected by the phase noise. Except for the
control voltage of the VCO, the main characteristic of the signals in the loop is a phase or a
frequency.
The main sources of noise in PLL are input noise, phase/frequency detector and VCO. The
transfer function of the loop typically has low-pass characteristic, therefore the input an phase
detector noise will dominate the output at the low frequency range. Beyond the loop
bandwidth, the VCO phase noise will pass to the output while the rest are attenuated.
There could be other significant noise contributors in the loop, but they often indicate an
unusual design or the problems with existing design.
As we discussed in [1], the phase noise or a noise during a transition when a signal is
sampled by a threshold will lead to a jitter. In communication system it will negatively impact
the transmission quality. Jitter generation by the circuit, jitter transfer by the circuit and jitter
tolerance by the locked PLL circuit are the common specification of PLL.
In the following example we will mainly concentrate on the jitter generation by the PLL itself*.
We assume the noiseless reference input signal and we measure the jitter at the output of the
circuit.
*To be augmented by a jitter transfer verification in the later revisions.
Design under investigation. 250MHz PLL circuit

The PLL was designed using 90nm PDK as a part of AMS methodology flow. The output
frequency is 250MHz, with 25MHz reference clock input. The top level schematic of the
original PLL is shown at Figure 2. We simplified an original testbench used in the demo of
AMS flow. For this Application Note we removed unneeded nose sources and complex
network of bond pads on the testbench, the same for the frequency detection instrumental
block. Inside PLL, we also reduced the loop filter by removing extra structures to simplify its
start and reset.
June 2006

Figure 2 250MHz PLL, original schematic with reduced LPF.
The input is the reference 25MHz signal, compared with the feedback from the frequency
divider, both are entering PFD. The RFD sends two sets of pulses using nodes UP and DN,
to speed up and slow down the VCO. Charge pump, receiving those pulses, generates a
current pulse, whose sign and duration will depend on the mismatch at the input of the PFD.
The current will be transformed into a smooth voltage wave at the VCONT node by the loop
filter. VCO will respond to the control and its output, which is expected to stabilize around
250MHz is passed through the frequency divider with the ratio of 10. The outputs from the
ring VCO are taken at 10 different phase offsets from the output that is used in the loop for
the feedback. The PD signal is used to reset the circuit functional blocks at the beginning of
the simulation.
Top level testbench to find locked operating conditions

Testbench
The PLL circuits described above is setup for timedomain transient simulation in Spectre.
Two DC voltage sources are used for VDD (power supply) and VSS (ground) signals. Current
example doesnt not include an actual reference oscillator with a pre-scaler counter. A voltage
June 2006

source V6 produces a pulse with the final rise and fall slopes. It represents the reference input
signal with constant frequency, 25MHZ, and amplitude, 2.5V. Another voltage source with a
single pulse at the beginning of the simulation will supply with the reset PD signal. After a
preset delay time, 1ns, it drops from initial VDD voltage to a ground value. To make it constant
afterwards, the period of the pulse is setup to be much longer than the expected simulation
time in transient analysis. All delays, slopes and periods of the waves are parametrized. The
typical value that are used in this circuit are 2.5V for a VDD power supply, 100ps. transition
from high to low and back at the reference signal (clock) input. The output of the VCO is
loaded with a capacitor at each of its 11 shifted in phase taps, see Figure 3. Only one of this
outputs is used for the feedback loop.
Figure 3 PLL testbench for the top level transient simulation.
Transient analysis
Setup a tran analysis to run for 12us, using moderate accuracy settings. Use Virtuoso ADE
Outputs menu to save the voltages on the nodes interconnecting PLL blocks and also the
currents at the terminals connected to the Vcont node. Select them directly from schematic
by descending inside the PLL circuit.
June 2006

Transient simulation results

Display results using Direct Plot. Plot control voltage to confirm the locking of the VCO,
seeFigure 4. The way to confirm that the frequency is stabilized would be to plot the output
of the VCO. Load the wave into calculator. Use cross function with 1.25V (half VDD)
threshold, plotting versus cycles, rising edge, multiple occurrences. Then on the
resulting curve, use the deriv function. The final results will be the period of the VCO output
versus cycle. It can be seen that the value is stable at 4ns.
Figure 4 VCO control voltage. TRAN analysis.
Also plot the current at the output terminal of the Charge Pump. At the beginning, before PLL
is settled, it is easy to read the value of the output current which will be used in the
behavioral model. In out case it is -55uA, see Figure 5. It could be also confirmed later, while
simulating the block alone.
June 2006

