Microwave Semiconductor Devices &

Circuits I
Amplifiers, Mixers, Oscillators

Lecture Notes

Summer Term

Institut für Hochfrequenztechnik (E-3)

Prof. Dr.-Ing. Arne F. Jacob

Exercises: M. Sc. Anton Sieganschin

Compiled on April 2, 2019

git rev. 0241186
Contents

1 Basics 2
1.1 Incident and reflected power waves . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Two-port parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Scattering parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.2 Transmission parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.3 Signal flow graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3.1 Stability circles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Gain definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4.1 Power relationships at generator and load . . . . . . . . . . . . . . . . . . 10
1.4.2 Power gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4.3 Available gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4.4 Transducer gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4.5 Achieving maximum gain . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4.6 Unilateral gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2 Microwave transistors 16
2.1 Bipolar-junction transistor (BJT) . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.1.1 Maximum frequency of oscillation . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Heterojunction bipolar transistor (HBT) . . . . . . . . . . . . . . . . . . . . . . . 22
2.3 Metal semiconductor field-effect transistor (MESFET) . . . . . . . . . . . . . . . 24
2.4 High electron mobility transistor (HEMT) . . . . . . . . . . . . . . . . . . . . . . 29
2.4.1 Classic device structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4.2 Pseudomorphic HEMT (PHEMT) . . . . . . . . . . . . . . . . . . . . . . 30
2.4.3 High performance materials . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3 Circuits 33
3.1 Matching Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1.1 Narrowband Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

i
ii Contents

3.1.2 Broadband Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

3.2 Balanced Amplifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.3 Low-noise amplifiers (LNAs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.4 Nonlinear distortions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.4.1 Weakly Nonlinear Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.4.2 Classes of nonlinear phenomena . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4.3 Intercept point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.5 Power Amplifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.5.1 Loadline Match Vs. Conjugate Match . . . . . . . . . . . . . . . . . . . . 47
3.5.2 Passive Load-pull Measurement . . . . . . . . . . . . . . . . . . . . . . . . 48
3.5.3 Active Load-pull Measurement . . . . . . . . . . . . . . . . . . . . . . . . 49
3.5.4 Performance Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.5.5 Conditions for 100% Efficiency . . . . . . . . . . . . . . . . . . . . . . . . 52
3.5.6 Switch-Mode Power Amplifiers . . . . . . . . . . . . . . . . . . . . . . . . 53
3.5.7 Class-F PA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.5.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

4 Mixer Basics 63
4.1 Energy-conservation relations in nonlinear reactances . . . . . . . . . . . . . . . . 63
4.1.1 Upconverter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.1.2 Downconverter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.2 Conversion matrix analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.2.1 Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.2.2 Example 1: conversion matrix with a single small signal . . . . . . . . . . 68
4.2.3 Example 2: conversion matrix with multiple small signals . . . . . . . . . 68

5 Mixer Devices 71
5.1 PN-diode (varactor) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.1.1 Normalized charge–voltage relationship . . . . . . . . . . . . . . . . . . . 73
5.1.2 Cutoff frequency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.2 Schottky diode (varactor / varistor) . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.3 Field-effect transistor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

6 Mixer Circuits 79
6.1 Schottky diode downconverter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.1.1 Conversion gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.1.2 Noise figure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
6.2 FET downconverter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Contents iii

7 Oscillator Basics [1] 100

7.1 Conditions for oscillation start-up . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
7.1.1 Case 1: One-port oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . 102
7.1.2 Case 2: Two-port oscillator [2] . . . . . . . . . . . . . . . . . . . . . . . . 103
7.2 Stationary operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
7.2.1 Two-port oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7.3 Stability of the stationary solution . . . . . . . . . . . . . . . . . . . . . . . . . . 106

