1

Fractional-N Frequency Synthesizer Design Using

The PLL Design Assistant and CppSim Programs

Michael H. Perrott
http://www.cppsim.com

July 2008

Copyright 2004-2008 by Michael H. Perrott
All rights reserved.


Table of Contents

Setup........................................................................................................................................................ 2
Introduction............................................................................................................................................. 2
A. GSM Synthesizer Specifications for a Transmitter Application.................................................... 2
B. Preliminary Synthesizer Design Specifications.............................................................................. 3
Noise Analysis using the PLL Design Assistant..................................................................................... 4
A. Performing Basic Noise Analysis................................................................................................... 4
B. Adjusting the PLL Architecture to Meet the Noise Specifications................................................ 5
C. Adding in the Effect of Detector Noise.......................................................................................... 6
D. Accounting for Parameter Variations............................................................................................. 7
Dynamic Analysis Using the PLL Design Assistant............................................................................... 9
A. Checking Stability.......................................................................................................................... 9
B. Checking the Settling Time.......................................................................................................... 10
C. Examining the Influence of Zero Variations................................................................................ 11
Preliminary CppSim Simulation Analysis............................................................................................ 13
A. Entering the Design into Sue2...................................................................................................... 13
B. Setting Up the CppSim Simulation File....................................................................................... 16
C. Examining the Simulated Step Response..................................................................................... 17
D. Examining the Simulated Phase Noise......................................................................................... 18
Advanced CppSim Simulation Analysis............................................................................................... 19
A. Observing Cycle Slipping............................................................................................................ 20
B. Examining the Effects of Charge Pump Mismatch...................................................................... 21
C. Moving the Nominal Phase Error Away from Zero..................................................................... 22
D. Producing Fractional Spurs.......................................................................................................... 25
Conclusion............................................................................................................................................. 26







2
Setup

Download and install the CppSim Version 3 package (i.e., download and run the self-extracting file
named setup_cppsim3.exe) located at:

http://www.cppsim.com

Upon completion of the installation, you will see icons on the Windows desktop corresponding to the
PLL Design Assistant, CppSimView, and Sue2. Please read the CppSim (Version 3) Primer
document, which is also at the same web address, to become acquainted with CppSim and its various
components. Please read the manual PLL Design Using the PLL Design Assistant Program,
which is located at http://www.cppsim.com, to obtain more information about the PLL Design
Assistant.

Introduction

In this tutorial we will focus on the design of a Sigma-Delta frequency synthesizer intended for a
direct conversion GSM cell phone transmitter. In this section we will review the key performance
specifications, and then present the initial PLL specifications for the first-pass design. Note that the
specifications are meant to be realistic, but are not necessarily the same as would be encountered for a
practical synthesizer for GSM applications (i.e., the design procedure is meant to serve only as an
example of doing fractional-N frequency synthesizer design).
A. GSM Synthesizer Specifications for a Transmitter Application

The key specifications are to achieve a settling time less than 150 microseconds with less than 10 ppm
of frequency error, to achieve better than -80 dBc spurious performance, and to meet the phase noise
requirements imposed by the GSM standard, which are shown below in both plot and table form.
Note that the 20 dB/decade slope in the plot is shown for reference the phase noise of the VCO will
have this slope for much of its frequency range of interest. Also note that practical GSM synthesizers
often strive for comparable settling times with less than 0.1 ppm of frequency error we chose 10
ppm here since it works well for the example presented (i.e., the goal is to provide a tutorial, not a
practical design).

3
10
5
10
6
10
7
160
140
120
100
80
60
Frequency (Hz)
d
B
c
/
H
z
GSM Transmit Phase Noise Requirements for Synthesizer
Mask Requirement
20 dB/decade slope


GSM Transmit Phase Noise Specification for Synthesizer
100
kHz
200
kHz
250
kHz
400
kHz
600
kHz
1.8
MHz
3.0
MHz
6.0
MHz
10
MHz
20
MHz
-52.3
dBc/Hz
-82.8
dBc/Hz
-85.8
dBc/Hz
-112.8
dBc/Hz
-112.8
dBc/Hz
-121
dBc/Hz
-123
dBc/Hz
-129.0
dBc/Hz
-150.0
dBc/Hz
-162.0
dBc/Hz