Figure 5 Charge Pump output current. TRAN analysis.
Then plot the inputs of the PFD and zoom in to see the last couple of transitions. Estimate
and record the input slopes of the signals coming into PFD. The simple visual estimate at the
rising edges gives us 100ps. The same observation for the input of the FD in the feedback
produces 150ps.
PLL blocks. Jitter parameters measurements

Split the circuit into the blocks, preserving all the sources connected to each block. In
addition, estimate the output load for each block, or use the first device from the next block
to represent the actual loading effects.
VCO
VCO testbench
The testbench, Figure 6, has the same DC sources as the original PLL schematic. In
addition, the input control voltage is supplies by a parametrized DC source V3, whose value
id fixed to vcntl. The load is the first inverting stage from the divider and possibly the PLL
output load.
June 2006

First, we find the operating condition that will provide the desirable output frequency of
250MHz. For more details, see Application Note for VCO [5].
Setup PSS analysis in ADE. Estimate the frequency, lets say 100M, better underestimate.
Use moderate accuracy settings. For tstab we use several estimated periods,
Initial conditions are used to start the oscillations for this ring oscillator. Open Simulation
menu and in convergence aids set up initial conditions to be 2.5V (high) on the odd or even
nodes between the stages of the ring oscillator. In PSS analysis Options check nodes for
initial condition parameters.
We sweep the control voltage. Use Sweep option on the PSS form. Setup Variable Name to
vcntl and set it in a reasonable range of 0.1V to 2.4V. Setup linear sweep with 10 or more
steps. Run simulation.
Figure 6 VCO testbench schematic.
Use Direct Plot to plot the tuning curve, Select Harmonic frequency and choose the first
June 2006

harmonic, see Figure 7.
Load the curve in calculator and find cross with 250MHz. The value is 1.255V which is close
to the stable value during the transient simulation of the original PLL.
From the same curve, we can estimate the additional parameters needed for the behavioral
model. They represent the linear tuning range:
Fmin = 97.65MHz; Fmax = 485MHz; Vmin = 0.79V; Vmax = 1.94V; Kvco = 420MHz/V.
Figure 7 VCO Tuning Curve.
VCO Phase Noise and Jitter.

Use PSS setup but disable the sweep option. Fix the value of vcntl parameter to 1.255V.
Also adjust the Beat Frequency estimate to 250MHz.
Select PNOISE analysis. We are interested in the noise near first harmonic at the output,
which contribute to the jitter through the phase noise component. Set relative sweep near 1
harmonic. Specify the sweep limits* from 10 to 125MHz with 10 points per decade to get a
June 2006

good smooth curve for integrating the jitter later on. Increase number of Sidebands to 100.**
Specify the voltage output at the node which is used for the feedback. Input source is not
important in this case. Enable jitter Noise Type. Run simulation.
Open Direct Plot and plot the Phase Noise from pnoise jitter analysis section of it. Then, to
see which type of noise is dominant in this circuit, one might add the -20dB/dec line, using
the same Direct Plot form. Select a frequency where the graph is seems to be straight. In this
circuit, it is pretty much over entire frequency range up to 100MHz. Type 1000 in Frequency
selector and push Plot. The curve will lay exactly on top of the most of the plot. This confirm
that there is no significant flicker noise in this VCO in the given range. The graph picks up in
the high frequency range, where the white or flicker PM noise is dominant.
Figure 8 VCO Phase Noise and Jitter.
Plot the period (i.e. cycle) jitter by using Jc button. Use rms type in seconds and select an
entire frequency range of 10 to 125MHz. Do not modify a Frequency Multiplier - it is only used
when the signal frequency is not the same as the PSS analysis fundamental, which would
happen when there is a frequency division or multiplication in the circuit. The final result is
close to the estimate that we get using the simple white jitter Equation 8 on page 33(see also
[2]):
(1) f sample = 100KHz ;f c = 250MHz ;S
June 2006
FM
(f
sample
) = 90.64 ( dBc Hz )
10