8 Oscillator Devices 109

8.1 IMPATT diode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
8.1.1 Principle of Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
8.1.2 Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
8.1.3 Diode Impedance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
8.1.4 Output Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
8.1.5 IMPATT Diode Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
8.2 Gunn-Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
8.2.1 Stationary Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
8.3 Field-effect transistor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

9 Oscillator Circuits 131

9.1 Resonator stabilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
9.2 YIG oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
9.2.1 YIG: single-crystal sphere of Y3 F e5 O12 . . . . . . . . . . . . . . . . . . . 139
Part I: Amplifiers

The first part of this lecture deals with amplifiers. The characteristics expected from an ideal
amplifier are, for example, linear amplification, sufficient bandwidth, and high gain. Especially
at microwave frequencies, simultaneously achieving these goals is not easy. When excited with
large-signal amplitudes, an amplifier generally yields a nonlinear response, and with small-signal
amplitudes, the amplifier’s usefulness is limited by electronic noise. Also, broadband amplification
is difficult to achieve, and, as will be explained later, increased bandwidth comes at a cost. Last
but not least, the gain of microwave transistors is quite low: It is often less than 20 dB, and may
be less than 10 dB when one approaches the limits of solid-state device technology.
Because one of these difficulties, there are different types of amplifiers that are optimized for
specific tasks:

• Low noise amplifiers (LNAs) are preamplifiers that are designed for a low noise figure and
moderate or high power gain. LNAs are usually the first amplifier stage in RF receiver
front-ends.

• Power amplifiers (PAs) are designed for maximum output power or efficiency. Because
PAs are large-signal circuits, linearity cannot be taken for granted and must be ensured by
proper design. PAs are usually the last amplification stage in RF transmitter front-ends.

• Broadband amplifiers may handle either small or large power levels. It is essential that their
characteristics are nearly constant with frequency (e.g., flat gain or flat output power).

A necessary condition for all types of amplifiers is stability because unstable amplifiers tend to
oscillate or otherwise act strangely (noisy/chaotic output spectra, hysteresis effects). Another
major point is impedance matching: maximum gain requires both input and output to be
conjugately matched. However, a mismatch at the input or output is sometimes unavoidable. In
this case, one should make sure that the voltage standing wave ratio (VSWR) at the mismatched
port is not excessively high. This is particularly important when the amplifier is interfacing with
components that expect a certain impedance level (e.g., Z0 = 50 Ω).
This part is structured into three main chapters. In the first chapter, we stay in the linear
regime and review the fundamentals of linear circuit analysis, but with microwave applications
in mind. The second section deals with device technologies. The ‘classic’ transistor technologies
(i.e., Si-BJT and MESFET) are reviewed, and the newer technologies (i.e., HBT and HEMT)
are first introduced. After laying these foundations, the third section discusses important circuit
topologies, nonlinear phenomena, and measurement methods that are often used in practice.

1
Chapter 1

Basics

Before dealing with nonlinear phenomena basic aspects of linear circuit analysis are reviewed. The
concept of scattering parameters [3, 4, 5], which is an important tool at microwave frequencies,
is first introduced and then applied to stability analysis and gain calculation.

1.1 Incident and reflected power waves

Two-port measurements using impedance (Z-) or admittance (Y -) parameters become more and
more inaccurate with increasing frequency. The reasons are the following:

• The loading conditions for which Z- and Y -parameters are defined (short and open circuits)
are difficult to realize at frequencies above ∼ 100 MHz.
• Parasitic oscillations may occur when active circuits are terminated with reactive loads.
• Transmission line aspects often need to be considered. For example, a coaxial line connecting
to the device under test (DUT) may cause an impedance transformation and/or shift of
the phase reference plane.

