Introduction to RF Simulation and its Application

Ken Kundert
Cadence Design Systems, San Jose, California, USA

Abstract — Radio-frequency (RF) circuits exhibit several distin- ference outside the desired channel. And, like transmitters,
guishing characteristics that make them difficult to simulate they must be frugal power consumers.
using traditional SPICE transient analysis. The various exten-
sions to the harmonic balance and shooting method simulation
A. Small Desired Signals
algorithms are able to exploit these characteristics to provide
rapid and accurate simulation for these circuits. Receivers must be very sensitive to detect small input signals.
This paper is an introduction to RF simulation methods and how Typically, receivers are expected to operate with as little as 1
they are applied to make common RF measurements. It µV at the input. The sensitivity of a receiver is limited by the
describes the unique characteristics of RF circuits, the methods noise generated in the input circuitry of the receiver. Thus,
developed to simulate these circuits, and the application of these
noise is a important concern in receivers and the ability to
methods.
predict noise by simulation is very important. As shown in
Index Terms — Circuit simulation, SPICE, harmonic balance,
Figure 1, a typical superheterodyne receiver first filters and
shooting methods, quasiperiodic methods, envelope methods,
cyclostationary noise, jitter, intermodulation distortion, inter- then amplifies its input with a low noise amplifier or LNA. It
channel interference, mixers, mixer noise, nonlinear oscillators, then translates the signal to the intermediate frequency or IF
phase noise. by mixing it with the first local oscillator or LO. The noise
performance of the front-end is determined mainly by the
I. THE RF INTERFACE LNA, the mixer, and the LO. While it is possible to use tradi-
tional SPICE noise analysis to find the noise of the LNA, it is
Wireless transmitters and receivers can be conceptually sepa-
useless on the mixer and the LO because the noise in these
rated into baseband and RF sections. Baseband is the range of
blocks is strongly influenced by the large LO signal.
frequencies over which transmitters take their input and
receivers produce their output. The width of the baseband cos(ωLO2t)
determines the underlying rate at which data can flow through
the system. There is a considerable amount of signal process- RF Filter LNA IF Filter
I

ing that occurs at baseband designed to improve the fidelity LF Filter

LF Filter
of the data stream being communicated and to reduce the load
Q
the transmitter places on the transmission medium for a par- LO1

ticular data rate. The RF section of the transmitter is responsi- sin(ωLO2t)

ble for converting the processed baseband signal up to the
assigned channel and injecting the signal into the medium. Fig. 1. A coherent superheterodyne receiver’s RF interface.
Conversely, the RF section of the receiver is responsible for The small input signal level requires that receivers must be
taking the signal from the medium and converting it back capable of a tremendous amount of amplification. Often as
down to baseband. much as 120 dB of gain is needed. With such high gain, any
With transmitters there are two primary design goals. First, coupling from the output back to the input can cause prob-
they must transmit a specified amount of power while con- lems. One important reason why the superheterodyne receiver
suming as little power as possible. Second, they must not architecture is used is to spread that gain over several fre-
interfere with transceivers operating on adjacent channels. quencies to reduce the chance of coupling. It also results in
For receivers, there are three primary design goals. First, they the first LO being at a different frequency than the input,
must faithfully recover small signals. They must reject inter- which prevents this large signal from contaminating the small
input signal. For various reasons, the direct conversion or
homodyne architecture is a candidate to replace the superhet-
Published in the IEEE Journal of Solid-State Circuits, vol. 34, no. 9
erodyne architecture in some wireless communication sys-
in September 1999. Last updated on March 29, 2001.
tems [1,16,44,45]. In this architecture the RF input signal is
Ken Kundert of Cadence Design Systems, San Jose, California, can
be reached by e-mail at kundert@cadence.com and by phone at directly converted to baseband in one step. Thus, most of the
(408) 428-5634. gain will be at baseband and the LO will be at the same fre-
quency as the input signal. In this case, the ability to deter-

1298
1299 IEEE JOURNAL OF SOLID STATE CIRCUITS, VOL. 34, NO. 9, SEPTEMBER 1999

mine the impact of small amounts of coupling is quite II. CHARACTERISTICS OF RF CIRCUITS
important and will require careful modeling of the stray sig- RF circuits have several unique characteristics that are barri-
nal paths, such as coupling through the substrate, between ers to the application of traditional circuit simulation tech-
package pins and bondwires, and through the supply lines. niques. Over the last decade, researchers have developed
many special purpose algorithms that overcome these barriers
B. Large Interfering Signals
to provide practical simulation for RF circuits, often by
Receivers must be sensitive to small signals even in the pres- exploiting the very characteristic that represented the barrier
ence of large interfering signals, often known as blockers. to traditional methods [28].
This situation arises when trying to receive a weak or distant
transmitter with a strong nearby transmitter broadcasting in A. Narrowband Signals
an adjacent channel. The interfering signal can be 60-70 dB RF circuits process narrowband signals in the form of modu-
larger than the desired signal and can act to block its recep- lated carriers. Modulated carriers are characterized as having
tion by overloading the input stages of the receiver or by a periodic high-frequency carrier signal and a low-frequency
increasing the amount of noise generated in the input stage. modulation signal that acts on either the amplitude, phase, or
Both of these problems result if the input stage is driven into a frequency of the carrier. For example, a typical cellular tele-
nonlinear region by the interferer. To avoid these problems, phone transmission has a 10-30 kHz modulation bandwidth
the front-end of a receiver must be very linear. Thus, linearity riding on a 1-2 GHz carrier. In general, the modulation is
is also an important concern in receivers. Receivers are nar- arbitrary, though it is common to use a sinusoid or a simple
rowband circuits and so the nonlinearity is quantified by mea- combination of sinusoids as test signals.
suring the intermodulation distortion. This involves driving
the input with two sinusoids that are in band and close to each The ratio between the lowest frequency present in the modu-
other in frequency and then measuring the intermodulation lation and the frequency of the carrier is a measure of the rel-
products. This is generally an expensive simulation with ative frequency resolution required of the simulation. General
SPICE because many cycles must be computed in order to purpose circuit simulators, such as SPICE, use transient analy-
have the frequency resolution necessary to see the distortion sis to predict the nonlinear behavior of a circuit. Transient
products. analysis is expensive when it is necessary to resolve low
modulation frequencies in the presence of a high carrier fre-
C. Adjacent Channel Interference quency because the high-frequency carrier forces a small
timestep while a low-frequency modulation forces a long
Distortion also plays an important role in the transmitter
simulation interval.
where nonlinearity in the output stages can cause the band-
width of the transmitted signal to spread out into adjacent Passing a narrowband signal though a nonlinear circuit results
channels. This is referred to as spectral regrowth because, as in a broadband signal whose spectrum is relatively sparse, as
shown in Figures 2 and 3, the bandwidth of the signal is lim- shown in Figure 3. In general, this spectrum consists of clus-

Serial to LPFs cos(ωLOt)

Parallel

I PA f
fc
in

Q
f
0 fc 2fc 3fc
sin(ωLOt)
Fig. 3. Spectrum of a narrowband signal centered at a carrier
frequency fc before (above) and after (below) passing though a
Fig. 2. A digital direct conversion transmitter’s RF interface. nonlinear circuit. The nonlinearity causes the signal to be replicated
at multiples of the carrier, an effect referred to as harmonic
ited before it reaches the transmitter’s power amplifier or PA, distortion, and adds a skirt to the signal that increases its bandwidth,
and intermodulation distortion in the PA causes the band- an effect referred to as intermodulation distortion. It is possible to
eliminate the effect of harmonic distortion with a bandpass filter,
width to increase again. If it increases too much, the transmit- however the frequency of the intermodulation distortion products
ter will not meet its adjacent channel power requirements. overlaps the frequency of the desired signal, and so cannot be
When transmitting digitally modulated signals, spectral completely removed with filtering.
regrowth is virtually impossible to predict with S PICE. The
transmission of around 1000 digital symbols must be simu- ters of frequencies near the harmonics of the carrier. These
lated to get a representative spectrum, and this combined with clusters take the form of a discrete set of frequencies if the
the high carrier frequency makes use of transient analysis modulation is periodic or quasiperiodic, and a continuous dis-
impractical. tribution of frequencies otherwise.
KUNDERT: INTRODUCTION TO RF SIMULATION AND ITS APPLICATION 1300

RF simulators exploit the sparse nature of this spectrum in This demonstrates that a linear periodically-varying transfer
various ways and with varying degrees of success. Steady- function implements frequency translation.
state methods (Section IV-A) are used when the spectrum is Often we can assume that the information signal is small
discrete, and transient methods (Section IV-C) are used when enough to allow the use of a linear approximation of the cir-
the spectrum is continuous. cuit from its input to its output. Thus, a small-signal analysis
can be performed, as long as it accounts for the periodically-
B. Time-Varying Linear Nature of the RF Signal Path
varying nature of the signal path, which is done by linearizing
Another important but less appreciated aspect of RF circuits about the periodic operating point. This is the idea behind the
is that they are generally designed to be as linear as possible small-signal analyses of Section IV-B. Traditional simulators
from input to output to prevent distortion of the modulation or such as SPICE provide several small-signal analyses, such as
information signal. Some circuits, such as mixers, are the AC and noise analyses, that are considered essential when
designed to translate signals from one frequency to another. analyzing amplifiers and filters. However, they start by lin-
To do so, they are driven by an additional signal, the LO, a earizing a nonlinear time-invariant circuit about a constant
large periodic signal the frequency of which equals the operating point, and so generate a linear time-invariant repre-
amount of frequency translation desired. For best perfor- sentation that cannot exhibit frequency translation. Lineariz-
mance, mixers are designed to respond in a strongly nonlinear ing a nonlinear circuit about a periodically-varying operating
fashion to the LO. Thus, mixers behave both near-linearly (to point extends small-signal analysis to clocked circuits, or cir-
the input) and strongly nonlinearly (to the LO). cuits that must have a periodic clock signal present to operate
A timing or clock signal, such as the LO, is independent of properly, such as mixers, switched filters, samplers, and oscil-
the information signal, and so may be considered to be part of lators (oscillators are self-clocked, so the clock signal is the
the circuit rather than an input to the circuit as shown in Fig- desired output of the oscillator and the information signal is
ure 4. This simple change of perspective allows the mixer to generally an undesired signal, such as the noise). In doing so,
a periodically-varying linear representation results, which
does exhibit frequency translation.
Input Output
All of the traditional small-signal analyses can be extended in
this manner, enabling a wide variety of applications (some of
which are described in [59]). In particular, a noise analysis
LO that accounts for noise folding and cyclostationary noise
sources can be implemented [40,52], which fills a critically
Input Output important need for RF circuits. When applied to oscillators, it
also accounts for oscillator phase noise [8,9,21,22].

C. Linear Passive Components

LO At the high frequencies present in RF circuits, the passive
Fig. 4. One can often approximate a nonlinear periodically-driven or
components, such as transmission lines, spiral inductors,
clocked circuit (above) with a linear periodically-varying circuit packages (including bond wires) and substrates, often play a
(below). significant role in the behavior of the circuit. The nature of
such components often make them difficult to include in the
be treated as having a single input and a near-linear, though simulation.
periodically time-varying, transfer function. As an example, Generally the passive components are linear and are modeled
consider a mixer made from an ideal multiplier and followed with phasors in the frequency-domain, using either analytical
by a low-pass filter. A multiplier is nonlinear and has two expressions or tables of S-parameters. This greatly simplifies
inputs. Applying an LO signal of cos ( ω LO t ) consumes one the modeling of distributed components such as transmission
input and results in a transfer function of lines. Large distributed structures, such as packages, spirals,
v out ( t ) = LPF { cos ( ω LO t )v in ( t ) } , (1) and substrates, often interface with the rest of the circuit
through a small number of ports. Thus, they can be easily
which is clearly time-varying and is easily shown to be linear replaced by a N-port macromodel that consists of the N 2
with respect to vin. If the input signal is transfer functions. These transfer functions are found by
v in ( t ) = m ( t ) cos ( ω c t ) , (2) reducing the large systems of equations that describe these
structures using Gaussian elimination, leaving only the equa-
then
tions that relate the signals at their ports. The relatively
v out ( t ) = LPF { m ( t ) cos ( ω c t ) cos ( ω LO t ) } (3) expensive reduction step is done once for each frequency as a
and preprocessing step. The resulting model is one that is efficient
to evaluate in a frequency-domain simulator if N is small.
v out ( t ) = m ( t ) cos ( ( ω c – ω LO )t ) . (4)
This is usually true for transmission lines and spirals, and less
true for packages and substrates.
1301 IEEE JOURNAL OF SOLID STATE CIRCUITS, VOL. 34, NO. 9, SEPTEMBER 1999

Time-domain simulators are formulated to solve sets of first-

order ordinary-differential equations (ODE). However, dis- Modulation Input
tributed components, such as transmission lines, are
described with partial-differential equations (PDE) and are
problematic for time-domain simulators. Generally, the PDEs
are converted to a set of ODEs using some form of discretiza- f
fm
tion [6,35]. Such approaches suffer from bandwidth limits. A
alternative approach is to compute the impulse response for a LO Input
distributed component from a frequency domain description
and use convolution to determine the response of the compo-
nent in the circuit [20,54,56]. Evaluating lossy or dispersive
transmission line models or tables of S-parameters with this f
fLO 2fLO 3fLO
approach is generally expensive and error-prone. Packages,
substrates and spirals can be modeled with large lumped net-
Output
works, but such systems can be too large to be efficiently
incorporated into a time domain simulation, and so some
form of reduction is necessary [11,42].