(2) Jc ( kTc ) =
sample 2
1
f
--- S ( f sample ) ------------------- = 528fs
3
2 FM
f
c
The difference is due to the assumption about an infinite bandwidth during the analytical
integration in Equation 6.
*> The limits will be affected by several factors: the bandwidth of the circuit or the system
requirements for the PLL are the obvious ones. There are several assumptions for the jitter
calculation from the phase noise, which assume that one will not include the noise too far from
the modulated carrier frequency. We will exclude the noise outside the f_osc/2 region where
it begins to change up and down around other harmonics.
** >This number might need an adjustment later on, to see if we included enough noise
frequency to cover the bandwidth of the circuit. Typically, one will increase the value until the
noise increase is significant to the desired accuracy. Then, one might reduce the number of
sidebands if the performance is important while the drop in accuracy could be tolerated. In
the current design, and increase to 200 sideband did not produce any significant change in
the results. There seems to be that even the much lesser number will work just fine.
LPF, transfer function and jitter.

Before computing the jitter for PFD/CP, we need to determine a noise transfer function of the
filter to know how it will affect the noise or a jitter generated by the charge pump. In addition,
we can confirm if the noise from the filter could be neglected in the process of modelling.
Since we eventually will run the transient analysis and not Noise or PNoise, the noise
generated by components of the LPF, if any, will have to be included explicitly into one of the
surrounding device models.
Setup PSS simulation using the simple testbench shown at Figure 9. The load is the first
stage of the VCO. Use a zero DC voltage source as a current probe. The bias and power
supply are two DC voltage sources. The input signal is a piece wise linear current source with
a 100ps rise and fall slope and a short flat regions of 20ps in the maximum (+55uA) and
minimum (-55uA). The PSS fundamental frequency is set to the reference frequency. Use
moderate accuracy settings.
Setup PXF analysis: wide frequency sweep from 10Hz to 10GHz, absolute type. The output
is the voltage across the load. We are interested in the baseband behavior, so the 0 maximum
sideband will be enough here. (If noise is not important, one can use XF analyses for such a
measurements.)
Add PNoise analysis. The same sweep as in PXF but 40 maximum sidebands, which is not
expensive here. The output is a voltage across the load. The Noise Type is set to sources.
June 2006
11

Specify the Input Source on PNoise setup form. Then select the input current source that we
use as a signal on the testbench. Set the Reference sideband to 0. *Run simulation.
Figure 9 LPF. Testbench schematic.
Use Direct Plot for PXF analysis to plot the frequency response (transfer function). Plot the
spectrum of the sideband 0. We can measure the 40 dB drop at about 680 MHz, see
Figure 10. Store this number as fmax to use for PFD/CP. Plot the output noise using PNoise
Direct Plot form.
We can get both transfer function and noise in a single PNoise analysis. From PNoise form,
select and plot the Transfer Function. Change the Y-axis to the log format. That will have the
same result as PXF spectrum for 0 sideband. In addition, to add LPF noise into the PDF/CP
model, we use Input Noise option for PNoise and integrate it up to fmax. See Figure 11.
June 2006
12

Figure 10 LPF. Output Noise and Transfer function.
Figure 11 LPF. Noise Gain and Input Referred Noise.
June 2006
13

PFD
PFD/CP testbench
The testbench, Figure 12, has two input voltage sources that represent the reference and
feedback signals coming into the PFD. Set them to be pulse type. Parametrize the periods,
rise and fall times, the delay time and pulse width. Use 100ps for transition, and reference
signal (25MHz) period for both inputs. The delay is set the same - 5ns - in one simulation. In
the second simulation, we introduce the mismatch of 2ns. in the delay. Power supply and bias
are DC sources - VDD, VCC, PD(set as dc=0), same as in the original PLL design. The load
is a voltage source that is biasing the output consistent with the operating conditions of
250MHz VCO - control voltage of 1.245V.
Setup PSS analysis. 25MHz Beat Frequency. Moderate accuracy settings, small
tstab=0.4u. Setup PNoise analysis with wide frequency range from 10Hz to 1GHz. A few
points per decade is enough, more could be added if needed. Specify output load as a
probe for the Output. Use sources type for the simulation. Run simulation.
June 2006
14