To overcome these difficulties the commonly used two-port (or more generally: N -port) parame-
ters are replaced by scattering parameters at high frequencies. The definition of this new set
of parameters is based on simple transmission line theory and given below for two-ports. It is
easily extendable to N -ports.
Consider an incident, forward-traveling (superscript ’+’) and a reflected, backward-traveling
(superscript ’−’) electromagnetic wave with voltage phasor V and current phasor I propagating
on a transmission line. The z-dependent voltage and current on the line can be written as

V (z) = V + (z) + V − (z) = V + · exp(−γz) + V − · exp(γz), and (1.1)

V+ V−
I(z) = I + (z) + I − (z) = · exp(−γz) − · exp(γz). (1.2)
Z0 Z0
The transmission line is described by its longitudinal impedance per unit length (R0 + jωL0 )
and its transverse admittance per unit length (G0 + jωC 0 ). The complex propagation constant γ
and the characteristic impedance Z0 are then given by

R0 + jωL0
s
q
γ = α + jβ = (R0 + jωL0 )(G0 + jωC 0 ) and Z0 = .
G0 + jωC 0

2
Chapter 1. Basics 3

For low-loss or lossless lines, Z0 is usually assumed to be real (Im{Z0 } ∼ 0).

From transmission line theory, the telegrapher’s equations in phasor √
notation (time-harmonic
steady-state) are known. Here, an additional normalization factor of Z0 is used:
1 dV (z) R0 + jωL0
√ =− √ I(z)
Z0 dz Z0
= −γ · I(z) · Z0 , and (1.3)
p

dI(z)
= − G0 + jωC 0 · Z0 V (z)
p
p
Z0 ·
dz
V (z)
= −γ · √ . (1.4)
Z0
When taking the sum (Σ) and difference (∆) of the above equations, one obtains
√ i
d √VZ + I Z0
h
V d(2a)
 p 
Σ: + γ √ + I Z0 = + γ(2a) = 0, and (1.5)
0 !
dz Z0 dz
√ i
d √VZ − I Z0
h
V d(2b)
 p 
∆: = − γ · (2b) = 0. (1.6)
!
0
− γ √ − I · Z0
dz Z0 dz
The solutions of the above differential equations are of the form
a(z) = A · exp(−γz) and (1.7)
b(z) = B · exp(γz). (1.8)
The newly defined quantities, a and b, are linear combinations of voltage and current; conversion
between (V , I) and (a, b) is possible with the following relationships:

1 V
 
a(z) = √ + I Z0 (1.9a) V (z) = (a(z) + b(z)) · Z0 (1.10a)
p p
2 Z0
1 (a(z) − b(z))
V I(z) = (1.10b)
 p 
b(z) = √ − I Z0 (1.9b) √
2 Z0 Z0
Plugging (1.1) and (1.2) into the conversion relationships (1.9a) and (1.9b) yields
V + (z)
a(z) = √ = I + (z) · Z0 and (1.11)
p
Z0
V (z)
−
b(z) = √ = −I − (z) · Z0 , (1.12)
p
Z0
which shows that a(z) is an incident wave, and b(z) is a reflected wave, or more generally forward-
and backward-traveling waves. The main advantage of the (a, b)-waves is that they can be easily
measured, e.g., using directional couplers. Also, a transformation of the phase reference plane is
simple. When a or b are known in a given reference plane z = z1 , they are transformed into the
plane z = z2 using
ai (z2 ) = ai (z1 )e−γ(z2 −z1 ) , and bi (z2 ) = bi (z1 )eγ(z2 −z1 ) . (1.13)
√
The dimension of a and b is V · A. For this reason, they are also called power waves. The real
power P + (z) and P − (z) carried at position z by the incident and reflected wave, respectively,
follows directly from the squared magnitude of a and b:
|V + (z)|2 |V − (z)|2
P + (z) = |a(z)|2 = , and P − (z) = |b(z)|2 = . (1.14)
Z0 Z0
4 1.2. Two-port parameters

Then, the total real power P (z) carried on the transmission line in z-direction at position z is
the difference between incident and reflected power:

P (z) = Re {V (z) · I ∗ (z)} = |a(z)|2 − |b(z)|2 . (1.15)