D. Semiconductor Models
f
The semiconductor models used by RF simuators must accu- Fig. 5. Signals at the inputs and outputs of an up-conversion mixer.
The modulation signal is mixed up to the upper and lower sidebands
rately model the high-frequency small-signal behavior of the of the LO and its harmonics.
devices to accurately perdict the behavior of RF circuits. BJTs
have long been used in high-frequency analog circuits and shown after passing through the limiter so that the output in
their models are well suited for RF circuits. With the advent the time-domain is simply the product of the inputs, or the
of submicron technologies, RF circuits are now being realized convolution of the two inputs in the frequency domain. The
in standard CMOS processes [1,16], however existing MOS information signal, here a modulation signal, is replicated at
models are inadequate for RF applications. In particular, the the output above and below each harmonic of the LO. These
distributed resistance in the gate and substrate are not well bands of signal above and below each harmonic are referred
modeled, which affects the driving point immitances, the to as sidebands. There are two sidebands associated with each
transfer functions, and perhaps most important, the noise harmonic of the LO. The ones immediately above the har-
[19]. In addition, flicker noise is not well modeled, which monics are referred to as the upper sidebands and the ones
plays a large roll in oscillator phase noise, and is particularly below are referred to as the lower sidebands. The sideband at
important for CMOS oscillators because of the large amount DC is referred to as the baseband.
of flicker noise produced by MOS devices [32]. When the LO has a rich harmonic content, an input signal at
any sideband will be replicated to each of the sidebands at the
III. BASIC RF BUILDING BLOCKS output. Usually, only one sideband is of interest and the oth-
RF systems are constructed primarily using four basic build- ers must be eliminated. If the desired sideband is the base-
ing blocks — amplifiers, filters, mixers, and oscillators. band, then the undesired sidebands are eliminated with a
Amplifiers and filters are common analog blocks and are well lowpass filter. Otherwise the undesired sidebands are
handled by SPICE. However, mixers and oscillators are not removed with a bandpass filter. This works well for sidebands
heavily used in analog circuits and SPICE has limited ability to of harmonics different from that of the desired sideband.
analyze them. What makes these blocks unique is presented However, special techniques are then required to eliminate
next. the remaining undesired sideband [44].
Consider a down-conversion mixer (as in a receiver) and
A. Mixers assume the mixer is followed by a filter. This filter is used to
Mixers translate signals from one frequency range to another. remove all but the desired channel. The output of the mixer/
They have two inputs and one output. One input is for the filter pair is sensitive to signals in each sideband of the LO.
information signal and the other is for the clock signal, the Associated with each sideband is a transfer function from that
LO. Ideally, the signal at the output is the same as that at the sideband to the output. The shape of the transfer function is
information signal input, except shifted in frequency by an determined largely by the filter. Thus, the bandwidth of the
amount equal to the frequency of the LO. As shown in Sec- passband is that of the filter. If the filter is a bandpass, then
tion II-B, a multiplier can act as a mixer. In fact, a multiplier the passband of the transfer function will be offset from the
is a reasonable model for a mixer except that the LO is passed LO or its harmonic by the center frequency of the filter. These
through a limiter, which is usually an integral part of the passbands are referred to as the images of the filter and are
mixer, to make the output less sensitive to noise on the LO. shown in Figure 6. Generally only one image is desired, the
rest are undesired. The most troubling is usually the one that
The input and output signals of a mixer used for up-conver-
shares the same harmonic as the desired image. Image-reject
KUNDERT: INTRODUCTION TO RF SIMULATION AND ITS APPLICATION 1302

x2
Input Output t0 ∆x(0)
t6
t0
t1 t1 ∆φ6 t5
x1

LO 2nd Harmonic
Baseband Image

2nd Upper Image

2nd Lower Image

LO 3rd Harmonic
LO Fundamental

3rd Upper Image

3rd Lower Image
t4

1st Upper Image

1st Lower Image
t6
t2
t2 t3
t3 t4 t5

Fig. 7. The trajectory of an oscillator shown in state space with and

without a perturbation ∆x. By observing the time stamps (t0,..., t6)
one can see that the deviation in amplitude dissipates while the
deviation in phase does not.
Since the oscillator is stable and the duration of the distur-
bance is finite, the deviation in amplitude eventually decays
Fig. 6. Images at the input of the first mixing stage of a typical
receiver. The images are frequency bands where the output is away and the oscillator returns to its stable orbit. In effect,
sensitive to signals at the input. there is a restoring force that tends to act against amplitude
noise. This restoring force is a natural consequence of the
mixers are designed to reduce the gain associated with this
nonlinear nature of the oscillator and at least partially sup-
undesired image [44].
presses amplitude variations, as shown in Figure 8. With lin-
Sidebands and images are related, but are not the same. Side-
bands are frequency bands in the signal actually produced at
the output of a mixer, whereas images are frequency bands at
the input of a mixer that have the potential to produce a
response at the output.

B. Oscillators
Oscillators generate a reference signal at a particular fre- Fig. 8. A linear oscillator along with its response to noise (left) and a
quency. In some oscillators, referred to as VCOs for voltage nonlinear oscillator with its response to noise (right). The arrows are
controlled oscillators, the frequency of the output varies pro- phasors that represents the unperturbed oscillator output, the carriers,
and the circles represent the response to perturbations in the form of
portionally to some input signal. Compared to mixers, oscilla- noise. With a linear oscillator the noise simply adds to the carrier. In
tors seem quite simple. That is an illusion. a nonlinear oscillator, the nonlinearities act to control the amplitude
Oscillators are generally used in RF circuits to generate the of the oscillator and so to suppress variations in amplitude, thereby
radially compressing the noise ball and converting it into
LO signal for mixers. The noise performance of the mixer is predominantly a variation in phase.
strongly affected by noise on the LO signal. The LO is always
passed through a limiter, which is generally built into the ear oscillators, there is no restoring force and so the
mixer, to make the mixer less sensitive to small variations in amplitude is arbitrary (i.e., they do not have stable orbits). As
the amplitude of the LO. However, the mixer is still sensitive such, linear oscillators exhibit equal amounts of amplitude
to variations in the phase of the LO. Thus, it is important to and phase noise because the amplitude noise is not sup-
minimize the phase noise produced by the oscillator. pressed.
Nonlinear oscillators naturally produce high levels of phase Since the oscillator is autonomous, any time-shifted version
noise. To see why, consider the trajectory of an oscillator’s of the solution is also a solution. Once the phase has shifted
stable periodic orbit in state space. Furthermore, consider dis- due to a perturbation, the oscillator continues on as if never
turbing the oscillator by applying an impulse u(t) = δ(t). The disturbed except for the shift in the phase of the oscillation.
oscillator responds by following a perturbed trajectory x(t) + There is no restoring force on the phase and so phase devia-
∆x(t) as shown in Figure 7, where x(t) represents the unper- tions accumulate. This is true for both linear an nonlinear
turbed solution and ∆x(t) is the perturbation in the response. oscillators. Notice that there is only one degree of freedom —
Decompose the perturbed response into amplitude and phase the phase of the oscillator as a whole. There is no restoring
variations. force when the phase of all signals associated with the oscilla-
φ(t ) tor shift together, however there would be a restoring force if
v ( t ) = x ( t ) + ∆x ( t ) = ( 1 + α ( t ) )x  t + ------------ (5) the phase of signals shifted relative to each other. This is
 2π f 
c important in oscillators with multiple outputs, such as quadra-
where v(t) represents the noisy output voltage of the oscilla- ture oscillators or ring oscillators. The dominant phase varia-
tor, α(t) represents the variation in amplitude, φ(t) is the vari- tions appear identically in all outputs, whereas relative phase
ation in phase, and fc is the oscillation frequency. variations between the outputs are naturally suppressed by the
1303 IEEE JOURNAL OF SOLID STATE CIRCUITS, VOL. 34, NO. 9, SEPTEMBER 1999

oscillator or added by subsequent circuitry and so tend to be

much smaller [8].

log Sφ

log Sv
After being disturbed by an impulse, the asymptotic response
of the amplitude deviation is α(t) → 0 as t → ∞. However, the
asymptotic response of the phase deviation is φ(t) → ∆φ. If
responses that decay away are neglected then the impulse log fm log fm
response of the phase deviation φ(t) can be approximated with Fig. 9. Two different ways of characterizing noise in the same
a unit step s(t). Thus, the phase shift over time for an arbitrary oscillator. Sφ is the spectral density of the phase and Sv is the spectral
input disturbance u is density of the voltage. Sv contains both amplitude and phase noise
components, but with oscillators the phase noise dominates except at
∞ t frequencies far from the carrier and its harmonics. Sv is directly
φ(t ) ∼
∫ s ( t – τ )u ( τ ) dτ = ∫ u ( τ ) dτ (6) observable on a spectrum analyzer, whereas Sφ is only observable if
the signal is first passed through a phase detector. Another measure
–∞ –∞ of oscillator noise is L, which is simply Sv normalized to the power
in the fundamental.
or the power spectral density (PSD) of the phase is
inversely proportional to a. Thus, the larger a, the more phase
Su( f )
S φ ( f ) ∼ ----------------2- . (7) noise, the broader the linewidth, and the lower signal ampli-
( 2πf ) tude within the linewidth. This happens because the phase
The disturbance u may be either deterministic or random in noise does not affect the total power in the signal, it only
character and may result from extraneous signals coupling affects its distribution. Without phase noise, Sv(f) is a series of
into the oscillator or from variations in the components that impulse functions at the harmonics of fc. With phase noise,
make up the oscillator, such as thermal, shot, and flicker the impulse functions spread, becoming fatter and shorter but
noise. retaining the same total power [9].

If Su(f) is white noise, then Sφ(f) is proportional to 1/(2πf)2. It is more common to report oscillator phase noise as L, the
This result has been shown here to apply at low frequencies, ratio of the single-sideband (SSB) phase noise power to the
but with a more detailed derivation it can also be shown to be power in the fundamental (in dBc/Hz)
true over a broad range of frequencies [21]. Assume u is
v c S (f + f ) m 1 f ∆
white and define a such that L ( f m ) = -----------------------------
- = --- -------------------- . (12)
X1 2 πf + f
2 2
∆ m
f c2
S φ ( f ) = a ------2 (8) At frequencies where the oscillator phase noise dominates
f over the amplitude noise, and that are also outside the line-
where fc = 1/T is the oscillation or carrier frequency. Sφ is the width (fm > f∆), the phase noise is approximated with†
PSD of the phase variable in (5). Phase cannot easily be f
∆ af2
c
observed directly, so instead one is often interested in Sv, the L ( f m ) = ---------
- = --------- = S φ ( f m ) for f ∆ « f « f c . (13)
πf m
2 2 fm
PSD of v. Near the fundamental [9,21,23,57]
a f c2 The roll-off in Sv(fc + fm) and L(fm) as fm → 0 is a result of
S v ( f c + f m ) = X 1 2 -------------------------------- , (9) the circuit responding in a nonlinear fashion to the noise
a 2 π 2 f c4 + f m 2
itself. As such, it cannot be anticipated by the small-signal
where fm is the frequency offset from the fundamental and X1 noise analyses that will be presented in Section IV-B. How-
is the first Fourier coefficient for x, ever, as can be seen from Figure 9, Sφ does not roll-off at low
∞ frequencies, so these analyses along with (13) can be used to

∑ j2πk f c t compute Sφ at low frequencies.

x(t ) = Xke . (10)
k = –∞
Phase noise acts to vary the period of the oscillation, a phe-
nomenon known as jitter. Assume that u is a white stationary
This spectrum is a Lorentzian as shown in Figure 9. The cor- process. Then its variance is constant and from (6) the vari-
ner frequency f∆ is known as the linewidth of the oscillator ance of φ increases linearly with time. Demir [9] shows that
and is given by f∆ = aπfc2, with the variance of a single period is aT where a is defined in (8)
X1 2 f ∆ and T = 1/fc. The jitter J k is the standard deviation of the
S v ( f c + f m ) = ------------ -------------------- . (11)
π f2 + f2 length of k periods, and so
∆ m
Jk = kaT . (14)
As t → ∞ the phase of the oscillator drifts without bound, and
so Sφ(fm) → ∞ as fm → 0. However, even as the phase drifts
without bound, the excursion in the voltage is limited by the
diameter of the limit cycle of the oscillator (represented by
the periodic function x in (5)). Therefore, as fm → 0 the PSD † Other references report that L(fm) = Sφ(fm)/2, which is true when
of v flattens out and S v (f c + f m ) → |X 1 | 2 /(πf ∆ ), which is
Sφ is the single-sided PSD [50,63]. Here, Sφ is the doubled-sided
PSD.
KUNDERT: INTRODUCTION TO RF SIMULATION AND ITS APPLICATION 1304

