You are on page 1of 4

SIMULATION OF CRYSTAL OSCILLATORS BY USING SIMULATORS

S. E: Overhoff1, D. Hildenhagen2, K. Solbach2
1
Siemens AG, ICM MP, Kamp-Lintfort, Germany
2
Gerhardt-Mercator-Universität, Duisburg, Germany

Abstract - The design of crystal oscillators in cell phone II. COMPONENT MODELLING
applications intended for high volume production is a challenge.
In the past, the development was based on empirical formulas First of all, for every component the operating conditions
and the designer’s experience. During development a good have to specified and measured in a crystal oscillator. The
prediction of the oscillator performance is desired. This resulted
resistors and capacitors are investigated to verify their
in a trial and error process. A deficiency of this method is that a
lot of impacts could not be studied easily. Today, it is
resistance locus diagram over frequency. Different
additionally possible to use modern design tools to ensure a so- parameters were also identified to have an impact on the
called “robust design”. To enhance the accuracy precise models capacitance diode resistance locus diagram. These parameters
of various circuit parts have to be available. In this paper were the applied DC voltage, the AC voltage swing across the
models for the circuit components in use and the layout diode and the ambient temperature. For the crystal the
influences will be presented. Also, models for the used resistors following parameters were determined to have some impact.
and the varactor diode will be described. The models for the First of all the ambient temperature and secondly the crystal
resistors and the layout influence will be derived from measured drive level. For the transistor following parameters were
data. The main focus of this paper will be the behaviour model
identified: The DC operating point, the AC voltage swing and
of the varactor diode under large signal conditions: Since, in the
crystal oscillator in use, the varactor works under large signal
the ambient temperature.
conditions, the simulation has to include capacitance variation First of all, the resistors and the capacitance diode were
due to high RF voltage swing. However, in commonly available investigated and modeled.
SPICE-models only a description of the small signal behaviour is
obtained and no large signal model is available. A. Resistor Model
Keywords - Capacitance diode, Crystal oscillator, Simulation,
Modeling For mass production typically leadless components are used.
These components are called “Surface Mount Devices”. First
I. INTRODUCTION of all measurement data of different resistors, used in the
crystal oscillator, were created. Fig. 1 shows the measurement
In the past the design of crystal oscillators was mostly based
data of a resistor with a resistance of 1 MΩ over a frequency
on empirical formulas and the designer’s experience. The
range from 10 kHz up to 30 MHz. The resistance locus
development was done with a trial and error process.
diagram shows a behavior like a series resonance circuit. The
Deducing the behavior of the oscillator from previous designs
reactance of a surface-mount resistor could not be neglected,
often completed this. A lack of this method is, that a lot of
also for low frequencies, like a few MHz. Every resistor used
impacts could not be studied easily. The impact of component
in the oscillator was measured. The measurement data show
variations on the oscillators behavior could only be studied by
that for increasing resistance values the reactance part will
using components with extreme values. A lot of
also increase. This means every resistor used in the crystal
measurements have also carried out to characterize the
oscillator has to be described with a model, containing a
performance of the oscillator. Today, it is additionally
resistance and a reactance component.
possible to use modern design tools like circuit simulators.
For simulations a commercial harmonic balance simulator is
Building up a crystal oscillator in circuit simulator is a
used. A common way to include measurement data in this
challenge. Because of getting a good match between
kind of simulators is to use so-called S-Parameter-files. The
simulation results and measurement data a good modeling of
resistor can be described as a one-port-device. Therefore a
the circuit components has to be done. To simulate the initial
one-port S-Parameter-file (S1P-file) is used. This S-
frequency of a crystal oscillator with normally used models
Parameter file contains the scattering parameter of a
for the circuit components will end up in differences of
component measured over frequency. For every resistor value
several hundred ppms between simulation data and
used in the crystal oscillator the S-Parameter files were
measurement. A good match between simulation and
constructed and used for simulations.
measurement is achieved, if the difference is smaller than the
specified initial frequency tolerance of the oscillator