1.2 Two-port parameters

1.2.1 Scattering parameters

Consider the two-port shown in Fig. 1.1. It is connected to two transmission lines with
characteristic impedance Z0 and own coordinate systems zi , i ∈ {1, 2}, which define the phase
reference planes. A power wave a1 incident on the left port results in a reflected power wave b1

2-port

Figure 1.1: Voltages, currents, and power waves at a two-port.

and a transmitted power wave b2 . If both ports are matched to Z0 , b1 and b2 travel along their
connecting transmission lines without causing further reflections. By the same argument, an
incident power wave a2 results in a reflected wave b2 and a transmitted wave b1 .
Separately considering a1 and a2 is a simplification. In reality, both waves may be present, and
the resulting power waves b1 and b2 are computed by superposition. This way, we can use the
waves ai and bi instead of the voltage V i and the current I i at port i (i ∈ {1, 2}) and obtain
new sets of linear two-port relations. In matrix notation these are expressed as
! ! !
b1 S11 S12 a
= · 1 ⇔ [b] = [S] · [a]. (1.16)
b2 S21 S22 a2
| {z }
Scattering matrix

The matrix elements Smn are called scattering parameters, or simply S-parameters. The scattering
parameters of a two-port are summarized below along with their physical interpretation:

b1
S11 = : input reflection coefficient (matched output), (1.17)
a1 a2 =0
b2
S22 = : output reflection coefficient (matched input), (1.18)
a2 a1 =0
b2
S21 = : forward transmission coefficient (matched output), (1.19)
a1 a2 =0
b1
and S12 = : backward transmission coefficient (matched input). (1.20)
a2 a1 =0

Reciprocal two-ports possess a symmetrical scattering matrix (S12 = S21 ).

Chapter 1. Basics 5

S-parameters can be measured when a1 = 0 or a2 = 0, that is, when port 1 and port 2 are
matched to the system impedance Z0 . In contrast to open or short circuits, this loading condition
is realizable at high frequencies and also less prone to causing parasitic oscillations in active
circuits.

generator splitter coupler coupler splitter termination

reference measurement measurement reference

receiver receiver receiver receiver

Figure 1.2: Scattering parameter measurement setup (network analyzer).

Fig. 1.2 shows a simplified block diagram of a network analyzer, which is commonly used for
measuring S-parameters. The generator creates a signal which is split equally by a 3 dB power
splitter. One half is delivered to the DUT, the other half is registered by the reference receiver
as a1 . The reflected and transmitted waves travel towards the couplers which feed a fraction to
the measurement receivers. If this fraction (the coupling factor C) is known, the waves b1 and
b2 can be determined. The illustrated setup measures S11 and S21 . S12 and S22 are measured
when the generator and termination switch locations.
Vector network analyzers detect both amplitude and phase of the waves and yield the complex
S-parameters. Scalar network analyzers, on the other hand, provide only the absolute values.

1.2.2 Transmission parameters

Similar to the ABCD-parameters for voltage and current, a transmission matrix T can be
defined for power waves:
! ! !
b1 T11 T12 a
= · 2 (1.21)
a1 T21 T22 b2
| {z }
Transmission matrix

Figure 1.3: Cascading of two-ports using the transmission matrix.

The transmission matrix is especially beneficial when many two-ports are cascaded. As shown
in Fig. 1.3, at the interface between the two-ports, incident and reflected waves exchange their
roles, and the resulting matrix T of the cascaded two-ports can be found by multiplying the
individual matrices T1 and T2 . One peculiar property of the T -matrix is that it is not necessarily
symmetric for reciprocal networks.
6 1.3. Stability

1.2.3 Signal flow graphs

Signal flow graphs [3, 4] are a graphical tool which is sometimes more convenient than the
algebraic manipulation of S-matrices. They consist of nodes which represent the waves ai , bi ,
and of directed line segments which represent the scattering parameters. Any node’s value is