In the case where u represents flicker noise, Su( f ) is gener- Such signals are called quasiperiodic and can be represented
ally pink or proportional to 1/f. Then Sφ( f ) would be propor- with a generalized Fourier series
tional to 1/f 3 at low frequencies [22]. In this case, there are no ∞ ∞
explicit formulas for f∆ and Jk or Sv and L at low offset fre-
quencies.
v(t ) = ∑ ∑ V kl e
j2π ( k f 1 + l f 2 )t
(15)
k = –∞ l = –∞
where Vkl are Fourier coefficients and f1 and f2 are fundamen-
IV. RF ANALYSES
tal frequencies. For simplicity, a 2-fundamental quasiperiodic
SPICE provides several different types of analyses that have waveform is shown in (15), though quasiperiodic signals can
proven themselves essential to designers of baseband circuits. have any finite number of fundamental frequencies. If there is
These same analyses are also needed by RF designers, except only one fundamental, the waveform is simply periodic. f1
they must extended to address the issues described in Section and f2 are assumed to be noncommensurate, which means that
II and the circuits of Section III. The basic SPICE analyses there exists no frequency f0 such that both f1 and f2 are exact
include DC, AC, noise, and transient. RF versions of each integer multiples of f0. If f1 and f2 are commensurate, then v(t)
have been developed in recent years based on two different is simply periodic.
foundations, harmonic balance and shooting methods. Both
The choice of the fundamental frequencies is not unique.
harmonic balance and shooting methods started off as meth-
Consider a down-conversion mixer that is driven with two
ods for computing the periodic steady-state solution of a cir-
periodic signals at fRF and fLO, with the desired output at fIF =
cuit, but have been generalized to provide all the functionality
fRF – fLO. The circuit responds with a 2-fundamental quasipe-
needed by RF designers. In their original forms they were
riodic steady-state response where the fundamental frequen-
limited to relatively small circuits. Recently, Krylov subspace
cies can be fRF and fLO, fLO and fIF, or fIF and fRF. Typically,
methods have been applied to accelerate both harmonic bal-
the drive frequencies are taken to be the fundamentals, which
ance and the shooting methods, which allows them to be
in this case are fRF and fLO, With an up-conversion mixer the
applied to much larger circuits [13,30,33,58,60,61].
fundamentals would likely be chosen to be fIF and fLO.
A. Periodic and Quasiperiodic Analysis As discussed in Section II-A, computing signals that have the
Periodic and quasiperiodic analyses can be thought of as RF form of (15) with traditional transient analysis would be very
extensions of SPICE’s DC analysis. In DC analysis one applies expensive if f1 and f2 are widely spaced so that min(f1, f2)/
constant signals to the circuit and it computes the steady-state max(f1, f2) « 1 or if they are closely spaced so that |f1 – f2|/
solution, which is the DC operating point about which subse- max(f1, f2) « 1. Large-signal steady-state analyses directly
quent small-signal analyses are performed. Sometimes, the compute the quasiperiodic solution without having to simu-
level of one of the input signals is swept over a range and the late through long time constants or long beat tones (the beat
DC analysis is used to determine the large-signal DC transfer tone is the lowest frequency present excluding DC). The
curves of the circuit. methods generally work by directly computing the Fourier
coefficients, Vkl. To make the computation tractable, it is nec-
With periodic and quasiperiodic analyses, the circuit is driven essary for all but a small number of Fourier coefficients to be
with one or more periodic waveforms and the steady-state negligible. These coefficients would be ignored. Generally,
response is computed. This solution point is used as a peri- we can assume that all but the first Ki harmonics and associ-
odic or quasiperiodic operating point for subsequent small- ated cross terms of each fundamental i are negligible. With
signal analyses. In addition, the level of one of the input sig- this assumption, K = Πi(2Ki+1) coefficients remain to be cal-
nals may be swept over a range to determine the power trans- culated, which is still a large number if the number of funda-
fer curves of the circuit. mentals is large. In practice, these methods are typically
Periodic and quasiperiodic analyses are generally used to pre- limited to a maximum of 3 or 4 fundamental frequencies.
dict the distortion of RF circuits and to compute the operating
point about which small-signal analyses are performed (pre- 1) Harmonic Balance: Harmonic balance [27,30,36,47] for-
sented later). When applied to oscillators, periodic analysis is mulates the circuit equations and their solution in the fre-
used to predict the operating frequency and power, and can quency domain. The solution is written as a Fourier series that
also be used to determine how changes in the load affect these cannot represent transient behavior, and so harmonic balance
characteristics (load pull). directly finds the steady-state solution. Consider
dq ( v ( t ) )
Quasiperiodic steady-state (QPSS) analyses compute the f ( v ( t ), t ) = i ( v ( t ) ) + --------------------- + u ( t ) = 0 . (16)
steady-state response of a circuit driven by one or more large dt
periodic signals. The steady-state or eventual response is the This equation is capable of modeling any lumped time-invari-
one that results after any transient effects have dissipated. ant nonlinear system, however it is convenient to think of it as
Such circuits respond in steady-state with signals that have a being generated from nodal analysis, and so representing a
discrete spectrum with frequency components at the drive fre- statement of Kirchhoff’s Current Law for a circuit containing
quencies, at their harmonics, and at the sum and difference nonlinear conductors, nonlinear capacitors, and current
frequencies of the drive frequencies and their harmonics. sources. In this case, v(t) ∈ RN is the vector of node voltages,
i(v(t)) ∈ RN represents the current out of the node from the
1305 IEEE JOURNAL OF SOLID STATE CIRCUITS, VOL. 34, NO. 9, SEPTEMBER 1999

conductors, q(v(t)) represents the charge out of the node from

∞ ∞
the capacitors, and u(t) represents the current out of the node
from the sources. To formulate the harmonic balance equa- x(t ) = ∑ ∑ X kl e
j2πl f 2 t j2πk f 1 t
e
(20)
tions, assume that v(t) and u(t) are T-periodic and reformulate k = –∞ l = –∞








the terms of (16) as a Fourier series. X̃ k ( t )
∞

∑ F k ( V )e j2πkf t = 0 , (17) Define x̂ ( t 1, t 2 ) , such that x ( t ) = x̂ ( t, t ) with x̂ ( t 1, t 2 )

being T1 periodic in t1 and T2 periodic in t2. In this way a two-
k = –∞
dimensional version of x is created where temporal dimen-
where f = 1/T is the fundamental frequency, and
sions are associated with the time scales of each of the funda-
F k ( V ) = j2πkf Q k ( V ) + I k ( V ) + U k . (18) mental frequencies. Then
Since e j2πkft and e j2πlft are linearly independent over a ∞ ∞
period T if k ≠ l then Fk(V) = 0 for each k individually and so x̂ ( t 1, t 2 ) = ∑ ∑ X kl e
j2πl f 2 t 2 j2πk f 1 t 1
e . (21)
(17) can be reformulated as a system of equations, one for k = –∞ l = –∞
each harmonic k. To make the problem numerically tractable,
This is a two-dimensional Fourier series, and so x̂ and X are
it is necessary to consider only the first K harmonics. The
related by a two-dimensional Fourier transform.
result is a set of K complex equations (Fk(V) = 0) and K com-
plex unknowns (Vk) that are typically solved using Newton’s Using these ideas, we can reformulate (16) in terms of t1 and
method [27]. t2
It is in general impossible to directly formulate models for ∂q ( v̂ ( t 1, t 2 ) ) ∂q ( v̂ ( t 1, t 2 ) )
------------------------------- + ------------------------------- + i ( v̂ ( t 1, t 2 ) ) + û ( t 1, t 2 ) = 0
nonlinear components in the frequency domain. To overcome ∂t 1 ∂t 2
this problem, nonlinear components are usually evaluated in
(22)
the time domain. Thus, the frequency domain voltage is con-
verted into the time domain using the inverse Fourier trans- or
form, the nonlinear component (i and q) is evaluated in the f ( v̂ ( t 1, t 2 ), t 1, t 2 ) = 0 . (23)
time domain, and the current or charge is converted back into
the frequency domain using the Fourier transform. Assuming v̂ and f of (23) take the form of (21) results in
∞ ∞
2) Autonomous Harmonic Balance: An extremely important
application of harmonic balance is determining the steady-
∑ ∑ F kl ( V )e
j2π ( k f 1 t 1 + l f 2 t 2 )
= 0, (24)
k = –∞ l = –∞
state behavior of oscillators. However, as presented, harmonic
balance is not suitable for autonomous circuits such as oscil- where
lators. The method was derived assuming the circuit was F kl ( V ) = j2π ( k f 1 + l f 2 )Q kl ( V ) + I kl ( V ) + U kl (25)
driven, which made it possible to know the operating fre-
The terms in (24) are linearly independent over all t assuming
quency in advance. Instead, it is necessary to modify har-
that f 1 and f 2 are noncommensurate (share no common
monic balance to directly compute the operating frequency by
period). So F kl (V) = 0 for each k, l. This becomes finite-
adding the oscillation frequency to the list of unknowns and
dimensional by bounding k < K and l < L. When evaluating I
adding an additional equation that constrains the phase of the
and Q the multidimensional discrete Fourier transform is
computed solution [27].
used.
3) Quasiperiodic Harmonic Balance: A t wo f u n d a m e n t a l Using a multidimensional Fourier transform is just one way
quasiperiodic signal takes the form of formulating harmonic balance for quasiperiodic problems
∞ ∞ [49,66]. It is used here because of its simple derivation and
x(t ) = ∑ ∑ X kl e
j2π ( k f 1 + l f 2 )t
, (19) because it introduces ideas that will be used later in Section
IV-C. An alternate approach that is generally preferred in
k = –∞ l = –∞
practice is the false frequency method, which is based on a
where f1 and f2 are the fundamental frequencies. Rearranging one-dimensional Fourier transform [18,27].
(19) shows this to be equivalent to constructing the waveform
as a conventional Fourier series where the frequency of each 4) Shooting Methods: Traditional S PICE transient analysis
term is an integer multiple of f1, except that the Fourier coef- solves initial-value problems. A shooting method is an itera-
ficients themselves are time-varying. In particular, the coeffi- tive procedure layered on top of transient analysis that is
cient X̃ k ( t ) is periodic with period T2 = 1/f2 and can itself be designed to solve boundary-value problems. Boundary-value
represented as a Fourier series. problems play an important role in RF simulation. For exam-
ple, assume that (16) is driven with a non-constant T-periodic
stimulus. The T-periodic steady state solution is the one that
also satisfies the two-point boundary constraint,
v(T ) – v(0) = 0 . (26)
KUNDERT: INTRODUCTION TO RF SIMULATION AND ITS APPLICATION 1306

Define the state transition function φT(v0, t0) as the solution to down-conversion mixer. Designate f1 as the carrier frequency
(16) at t0 + T given that it starts at the initial state v0 at t0. In and f2 as the modulation frequency. Then v(t) is the quasiperi-
general, one writes odic response, where
v ( t 0 + T ) = φ T ( v ( t 0 ), t 0 ) . (27) v(t ) = ∑k ∑l V kl e j2π ( k f 1 + l f 2 )t
. (29)
Shooting methods combine (26) and (27) into
φ T ( v ( 0 ), 0 ) = v(0) , (28) Consider sampling the signal v(t) at the carrier frequency. The
sampled signal is referred to as the sample envelope and is
which is a nonlinear algebraic problem and so Newton meth- related to the continuous signal by v̂ n = v(nT 1) , where T1 =
ods can be used to solve for v(0). The combination of the 1/f1. v̂ n represents a sampled and perhaps scaled version of
Newton and shooting methods are referred to as the shooting- the modulation signal.
Newton algorithm. The MFT method works by computing the discrete sequence
When applying Newton's method to (28), it is necessary to v̂ instead of the continuous waveform v. Notice that if every
compute both the response of the circuit over one period and v̂ n is related to the subsequent sample point v̂ n + 1 by
the sensitivity of the final state v(T) with respect to changes in v̂ n + 1 = φ T ( v̂ n, nT 1 ) , (30)
the initial state v(0). The sensitivity is used to determine how 1

to correct the initial state to reduce the difference between the then all the v̂ n will satisfy the circuit equations. The transi-
initial and final state [2,58]. tion function in (30) can be computed by standard SPICE tran-
sient analysis and serves to translate between the continuous
5) Autonomous Shooting Methods: As with harmonic bal- signal and the discrete representation. The key to the MFT
ance, it is extremely important to be able to determine the method is to require that the samples v̂ n represent a sampled
steady-state behavior of oscillators. To do so it is necessary to quasiperiodic signal. This requirement is easily enforced
modify shooting methods to directly compute the period of because, as shown in Figure 10, sampling a 2-fundamental
the oscillator. To do so, the period is added as an extra
unknown and an additional equation is added that constrains Sample Envelope
the phase of the computed solution [27].

6) Quasiperiodic Shooting Methods: As shown in (20), a 2-

fundamental quasiperiodic signal can be interpreted as a peri-
odically modulated periodic signal. Designate the high fre-
quency signal as the carrier and the low frequency signal as Fig. 10. The sample envelope is the waveform that results from
the modulation. If the carrier is much higher in frequency sampling a signal with a period equal to that of the carrier.
than the modulation, then the carrier will appear to vary only
slightly from cycle to cycle. In this case, the complete wave- quasiperiodic signal at the carrier frequency results in a sam-
form can be inferred from knowledge of a small number of pled waveform that is 1-fundamental quasiperiodic, or simply
cycles of the carrier appropriately distributed over one period periodic, at the modulation frequency. In other words, the
of the modulation. The number of cycles needed can be deter- sampled waveform can be written as a Fourier series with the
mined from the bandwidth of the modulation signal. If the carrier removed,
modulation signal can be represented using K harmonics, then ∞
the entire quasiperiodic signal can be recovered by knowing
the waveform over 2K + 1 cycles of the carrier that are evenly
v̂ n = v(nT 1) = ∑
l = –∞
V̂ k e
j2πln f T 2 1 . (31)

distributed over the period of the modulation. This is the basic

idea behind the Mixed Frequency-Time or MFT method Alternatively, one can write
[13,26,27]. v̂ = F – 1 V̂ , (32)
Consider a circuit driven by two periodic signals that which states that v̂ is the inverse Fourier transform of V̂ .
responds in steady-state by producing 2-fundamental quasip- Consider the nth sample interval and let x n = v̂ n be the solu-
eriodic waveforms as in (15). Designate the fundamental fre- tion at the start of the interval and y n = v̂ n + 1 = x n + 1 be
quencies as f1 and f2 and consider the case where f1 » f2.‡ This the solution at the end. Then, (30) uses the circuit equations to
may be because one input is a high frequency signal and the relate the solution at both ends of the interval,
other is a low frequency signal, as would be the case with an
y n = φ T ( x n, nT 1 ) . (33)
up-conversion mixer. Or it may be that both inputs are high 1
frequency signals, but that their frequencies are close to each
Define Φ as the function that maps the sequence x to the
other and so they generate a low beat frequency, as with a
sequence y by repeated application of (33),
y = Φ(x). (34)
‡ The MFT method does not require that f1 » f2. However, if true, Let X = Fx and Y = Fy (X and Y are the Fourier transforms of
using MFT gives significant performance advantages over traditional x and y). Then, from (31) and since y n = x n + 1 ,
transient analysis. – j 2πl f 2 T 1
Xl = e Yl , (35)
1307 IEEE JOURNAL OF SOLID STATE CIRCUITS, VOL. 34, NO. 9, SEPTEMBER 1999