B. Preliminary Synthesizer Design Specifications

The local PLL expert in your company has suggested that you try implementing the synthesizer as a
fractional-N topology with first-cut specifications as described below:
PLL System specifications
o Bandwidth: 100 kHz (hopefully this will be high enough to meet the settling time
requirement)
o Order: 2 (we want a simple implementation)
o Shape: Butterworth (theres no reason to think the other shapes are better at this point)
o Type: 2 with fz/fo =1/8 (these are fairly standard values for practical PLLs)
Implementation details
o Third order MASH - (avoid second order due to its problem with fractional spurs)
o Reference Frequency: 26 MHz (this is standard for GSM applications)
o Output Frequency: 900 MHz (this is set by having a direct conversion transmitter)
PLL noise parameters
4
o PFD-referred noise: the PLL expert wasnt sure what you need here
o VCO: -165 dBc/Hz at 20 MHz frequency offset (youll need this to meet the GSM
noise specification with a bit of margin)

Your job is to examine the suitability of using a fractional-N synthesizer architecture with the given
specifications, and to investigate the impact of several non-idealities at the system level. Fortunately,
you just learned of some tools to make this task easier the PLL Design Assistant and the CppSim
behavioral simulator.

Noise Analysis using the PLL Design Assistant

To check the suitability of the above architecture, we will do noise analysis in four steps. The first is
to do a basic check of noise performance with the system parameters given to us. The next is to tweak
the system to account for any problems we encounter in meeting the noise specification. We will then
investigate what levels of PFD-referred noise will be required to meet the specification. Finally, we
will look at the impact of parameter variations on the noise performance to be sure that process and
temperature variations do not push us out of spec.

A. Performing Basic Noise Analysis

Put in the above parameters into the PLL Design Assistant, and then click on the Noise Plot radio
button to see the estimated phase noise of the synthesizer. Now set the axis values of the noise plot to

Noise axis limits: 10e3 20e6 -170 -80

As a check to your work, the figures below illustrate what the PLL Design Assistant and resulting
noise plot should look like upon completion of this task.



5

B. Adjusting the PLL Architecture to Meet the Noise Specifications

Notice that the GSM phase noise specifications are not met in the above design due to the
quantization noise. We will now examine two different approaches to fix this issue.

Strategy 1: try changing the PLL order to be 3 instead of 2

o The GSM phase noise specs are met, but now the PLL loop filter is much harder to
implement

Strategy 2: try adding some parasitic poles on purpose one at 500 kHz, and one at 1 MHz

o The GSM phase noise specs are again met, and this type of loop filter is fairly easy to
implement (you just need to cascade the base loop filter with two RC filters). Note that
the choice of frequencies here (i.e., 500 kHz and 1 MHz) is fairly arbitrary you could
also make these two frequencies the same value. A chief constraint is that parasitic
pole frequencies should generally be a factor of 10 higher than the PLL bandwidth to
allow proper pole placement of the dominant poles of the PLL. Putting both poles at
500 kHz might be pushing it, but you can try.

Lets go with Strategy 2, in which case the phase noise of the synthesizer should appear as shown
below.
6

C. Adding in the Effect of Detector Noise

Now lets consider the performance requirements of PFD-referred noise caused by charge pump noise,
divider jitter, and reference jitter. We will do so by using the PLL Design Assistant to determine the
allowable detector noise for which the GSM transmit phase noise specification is still met. For
convenience, the PLL Design Assistant specifies detector noise referenced to the output of the PLL.
Once the detector noise value and PLL parameters (such as the charge pump current, etc.) are known,
it is fairly straightforward to refer this noise to the charge pump current sources and/or
reference/divider input nodes. Please see the manual, PLL Design Using the PLL Design Assistnant
Program, for more details on this issue.

Enter -85 dBc/Hz for Detector noise in the PLL Design Assistant.

o Notice that the GSM phase noise specification is not met at the 400 kHz offset.

Now enter -90 dBc/Hz for Detector noise

o Verify that the GSM phase noise specification is now met at all frequencies. The noise
plot from the PLL Design Assistant should look as shown below.