B. This means the AC voltage DC Voltage [V] applied to capacitance diode] swing across the capacitance diode is small compared to the applied DC voltage. if the applied DC voltage is larger C S (U ) = ∑a U i i (1) than 1 V.0 Figure 1: Resistance locus diagram of a 1 MΩ resistor measured over DC Voltage [V] applied to capacitance diode frequency.2 1. The series resistance and the capacitance were modeled against the applied DC voltage. Therefore two different and a voltage dependent resistor in a serial connection. Measurement frequency was 26 MHz. This frequency shift is in the Figure 4: Behavior model of the capacitance diode.5 1. 4 fifth order polynomials were used to model the series shows the model.5 0.6 1. For DC voltages below 1 V the diode capacitance series resistance increased up to 60 Ω. range of several hundred ppms.7 1.1 1.4 1. used for cell phone applications.7 0. 20 Available SPICE-models for capacitance diodes are only 10 valid for small-signal applications. Start frequency is 10 kHz and the stop frequency is 30 MHz. A good match between simulation data and measurement results could be achieved 0 0 1 2 3 4 5 only for small-signal applications. by using the method of least capacitance diode not only with a constant resistor value but error squares. Fig. simulation data and measurement results for large-signal conditions are presented. the I =0 resistance is not negligible and has to be modeled with a voltage dependency.0 1. Especially for low DC voltages the measurement data show a larger capacitance value than the simulation predicts. In normal SPICE-models the series resistor of a capacitance diode is described with a resistor value of some few ohms. But for applied DC voltages lower than 1 V. The voltage dependent resistor was also modeled with a fifth Therefore a new behavior model of the capacitance diode was order polynomial.6 0. used for the capacitance diode. This model contains a voltage dependent capacitor increased for DC voltages below 1 V. voltages below 1 V. Re{Z} in kOhm 0 200 400 600 800 1000 1200 0 35 -50 10 kHz Simulation 30 -100 Measurement data 30 MHz 25 Capacitance [pF] -150 Im{Z} in kOhm -200 20 -250 -300 15 -350 10 -400 Impedance 5 -450 -500 0 0. Figure 2: Simulated and measured capacitance of the diode in use.8 0. 2 shows a large discrepancy between simulation results and measurement data. As mentioned before. the resistance value created. The coefficients of both . For DC In crystal oscillator. it was obvious to model the series resistance of the determined with curve fitting. also with a voltage dependent resistor. Figure 3: Measured series resistance of the capacitance diode in use. The applied AC voltage swing during the measurement complies with AC voltage swing measured in the oscillator.9 2. With this measurement The coefficient values for this fifth order polynomial were results. A large difference between simulation data and measurement results is obvious. In Fig. To determine the initial frequency and the tuning 40 behavior of the crystal oscillator a precise modeling of the 30 capacitance characteristic has to be achieved. Fig.3 1. Modeling the capacitance diode 60 The crystal oscillator used in cell phone applications contains Measured series Series resistance [Ohm] resistance a capacitance diode to control the frequency of the crystal 50 oscillator. The resistor behaves like a series resonance circuit over frequency. Fig. 2 neglected. resistance of the capacitance diode.9 1.8 1. This is true 5 for this capacitance diode. 3 shows the series resistance of the capacitance diode Equation (1) is a fifth order polynomial used for modeling the measured over DC voltage. This results in a lower frequency of the crystal oscillator than expected. the series resistance of the diodes could not be normally large-signal conditions predominate.