or detectors. These circuits, which are referred to as a group as

X = DT Y , (36) clocked circuits, require the traditional small-signal analyses
1
to be extended such that the circuit is linearized about a peri-
where D T is referred to as the delay matrix. It is a diagonal odically-varying operating point. Such analyses are referred
1 – j 2πl f 2 T 1
matrix with e being the lth diagonal element. Equa- to as linear periodically-varying or LPV analyses.
tion (36) is written in the time domain as
A great deal of useful information can be acquired by per-
x = F –1 D
T 1F y , (37) forming a small-signal analysis about the time-varying oper-
ating point of the circuit. LPV analyses start by performing a
Together, (34) and (37) make up the MFT method where (34)
periodic analysis to compute the periodic operating point with
stems solely from the circuit equations and (37) solely from
only the large clock signal applied (the LO, the clock, the car-
the requirement that v̂ represent a sampled quasiperiodic
waveform. They can be combined into rier, etc.). The circuit is then linearized about this time-vary-
ing operating point (expand about the periodic equilibrium
x = F – 1 D T 1 F Φ(x) , (38) point with a Taylor series and discard all but the first-order
term) and the small information signal is applied. The
or
response is calculated using linear time-varying analysis.
v̂ = F – 1 D T 1 F Φ(v̂) . (39)
Consider a circuit whose input is the sum of two periodic sig-
Equation (39) is a implicit nonlinear equation that can be nals, u(t) = uL(t) + us(t), where uL(t) is an arbitrary periodic
solved for v̂ using Newton’s method. waveform with period TL and us(t) is a sinusoidal waveform
In practice, the modulation signals in the circuit are band-lim- of frequency fs whose amplitude is small. In this case, uL(t)
ited, and so only a finite number of harmonics are needed. represents the large clock signal and us(t) represents the small
Thus the envelope shown in Figure 10 can be completely information signal.
specified by only a few of the sample points v̂ n . With only K Let vL(t) be the steady-state solution waveform when us(t) is
harmonics needed, (39) is solved over 2K+1 distinct intervals zero. Then allow us(t) to be nonzero but small. We can con-
using shooting methods. In particular, if the circuit is driven sider the new solution v(t) to be a perturbation vs(t) on vL(t),
with one large high frequency periodic signal at f1 and one as in v(t) = vL(t) + vs(t). The small-signal solution vs(t) is
moderately sized sinusoid at f2, then the number of harmonics computed by linearizing the circuit about vL(t), applying us(t),
needed, K, is small and the method is efficient. The total sim- and then finding the steady-state solution. Given that
ulation time is proportional to the number of harmonics j2πf s t
needed to represent the sampled modulation waveform and is u s ( t ) = Us e (40)
independent of the period of the low-frequency beat tone or
the perturbation in steady-state response is given by
the harmonics needed to represent the carrier.
∞
B. Small-Signal Analyses vs ( t ) = ∑
k = –∞
Vsk e
j2π ( f s + k f L )t
(41)
The AC and noise analyses in SPICE are referred to as small-
signal analyses. They assume that a small signal is applied to where fL = 1/TL is the large signal fundamental frequency
a circuit that is otherwise at its DC operating point. Since the [39,61]. Vsk represents the sideband for the kth harmonic of
input signal is small, the response can be computed by linear- VL. In this situation, shown in Figure 11, there is only one
izing the circuit about its DC operating point (apply a Taylor sideband per harmonic because Us is a single frequency com-
series expansion about the DC equilibrium point and discard plex exponential and the circuit has been linearized. This rep-
all but the first-order term). Superposition holds, so the resentation has terms at negative frequencies. If these terms
response at each frequency can be computed independently. are mapped to positive frequencies, then the sidebands with k
Such analyses are useful for computing the characteristics of < 0 become lower sidebands of the harmonics of vL and those
circuits that are expected to respond in a near-linear fashion to with k > 0 become upper sidebands.
an input signal and that operate about a DC operating point.
Vsk/Us is the transfer function for the input at fs to the output
This describes most “linear” amplifiers and continuous-time
at fs + kfL. Notice that with periodically-varying linear sys-
filters.
tems there are an infinite number of transfer functions
The assumption that the circuit operates about a DC operating between any particular input and output. Each represents a
point makes these analyses unsuitable for circuits that are different frequency translation.
expected to respond in a near-linear fashion to an input signal
Versions of this type of small-signal analysis exists for both
but that require some type of clock signal to operate. Mixers
harmonic balance [17,24,31] and shooting methods
fit this description, and if one considers noise to be the input,
[39,40,61].
oscillators also fit. However, there is a wide variety of other
circuits for which these assumptions also apply. Circuits such There are two different ways of formulating a small-signal
as samplers and sample-and-holds, switched-capacitor and analysis that computes transfer functions [59,61]. The first is
switched-current filters, chopper-stabilized and parametric akin to traditional AC analysis, and is referred to here as a
amplifiers, frequency multipliers and dividers, and phase “periodic AC” or PAC analysis. In this case, a small-signal is
applied to a particular point in the circuit at a particular fre-
KUNDERT: INTRODUCTION TO RF SIMULATION AND ITS APPLICATION 1308

All of the small-signal analyses are extensible to the case

Input Output where the operating point is quasiperiodic. This is important
when predicting the effect of large interferers or blockers and
LO
is discussed further in Section VI-C. Such analyses are
referred to as linear quasiperiodically-varying or LQPV anal-
yses as a group, or individually as QPAC, QPXF, QPNoise,
Input Signals
etc.

C. Transient-Envelope Analyses
Transient-envelope analyses are applied to simulate modu-
Output lated carrier systems when the modulation waveform is some-
thing other than a simple sinusoid or combination of
sinusoids. It does so by performing a series of linked large-
signal pseudo-periodic analyses, which are periodic analyses
that have been modified to account for slow variations in the
Output (mapped to envelope over the course of each period of the carrier as a
positive frequencies) result of the modulation. The pseudo-periodic analyses must
be performed often enough to follow the changes in the enve-
lope. In effect, transient-envelope methods wrap a conven-
Fig. 11. The steady-state response of a linear periodically-varying
tional transient analysis algorithm around a modified version
system to a small complex exponential stimulus. The large signals of a periodic analysis. Thus the time required for the analysis
are represented with solid arrows and the small signals with hollow is roughly equal to the time for a single periodic analysis mul-
arrows. tiplied by the number of time points needed to represent the
quency, and the response at all points in the circuit and at all envelope. If the envelope changes slowly relative to the
frequencies is computed. Thus, in one step one can compute period of the carrier, then transient-envelope simulation can
the transfer function from one input to any output. It is also be very efficient relative to traditional transient analysis.
possible to do the reverse, compute the transfer functions Transient-envelope methods have two primary applications.
from any input to a single output in one step using an The first is predicting the response of a circuit when it is
‘adjoint’ analysis. This is referred to as a “periodic transfer driven with a complicated digital modulation. An important
function” or PXF analysis. PAC is useful for predicting the problem is to determine the interchannel interference that
output sidebands produced by a particular input signal, results from intermodulation distortion. Simple intermodula-
whereas PXF is best at predicting the input images for a par- tion tests involving a small number of sinusoids as can be per-
ticular output. formed with quasiperiodic analysis are not a good indicator of
Small-signal analysis is also used to perform cyclostationary how the nonlinearity of the circuit couples digitally modu-
noise analysis [8,40,52], which is an extremely important lated signals between adjacent channels. Instead, one must
capability for RF designers [59]. It is referred to as a “peri- apply the digital modulation, simulate with transient-enve-
odic noise” or PNoise analysis, and is used to predict the lope methods, and then determine how the modulation spec-
noise figure of mixers. PNoise analysis is also used to predict trum spreads into adjacent channels.
the phase noise of oscillators, however this is a numerically The second important application of transient-envelope meth-
ill-conditioned problem that requires special techniques in ods is to predict the long term transient behavior of certain RF
order to overcome the ill-conditioning and accurately com- circuits. Examples include the turn-on behavior of oscillators,
pute close-in phase noise [21,22]. power supply droop or thermal transients in power amplifiers,
LPV analyses provides significant advantages over trying to and the capture and lock behavior of phase-locked loops.
get the same information from equivalent large signal analy- Another important example is determining the turn-on and
ses. First, they can be much faster. Second, a wider variety of turn-off behavior of TDMA transmitters. TDMA (time-divi-
analyses are available. For example, noise analysis is much sion multiple access) transmitters broadcast during a narrow
easier to implement as a small-signal analysis. Finally, they slice of time. During that interval the transmitter must power
can be more accurate if the small signals are very small rela- up, stabilize, send the message, and then power down. If it
tive to the large signals. Small signals applied in a large signal powers up and down too slowly, the transmitter does not work
analysis can be overwhelmed by errors that stem from the properly. If it powers up and down too quickly, the resulting
large signals. In a small-signal analysis, the large and small spectrum will be too wide to fit in the allotted channel. Simu-
signals are applied in different phases of the analysis. Small lating with traditional transient analysis would be prohibi-
errors in the large signal phase typically have only a minor tively expensive because the time slice lasts on the order of
effect on the linearization and hence the accuracy of the 10-100 ms and the carrier frequency is typically at 1 GHz or
small-signal results. greater.
1309 IEEE JOURNAL OF SOLID STATE CIRCUITS, VOL. 34, NO. 9, SEPTEMBER 1999

1) Fourier-Envelope Method: With the Fourier-envelope waveforms back into the frequency domain using the forward
method the envelope is represented by slowly-varying Fou- Fourier transform. This procedure relies on the envelope
rier coefficients. First developed by Sharrit and referred to as being essentially constant over the length of a cycle of the
circuit envelope [55], the Fourier-envelope method is a tran- carrier.
sient-envelope method based on harmonic balance. In the One of the important strengths of harmonic balance is its abil-
Section IV-A, the concept of harmonic balance with time- ity to easily incorporate frequency-domain models for the lin-
varying Fourier coefficients was introduced. In that case, the ear components such as lossy or dispersive transmission lines.
Fourier coefficients were assumed to be periodic, with the Unfortunately, this is not true with the Fourier-envelope
result that signals themselves were quasiperiodic. With the method. The transient nature of the modulations Ṽ k ( t ) intro-
Fourier-envelope method [12,38,48], the Fourier coefficients duces the same difficulties that are present with distributed
in (20) are time varying, but are not necessarily periodic. components in transient analysis, which are addressed using
Instead, the Fourier coefficients X̃ k ( t ) are taken to be slowly similar techniques. In particular, one can use convolution
varying transient waveforms. Thus, signals take the form [20,54], or the model for the distributed component can be
∞ separated into delay and dispersion, with the dispersion being
x(t ) = ∑ X̃ k ( t )e j2πkf t , (42) replaced by a lumped approximation [56].
k = –∞
As the distributed components are linear, the sidebands for
where f is the fundamental frequency of the base Fourier each harmonic k can be treated individually. Thus, a separate
series. X̃ k ( t ) represents the complex modulation of the kth model is generated for each harmonic k, which greatly
harmonic. X̃ k ( t ) must vary slowly relative to f because if the reduces the bandwidth requirements on the models. The
bandwidth of X̃ k becomes greater than f/2 then the sidebands model for each harmonic must only be valid over the band-
of adjacent harmonics begin to overlap and the representation width of the sidebands associated with that harmonic. In RF
is not unique. circuits, the bandwidths of the sidebands are usually small
Now, rewrite (16) assuming v and f take the form of (42), relative to the carrier frequency, and so generating models of
distributed components for use in the Fourier-envelope
∞

∑
method is much easier than for conventional transient analy-
F̃ k ( Ṽ ( t ), t )e j2πkf t = 0 , (43)
sis. In fact, it is not uncommon for the bandwidth of an RF
k = –∞
circuit to be so small that the transfer function of a distributed
where component does not change appreciably over the bandwidth
dQ̃ k ( Ṽ ( t ) ) of the sidebands. In this case, the transfer function is taken to
F k ( Ṽ ( t ), t ) = -------------------------- + j2πkf Q̃ k ( Ṽ ( t ) ) + Ĩ k ( Ṽ ( t ) ) + Ũ k be constant.
dt
(44) 2) Sample-Envelope Method: With the sample envelope
Assume that the variations in Ṽ ( t ) are slow enough so that method the envelope is represented by slowly-varying sam-
the bandwidth of each term in (44) is much less than f/2, then pled version of the waveform as shown in Figure 10. First
the terms associated with each harmonic k will sum to zero developed by Petzold and referred to envelope following [41],
individually. Then, F̃ k ( Ṽ ( t ), t ) = 0 for each k, or in vector it is a transient-envelope method based on shooting methods.
form It approximates the sample envelope as a piecewise polyno-
mial [25,41] in a manner that is analogous to conventional
dQ̃ (Ṽ ( t ))
F̃ (Ṽ ( t ), t ) = ----------------------- + ΩQ̃ (Ṽ ( t )) + Ĩ (Ṽ ( t )) + Ũ ( t ) = 0 transient analysis. This approach is efficient if the sequence
dt
formed by sampling the state at the beginning of each clock
(45) cycle, v(0), v(T), v(2T),..., v(mT), changes slowly as a func-
where Ω is a diagonal matrix with j2πkf on the kth diagonal. tion of m. A “differential-like” equation is formed from (27)
As with transient analysis, discretization methods such as ∆v ( mT ) = φ T ( v ( mT ), mT ) , (47)
trapezoidal rule or the backward difference formulae replace
dQ/dt with a finite-difference approximation, converting (45) where ∆v ( mT ) = v ( ( m + 1 )T ) – v ( mT ) is a measure of the
to a system of nonlinear algebraic equations that is solved time-derivative of the sample envelope at mT. We can apply
with Newton’s method. For example, applying backward traditional integration methods to compute an approximation
Euler converts (45) to to the solution using a procedure that involves solving (47) at
isolated time points. If the sample envelope is accurately
Q̃ (Ṽ ( t m )) – Q̃ (Ṽ (t m – 1))
------------------------------------------------------- + ΩQ̃ (Ṽ ( t m )) + Ĩ (Ṽ ( t m )) + Ũ ( t m ) = 0 approximated by a low-order polynomial, then this procedure
tm – tm – 1 should allow us to skip many cycles, and so find the solution
(46) over a vast number of cycles in an efficient manner. For RF
Ĩ (Ṽ ( t m )) and Q̃ (Ṽ ( t m )) are evaluated at t s by converting circuits, (47) is stiff and so requires implicit integration meth-
Ṽ ( t m ) into the time domain using the inverse Fourier trans- ods such as backward Euler, which can be written as
v ( mT ) – v ( ( m – l )T )
form, passing the time-domain voltage waveform through ∆v ( mT ) ≈ --------------------------------------------------- , (48)
i( ⋅ ) and q( ⋅ ), and converting the resulting current and charge l
KUNDERT: INTRODUCTION TO RF SIMULATION AND ITS APPLICATION 1310