7


D. Accounting for Parameter Variations

In reality, there will be parameter variations in the implemented synthesizer due to process and
temperature variations, and it will be important to check that such variations do not prevent the phase
noise specification from being met. Given your experience with the integrated circuit process and
PLL blocks you are dealing with, you decide that the following are reasonable values

+/- 40 % variation of the PLL open loop gain, K
o Note that the open loop gain is a function of the gain of each individual PLL block,
including the VCO (K
v
), the loop filter (gain), and the charge pump (i
cp
)
+/-30 % variation of the loop filter pole, f
p
.
+/-30 % variation of the parasitic poles at 500 kHz and 1 MHz.

Note that, in an actual design, the values of variation would be determined from detailed SPICE
simulations of the actual transistor-level PLL blocks run across the different process and temperature
corners of interest. Since we need to design the circuitry for these blocks after the preliminary system
design has been completed, we will assume the above with knowledge that we will need to go back
later in the design process and update the above parameters with better approximations. We see that
the design process is iterative between system and circuit level analysis and simulations!

Enter in the above variations into the PLL Design Assistant and then hit Apply to see the
impact of these variations on the noise performance. As an example, to examine the impact of
the open loop gain having two different values, 40% below nominal value and 40% above
nominal value, you enter the expression [-0.4 0.4] into the alter: entry of the PLL Design
Assistant. Since the program must consider all combinations, it may take some time for the
results to appear. As a check to your work, the figures below show how the PLL Design
Assistant should look upon entry of these values and the resulting phase noise plot.
8
o Note that it is important to put the nominal value of parasitic poles first when they are
entered with variation values since only the first value influences the computation of
the open loop PLL parameters (i.e., K, fp, fz). In the example figure below, parasitic
pole values of 500e3 and 1e6 are assumed when computing the open loop PLL
parameters since the parasitic poles are entered as 500e3*[1.0 0.7 1.3] and 1e6*[1.0 0.7
1.3].




We see from the above results that the GSM transmit noise specification is not met at 400 kHz, so that
we must adjust our design. We have several options

Lower the PLL bandwidth to get more suppression of the Detector noise at 400 kHz
Lower the Detector noise to a suitable value

9
Due to our design requirement of having a settling time less than 150 microseconds, we will see later
that we really cant afford to lower the PLL bandwidth further. Therefore, we will assume at this
point that we can achieve lower Detector noise than -90 dBc/Hz (referred to the PLL output), and
examine what value is necessary to meet the desired GSM specifications. In practice, the designer
would need to run SPICE simulations on transistor level implementations of the charge pump and
other circuits (and perform other design tasks) to determine what level of detector noise is practical.
We see again that one must iterate between system and circuit level simulations and analysis in a
practical design process.

A few iterations of changing the Detector noise entry of the PLL Design Assistant reveals that we
need -95 dBc/Hz for the detector noise (referred to the PLL output) to achieve worse case PLL phase
noise performance of -113 dBc/Hz at 400 kHz offset. (Note that is helpful to change the axis limits to
verify this fact simply change the noise plot axis limits to 399e3 401e3 -120 -110).

Dynamic Analysis Using the PLL Design Assistant

Now that we have met the GSM noise specifications, let us examine the issue of settling time. Here
we will assume simplistically that we only need to consider the settling time in the case that the PLL
starts and remains in frequency lock. This scenario is not realistic for a synthesizer that must settle
from a power-up condition, but is reasonable for a fractional-N PLL that is already in lock (recall that
the divide value of a fractional-N synthesizer can be ramped gradually to avoid loss of frequency
lock).
A. Checking Stability

In the PLL Design Assistant, click on the Step Response radio button to see the synthesizer
time domain response to a unit step function.

o Key in axis limits of: 0 150e-6 0 1.8

Your results should match the figures below


10





The above step response waveforms reveal that the PLL is stable across all parameter variations, and
that it appears to settle quickly in the time window examined.
B. Checking the Settling Time

Although we have now verified that the synthesizer is stable across all parameter variations, we have
not verified that it settles within 10 ppm of its target frequency in less than 150 microseconds. To do
so, we simply need to change the scale parameters in the PLL Design Assistant on the y-axis to +/- 50
ppm about the final value of 1.