(a) (b)

Figure 1.4: Signal flow graphs of (a) a two-port and (b) a load.

determined by summing up the contributions of all segments directed at it. Each contribution is
the product of the segment’s S-parameter and its origin node value. Fig. 1.4 illustrates the case
of a two-port and one-port (a load with reflection coefficient ΓL ). For instance, the node b2 is
calculated as b2 = S21 a1 + S22 a2 .
When working with signal flow graphs, the rules depicted in Fig. 1.5 are often helpful:

(a)

(b)

(c)

(d)

Figure 1.5: Signal flow graph rules — (a) series rule, (b) parallel rule, (c) self-loop rule, and (d)
splitting rule.

1.3 Stability

In RF amplifier design, stability [3, 4, 5] is an important matter. The parasitic reactive elements
that are present in most active devices (e.g., the gate–drain capacitance Cgd in a FET) lead to
feedback within the device. The influence of device parasitics is often negligible at low frequencies
but becomes significantly large at microwave frequencies. If the overall feedback is positive,
oscillation due to self-excitation may occur.
Chapter 1. Basics 7

To prevent such unwanted oscillations, a stability criterion is needed. General stability criteria
such as the Nyquist criterion or Bode plot gain-/phase-margin analysis rely on knowledge of both
open-loop and closed-loop transfer functions, which are difficult to measure when the feedback is
internal to the device. These practical difficulties lead to the introduction of a special stability
criterion based on S-parameters.
Consider the two-port in Fig. 1.6: the input port (1) is loaded with the source impedance ZS
(reflection coefficient ΓS ), and the output port (2) is loaded with a load impedance ZL (reflection
coefficient ΓL ). When looking into the input of the two-port, one sees the input reflection
coefficient Γin , and likewise, one can define an output reflection coefficient Γout .

1 2

Figure 1.6: Amplifier with input and output reflection coefficients.

The input reflection coefficient depends on S and the load impedance. Similarly, the output
reflection coefficient depends on S and the source impedance. They are calculated as
b1 S12 S21 ΓL
Γin = = S11 + = f ([S], ΓL ), and (1.22)
a1 1 − S22 ΓL
b S12 S21 ΓS
Γout = 2 = S22 + = g([S], ΓS ). (1.23)
a2 1 − S11 ΓS

Oscillations may occur if either the input impedance or the output impedance of the two-port
has a negative real part. In terms of the reflection coefficients, this condition is written as
|Γin | > 1 or |Γout | > 1, (1.24)
i.e., the input/output reflection coefficient’s magnitude becomes greater than one.
Thus, the amplifier is stable if
S12 S21 ΓL
|Γin | = S11 + < 1 and (1.25)
1 − S22 ΓL
S12 S21 ΓS
|Γout | = S22 + < 1. (1.26)
1 − S11 ΓS

Because stability depends on the source and load impedances, one can define two types of
stability:

Absolute Stability: A network is absolutely stable or unconditionally stable if it is stable for

all passive source impedances ZS and all passive load impedances ZL .
Conditional Stability: A network is conditionally stable if it is stable only for a subset of
passive source impedances ZS or a subset of passive load impedances ZL , and unstable
otherwise.

In practice the aim is to achieve absolute stability, but conditional stability may be sufficient if
the source/load impedance constraints can be met.
8 1.3. Stability

1.3.1 Stability circles

Until now, we have only found the conditions for Γin and Γout which lead to a stable design.
However, when designing an amplifier, one is more interested in determining suitable source and
load impedances because this information is needed when designing the matching networks.
The relationship Γin = f (ΓL ) is a bilinear transformation which maps circles in the ΓL -plane onto
circles in the Γin -plane. The inverse transformation ΓL = f −1 (Γin ) maps the circle inequality
|Γin | < 1 onto a corresponding circle inequality in ΓL and can be found with algebraic methods.
Since circle equations are more convenient than circle inequalities, only the limiting case of a unit
circle |Γin | = 1 is now considered. The corresponding circle in the ΓL -plane defines the boundary
between stable and unstable load reflection coefficients and is hence called load stability circle.
Likewise, a source stability circle can be defined. Fig. 1.7 illustrates the load stability circle and
j j
input-stable
loads
passive input
reflection coefficients
0 mapping 0

load
stability
circle
-j -j
unit circle

Figure 1.7: Bilinear transformations between Γin - and ΓL -planes; definition of the stability circle.