where l is the timestep, which is measured in terms of cycles. A. Linear Passive Component Models
This equation represents a 2-point boundary constraint on The main strength of harmonic balance is its natural support
(47), and so together they can be solved with shooting-New- for linear frequency-domain models. Distributed components
ton methods to find v(mT). If desired, other integration meth- such as lossy and dispersive transmission lines and interpo-
ods can be used, such as the backward-difference formulae. lated tables of S-parameters from either measurements or
As with transient analysis, once v(mT) is computed, it is nec- electromagnetic simulators are examples of linear models that
essary to check that the trajectory is following the low-order are handled easily and efficiently with harmonic balance.
polynomial as assumed. If not, the point should be discarded The difficulty with which shooting methods handle distrib-
and the step l should be reduced. If rapid changes in the enve- uted component models contrasts sharply with harmonic bal-
lope are encountered, envelope following can reduce its step- ance. The problem is that the state vector associated with
size down to the point where no cycles are skipped, in which distributed components is infinite dimensional. The state vec-
case envelope following reduces to simple transient analysis. tor must somehow be discretized before shooting methods
Thus, envelope following does not suffer the accuracy prob- can be applied. However, even then shooting methods will be
lems of the Fourier-envelope method when small timesteps expensive if the state vector is large [27,28]. This disadvan-
are taken to resolve a rapidly changing envelope. tage explains why shooting method-based RF simulation
techniques have mainly been applied to RFICs. Most RFICs
D. Other Methods
can be modeled completely with lumped components. New
The methods described above are either currently available, approaches for generating lumped equivalent models for dis-
or expected to be available soon, in the mainstream commer- tributed components [35,42], and components described in
cial RF simulators. However, there other methods that have the frequency domain such as with tables of S-parameters [6],
the potential to become significant to RF designers. In partic- are becoming available that are more reliable and effective
ular, two families of methods seem promising: the Volterra than existing methods, allowing shooting methods to be
methods [31] and the MPDE methods [4,51]. applied to circuits that contain a small number of distributed
The Volterra methods are similar in concept to the small-sig- components.
nal analyses in that they represent the circuit using a Taylor
series expansion, except they take into account more than just B. Nonlinearity
the first term in the expansion. In this way, Volterra methods Harmonic balance is very accurate and very efficient if the
can efficiently compute the response of circuits exhibiting a circuit is near linear and the voltage and current waveforms
small amount of distortion. are near sinusoidal. In fact, assuming the component models
MPDE is an abbreviation for Multirate Partial Differential are correct, harmonic balance becomes exact in the limit
Equation. It represents a family of methods based on the idea where the circuit is linear and the stimulus are sinusoidal.
of replacing the single time variable with a sum of time vari- This is not true for shooting methods. However this feature is
ables, one for each of the time scales in the circuit. Consider a generally only significant when trying to determine the dis-
mixer with a 1GHz LO and a 100MHz IF. Then t would be tortion of low distortion amplifiers and filters. It does not help
replaced with t = t1 + t2 where t1 is associated with the LO when analyzing mixers, oscillators, and sampling circuits
and t2 is associated with the IF. The underlying ordinary dif- because these circuits contain signals that are far from sinuso-
ferential equations that describe the circuit are reformulated idal.
as partial differential equations in t 1 and t 2 . The various Harmonic balance can struggle on strongly nonlinear circuits
MPDE methods are formulated by applying particular bound- or circuits that contain signals with abrupt transitions. Such
ary conditions and numerical methods to the t1 and t2 dimen- signals are common in RF circuits. For example, mixers are
sions. For example, quasiperiodic harmonic balance from driven with an LO that resembles a square wave, and even
Section IV-A is a MPDE method that applies periodic bound- sinusoidal oscillators contain current waveforms that are nar-
ary conditions and harmonic balance to both dimensions. row pulses. In this case, many frequencies are needed to accu-
Similarly, the Fourier-envelope method from Section IV-C is rately represent the signal, which increases the expense of
a MPDE method that applies a periodic boundary condition harmonic balance. In addition, the magnitude of the harmon-
and harmonic balance to the t1 dimension and an initial condi- ics drop slowly for signals with sharp transitions, making it
tion and transient analysis to the t2 dimension. Many other difficult to know how many harmonics must be computed by
variations are possible. harmonic balance. If too few harmonics are included, the
results are inaccurate, if too many are included, the simula-
V. COMPARING THE METHODS tions can be impractical.
All of the methods presented can be grouped into two broad Harmonic balance is also susceptible to convergence prob-
families, those methods based on harmonic balance and those lems when applied to strongly nonlinear circuits. Conver-
based on shooting methods. Most of the differences between gence can be improved by employing continuation or
the methods emanate from the attributes of the base methods: homotopy methods [3,27] These methods initially reduce the
harmonic balance and shooting methods. So only the base power of the input signal until convergence is achieved. Then
methods will be compared. the power is stepped up in a sequence of harmonic balance
1311 IEEE JOURNAL OF SOLID STATE CIRCUITS, VOL. 34, NO. 9, SEPTEMBER 1999

analyses where the result computed at one step is used as the VI. RF MEASUREMENTS
starting point for the next to improve convergence. In this Simulators are used to predict the performance of RF circuits
way, harmonic balance can be made robust. However, before they are actually constructed. This section introduces
because continuation methods end up calling harmonic bal- several of the most common RF measurements used to verify
ance tens, or perhaps hundreds, of times, they can be slow. performance, with a description of how these measurements
In contrast to harmonic balance, the ability of shooting meth- are made using an RF simulator. The measurements presented
ods to handle strongly nonlinear circuits is quite good. The are representative of the most important and common mea-
strengths of shooting methods stem from the properties of its surements being made on RF circuits.
underlying transient analysis. In particular, it chooses nonuni-
form timesteps in order to control error, and it has excellent A. Transfer Functions
convergence properties.
1) Conversion Gain: Conversion gain is the generalization of
The ability of transient analysis, and so shooting methods, to gain to periodically-varying circuits such as mixers. It is sim-
place time points in a nonuniform manner allows it to accu- ply the small-signal gain through a mixer as a function of fre-
rately and efficiently follow abruptly discontinuous wave- quency. Typically, conversion gain refers to the transfer
forms. Small timesteps can be used to accurately resolve function from the desired input to the desired output. But
rapid transitions without taking small steps everywhere. This there are many other transfer functions of interest, such as the
is very important for circuits such as mixers, relaxation oscil- gain from an undesired image or from an undesired input
lators, switched-capacitor filters, and sample-and-holds. In such as the LO, power, and bias supplies.
addition, the timestep is automatically chosen to control error.
With harmonic balance, the timestep is constrained to be uni- Remember that the output signal for a periodically-varying
form by the FFT, however there is new work that explores the circuit such as a mixer may be at a different frequency than
possibility of using new FFT algorithms that do not require the input signal. The transfer functions must account for this
equally spaced points with harmonic balance [10, 37]. frequency conversion. As described earlier, these circuits may
have many images, and so for a single output frequency there
The strong convergence properties of shooting methods result may be many transfer functions from each input.
from its implementation as a multilevel Newton method, and
not from the fact it is a time-domain method. Indeed, it is pos- One measures a transfer function of a mixer by applying the
sible to formulate harmonic balance as a time-domain method LO, computing the steady-state response to the LO alone, lin-
[27,60], yet its convergence properties do not fundamentally earizing the circuit about the LO, applying a small sinusoid,
change. As described in Section IV-A, shooting methods and performing one of the LPV analyses described in Section
apply Newton’s method to solve IV-B, such as PAC or PXF. One might also want to measure
the transfer function with a large interferer present. If the
φ T ( v ( 0 ), 0 ) = v(0) , (49)
interferer is assumed periodic, then the circuit would be lin-
for v(0). φT(v(0), 0) relates the initial state of the circuit, v(0), earized about the quasiperiodic response to both the LO and
to the state one period later. Newton’s method is applied to the interferer and a LQPV analysis such as QPAC or QPXF is
solve (49) and is both efficient and reliable if φT is a near lin- performed.
ear function. This is usually the case even when the underly- Actual measurements on mixers have shown that it is possible
ing circuit is behaving in a strongly nonlinear fashion because to predict conversion gain to within 0.25 dB [7].
φT is evaluated over exactly a period of the large periodic
clock signal, the signal that is driving the circuit to behave 2) AM and PM Conversion: As shown in Figure 11, when a
nonlinearly. Evaluating φ T itself still involves solving small sinusoid is applied to a periodically-driven or clocked
strongly nonlinear sets of equations, however that is done circuit, the circuit responds by generating both the upper and
using transient analysis, a natural continuation method, and lower sidebands for each harmonic. The sidebands act to
so is quite robust. modulate the harmonics, or carrier, and the relationship
The ability of shooting methods to converge on a large class between the sidebands and the carrier determines the charac-
of strongly nonlinear circuits without the need for continua- ter of the modulation. In Figure 12, both the carrier and its
tion methods or other convergence aids represents a signifi- sidebands are shown as phasors [50]. Assume that the side-
cant advantage in efficiency over harmonic balance. bands are small relative to the carrier and that the circuit is
driven at baseband with a small sinusoid with a frequency of
With shooting methods, it is natural to perform transient anal- fm. The sideband phasors rotate around the end of the carrier
ysis for a while before starting the shooting iteration in order phasor at a rate of fm, with the upper sideband rotating one
to generate a good starting point. This is usually sufficient to way and the lower rotating the other. The composite of the
get convergence even on troublesome circuits except when sideband phasors traces out an ellipse as shown in Figure
the time constants in the circuit are much larger than the 12(b). However, if the two sidebands have identical ampli-
period of the signal. If this is not sufficient, one can also use tudes and their phase is such that they align when parallel to
continuation methods with shooting methods. The initial tran- the carrier, the phase variations from each sideband cancels
sient analysis has the side benefit that it helps to identify cir- with the result being pure amplitude modulation (AM) as
cuits that are unexpectedly unstable. shown in Figure 12(c). If instead the amplitudes are identical
KUNDERT: INTRODUCTION TO RF SIMULATION AND ITS APPLICATION 1312

j2π ( f m – f c )t
Upper and Lower Sidebands Shown Separately l ( t ) = Le (51)
j2π ( f m + f c )t
u(t ) = U e (52)
– j φc jφ c
where L = ( A – j Φ )e and U = ( A + jΦ )e . These
can be rearranged to give Α and Φ in terms of L and U,
jφ c – j φc
A = ( Le + Ue )⁄2 (53)
jφ c – j φc
Sum of Upper and Lower Sidebands Φ = j ( Le – Ue )⁄2. (54)
Thus, given the phase of the carrier, φc, which can be com-
puted with a periodic steady-state analysis, and the transfer
functions from the input to the upper and lower sidebands, L
and U, which can be computed with a periodic AC analysis,
one can compute the to-AM (A) and to-PM (Φ) transfer func-
tions.
SSB DSB AM PM
(a) (b) (c) (d) If the to-FM transfer function is desired instead, let
j2π f m t
Fig. 12. How the amplitude and phase relationship between
ω(t) = Ωe be the modulation signal where in (50) φ
sidebands cause AM and PM variations in a carrier. The phasors with becomes

∫ ω(t) dt .
the hollow tips represents the carrier, the phasors with the solid tips φ(t) = (55)
represent the sidebands. The upper sideband rotates in the clockwise
direction and the lower in the counterclockwise direction. The
composite trajectory is shown below the individual components. (a) Then the to-FM transfer function is Φ = jωmΦ, or
– j φc jφ c
Single-sideband modulation (only upper sideband). (b) Arbitrary Ω = 2πf m ( U e – Le )⁄2. (56)
double-sideband modulation where there is no special relationship
between the sidebands. (c) Amplitude modulation (identical
magnitudes and phase such that phasors point in same direction 3) Oscillator Load Pull: Load pull refers to shifts in the fre-
when parallel to carrier). (d) Phase modulation (identical magnitudes quency of an oscillator as a function of changes in load
and phase such that phasors point in same direction when impedance, supply voltage, substrate, bias lines, etc. A
perpendicular to carrier).
change in load impedance represents a parametric change in
but the phases align when perpendicular to the carrier, then the circuit and so requires a full periodic steady-state analysis
the amplitude variations cancel and the result is almost purely to compute the response in the oscillation frequency. How-
a phase modulation (PM) as shown in Figure 12(d) (assuming ever, as long as changes in the signal levels on supplies, sub-
the sidebands are small). The DSB modulation shown in Fig- strates, and bias lines are small, sensitivity of the oscillator
ure 12(b) can be considered a combination of both AM and frequency to perturbations of this type can be computed using
PM modulation. the technique given above for computing the to-FM transfer
AM and PM conversion occurs either when a tone is injected function.
at either baseband or at a sideband. The former is referred to
B. Cyclostationary Noise
as baseband to AM/PM conversion and the latter is SSB to
AM/PM conversion. Both cases were demonstrated in the With clocked systems, there are two effects that act to compli-
case of an oscillator by Razavi [43]. cate noise analysis. First, for noise sources that are bias
dependent, such as shot noise sources in BJTs or the thermal
A PAC analysis directly computes the transfer function from
noise of MOSFETs, the time-varying operating point acts to
some small input signal to the upper and lower sidebands
modulate the noise sources. Such noise sources are referred to
components of a modulated carrier. It is also possible, using a
as being cyclostationary. Second, the transfer function from
change of basis, to recast these transfer functions in terms of
the noise source to the output is also periodically-varying and
the AM and PM components of the modulation [53]. To show
so acts to modulate the contribution of the noise source to the
this, consider a circuit that is generating a sinusoidal carrier.
output. In this case, even if the noise source were stationary,
Assume that the carrier is both amplitude and phase modu-
as it would be for thermal noise of a linear time-invariant
lated by small complex exponentials at the same frequency
resistor, the noise at the output is cyclostationary.
fm. The resulting signal would take the form
v m ( t ) = A c(1 + α(t)) cos ( 2πf c t + φ c + φ ( t ) ) (50) Modulation is a multiplication of signals in the time-domain
and so in the frequency-domain the spectrum of the noise
j2πf t
where α(t) = Ae m is the amplitude modulation and
source is convolved with the spectrum of the transfer function
j2πf m t
φ(t) = Φe is the phase modulation. Both A and Φ are [68]. The transfer function is periodic or quasiperiodic and so
complex coefficients. Using the narrowband angle modula- has a discrete line spectrum. Convolution with a discrete
tion approximation [68], (50) can be expanded into a sum of spectrum involves a series of scale, shift, and sum operations
complex exponentials in order to identify the upper and lower as shown for a mixer in Figure 13. The final result is the sum
sidebands. of the noise contributions from each source both up-converted
1313 IEEE JOURNAL OF SOLID STATE CIRCUITS, VOL. 34, NO. 9, SEPTEMBER 1999