Alter the axis values to: 0 150e-6 1-50e-6 1+50e-6

11


You should obtain a settling time plot as shown below.


We see that the settling response meets the 10 ppm requirement within 150 microseconds.
C. Examining the Influence of Zero Variations

After giving more thought to the design, you realize that you forgot to include the impact of the
variations of the PLL zero due to process and temperature variations. It seems reasonable to assume
similar variation as for the f
p
parameter, which is +/-30 %.

Enter +/-30 % variation into the alter: form for f
z
, as shown below.
10 ppm
12

You should see a plot similar to that shown below, which unfortunately reveals that the settling time
specification is no longer met! Even though it is close, we dont want to live on the edge in an actual
production situation.

At this point, you have several options to correct this problem:
Increase the PLL bandwidth (and keep G(f) second order)
o Unfortunately, this leads to an increase in the impact of both Detector and
quantization noise on the output PLL phase noise.
Increase the PLL bandwidth, but also make G(f) third order
o Unfortunately, making G(f) third order will greatly increase the implementation
complexity of the system. Also, going to third order adds yet another parameter, Q
p
,
which will also be sensitive to process and temperature variations.
10 ppm
13
Try to create circuit designs for the loop filter implementation that have less variation than the
assumed value of +/-30 %, and/or strive for less variation in the open loop gain.
o This is a nice approach, but may not be easy!
Apply architectural innovation to solve this problem
o This is usually the best way to go. One common approach is to dynamically change
the PLL bandwidth such that it is increased during settling, and lowered when in lock
(to achieve the required noise performance). One recent example of such an approach
is found in:
M. Keaveney, P. Walsh, M. Tuthill, C. Lyden, B. Hunt, A 10 s Fast Switching PLL
Synthesizer for GSM/EDGE Base-Stations, ISSCC, Feb. 2004
Given the above options, we will assume that further investigation must be done to solve the settling
time issue. Note that practical GSM synthesizers will often have even more stringent settling time
requirements than specified in this tutorial, such as demanding <0.1 ppm of frequency error in less
than 150 microseconds. Therefore, we see that the settling time performance metric is not a trivial
one to meet.
We want to be aware of any other issues that must be kept in mind before circuit design begins, so we
now embark on further system level investigation of the system through the use of behavioral
simulation.