Figure 12 PFD/CP testbench schematic.
June 2006
15

Figure 13 PFD/CP Output Current, 2ns inputs mismatch.
Open Direct Plot. On PSS form, select Current. Select the current at the terminal of the load.
We confirm that the Imax is 55uA just as in the original PLL transient simulation, see
Figure 13.
PFD/CP Noise Current and input-referred Jitter.
The VCO will be sensitive to the voltage level on the control node, not the timing of its
variation. As a result we are interested in the voltage noise as the node.
Use Direct Plot, PNoise form to plot the Output Noise. On Figure 14, you see the compared
noise curves for no mismatch and for 2ns delay between PFD inputs. The units are current
squared per Hz. Integrated over the total frequency range, the current noise is translated to
the input noise in seconds, using Kdet which is defined as transfer function from relative time
delay on the input of PFD/CP to the output current of the block.
June 2006
16

Figure 14 PFD/CP Noise Current and input referred Jitter.
Open the wave in Calculator tool and implement the following operations to determine the
input referred jitter:
T 1
- --(3) J ee PFD CP = --------K det 2
f max
S ( f ) df,
f min
where the Kdet equal the charge pump current (55uA) and T is the reference signal period
(40ns). The integration is up to the 40dB drop limit that we found in the LPF simulation 680MHz.
The jitter value depends on the operating condition - the noise is larger as the phase
mismatch in inputs is increasing. For no mismatch, the estimate is 10ps while for 2ns
advanced feedback signal, it grows to 16ps. Depending on the situation, we can select the
worst case or the locked (no mismatch) number.
Frequency divider.
The jitter extraction for dividers is based on the notion that the output of the divider will
June 2006
17

typically enter the logical or digital circuits with the threshold sensitive behavior. The time of
the threshold crossing will be affected by the additive noise. The time of the crossing will be
affected by the synchronous jitter [2]. We need to find the noise at the location of the crossing
by the noiseless signal and use the slew rate to convert it into the jitter. In the current
SpectreRF implementation, it is done automatically as the jitter measurements for driven
circuits.
The testbench for the divider is shown at Figure 15. The input is the voltage source with the
typical transition times (100ps). The Period of the pulse corresponds to the VCO output of
250MHz. The load represent the input stage of the PFD.
Setup PSS analyses the fundamental frequency equal to the output frequency of FD, 25MHz,
or period of 40ns. Short tstab of 100ns is enough. Add PNoise analysis with the frequency
sweep over the wide range, up to half of the output frequency, which is 12.5MHz. Set the
sweep to be relative to the first harmonic. Output voltage is from output node FBCLK to the
ground. Activate Jitter option for PNoise and specify the threshold value (VDD/2 = 1.25V) and
the rise direction for the signals transition. Set Maximum sidebands to 400. Run simulation.
Figure 15 Frequency Divider. Testbench schematic.
Use Direct Plot. Open pnoise jitter form to plot PSD of the time jitter. Jee represents the
June 2006
18

edge to edge jitter from the internal noise sources of the divider. The complication arises from
the fact that the aliasing due to sampling of the divider output requires very high number of
the sidebands. The Figure 16 depicts the increase in the value from 400 to 800 sidebands.
The simulation time could be very long for 800 sidebands.
Figure 16 Frequency Divider. PSD of the time Jitter.
Input Jitter from the reference signal

In addition to the original example, where reference signal is an ideal voltage source, we
present additional results for the circuit with a reference oscillator present. Since no transistor
level design was available, we used some typical values for the reference jitter and other
parameters of the model.
The diagram for the modified PLL is shown at Figure 17.
June 2006
19

Figure 17 The block diagram of a PLL with reference oscillator.
R e f. O S C
V in
P h a s e /F re q
D e te c to r
(P F D )
C h a rg e
Pum p
(C P )
Ic p
L o w -P a s s
F ilte r
(L P F )
V c n tl
V o lta g e C o n tro l
O s c illa to r
(V C O )
O u tp u t
V o lta g e
V fb
H L P (s )
Vosc
F re q u e n c y
d iv id e r
(F D )
Power supply and substrate contribution to the jitter

For a simple stationary noise on the power supply or a substrate in the VCO, one can use a
modulated PXF analyses to extract the phase noise due to such noise source.
Such noise could be added to the internal noise of the VCO computed above.
The usage of PXF for jitter contribution of the power supply and substrates will be
demonstrated on the future examples.
Behavioral models
A small library of the VerilogA models with corresponding symbols was created for this PLL
circuit. The blocks are generic and could be always improved or their functionality augmented
for more accurate representation of the PLL components and to include more noise/jitter
effects.
June 2006
20