and so double-sideband (DSB) NF is employed. In this case,

Noise Output the images associated with the carrier overlap, and SNRin
includes the noise power from both. In both cases, SNRout
includes the affect of the input noise from all images, how-
LO
ever, it excludes the noise generated in the load at the output
LO
frequency.
Input Signals
Noise In (57) the signal power is both in the numerator and the
denominator and so cancels out. Thus, (57) can be rewritten
PO – PL
NF = 10 log -------------------- (58)
Replicate PS
Individual Noise Translate
Contributions
where PO is the total output noise power, PL is the output
noise power that results from noise generated by the load at
the output frequency, and PS is the output noise power that
results from noise generated by the source at the input fre-
quency. Generally, the PS used when computing DSB NF is
Sum twice as large as the one used with computing SSB NF,
Total Output Noise whereas PO and PL are both the same, thus the DSB NF is
usually 3 dB smaller than the SSB NF for the same circuit.
One computes the noise of a mixer by applying the LO, com-
puting the steady-state response to the LO alone, linearizing
Fig. 13. How noise is moved around by a mixer. The noise is
replicated and translated by each harmonic of the LO, resulting in the circuit about the LO, and applying the PNoise analysis of
correlations at frequencies separated by kfLO. Section IV-B. Actual measurements on bipolar mixers have
shown that noise figure can be predicted to within 0.25 dB
and down-converted by the harmonics of the LO to the
[7,34]. Results are not expected to be as accurate on CMOS
desired output frequency. This is referred to as noise folding.
mixers because the noise model for MOS transistors is not as
Periodic modulation of a stationary noise source, either from accurate as the one for bipolar transistors.
a periodic bias or from a periodically-varying signal path
from the source to the output, results in cyclostationary noise 2) Impact of Cyclostationarity on Subsequent Stages: In gen-
at the output. In stationary noise, there is no correlation eral, clocked circuits such as mixers produce cyclostationary
between noise at different frequencies. As can be seen from noise, which implies that the noise statistics, such as the
Figure 13, at frequencies separated by kf cyclostationary power spectral density (PSD), denoted S(f,t), varies as a peri-
noise is correlated, where f is the modulation frequency and k odic or quasiperiodic function of time. If a spectrum analyzer
is an integer [15]. is used to observe this noise, and if the frequency of the noise
variation is much faster than the analyzer can track, then the
1) Noise Figure: Noise is a critical concern in receivers spectrum analyzer will measure the time average PSD. Of
because of the small input signals. Typically designers char- course, the time average PSD is not a complete characteriza-
acterize the noise of individual blocks using the noise figure tion of the noise, but often it is sufficient. An important ques-
of the block because it is relatively simple to combine the tion is: when is the time average PSD sufficient to
noise figure of cascaded blocks to determine the noise figure characterize the output noise of a circuit, and when it not?
of the entire receiver [44,65]. The noise figure (NF) of a block
This question can be answered with the help of the following
is a measure of how much the signal-to-noise ratio (SNR)
observation. If an uncertainty in time is introduced into the
degrades as the signal passes through the block. It is defined
cyclostationary process — a uniformly distributed random
as
variable from 0 to T is added to t — the resulting process is
SNR in stationary and its statistics are the time average of the statis-
NF = 10 log ----------------- (57)
SNR out tics of the cyclostationary process [15]. Similarly, if the
cyclostationary process is input to a system that does not
At the input of a receiver the SNR is defined as the signal track the variation of the PSD with time, then the phase of the
power relative to the background noise power picked up by variation is unknown to the system. In the absence of infor-
the antenna. Figure 6 shows that a receiver is sensitive to mation about the phase of the variation, the process becomes
noise at the input at each of its images. SNRin only includes stationary, with the PSD equal to the time average of S(f,t)
the noise power in the images where the input signal is found. [62].
In heterodyne receivers, the input signal is found in a single
sideband, and so single-sideband (SSB) NF is employed. In There are two common situations that would cause a subse-
this case, SNRin includes only the noise power in the image quent stage to track the variations of a cyclostationary pro-
associated with the input. In homodyne, or direct-conversion, cess. The first occurs if the signal driving the subsequent
receivers, the signal is found in both sidebands of the carrier, stage is large enough to generate a nonlinear response. This
KUNDERT: INTRODUCTION TO RF SIMULATION AND ITS APPLICATION 1314

happens, for example, when an oscillator drives a limiter. The decomposed into AM and PM noise, but there will always be
oscillator signal is large enough to drive the limiter into a equal amounts of both.
nonlinear region, causing the characteristics of the limiter to To find the AM or PM noise of a carrier, one must perform
track the variations in the cyclostationary noise produced by PNoise analysis and output both the noise at the upper and
the oscillator. The same is true when an oscillator drives a lower sidebands of the carrier along with their complex corre-
mixer. The second situation is when both circuits are being lation. The AM and PM components of the noise can then be
driven by large signals derived from the same reference. This computed using (53) and (54).
would occur if, for example, the output of one mixer were fed
to the input of another, and both were driven by the same LO. 4) Oscillator Phase Noise: One can apply the PNoise analy-
Because they are driven by the same LO, the second mixer is ses of Section IV-B to oscillators to compute their phase
synchronous with, and tracks the variations in the cyclosta- noise. Or one can apply the PXF analysis to determine the
tionary noise of, the first mixer. In both of these cases, just sensitivity of the output to small interfering signals such as
knowing the time average of S(f,t) is not sufficient to predict those on the power supply. And as indicated earlier in this
the noise performance of the entire system. In particular, section, these analyses are able to properly account for fre-
knowing the time-averaged noise figure of each of the two quency conversions and for the fact that the noise in the out-
mixers does not give sufficient information to predict the put manifests itself largely as changes in the phase of the
noise figure of the cascaded pair. output.
However, in the case of the two mixers, if the second mixer These analyses are small-signal analyses that assume that the
were driven with an independent LO, even one that was close circuit being analyzed does not respond in a nonlinear way to
in frequency to the first LO, then the phase drift between the the small-signal inputs. However, (7) indicates that even small
two LOs would cause the synchronism between the two mix- inputs can generate large changes in the phase if they are
ers to be lost, with the result that using the time-average noise close in frequency to the fundamental or to one of its harmon-
statistics is adequate when predicting the noise performance ics. While the deviation in oscillator phase is generally a lin-
of the whole system. Thus, in this case, if one knows the time- ear function of the input, the output voltage or current is a
averaged noise figures for the two mixers, then one can pre- linear function of the phase only when the deviations in phase
dict the noise figure of the combination using standard formu- are small. If the phase changes by a significant fraction of a
las [44,65]. period, the small-signal assumption is violated and the
It is common to combine two mixers in cascade in a superhet- response becomes a nonlinear function of the input. It is this
erodyne receiver architecture, and to generate the LO signals nonlinear response that causes the linewidth of Sv and L (the
for the two mixers by using two PLLs with a common refer- roll-off at very low offset frequencies that is given in (12)). As
ence frequency. However, it is interesting to note that even in a result, the small-signal analysis results do not predict the
this case, if the ratio of the LO frequencies is m/n where m linewidth and so are inaccurate for computing Sv and L at fre-
and n are relatively large integers (often m,n ≥ 4 is sufficient) quencies very close to the carrier or its harmonics. Thus, the
with no common factors, then using the time average of the small-signal PNoise analysis can only be used to compute Sv
noise at the output of the first mixer will usually introduce lit- and L for fm well above f∆. Usually, f∆ is quite small, and so
tle error when estimating the noise at the output of the second this is not considered an issue. PNoise analysis can be used to
mixer [62]. In addition, interstage filtering also acts to reduce compute Sφ for all frequencies.
the chance of error. Remember that noise sidebands must be Actual measurements on oscillators have shown that it is pos-
correlated for the noise to be cyclostationary. Filtering can sible to predict oscillator phase noise on bipolar resonant
convert a cyclostationary signal to a stationary signal if the oscillators to within 2 dB [67].
filter’s bandwidth and center frequency are such that it elim-
nates all but one sideband. 5) Jitter and Phase Noise of PLLs: Oscillators are frequently
encapsulated in phase-locked loops in order to stabilize their
3) AM and PM Noise: As shown in Figure 13, clocked cir- output frequency and reduce their phase noise and jitter by
cuits generate noise with correlated sidebands. And as shown locking them to a more stable reference. The reference is usu-
in Figure 12, depending on the magnitude and phase of the ally a fixed frequency, whereas the oscillator may need to
correlation, the noise at the output of the circuit can be AM change its frequency, perhaps to allow the receiver to tune
noise, PM noise, or some combination of the two. For exam- over a range of channels. Complex feedback schemes are
ple, oscillators almost exclusively generate PM noise near the often necessary to satisfy the often competing requirements
carrier whereas noise on the control input to a variable gain of frequency tuning resolution and noise performance [14].
amplifier results almost completely in AM noise at the output Predicting the phase noise and jitter of such circuits can be
of the amplifier. quite difficult. They rarely operate with simple periodic or
This ability to emphasize one type of noise over another is a quasiperiodic signals, and so the PNoise and QPNoise analy-
characteristic of clocked circuits and cyclostationary noise. ses cannot be directly applied. Even if they are periodic, the
Linear time-invariant circuits driven by stationary noise frequency ratio between the oscillator and the reference can
sources can only produce additive noise, which can be be so large as to make these techniques impractical. Tran-
sient-envelope methods could be applied, but like simple
1315 IEEE JOURNAL OF SOLID STATE CIRCUITS, VOL. 34, NO. 9, SEPTEMBER 1999

transient analyses, the transient-envelope methods are not and as such it is not necessary to specify the signal level at
setup to include the effect of component noise sources, which which the circuit was characterized.
play a large role in the noise and jitter of most PLLs. To measure the compression point of a circuit, apply a sinu-
Attempts to include device noise in generic transient and tran- soid to its input and plot the output power as a function of the
sient-envelope analyses generally fail because of the small power of the input. The 1 dB compression point is the point
amplitude and wide bandwidth of device noise. Device noise where the gain of the circuit has dropped 1 dB from it small-
is usually much smaller than the simulator error that masquer- signal asymptotic value. This is illustrated in Figure 15.
ades as noise. To avoid this problem, tolerances must be set iCP1dB is the input power and oCP1dB is the output power
very tight. To accurately model wideband noise without that corresponds to the 1 dB compression point. iCP is nor-
excessively coloring it, generally a very small and usually
fixed timestep is needed. Both of these combine to make this

Output Power
approach impracticable in most cases.
IP3

PLLs simulations often occur at the behavioral level rather

than at the transistor level because the transistor level simula- Fund CP
tions are very expensive and because behavioral simulation of 3rd
PLLs can be quite effective. If the behavioral models can be Input Power
made to include the effect of component noise sources, they
Fig. 15. The 1 dB compression point (CP1) is the point where the
can be used to efficiently predict the noise and jitter of PLLs output power of the fundamental crosses the line that represents the
[8,29]. To do so, it is first necessary to individually character- output power extrapolated from small-signal conditions minus 1 dB.
ize the noise behavior of the blocks that make up the PLL The 3rd order intercept point (IP3) is the point where the third-order
using transistor-level simulation. For each block, representa- term as extrapolated from small-signal conditions crosses the
tive periodic signals are applied and a PNoise analysis is per- extrapolated power of the fundamental.
formed. Then the jitter is extracted and provided as a
mally used for receivers and oCP for transmitters. Compres-
parameter to behavioral models for inclusion in a high-level
sion point is typically measured in dBm, which is decibels
simulation of the entire PLL. The noise and jitter of the PLL
relative to one milliwatt.
can be obtained by observing the transition times in simulated
output. To measure a two-tone nth-order IMD intercept point, IPn,
apply two sinusoids to the circuit’s input at f1 and f2. Make
C. Intermodulation Distortion the amplitude of the two sinusoids the same, and increase
Distortion is commonly measured in narrowband circuits by their power while plotting the power at the output in a funda-
applying two pure sinusoids with frequencies well within the mental (either f1 or f2) and in an nth-order intermodulation
bandwidth of the circuit (call these frequencies f1 and f2). The product (for IP3 use either 2f1 – f2 or 2f2 – f1). This is illus-
harmonics of these two frequencies would be outside the trated for IP3 in Figure 15. The nth-order intercept point IPn is
bandwidth of the circuit, however there are distortion prod- defined in terms of the power levels of the fundamentals and
ucts that fall at the frequencies 2f1 – f2, 2f2 – f1, 3f1 – 2f2, 3f2 – the nth-order products as extrapolated from their asymptotic
2f1, etc. As shown in Figure 14 these frequencies should also small-signal behavior. When the input signal is small, a dou-
bling of the input power doubles the fundamental output
power and multiplies the output power of the nth-order prod-
ucts by 2n. Thus, the asymptotic slope of the fundamental is 1
dB/dB and the asymptotic slope of the nth-order products is n
dB/dB. The n th -order intercept point (IP n ) is where the
Fig. 14. A narrowband circuit driven with two closely spaced asymptotes for the nth-order intermodulation product and the
sinusoidal tones ( ) responds by generating harmonics ( ) fundamental cross. iIPn is the input power and oIPn is the out-
and intermodulation ( ) tones. Distortion of the output signal put power corresponding to the intercept point. They are gen-
results because several of the odd-order intermodulation tone fall erally measured in dBm.
within the bandwidth of the circuit.
In practice it is only necessary to sweep the input power to
determine an appropriate input power P in for an accurate
be well within the bandwidth of the circuit and so can be used
extrapolation. Pin should be chosen small enough that P1, the
to measure accurately the intermodulation distortion, or IMD,
output power of the first order term, and Pn, the output power
produced by the circuit.
of the nth-order term, are both in their asymptotic ranges. And
1) Compression and Intercept Points: At low frequencies, it Pin should be chosen large enough so that P1 and Pn are com-
is common to describe the distortion of a circuit by indicating puted accurately by the simulator. Once an appropriate Pin is
the distortion in the output signal when driven by a sinusoid applied and the corresponding P1 and Pn found, the output
to achieve a certain output level. At high frequencies it is intercept point is computed with
more common to characterize the distortion produced by a ∆P
I P n = P + ------------ (59)
circuit in terms of a compression point or an intercept point. n–1
These metrics characterize the circuit rather than the signal,
KUNDERT: INTRODUCTION TO RF SIMULATION AND ITS APPLICATION 1316

where IP n is the nth-order intercept point (dBm), P is the receiver. If it is large enough to drive the front-end into com-
power in the fundamental in dBm, and ∆P is the difference pression, the effective gain for the desired signal is reduced,
between the desired output signal (P1) and the undesired nth- which reduces the sensitivity of the receiver. The effect of a
order output product (Pn) in dB. P is the input power Pin if blocker on gain can be determined by applying the interferer,
iIPn is desired and it is the output power P1 if oIPn is desired using a PSS or QPSS analysis to compute the time-varying
[65]. operating point, and then performing either a PAC or QPAC
IP3 is the most commonly used intercept point, but others, analysis to compute the gain for the desired signal.
including IP2, IP5, and IP7, are also of interest. To see how a blocker acts to increase the noise floor, consider
Actual measurements on receivers have shown that it is possi- its effect when it interacts with the phase noise from the local
ble to predict 1 dB compression point and IP3 to within 0.5 oscillator of the mixer used in the front-end of the receiver as
dB [7]. shown in Figure 16. The phase noise of the LO is directly