the mapping between the planes. The stability circle is determined by its center and radius
(CL , RL ) and (CS , RS ) for load and source stability circles, respectively. Here, the derivation
[3, 4] is omitted, and only the results for the load stability circle are presented:
(S22 − ∆S11 ∗ )∗
CL = , with ∆ = S11 S22 − S12 S21 , (1.27)
|S22 |2 − |∆|2
S12 S21
and RL = . (1.28)
|S22 |2 − |∆|2

By reducing the inequality |Γin | < 1 to the equality |Γin | = 1, information is lost; the question
whether the interior or the exterior of the stability circle is the stable region still needs to be
solved.
One approach is to pick an arbitrary point in the ΓL -plane and calculate |Γin |. A convenient
choice is ΓL = 0, which leads to the simple condition |Γin | = |S11 | < 1. In words:

• If the device input reflection coefficient S11 has a magnitude of less than 1, the origin of
the ΓL -plane belongs to the stable region.
• Otherwise, the origin of the ΓL -plane is an unstable load and therefore marks the unstable
region.

Figure 1.8 shows four possible constellations for the load stability circle when assuming |S11 | < 1.
Again, the depicted relations are analogous for source stability circles.
Chapter 1. Basics 9

j j

0 0

(a) -j (b) -j

j j

0 0

-j -j
(c) (d)

stable loads stable+passive loads

Figure 1.8: Four cases of stability when |S11 | < 1 — (a) conditionally stable with stable region =
interior; (b) conditionally stable with stable region = exterior; (c) absolutely stable
with stable region = interior; (d) absolutely stable with stable region = exterior.

Conditions for Absolute Stability

Stability circles are a useful tool when load or source terminations at a few frequencies are
analyzed. When stability circles are plotted for a broad range of frequencies, the Smith Chart
gets crowded with circles. Algebraic criteria are then often preferred, although they can only
determine the type of stability (absolute or conditional).
By Rollet’s condition, also known as K–∆ test, a two-port is absolutely stable if and only if

1 − |S11 |2 − |S22 |2 + |∆|2

K= > 1 and (1.29)
2|S12 S21 |
|∆| = |S11 S22 − S12 S21 | < 1 are both fulfilled. (1.30)

K is called Rollet- or Linvill stability factor. The condition is based on the pair (K, ∆), which
does not always allow to compare two designs. For example, if one amplifier has (K1 , ∆1 ) and
another has (K2 , ∆2 ), the first may have a better stability factor, (K1 > K2 ), and the second
may have a better ∆, (∆2 < ∆1 ). In this case one cannot tell which amplifier is better in terms
of stability, and one may instead use the single-valued µ stability factor.
10 1.4. Gain definitions

Equivalently, a two-port is absolutely stable if and only if

1 − |S11 |2
µ= ∗ | + |S S | > 1, (1.31)
|S22 − ∆S11 12 21

where greater values of µ refer to a greater stability margin.

Let us conclude the discussion of stability with a few remarks:

• Picking an ’unstable’ load or source termination does not automatically lead to oscillations;
it is only a necessary condition for it. Conversely, picking a stable termination is a sufficient
condition for a non-oscillating circuit.