VCO
The simple VCO model is presented below. The parameters from the VCO simulation are
Fmin, Fmax, Vmin, Vmax, transition rate tt and the period jitter. The period of the VCO output
signal will be stored in the file in the Matlab format. We set the time to start the output using
a parameter outStart.
Inside the model, the period jitter is converted into the random frequency variations that lead
to the variations in the phase of the output and its transition timing.
Listing 1 VCO model

// Verilog-A model for VCO with accumulating jitter.
// The user accepts full responsibility for the use of this model.
ìnclude "disciplines.h"
ìnclude "constants.h"
module PLL_VCO(vco_clk_ph_0, Vcontrol);
input Vcontrol;
output vco_clk_ph_0;
electrical Vcontrol, vco_clk_ph_0;
parameter real Vmin=0;
parameter real Vmax=Vmin+1 from (Vmin:inf);
parameter real Fmin=1M from (0:inf);
parameter real Fmax=2*Fmin from (Fmin:inf);
parameter real Vhi=1;
parameter real Vlo=0;
parameter real tt=0.01/Fmax from (0:inf);
parameter real jitter=0 from [0:0.25/Fmax);
parameter outStart=10m from (1/Fmin:inf);
VCO to a file
parameter real ttol=1u/Fmax from (0:1/Fmax);
real freq, phase, dT, prev, curPeriod, diffPeriod, prevPeriod;
integer n, seed, fp, fjcc;
// minimum output frequency

// maximum output frequency
// high output voltage
// low output voltage
// output transition time
// period jitter (produces white accumulating jitter)
// start saving periods of
// time tolerance
analog begin
@(initial_step) begin
seed = -561;
fp = $fopen("periods.m");
fjcc = $fopen("jcc.m");
end
// compute the freq from the input voltage
freq = (V(Vcontrol) - Vmin)*(Fmax - Fmin) / (Vmax - Vmin) + Fmin;
// bound the frequency (this is optional)
if (freq > Fmax) freq = Fmax;
if (freq < Fmin) freq = Fmin;
// add the phase noise
freq = freq/(1 + dT*freq);
// phase is the integral of the freq modulo 2p
phase = 2*`M_PI*idtmod(freq, 0.0, 1.0, -0.5);
// update jitter twice per period
// `M_SQRT2=sqrt(K), K=2 jitter updates/period
@(cross(phase + `M_PI/2, +1, ttol) or cross(phase - `M_PI/2, +1, ttol)) begin
dT = `M_SQRT2*jitter*$dist_normal(seed,0, 1);
n = (phase >= -`M_PI/2) && (phase < `M_PI/2);
if(n == 1) begin
if($abstime >= outStart) begin
curPeriod = $abstime - prev;
diffPeriod = curPeriod - prevPeriod;
June 2006
21

$fstrobe(fp, "%0.10e", curPeriod);
$fstrobe(fjcc, "%0.10e", diffPeriod);
end
prev = $abstime;
prevPeriod = curPeriod;
end
end
// generate the output
V(vco_clk_ph_0) <+ transition(n ? Vhi : Vlo, 0, tt);
end
endmodule
Frequency divider.
The frequency divider model has typical timing parameters as transition times and time delay
from input to the output. The maximum, minimum and threshold voltages are the other set of
parameters. The synchronous edge to edge jitter from the above simulation of FD is used to
generate the random dither at the output.
The model is a counter of the crossing transition in one preset direction (another parameter),
generates random jitter and uses the transition function to generate the continues transitions
at the output when the counter is matching the divider ratio.
Listing 2 Frequency Divider model

// Verilog-A model of frequency divider with synchronous jitter.
module PLL_250MHZ_MDIV(clkfb, clkvco);
output clkfb; // output
input
clkvco; // input (edge triggered)
electrical clkfb, clkvco;
parameter integer ratio=2 from [2:inf); // divide ratio
parameter real Vhi=+1;
// output voltage in high state
parameter real Vlo=0;
// output voltage in low state
parameter real Vth=(Vhi+Vlo)/2; // threshold voltage at input
parameter integer dir=1 from [-1:1] exclude 0;
// dir=1 for positive edge trigger
// dir=-1 for negative edge trigger
parameter real tt=100p from (0:inf);
// transition time of output signal
parameter real td=0 from (0:inf);
// average delay from input to output
parameter real jitter=0 from [0:td/5); // edge-to-edge jitter
parameter real ttol=1p from (0:td/5); // time tolerance, recommend ttol << jitter
integer count, n, seed;
real dt;
analog begin
@(initial_step) seed = -311;
@(cross(V(clkvco) - Vth, dir, ttol)) begin
count = count + 1; // count input transitions
if (count >= ratio)
count = 0;
n = (2*count >= ratio);
dt = jitter*$dist_normal(seed,0,1); // add jitter
end
June 2006
22