2) Rapid IP2 and IP3 Prediction: The traditional approach to RF Interfering

LO Channel
measuring intermodulation distortion is to apply two large
closely spaced test tones to the input and measure the inter-
modulation products. Both tones are large enough to generate f
a nonlinear response from the circuit. Thus, at a minimum, a Desired
two tone QPSS analysis is required. If the circuit is clocked, IF Channel
such as a mixer, a third tone is also needed, in this case the Interfering IF
LO. As such, it can be an expensive simulation. However, in
those cases where the intermodulation products of interest fall
at the frequencies (n–1)f1 ± f2 it is not necessary for both test f
Desired IF
tones to be large to measure intercept point. One tone (f1) Fig. 16. In a receiver, the phase noise of the LO can mix with a large
must be large to drive the circuit into a nonlinear region. The interfering signal from a neighboring channel and swamp out the
second tone (f 2 ) is needed only to produce an in-band signal from the desired channel even though most of the power in the
response by mixing with the distortion products generated by interfering IF is removed by the IF filter. This process is referred to
the first tone. Thus, it is possible measure IP2 or IP3 using a as reciprocal mixing.
small-signal analyses [58], and doing so is considerably more translated onto the mixer products in a process referred to as
efficient in terms of both time and memory than using a full reciprocal mixing. Although the IF filtering at the output of
large-signal quasiperiodic analysis. the mixer may be sufficient to remove the blocker’s main
To compute IP2 or IP3 of an amplifier using this approach, mixing product, the desired signal is masked by the phase
apply the first tone with power P (in dBm) and perform a PSS noise of the LO that was down-converted by the blocker.
analysis to compute P1 (in dBm), the power at the output in Thus, the blocker acts to increase the noise floor of the
the fundamental frequency. Then apply the second tone as a receiver.
small complex exponential at a sideband of the first tone and Reciprocal mixing is currently a difficult simulation because
use a PAC analysis to compute ∆P (in dB), the ratio between it is a semi-autonomous analysis. In other words, it is a quasi-
the power at the output in the fundamental frequency of the periodic analysis where one fundamental, the blocker, is
small tone and the power at the output in the nth-order inter- driven, and one fundamental, the LO, is autonomous. While
modulation product. IPn is then computed with (59). conceptually possible, no simulator is currently available that
The intermodulation distortion of a clocked circuit such as provides a semi-autonomous analysis. So instead, one must
mixer is measured in a similar manner, except the both the replace the oscillator with a nonautonomous equivalent
LO and the first tone are applied together and the response model and then perform a QPNoise analysis. The model must
computed with a QPSS analysis. The response to the second mimic the phase noise produced by the oscillator. A Verilog-
tone is computed using a QPAC analysis. A [64] model that includes phase noise is given in Listing I.
In practice this approach is limited to computing IP2 and IP3, This particular module does not include flicker noise, even
but given that restriction, it can be used in the same situations though it uses the flicker_noise function. Notice that the slope
as the traditional approach and gets the same answer, but is (the second argument) is 2. This causes the function to gener-
considerably faster. ate noise whose power is proportional to f–2, which matches
the characteristic of oscillator noise that stems from white
3) Blockers: A blocker is a large interfering signal in a nearby noise sources. It is not difficult to generalize this model to
channel that acts to degrade the reception of the desired sig- include the effect of flicker noise, simply add in another
nal. It does so in two ways, by reducing the gain and increas- flicker_noise function, this time with a slope of 3.
ing the noise floor of the receiver. A typical receiver, shown
in Figure 1, has both an RF and IF filter. The RF filter is 4) Spurious Responses: Spurious responses are undesired
always broader than the IF filter. Thus, a blocker is often not responses to signals at the input of the receiver that occur at
eliminated until it has passed through the front-end of the frequencies that are different from the desired receive fre-
1317 IEEE JOURNAL OF SOLID STATE CIRCUITS, VOL. 34, NO. 9, SEPTEMBER 1999

LISTING I
0 dB
VERILOG-A OSCILLATOR MODEL THAT INCLUDES PHASE NOISE.
–10 dB
// Oscillator with Phase Noise –20 dB
‘include “discipline.h” –30 dB
‘include “constants.h”
–40 dB
module osc (out); –50 dB
output out; electrical out; –4 MHz –2 MHz 0 Hz 2 MHz 4 MHz
parameter real freq=1 from (0:inf); Fig. 17. Intermodulation distortion in a power amplifier spreads the
parameter real ampl=1; bandwidth of the transmitted signal to the point where it can interfere
with adjacent channels.
parameter real pn=0;
is difficult to quantify using simulation. Simple two-tone
real phase, phase_noise;
intermodulation tests are not representative of digitally modu-
analog begin lated signals. Instead, the transmission of a long pseudoran-
phase = 2*‘M_PI*freq*$realtime; dom sequence of symbols is simulated. The output spectrum
phase_noise = flicker_noise(ph,2); is calculated from a sequence that typically contains between
V(out) <+ ampl*cos(phase + phase_noise); 1k and 4k symbols. The adjacent channel power is then char-
end
acterized with the adjacent channel power ratio, or ACPR,
endmodule
P adj
quency. The undesired images of Section III-A are an exam- ACPR = --------- (61)
P in
ple of spurious responses. However, there are additional
frequency ranges that can result in spurious responses if the where Padj is the total power in the adjacent channel and Pin
interfering signals encountered are at sufficiently high level. is the power in the desired channel.
Any RF signal with a frequency that satisfies the following The carrier frequencies are typically in the 1-5 GHz range and
relationship for any integers m and n could result in a spuri- the symbols typically have a rate of 10-300 kHz. Such a sim-
ous response, ulation is clearly impractical for traditional transient analysis.
± m f in ± n f LO = ± f IF , (60) Instead, the transient-envelope methods of Section IV-C are
used. However, simulating a 1-4k symbol sequence still
where fin is any input frequency, fLO is the LO frequency, fIF
requires between 10k and 100k simulation points, each of
is the desired IF frequency. Of particular importance is the
which represents a harmonic balance or shooting method
half-IF spurious response in which (m,n) is either (2, – 2 ) or (–
solve, and so even the transient-envelope methods are very
2,2). This occurs at fin = (fRF + fLO)/2, midway between the
expensive for this type of simulation, particularly for a circuit
desired input frequency, fRF, and that of the LO. There are two
with more than just a few components.
possible causes for a half IF response [44]. First, if the inter-
fering input signal is subject to significant amounts of second- An alternative to transient-envelope simulations is to extract a
order distortion and the LO contains a significant second har- behavioral baseband-equivalent table model of the transmitter
monic. In this case the second harmonics of the interferer and using one of the large-signal steady-state methods described
the LO will mix and generate a response at the IF. Second, if in Section IV-A [5]. Because the behavioral model abstracts
the fundamental of the interferer mixes with the fundamental away the carrier and unnecessary circuit details, the ACPR
of the LO and the product, at fIF/2, is subject to significant calculation step is fast regardless of circuit size or complexity.
second order distortion. The half-IF response can be predicted In transmitter circuits the input baseband signal is usually
from IP2 measurements [65]. well within the transmitter’s bandwidth, so a memoryless
model often suffices. This is important because it makes gen-
5) Spectral Regrowth and ACPR: A very important issue erating the behavioral models fairly easy. A table model is
when transmitting digitally modulated signals is adjacent constructed by simply exercising the circuit over a range of
channel power. A transmitter should only emit power in its input amplitudes and phases and computing the steady-state
designated channel. Any power emitted in adjacent channels responses. This approach is accurate if the bandwidth of the
can interfere with the proper operation of nearby receivers circuit is wide compared with the bandwidth of the signal.
that are attempting to receive signals from distant transmit- Given the high carrier frequencies (1 GHz and above) and the
ters. As such, transmitters have strict adjacent channel inter- low signal bandwidths (1.23 MHz for the relatively broad-
ference (ACI) or adjacent channel power (ACP) requirements band IS-95 CDMA standard) this assumption is relatively
that they must satisfy. The lowpass filters in the transmitter of safe. There is, however, one situation that will cause this
Figure 2 are designed to limit the bandwidth of the transmit- assumption to be violated, if low frequency dynamics in
ted sequence. However, if the mixers or power amplifier are either the power supply or the bias supply lines significantly
nonlinear, intermodulation distortion can cause the bandwidth affect the characteristics of the transmitter’s signal path. It is
to grow back, as shown in Figures 3 and 17. This effect is easy to check for this situation using a periodic transfer func-
referred to as spectral regrowth. Unfortunately, this situation tion analysis. Simply apply a large periodic signal to repre-
KUNDERT: INTRODUCTION TO RF SIMULATION AND ITS APPLICATION 1318

sent an unmodulated carrier, and then perform the PXF off. SFDR or NPR is the ratio of the power in that channel
analysis over a range of frequencies centered about the car- when the transmitter is on relative to the power when it is off.
rier. If the transfer functions are relatively constant over the When the transmitter is off, the power in the channel is due to
signal bandwidth, it is appropriate to use a memoryless table intermodulation distortion cause by the other channels. SFDR
model. Otherwise a more sophisticated model or a transient- or NPR represents the dynamic range available in that chan-
envelope simulation is needed. nel.
In cable multichannel systems simple transient analysis is
6) Triple Beats: The term triple-beat refers to three-tone,
used to compute SFDR/NPR. However, in wireless multi-
third-order intermodulation distortion. Three-tone IMD
channel systems the frequency of the lowest channel is much
involves terms of the form f1 ± f2 ± f3 where as the two-tone
greater than the channel spacing and so transient-envelope
IMD discussed earlier involves terms of the form 2f1 ± f2 and
methods would be used. However, as with ACPR, this can be
f1 ± 2f2. Interference caused by triple beats is 6 dB higher
a very expensive simulation, particularly if the carriers are
than that caused by two-tone IMD [65]. Thus, if there are
modulated. And as with ACPR, there are many cases where
only three tones, one can determine the size of the triple beats
the multichannel system being tested is broadband. In this
rather simply from a two-tone IMD measurement. However,
case it is possible to generate a memoryless table model as
if the number of tones becomes large, the number of triple
before and efficiently evaluate the table model to predict
beats becomes very large. The number of triple beats gener-
SFDR/NPR. And again, PXF analysis can be used to verify
ated is n(n–1)(n–2)/2, where n is the number of tones. Even
the broadband assumption.
for a moderate number of tones it becomes an arduous task to
track the triple beats because of their numbers, which makes
estimation of triple-beat interference using this approach dif- ACKNOWLEDGMENTS
ficult if not impossible when there are a large number of I would like to thank the following people for the many
tones. enlightening conversations about RF simulation and RF cir-
cuits that contributed to this article: Joel Phillips, Jacob
7) Multichannel Systems: In modern cellular phone systems, White, Manolis Terrovitis, Dan Feng, Keith Nabors, Jess
the handset transmitters are power controlled, so the power Chen, Asad Abidi, Frank Carr, Kevin Gard, Geoff Dawe and
received at the basestation for each channel is roughly the Alper Demir.
same. The receivers used in cable tuners and repeaters and
satellite communications also must operate properly in the
REFERENCES
presence of a multitude of large carriers. In this setting, dis-
tortion performance is often difficult to quantify because of [1] A. Abidi. Radio frequency integrated circuits for portable com-
munications. Proceedings of the 1994 IEEE Custom Integrated
the large number of intermodulation distortion products. If we
Circuits Conference, May 1994.
consider only third-order nonlinearities, there are two types of
[2] T. Aprille and T. Trick. Steady-state analysis of nonlinear cir-
intermodulation distortion of concern, two-tone IMD and cuits with periodic inputs. Proceedings of the IEEE, vol. 60, no.
three-tone IMD (triple-beats). If there are a large number of 1, pp. 108-114, January 1972.
equally spaced and equally sized carriers then two-tone IMD [3] J. Bolstad and H. Keller. A multigrid continuation method for
is constant over the band while triple-beat distortion peaks in elliptic problems with folds. SIAM Journal on Scientific and
the middle of the band [65]. In addition, there are many more Statistical Computing, vol. 7, no. 4, pp. 1081-1104, Oct. 1986.
triple-beat products than two-tone products. These, combined [4] H. Brachtendorf, G. Welsch, R. Laur and A. Bunse-Gerstner.
with the triple-beats being twice as large as the two-tone Numerical steady state analysis of electronic circuits driven by
products, results in triple-beat IMD limiting the dynamic multi-tone signals. Electrical Engineering, Springer-Verlag,
vol. 79, pp. 103-112, 1996.
range of multichannel systems more than two-tone IMD, par-
[5] J. Chen, D. Feng, J. Phillips, and K. Kundert. Simulation and
ticularly in the center of the band.
modeling of intermodulation distortion in communication cir-
Dynamic range of multichannel systems is characterized cuits. Proceedings of the 1999 IEEE Custom Integrated Cir-
using a spurious-free dynamic range (SFDR) or noise-power cuits Conference, May 1999.
ratio (NPR) measurement, which is illustrated in Figure 18. [6] C. Coelho, J. Phillips, M. Silveira. Robust rational function ap-
proximation algorithm for model generation. The 36th Design
Interferers Automation Conference, June 1999.
SFDR [7] G. Dawe, J.M. Mourant, and P. Brokaw. A 2.7V DECT RF
or NPR transceiver with integrated VCO. 1997 IEEE International Sol-
IMD floor id-State Circuits Conference Digest of Technical Papers and
Slide Supplement. February 1997.
Fig. 18. Spurious-free dynamic range or noise power ratio of a [8] A. Demir and A. Sangiovanni-Vincentelli. Analysis and Simu-
multichannel system. lation of Noise in Nonlinear Electronic Circuits and Systems.
Kluwer Academic Publishers, 1997.
This measurement mimics the situation where all channels
[9] A. Demir, A. Mehrotra, and J. Roychowdhury. Phase noise in
are being received at their maximum level. The transmitter oscillators: a unifying theory and numerical methods for char-
assigned to one channel near the middle of the band (where acterisation. Proceedings of the 35th Design Automation Con-
triple-beat distortion is at its highest) is then turned on and ference, June 1998.
1319 IEEE JOURNAL OF SOLID STATE CIRCUITS, VOL. 34, NO. 9, SEPTEMBER 1999