Preliminary CppSim Simulation Analysis

CppSim has several synthesizer examples as part of its installation which can be leveraged in our
system-level investigation of the synthesizer. In this section, we first modify one of those examples in
Sue2 to properly reflect the parameters of our system, and then create a CppSim simulation file to
produce signals for both dynamic and noise analysis. We will then run CppSim, plot the simulated
step response and phase noise of the synthesizer, and compare our results to that obtained from the
PLL Design Assistant.
A. Entering the Design into Sue2
Start up Sue2, and select the sd_synth_tristate_fast cell within the Synthesizer_Examples
library.
Select File->Save as and then save the file as project_synth within the
Synthesizer_Examples directory.
The PLL Design Assistant has provided the overall system parameters for the synthesizer, but there
are still many detailed PLL parameters to consider. You go back to the expert PLL designer, and she
provides you with first-cut values for the PLL parameters as listed below. In practice, these values are
adjusted as circuit level design of the individual components progresses, so we again catch a glimpse
of the iterative cycle between system and circuit level design that must occur for a practical design.
Reference frequency source:
14
o Center frequency =26 MHz (this value is fixed for most GSM applications)
o Kv =1 (this is arbitrary, and doesnt matter for the simulator since the phase input is
always 0 for this block)
VCO
o Center frequency =900 MHz (fixed by GSM standard)
o K
v
=50 MHz/V (this would be determined by VCO implementation)
o Noise: to meet the GSM specification with some margin, let us assume that we can
achieve -165 dBc/Hz at 20 MHz offset.
Divider this is buried within the VCO implementation in the CppSim schematic. Its nominal
value will be set by the input to the sd_modulator block, which is, in turn, fed from the step
block.
o Nominal divide value =900 MHz/26 MHz =34.6154
Therefore, we will set in_gl of the step block to 34.6154 within the test.par file
Phase-Frequency Detector Tristate design
o =1 for given topology (see PLL Design Assistant manual for explanation of the
meaning of this parameter). Well use this value in the calculation of the loop filter
gain below.
o reset_delay =2.5 ns (we would need to decrease the time step, Ts, of the simulator to
decrease this further. For now, this is OK, but would need to adjust this value once
SPICE simulations of the PFD reveals its true value).
Charge pump
o i_val =100 microamps for both the up and down current sources (this is a function of
desired noise performance you need and the amount of loop filter capacitance available
to you you need to do noise analysis of the charge pump transistors in SPICE, and
other circuit level design tasks, to get a good estimate of this. A value of 100
microamps is a reasonable assumption to begin with).
o i_variance =0 (lets ignore detector noise for now)
Sigma-Delta Modulator assume MASH structure (as currently implemented in the
schematic)
o order =3
Loop Filter consists of a lead/lag filter followed by two RC filters
o RC filters (rcfilter block in CppSimModules)
One with fo =500 kHz, and one with fo =1 MHz (as previously assumed)
An important point here: cascading two rcfilter blocks in CppSim is not
the same as cascading two RC networks CppSim does not account for
loading between blocks. CppSim is simply implementing two first
order poles that are placed at the values specified 500 MHz and 1
MHz. If you cascade two RC networks whose isolated pole frequencies
were located at these values, then the frequencies would shift when you
15
connected them due to loading effects. One can figure out how to set
the RC values of a cascaded RC filter network to match the desired pole
values by simply solving for the transfer function of the overall network
this strategy should be followed in the circuit implementation of the
loop filter.
o Lead/lag filter use values from PLL Design Assistant calculated previously
fp =217.3 kHz, fz =12.5 kHz (from the PLL Design Assistant)
gain must be calculated from PLL parameters
KvIcp
Nnom
K gain

=
K =3.272e10 (from the PLL Design Assistant)
Nnom =34.6, =1, Kv =50e6, Icp =100e-6 (from above)
Result: gain =1/(4.42e-9)
o This corresponds to 4.4 nF of capacitance in the loop filter!
Therefore, the loop filter capacitance would likely need to be
implemented off-chip in this design. Note that the above formula
allows us to see that reduction of the charge pump current, Icp,
would also allow reduction of the capacitance value. However,
reduction of Icp would, in turn, lead to an increased impact of
charge pump noise on the PLL output. The best value for Icp, and
loop filter capacitance, would need to be decided later after SPICE
simulations of the charge pump and other PLL blocks are
performed.
Implementation of the above parameters and PLL configuration into Sue2 should yield a circuit whose
primary circuit blocks are similar to those shown below.

16
B. Setting Up the CppSim Simulation File
Within the Sue2 window, select Tools CppSim Simulation. You should see the Cppsim
Run Menu that pops up. Left click on the Edit Sim File button, and an Emacs window should
pop up. Now open CppSimView and make sure it synchronizes to your new schematic. If
not, left-click the Synch button on the CppSimView window to do it.


Within the Emacs windows that pops up, enter the following values

// Number of time steps needs to be large enough to achieve noise analysis
num_sim_steps: 3e6
// Choose a time step such that its inverse is over a factor ten higher than the reference
frequency
Ts: 1/400e6
// For the transient response, look at a 150 microsecond span starting at 50 microseconds
output: test start_time=50e-6 end_time=200e-6
probe: vin sd_in
// For the noise performance, start recording after transients have died away
17
output: test_noise start_time=200e-6
probe: noiseout
// Set the nominal divide value to 34.6154, and apply a step of 0.1 at 50 microseconds
global_param: in_gl=34.6154 delta_gl=0.1 step_time_gl=50e-6

C. Examining the Simulated Step Response

Now run the CppSim simulation, and then plot signal vin to observe the transient response.
You should obtain a plot as shown below.