V(clkfb) <+ transition(n ? Vhi : Vlo, td+dt, tt);
end
endmodule
PFD/CP.
The model below represents an example of a simple three state phase-frequency detector.
The amplitude of the output current Iout represent the Imax that was measured in the
simulation for the block. the synchronous gaussian jitter is added to the output and it changes
the shape of the transition during the rise and fall.
Listing 3 PFD/CP model.

// Verilog-A model for PFD/CP with synchronous jitter.
module PLL_PFD_CP(out, refclk, fbclk);
input refclk, fbclk;
output out;
electrical refclk, fbclk, out;
parameter real Iout=50u;
parameter real Vtrans= 0 from [-10:10]; // voltage above which the transition is
parameter integer dir=1 from [-1:1] exclude 0;
// dir=1 for positive edge trigger

// dir= 1 for negative edge trigger
parameter real tt=1n from (0:inf);

parameter real td=0 from (0:inf);
parameter real jitter=0 from [0:td/5); // edge-to-edge jitter
parameter real ttol=1p from (0:td/5); // recommend ttol << jitter
integer state, seed, fdbg;
real dt;
analog begin
//
seed = 716;
fdbg = $fopen("PFDCPdebug.txt");
end
@(cross(V(refclk) - Vtrans, dir, ttol)) begin
if (state > -1) state = state -1;
dt = jitter*$dist_normal(seed,0,1);
end
//
//
@(cross(V(fbclk) - Vtrans, dir, ttol)) begin

if (state < 1) state = state + 1;
dt = jitter*$dist_normal(seed,0,1);
end
$fstrobe(fdbg, "%d\n", state);
$fstrobe(fdbg, "%0.10e\n\n", dt);
I(out) <+ transition(Iout*state, td + dt, tt);
end
endmodule
June 2006
23

Reference oscillator.
The simple OSC model is presented below. The parameters from the VCO simulation are
Fmin, Fmax, Vmin, Vmax, transition rate tt and the period jitter. The period of the VCO output
signal will be stored in the file in the Matlab format. We set the time to start the output using
a parameter outStart.
Inside the model, the period jitter is converted into the random frequency variations that lead
to the variations in the phase of the output and its transition timing..
Listing 4 OSC model.
//
// VerilogA model for OSC with accumulating jitter.
// Original author: Ken Kundert
//
ìnclude "discipline.h"
ìnclude "constants.h"
module osc( out);
output out;
electrical out;
parameter real freq=1M from (0:inf);
parameter real Vlo=0, Vhi=1;
parameter real tt=0.01/freq from (0:inf);
parameter real jitter=0 from [0:0.1/freq); // period jitter
//
integer n, seed;
real next, dT;
analog begin
seed = 286;
next = 0.5/freq + $abstime;
end
@(timer(next)) begin
n = !n;
dT = jitter*$dist_normal(seed,0,1);
next = next + 0.5/freq + 0.707*dT;
end
V(out) <+ transition(n ? Vhi : Vlo, 0, tt);
end
endmodule //osc
June 2006
24

PLL jitter measurement using behavioral models.

Testbench.
Final jitter measurement behavioral model for PLL is presented at Figure 18. The only analog
block is the original loop filter. The input is the reference frequency voltage source, as in
original PLL.
Figure 18 PLL block diagram for the behavioral modeling.
fin
fout
Input
Signal
PFD/CP
VCO
LPF
ffbck
FD
1/N
The schematic of the PLL is shown on Figure 19. The PFD and CP are joined in a single
model which symbol view is shown on Figure 20.
June 2006
25

Figure 19 PLL schematic with VerilogA models for its blocks.
Figure 20 Combined PFD/CP VerilogA block.
The final testbench is similar to the original PLL testbench, Figure 21. The sources are the
same. The load represents only a single phase output of PLL.
June 2006
26

Figure 21 Testbench for PLL simulation using VerilogA models.
Transient jitter simulation.