[10] A. Dutt and V. Rokhlin. Fast Fourier transforms for nonequis- [28] K. Kundert. Simulation Methods for RF Integrated Circuits.
paced data. SIAM Journal of Scientific Computing, vol. 14, no. IEEE ICCAD-97 International Conference on Computer-Aided
6, Nov. 1993. Design, November 1997.
[11] P. Feldmann and R. Freund. Efficient linear circuit analysis by [29] K. Kundert. Modeling and simulation of jitter in phase-locked
Padé Approximation via the Lanczos Process. IEEE Transac- loops. In Analog Circuit Design: RF Analog-to-Digital Con-
tions on Computer-Aided Design of Integrated Circuits and verters; Sensor and Actuator Interfaces; Low-Noise Oscilla-
Systems, vol. 14, no. 5, pp. 639-649, May 1995. tors, PLLs and Synthesizers, Rudy J. van de Plassche, Johan H.
[12] P. Feldmann and J. Roychowdhury. Computation of circuit Huijsing, Willy M.C. Sansen, Kluwer Academic Publishers,
waveform envelopes using an efficient, matrix-decomposed November 1997.
harmonic balance algorithm. IEEE International Conference [30] D. Long, R. Melville, K. Ashby and B. Horton. Full-chip har-
on Computer-Aided Design: Digest of Technical Papers, No- monic balance. Proceedings of the IEEE Custom Integrated
vember 1996. Circuits Conference, May 1997.
[13] D. Feng, J. Phillips, K. Nabors, K. Kundert, and J. White. Effi- [31] S. Maas. Nonlinear Microwave Circuits. Artech House, 1988.
cient computation of quasi-periodic circuit operating condi- [32] S. Martin, G.P. Li, H. Guan, S. D'Souza, M. Matloubian, G.
tions via a mixed frequency/time approach. The 36th Design Claudius, C. Compton. BSIM3 based RTS and 1/f noise models
Automation Conference, June 1999. suitable for circuit simulators. IEDM '98 Technical Digest, pp.
[14] F. Gardner. Phaselock Techniques. John Wiley & Sons, 1979. 851-884, Dec. 1998
[15] W. Gardner. Introduction to Random Processes: with Applica- [33] R. Melville, P. Feldmann and J. Roychowdhury. Efficient
tions to Signals and Systems. McGraw-Hill, 1989. multi-tone distortion analysis of analog integrated circuits.
[16] P. Gray and R. Meyer. Future directions in silicon ICs for RF Proceedings of the IEEE Custom Integrated Circuits Confer-
personal communications. Proceedings of the 1995 IEEE Cus- ence, May 1995.
tom Integrated Circuits Conference, May 1995. [34] R. Meyer, W. Mack, and J. Hageraats. A 2.5GHz BiCMOS
[17] D. Held and A. Kerr. Conversion loss and noise of microwave transceiver for wireless LAN. 1997 IEEE International Solid-
and millimeter-wave mixers: part 1 — theory. IEEE Transac- State Circuits Conference Digest of Technical Papers and Slide
tions on Microwave Theory and Techniques, vol. MTT-26, no. Supplement. February 1997.
2, pp. 49-55, February 1978. [35] K. Nabors, T. Fang, H. Chang, K. Kundert, and J. White.
[18] D. Hente and R. Jansen. Frequency-domain continuation meth- Lumped interconnect models via Gaussian quadrature. Pro-
od for the analysis and stability investigation of nonlinear mi- ceedings of the 34th Design Automation Conference, June
crowave circuits. IEE Proceedings, part H, vol. 133, no. 5, pp. 1997.
351–362, October 1986. [36] M. Nakhla and J. Vlach. A piecewise harmonic balance tech-
[19] X. Jin, J.-J. Ou, C.-H. Chen, W. Liu; M. J. Deen, P. R. Gray, C. nique for determination of periodic response of nonlinear sys-
Hu. An effective gate resistance model for CMOS RF and noise tems. IEEE Transactions of Circuits and Systems. vol. CAS-23,
modeling. IEDM '98 Technical Digest, pp. 961-964, Dec. 1998 no. 2, pp. 85-91, February 1976.
[20] S. Kapur, D. Long and J. Roychowdhury. Efficient time-do- [37] O. Nastov, J. White. Grid Selection Strategies for Time-
main simulation of frequency-dependent elements. IEEE/ACM Mapped Harmonic Balance Simulation of Circuits with Rapid
International Conference on Computer-Aided Design: Digest Transitions. Proceedings of the 1999 IEEE Custom Integrated
of Technical Papers, November 1996. Circuits Conference, May 1999.
[21] F. Käertner. Determination of the correlation spectrum of oscil- [38] E. Ngoya and R. Larcheveque. Envelop (sic) transient analysis:
lators with low noise. IEEE Transactions on Microwave Theo- a new method for the transient and steady state analysis of mi-
ry and Techniques, vol. 37, no. 1, pp. 90-101, Jan. 1989. crowave communication circuits and systems. IEEE Micro-
wave Theory and Techniques Symposium Digest (MTTS), pp.
[22] F. Käertner. Analysis of white and f–a noise in oscillators. In- 1365-1368, June 1996.
ternational Journal of Circuit Theory and Applications, vol.
18, pp. 485-519, 1990. [39] M. Okumura, T. Sugawara and H. Tanimoto. An efficient small
signal frequency analysis method for nonlinear circuits with
[23] F. Käertner. Noise in oscillating systems. Proceedings of the two frequency excitations. IEEE Transactions of Computer-
Integrated Nonlinear Microwave and Millimeter Wave Circuits Aided Design of Integrated Circuits and Systems, vol. 9, no. 3,
Conference, 1992. pp. 225-235, March 1990.
[24] A. Kerr. Noise and loss in balanced and subharmonically [40] M. Okumura, H. Tanimoto, T, Itakura and T. Sugawara. Nu-
pumped mixers: part 1 — theory. IEEE Transactions on Micro- merical noise analysis for nonlinear circuits with a periodic
wave Theory and Techniques, vol. MTT-27, no. 12, pp. 938- large signal excitation including cyclostationary noise sources.
950, December 1979. IEEE Transactions on Circuits and Systems — I. Fundamental
[25] K. Kundert, J. White and A. Sangiovanni-Vincentelli. An enve- Theory and Applications, vol. 40, no. 9, pp. 581-590, Septem-
lope-following method for the efficient transient simulation of ber 1993.
switching power and filter circuits. IEEE International Confer- [41] L. Petzold. An efficient numerical method for highly oscillato-
ence on Computer-Aided Design: Digest of Technical Papers, ry ordinary differential equations. SIAM Journal of Numerical
November 1988. Analysis, vol. 18, no. 3, pp. 455-479, June 1981.
[26] K. Kundert, J. White and A. Sangiovanni-Vincentelli. A mixed [42] J. Phillips, E. Chiprout and D. Ling. Efficient full-wave electro-
frequency-time approach for distortion analysis of switching magnetic analysis via model-order reduction of fast integral
filter circuits. IEEE Journal of Solid-State Circuits, vol. 24, no. transforms. Proceedings of the 33nd Design Automation Con-
2, pp. 443-451, April 1989. ference, June 1996.
[27] K. Kundert, J. White and A. Sangiovanni-Vincentelli. Steady- [43] B. Razavi. A study of phase noise in CMOS oscillators. IEEE
State Methods for Simulating Analog and Microwave Circuits. Journal of Solid-State Circuits, vol. 31, no. 3, pp. 331-343,
Kluwer Academic Publishers, 1990. March 1996.
[44] B. Razavi. RF Microelectronics. Prentice Hall, 1998.
KUNDERT: INTRODUCTION TO RF SIMULATION AND ITS APPLICATION 1320

[45] B. Razavi. RF IC design challenges. Proceedings of the 35th [59] R. Telichevesky, K. Kundert and J. White. Receiver character-
Design Automation Conference, June 1998. ization using periodic small-signal analysis. Proceedings of the
[46] G. Rhyne, M. Steer and B. Bates. Frequency-domain nonlinear IEEE Custom Integrated Circuits Conference, May 1996.
circuit analysis using generalized power series. IEEE Transac- [60] R. Telichevesky, K. Kundert, I. El-Fadel and J. White. Fast
tions on Microwave Theory and Techniques, vol. 36, no. 2, pp. simulation algorithms for RF circuits. Proceedings of the 1996
717-720, February 1988. IEEE Custom Integrated Circuits Conference, May 1996.
[47] V. Rizzoli and A. Neri. State of the art and present trends in [61] R. Telichevesky, K. Kundert and J. White. Efficient AC and
nonlinear microwave CAD techniques. IEEE Transactions on noise analysis of two-tone RF circuits. Proceedings of the 33rd
Microwave Theory and Techniques, vol. 36, no. 2, pp. 343-365, Design Automation Conference, June 1996.
February 1988. [62] M. Terrovitis, R. Meyer. Cyclostationary Noise in Communica-
[48] V. Rizzoli, A. Neri and F. Mastri. A modulation-oriented piece- tion Systems. ERL memo M99/36, 1999. Available through
wise harmonic-balance technique suitable for transient analysis Electronics Research Laboratory Publications, University of
and digitally modulated signals. Proc. 26th Microwave Conf., California, Berkeley.
Prague, pp. 546-550, Sept. 1996. [63] G. Vendelin, A. Pavio, U. Rohde. Microwave Circuit Design
[49] V. Rizzoli, C. Cechetti, A. Lipparini and F. Mastri. General- using Linear and Nonlinear Techniques. Wiley-Interscience,
purpose harmonic balance analysis of nonlinear microwave cir- 1990.
cuits under multitone excitation. IEEE Transactions on Micro- [64] Verilog-A Language Reference Manual: Analog Extensions to
wave Theory and Techniques, vol. 36, no. 12, pp. 1650–1660, Verilog-HDL, version 1.0. Open Verilog International, 1996.
December 1988. Available from www.ovi.org.
[50] W. Robins. Phase Noise in Signal Sources (Theory and Appli- [65] P. Vizmuller. RF Design Guide: Systems, Circuits and Equa-
cation). IEE Telecommunications Series, 1996. tions. Artech House, 1995.
[51] J. Roychowdhury. Efficient methods for simulating highly non- [66] A. Usihda, L. Chua and T. Sugawara. A substitution algorithm
linear multi-rate circuits. Proceedings of the 34th Design Auto- for solving nonlinear circuits with multi-frequency compo-
mation Conference, June 1997. nents. International Journal on Circuit Theory and Applica-
[52] J. Roychowdhury, D. Long and P. Feldmann. Cyclostationary tion, vol. 15, pp. 327–355, 1987.
noise analysis of large RF circuits with multi-tone excitations. [67] M. Zannoth, B. Kolb, J. Fenk, and R. Weigel. A fully integrated
Journal of Solid-State Circuits, March 1998. VCO at 2GHz. 1998 IEEE International Solid-State Circuits
[53] C. Samori, A. Lacaita, F. Villa, and F. Zappa. Spectrum folding Conference Digest of Technical Papers. February 1998.
and phase noise in LC tuned oscillators. IEEE Transactions on [68] R. Ziemer and R. Tranter. Principles of Communications: Sys-
Circuits and Systems–II: Analog and Digital Signal Process- tems, Modulation, and Noise. Houghton Miffin, 1976.
ing, vol. 45, no. 7, pp. 781-790, July 1998.
[54] J. Schutt-Aine and R. Mittra. Scattering parameter transient
analysis of transmission lines loaded with nonlinear termina- Kenneth S. Kundert received the B. S., M.
tions. IEEE Transactions on Microwave Theory and Tech- Eng., and Ph. D. degrees in electrical engineer-
niques, vol. MTT-36, no. 3, pp. 529-536, March 1988. ing and computer sciences from the University
[55] D. Sharrit. New method of analysis of communication systems. of California, Berkeley in 1979, 1983, and
MTTS’96 WMFA: Nonlinear CAD Workshop, June 1996. 1988 respectively.
[56] M. Silveira, I. El-Fadel, J. White, M. Chilukuri, and K. He specialized in circuit simulation and analog
Kundert. Efficient frequency-domain modeling and circuit circuit design. He is a Fellow with Cadence
simulation of transmission lines. IEEE Transactions on Com- Design Systems, San Jose, CA. and is the prin-
ponents, Packaging, and Manufacturing Technology — Part B: cipal architect of the Spectre circuit simulation
Advanced Packaging, vol. 17, no. 4, pp. 505-513, Nov. 1994. family. As such, he has led the development of Spectre, Spectre-
HDL, and SpectreRF. He also played a key role in developing
[57] J. Stewart, The power spectrum of a carrier frequency modulat-
Hewlett-Packard’s MNS harmonic balance simulator as well as the
ed by Gaussian noise. Proceedings of the IRE, vol. 42, no. 10,
Verilog-AMS and VHDL-AMS analog hardware description lan-
pp. 1539-1542, October 1954.
guages. Finally, he is the author of two books on circuit simulation,
[58] R. Telichevesky, K. Kundert and J. White. Efficient steady- Steady-State Methods for Simulating Analog and Microwave Cir-
state analysis based on matrix-free Krylov-subspace methods. cuits (Kluwer Academic Publishers, 1990) and The Designer’s
Proceedings of the 32nd Design Automation Conference, June Guide to SPICE and Spectre (Kluwer Academic Publishers, 1995).
1995.