Compare the above step response to that produced by the PLL Design Assistant.
To do so, go back to the PLL Design Assistant, turn off the alter: statements, and remove
the variations from the parasitic poles.
The PLL Design Assistant should then look as shown below, and the corresponding step response
should match that of CppSim quite well.
18


D. Examining the Simulated Phase Noise

In CppSimView, switch to the test_noise.tr0 output file, and then run the plot function
plot_pll_phasenoise(x,10e3,20e6,noiseout)
The above function specifies that we look at the output phase noise of the PLL at offset frequencies
that span from 10 kHz to 20 MHz. You should see a plot similar to that shown below. Note that the
frequency does not span all the way down to 10 kHz due to the limited number of points at
frequencies in that range. To see this, rerun the plot function and have it span from 1 kHz to 20 MHz,
and then left-click on the Zoom button and then on (L)ineStyle you should see that there are very
few points at low frequencies, and that the lowest frequency point is quite inaccurate due to the
spectral density calculation method employed. The issue of limited low frequency points is improved
if we substantially increase the number of simulation sample points (and decimate the resulting
signals using the trigger command in the test.par file to keep a reasonable output file size), but doing
so carries the negative effect of increasing the time for the simulation.
Note that the very low frequency sample points in the simulated phase noise plot have a large amount
of variance compared to their high frequency counterparts, and are therefore unreliable in their
accuracy (i.e., dont make conclusions that you are seeing 1/f noise or other phenomenon based on the
lowest frequency portion of the simulated phase noise plots). To get better accuracy, you need to
increase the number of simulation sample points (at the expense of longer simulation times).

19


Now go back to the PLL Design Assistant, and plot the phase noise of the synthesizer with
the detector noise turned off.
The PLL Design Assistant should look as shown below, and the results between CppSim and PLL
Design Assistant should match quite well.



Advanced CppSim Simulation Analysis

20
Let us now examine some non-ideal effects that are not predicted by the PLL Design Assistant, but
can be captured by the time-domain simulation approach that CppSim offers. We will first observe the
impact of increasing the divider step value so that the PLL loses frequency lock and creates a
condition of cycle slipping. Mismatch between the up and down current sources will then be
introduced in the charge pump, which will reveal increased phase noise due to noise folding of the
quantization noise. To deal with such mismatch, it is common to purposefully introduce a phase
offset into the synthesizer through the addition of current at the charge pump output we will show
the positive and negative impact of this approach. Finally, we will change the input value to the
modulator in order to observe fractional spurs that are produced.
A. Observing Cycle Slipping
Let us look at the effect of introducing a larger step value in the divide value on the transient
performance of the synthesizer.
Click on the Edit Sim File button in CppSim Run Menu, and change the delta_gl
parameter in test.par from 0.1 to 1.
Rerun the CppSim simulation, load the test.tr0 (rather than test_noise.tr0) output file, and
then run the plotsig() function on signal vin.
The resulting step response should be similar to that shown below it no longer looks as predicted by
the PLL Design Assistant! The reason for the discrepancy is that the PLL Design Assistant assumes a
linearized PLL model that is valid only when the PLL is in lock. In this case, we have put a large
enough step to take the synthesizer out of lock, and are seeing the effects of cycle slipping that occurs
before it eventually returns to a locked state.



21
B. Examining the Effects of Charge Pump Mismatch
Now let us examine the impact of charge pump current mismatch on the phase noise of the
synthesizer.
Within Sue2, change the bottom charge current from 100 microamps to 110 microamps.
This change causes there to be a mismatch between the up and down charge pump
currents. Note that the updated Sue2 schematic should match that shown below.



Rerun the CppSim simulation, load in the test_noise.tr0 output file, and run the
plot_pll_phasenoise() function on signal noiseout. This time, increase the frequency
scale in plot_pll_phasenoise() to span from 10 kHz to 30 MHz.
The resulting phase noise plot should be similar to that shown below. Comparing this plot to the
earlier one (with matched up and down current sources) reveals significantly more noise at low
frequencies. This effect is caused by the effective nonlinearity introduced by the mismatched current,
which causes the highpass shaped Sigma-Delta quantization noise to be folded down into lower
frequencies.