Setup a long transient simulation, in our example we use 40ms. The output of the periods by
VCO model will begin at the preset time outStart, set to 10ms. We select in ADE to output a
minimum number of nodes to save time and disk space. Use moderate accuracy settings.
Run simulation.
Load the results in Direct Plot or Results Browser. We plot several parts of the control voltage
output to see the locking happening, see Figure 22. The frequency is also settled before
10ms to a reasonable value.
June 2006
27

Figure 22 Tran analysis. VCO control voltage.
Post processing.
The output file that contain the periods of the output signal is processed using MATLAB tool
by MathWorks. The original processing script is written by Ken Kundert in his work [2]. The
period jitter is computed according to the definition[1]. The power spectral density of the
output phase is computed using standard psd function, Listing 5. It uses Welchs method and
it requires to select the number of samples to use for computations. After several experiments
we use a large number of samples to get the stable result, Figure 23.
Listing 5 processing.m
echo off;
nfft=32768;
% should be power of two
winLength=nfft;
overlap=nfft/2;
winNBW=1.5;
% Noise bandwidth given in bins
% Load the data from the file generated by the VCO
load periods.m;
% output estimates of period and jitter
T=mean(periods);
J=std(periods);
maxdT = max(abs(periods-T))/T;
fprintf('T = %.3gs, F = %.3gHz\n',T, 1/T);
fprintf('Jabs = %.3gs, Jrel = %.2g%%\n', J, 100*J/T);
fprintf('max dT = %.2g%%\n', 100*maxdT);
fprintf('periods = %d, nfft = %d\n', length(periods), nfft);
% compute the cumulative phase of each transition
phases=2*pi*cumsum(periods)/T;
June 2006
28

% compute power spectral density of phase
[Sphi,f]=psd(phases,nfft,1/T,winLength,overlap,'linear');
% correct for scaling in PSD due to FFT and window
Sphi=winNBW*Sphi/nfft;
% plot the results (except at DC)
K = length(f);
semilogx(f(2:K),10*log10(Sphi(2:K)));
title('Noise Power Spectral Density at the output of the PLL');
xlabel('Frequency (Hz)');
ylabel('S phi (dB/Hz)');
rbw = winNBW/(T*nfft);
RBW=sprintf('Resolution Bandwidth = %.0f Hz (%.0f dB)', rbw, 10*log10(rbw));
line1=sprintf('Period = %.3gs, Frequency = %.3gHz\n',T, 1/T);
line2=sprintf('Jitter = %.3gs',J);
line3=sprintf('max dT = %.2g%%\n', 100*maxdT);
line4=sprintf('periods = %d, Nfft = %d\n', length(periods), nfft);
imtext(0.3,0.11, line4);
imtext(0.3,0.07, RBW);
Figure 23 Phase jitter Power Spectral Density. Matlab postprocessing.
June 2006
29

Results discussion.
The VCO noise is low and the models of the example did not include the flicker noise. Most
of the output results are dominated by the PFD/CP noise. It is shaped by the loop phase noise
transfer function. Most important noise sources which will need to be considered in the future
are the input jitter from reference oscillator, power and substrate noise sources.
The behavioral simulation is slow since the LPF has components that are causing the small
time step. Optimally, they will be replaced by the equivalent simplified models.
Improved behavioral models.

VCO and frequency divider. Reference OSC with frequency divider.
In the more complete examples, we will present the effect of the input jitter from the reference
oscillator, using an additional model for a free running oscillator as the input of the PLL,
Figure 24.
Figure 24 Complete frequency multiplier (synthesizer).
fin
Ref. OSC
fout
FD
1/M
PFD/CP
VCO
LPF
ffbck
FD
1/N
Additional runtime improvement will be done by including the frequency divider model into the
oscillators models, Figure 25. This will save time for simulation and simplify the models.
Certain restrictions apply, and additional postprocessing of results will be also needed.
June 2006
30

Figure 25 Optimized PLL, dividers included into oscillators.
fin
Ref. OSC
fout
FD
1/M
PFD/CP
VCO
FD
1/N
LPF
ffbck
PFD/CP with jitter referred to the reference oscillator.