Table 1 shows different applied AC voltages. The resistance and No.57 The capacitance diode was also described with a model for No. measured and also simulated with a field simulator. Fig.00 voltage dependent resistor and capacitor.000071 -1. The measurement data and the simulation modeling the series resistance of the capacitance diode. 2 26. results were compared. the capacitance will increase. 5 1. One because the depletion layer became smaller.3 V the this stray capacitance. the depletion IV. 5 RS1 (U ) = ∑b U i i U ≤ 1V (2) I =0 30 Difference between simulation and 5 RS 2 (U ) = ∑c U i i U > 1V (3) 25 measurement in [ppm] Difference I =0 20 15 C.46 large signal behavior.4 1.8 1.000118 2. Different circuit components were identified to improve the Table 1: Comparison between simulation result and measurement data.999944 25. The simulation accuracy could be achieved by describing the Initial frequency Initial frequency Difference resistors with a reactive part even for low frequencies like a measured fmeas calculated fsim fmeas and fsim [MHz] [MHz ] [ppm] few MHz. Equations (2) and (3) are used for capacitance diode. This will have impact on the accuracy of our simulation. No. The stray capacitance were increase up to – 2 ppm.2 1.23 pF up to 1 pF depending on pad geometry and distance applied DC voltages. The transistor was modeled with a different for different DC voltages. because the capacitance SPICE-model. difference increases drastically.000109 26. which pads caused -2 ppm. For DC voltages below 1. accuracy of the simulation. when the To investigate. But for applied DC voltages below 1. These capacitors were included additionally in the between simulation and measurement is in the range of simulation.000064 26. Several reasons could be identified. if the modeled components will improve the positive alternation occurs. The diode was oscillators. closely to those components. 5. 1 26. RESULTS layer of the capacitance diode will be reduced.77 25. Normally the crystal oscillator is 0 mounted either on the top layer or on the bottom layer of such -5 a circuit board.65 equation. measurement data and simulation results were done. a comparison between capacitance of the diode. the simulation and measurement results.3 V.6 0. After performing a circuit analysis Applied DC voltage in [V] different pads can be identified with a stray capacitance against the ground layer. provided by the transistor supplier.polynomials were determined in such a way that at 1 V the Afterwards the applied DC voltage was varied at the function was continuous. Modeling the stray capacitances 10 In typical cell phone applications a multi-layer printed circuit 5 board (PCB) is used.3 V the difference increases drastically. A good match between simulation and measurement could be achieved for the specified DC . Special attention was turned to the Compared parameter was the initial frequency of the crystal oscillator resistors in use and the capacitance diode.000105 1. The of the diode changes. shows the difference between simulation and measurement. Therefore in five different oscillators the crystals modeled with an AC voltage swing of 1 V. the difference capacitor.2 0. The stray capacitances differ from Figure 5: Difference between simulation and measurement for different 0. 4 26. but: the voltage were measured and every crystal was described with its own drop across different components in the circuit will be equivalent circuit diagram. First of all. Also the different stray by enlarging the models for the capacitance and resistance for capacitances are included in the simulation. The V.8 2.999987 capacitance were modeled with a fifth order polynomial No. CONCLUSION difference is in the interval of ± 2 ppm. This means the behavior model of the resistors and the capacitance diode were modeled with the capacitance diode has to be improved for lower DC voltages above described behavior models.0 0.0 grounding of the circuit.000091 1.3V the difference will between pad and ground layer.0 1. This model was described with a No. For DC voltages larger than 1. Some inner layers are used to provide the 0.000045 26.000066 26. The different stray capacitances are modeled with discrete For applied DC voltages larger than 1.6 1. A good match between simulation and measurement could be achieved. A second reason parameter was to determine the initial frequency of different can be found in the modeling of the diode.4 0. 3 26.

Therefore the behavior model of the capacitance diode has to be improved. Further investigations will focus on the validation of the capacitance diode model for different AC voltages and DC voltages. Neubig and W. Varactor SPICE models for RF VCO applications. Skyworks (former Alpha Industries) [4] ADS 2002. Feldkirchen. 1997 [3] Application note APN1004. Simulation of the temperature dependency of a crystal oscillator will be a further target. For larger applied DC voltages a good matching between simulation and measurement could be found.operating point of the capacitance diode. Agilent Technologies . ACKNOWLEDGMENT The authors would thank the Siemens AG for supporting their work. Also the transistor model has to be improved. Franzis-Verlag. Das grosse Quarzkochbuch. by adding a AC voltage dependency of the capacitance and the series resistance of the diode. Hildenhagen. 2002 [2] B. but for lower DC voltages the differences between simulation and measurement increased drastically. The capacitance diode model has also to be described with a “non-behavior model” This means components with physical representation should be used. Briese. Gerhard-Mercator-Universität Duisburg. REFERENCES [1] D. Online Manual. Simulation eines Quarzoszillators mit ADS. Therefore the temperature dependency of the capacitance diode has to be included.