22

C. Moving the Nominal Phase Error Away from Zero
Now go back to the Sue2 schematic, and add in a constant offset current into the charge
pump output of 30 microamps (as shown in the schematic figure below).
The effect of this offset current is to shift the relative phase of the reference and divider outputs, so
that we are no longer in the portion of the tristate phase detector characteristic that experiences
variable up and down pulses. In the case below, the down pulse will become fixed in width according
to the reset delay, and the up pulse be expanded such that its average width is roughly 30% of the
reference period (i.e., not considering the effect of the small-width down pulse, the up pulse must be
on 30% of the time in order to offset the 30 microamp offset and make the input to the leadlagfilter
have a zero average). This schematic change will have some positive and negative effects, as we will
now explore.

23


Run CppSim, and then use the plotsig() function to plot signal vin.
Your plot should match the one shown below, and reveals that the PLL is no longer locking properly.
The reason for this behavior is that the offset added to the charge pump output counteracts the offset
added by the frequency detection circuitry of the PFD when the PLL is out of lock. In this case, the
charge pump offset value is so large (at 30 microamps) that it overrides the frequency detector action
to the point that frequency lock is not achieved. In practice, one would not put such a large offset into
the charge pump output, but this exercise hopefully illustrates one of the impacts of doing so.


24

Now click on Edit Sim File in CppSimView, and change the delta_gl parameter in the
test.par file to be zero.
Rerun the CppSim simulation and again plot vin as above.
You should see a plot as shown below, which reveals that the PLL becomes locked within 200
microseconds. In this case, we started close enough to lock at the beginning of the simulation that the
offset current does not pose a problem. Again, in a practical design, you could not accommodate such
a large current offset at the charge pump output as you would need to be able to regain lock with a
large divide value change (especially at power-up). We are simply following through on this exercise
for pedagogical reasons.


Now re-examine the phase noise of the synthesizer by selecting the test_noise.tr0 output file
and then running the plot_pll_phasenoise() function (from 10 kHz to 30 MHz) on signal
noiseout.
The resulting plot should appear as shown below. Notice that the mismatch between the up and down
current pulses no longer causes significant folding of the Sigma-Delta quantization noise, but that a
spur now appears at the reference frequency of 26 MHz. Fortunately, there is a great deal of filtering
in the loop filter by the time you hit the reference frequency, so that is still below -100 dBc and not of
consequence.
25

D. Producing Fractional Spurs
Now click on Edit Sim File in CppSimView, and change the in_gl parameter to 34.65.
Rerun CppSim, and again plot the phase noise of the synthesizer as before.
The plot should appear as shown below, and reveals the presence of fractional spurs due to the
updated value of 34.65 going into the Sigma-Delta modulator. In practice, if possible, a proper job of
frequency planning in the system can avoid having to hit such undesirable input values to the
modulator. Otherwise, dithering methods are a possibility to prevent the creation of such spurs.
Fortunately, in this case, we see that the spurs have magnitude less than -80 dBc, and shouldnt pose a
problem since they meet the goal of achieving spur performance better than -80 dBc. However, this
may not be the case for other Sigma-Delta input values, and thorough investigation of this issue is
warranted in an actual design.
Also, please note that the spur peaks shown in the phase noise plot are sampled values of waveforms
of very narrow width the peak that you see may not be the actual peak of the waveform. Therefore,
the actual value of the spur may be higher than actually observed here. Phase noise measurement
systems perform an algorithm to estimate the spur values, which we havent bothered to do (i.e., we
just provide the right-hand scale to give you a reasonable sense of the spur level values).
26


Conclusion

In this tutorial, we have used the PLL Design Assistant and CppSim programs to design a frequency
synthesizer at the system level that is intended for a GSM transmitter application. Using the PLL
Design Assistant, we were able to examine the noise performance of the synthesizer under different
configurations and thereby select a candidate approach that met the GSM noise specifications and also
met the settling time requirements under nominal conditions. The impact of parameter variations was
then assessed, and its effects on phase noise and settling time examined. Using the CppSim
behavioral simulator, we were then able to verify the noise and dynamic analysis of the PLL Design
Assistant under nominal conditions. CppSim also allowed for examination of non-ideal behavior such
as cycle slipping, the impact of charge pump mismatch, and the presence of fractional spurs for a
given modulator input setting. Hopefully, this exercise has given you helpful insights not only
for fractional-N synthesizer design, but also for PLL design in general.