Another speed up could be achieved by reducing the number of events due to the jitter on the
transitions. It can be done if we combine the jitter from several sources into a fewer number
of blocks.
Conclusions.
We demonstrated the flow that can be used for a time domain simulation of the PLL jitter. The
flow is as generic as the behavioral models could be. It uses latest updated jitter
measurements for autonomous and driven circuits in SpectreRF. Additional upgrades to the
flow will be introduced by using the new sampled small signal analyses. It will also benefit
from the development of the models in the future.
Future development.
External noise sources.
The external uncorrelated stationary noise sources could be currently added to the flow by
using the noisefile feature for external sources or ports. They will be properly modelled by
June 2006
31

PNoise analysis and its variations.
Correlated external noise sources.
Correlated external noise source could be added by using PXF, modulated or sampled. The
transfer functions from particular source will have to be later multiplied with the noise PSD,
summed up by correlated group and added to the PNoise results.
When automated in MMSIM 6.2 release, it will allow to use more advanced version of PNoise/
Jitter analysis setup and post processed by enhanced Direct Plot.
Appendix.
Jitter estimate.
a) PM Jitter for driven circuits.
GUI supports the ability to specify the timing event by selecting the value and a direction
of the crossing event for the output signal. PSS will produce results which will contain the
breakpoint for this event, which guarantees the more accurate slew rate and noise
computations for the time of the event (t x ). Using the relation between the simple PM jitter
(rms) in time and time-domain noise at the time of crossing:
(4) Jc =
2var ( n v, t x )
2var ( j PM ) ( tx ) = -----------------------------------d
v ( tx )
dt
where var ( n v, t x ) is the total instant noise, the result of the sampled PNoise analysis for
this particular crossing event. The jitter will be made available in the different metrics and
using different units.
User sets up a regular PSS and PNoise analysis. After Jitter measurement part of the
PNoise form is activated, the user will specify all the details for the events of the interest.
(If jitter histogram is needed, there could be some extra information needed to limit the
amount of the calculations. - Not yet implemented.)
b) Simple FM Jitter for autonomous circuits.
This calculations will be based on results of the phase noise measurements. User will be
June 2006
32

able to choose the frequency range to do the phase noise calculation. Later in the postprocessing, a user will be able to select the frequency integration range to estimate the
FM jitter. The regular PNoise analysis could be also used instead of the more expansive
modulated phase noise computations. This will be accurate if the final measurements is
done for the frequency range where the phase noise is by far greater than its AM
counterpart.
The phase noise PSD, which is used later on, is assumed to be SSB and is twice the value
of the Leeson Number which is computed by the regular PNoise or twice the PM
component computed by the modulated PNoise.
Assuming the simple FM Jitter is present, the standard deviation of the variation in the
period is:
(5) J c ( kT c ) =
akT c =
ak f c
The period jitter is derived from the noise process PSD as:
(6)
2
Jc
1
( kT c ) = ---------------2- S ( f ) sin2 ( kfT c ) df
( fc )
where f is an offset frequency from the carrier frequency. The phase noise (or one could
also use a ratio of the total voltage noise to the power in the carrier signal, with less
accuracy) is used to calculate the a via the following:
2
(7) S
2af c
( f ) = ----------2
FM
f
User selects the frequency point fsample to measure the phase noise PSD. To assist in
selection the frequency where the slope corresponds to the white FM noise, we provide
the function that plots the -20dB/dec slope. The final expression could be written as:
(8) Jc ( kTc ) =
sample 2
1
f
--- S ( f sample ) -----------------3
2 FM
f
c
June 2006
33

References
[1] Cadence Application Note, Jitter measurements using SpectreRF, 2005.
[2] Ken Kundert,Predicting the Phase Noise and Jitter of PLL-Based Frequency
Synthesizers, The Designers Guide, www.designers-guide.org, 2005
[3] W.A. Gardner, Introduction to Random Processes with Applications to Signals
and Systems, 2nd ed. New York: McGraw Hill, 1989
[4] Cadence Application Note, Oscillator Noise Analyses in SpectreRF.
[5] Cadence Application Note, VCO Design Using SpectreRF, 2004.
June 2006
34

Application Note. PLL Jitter Measurements.: Purpose

Uploaded by

Copyright:

Available Formats

Application Note. PLL Jitter Measurements.: Purpose

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Application Note. PLL Jitter Measurements.: Purpose

Uploaded by

Copyright:

Available Formats

PLL jitter measurements.