• It can be shown that absolute stability at the input is equivalent to absolute stability at the
output. In this case, checking one side of the two-port is sufficient. Similarly, in the case
of conditional stability, it is sufficient to enforce stability only at one port. For example,
when picking an input-stable load (i.e., a ΓL leading to |Γin | < 1) and assuming a passive
source impedance |ΓS | < 1, the input port is stable. Even if now |Γout | > 1, oscillation
cannot start up. This is because an exponentially growing signal at one node must lead to
exponentially growing signals at all other connected nodes. Thus, by prohibiting oscillations
at a particular node one makes them impossible at all other nodes.

• S-parameters are frequency-dependent, and so are stability criteria. Checking for stability
only at the operating frequency is an unsafe design practice! Oscillations start from noise,
which is broadband and present at all circuit nodes. One needs to check for stability at all
frequencies where the device is active (i.e., until the maxmimum frequency of oscillation,
fmax , which is defined later).

• Absolute stability may not be achievable for all frequencies of interest. For these frequencies,
simultaneous conjugate matching at both ports is not possible.

1.4 Gain definitions

In this section, different definitions for the gain of a two-port are introduced [3, 4, 5]. These
definitions are often used in amplifier design. The following derivations use the nomenclature of
Fig. 1.6 for reflection coefficients or impedances.

1.4.1 Power relationships at generator and load

Before calculating the gain, let us first consider the power relationships at the generator and
load attached to the amplifier. Starting with the generator, the amplifier circuit is simplified
by replacing the two-port and load with the input impedance Zin seen by the generator. The
generator is modeled as a voltage source with phasor V S and source impedance ZS . The
corresponding signal flow graph shows a generator wave bS fed into a loop. The loop accounts
for multiple reflections between the source impedance and the amplifier input. In the case of
Γin = 0 (i.e., when Zin = Z0 ), the physical meaning of bS becomes obvious: |bS |2 is the power
delivered to the amplifier when its input is matched to the (real) system impedance. Now one
can relate bS to the circuit quantities:
Z0
|bS |2 = Pdel,in = · |V S |2 . (1.32)
Zin =Z0 |ZS + Z0 |2
Chapter 1. Basics 11

Figure 1.9: Generator subcircuit and corresponding signal flow graph.

In the general case, where Zin 6= Z0 , the power delivered to the amplifier is

Pdel,in = |ain |2 − |bin |2 = |ain |2 (1 − |Γin |2 ), (1.33)

and ain is found by applying the self-loop rule (Fig. 1.5c), or by algebraic methods:

bS
ain = . (1.34)
1 − (ΓS Γin )

The power delivered to the amplifier can now be expressed as

1 − |Γin |2
Pdel,in = · |bS |2 . (1.35)
|1 − ΓS Γin |2

The power available from the generator, Pav,S , is the maximum power that the generator can
deliver to the amplifier input. It is found when the input is conjugately matched (Zin = ZS∗ ):

|bS |2
Pav,S = Pdel,in = . (1.36)
Γin =Γ∗S 1 − |ΓS |2

(a) (b)

Figure 1.10: Amplifier flow graphs — (a) signal flow graph of the complete amplifier; (b)
simplified flow graph at the load.

To determine the power relationships at the load side, the signal flow graph of the whole amplifier
is needed (Fig. 1.10a). Solving it from scratch is quite cumbersome. However, by using the
general expression for ain (1.34), the load side can be simplified, and after applying the splitting
rule (Fig. 1.10b) and the self-loop rule, one obtains for the power incident to the load:
2
S21 |S21 |2
|aL |2 = ·a = · |bS |2 (1.37)
1 − S22 ΓL in |1 − ΓS Γin |2 · |1 − S22 ΓL |2

The power delivered to the load, Pdel,L , is obtained by subtracting the reflected power from the
incident power:
Pdel,L = |aL |2 − |bL |2 = |aL |2 (1 − |ΓL |2 ). (1.38)
12 1.4. Gain definitions

Lastly, the power available from the amplifier output, Pav,out , is attained when the output is
conjugately matched (Zout = ZL∗ ):

|S21 |2
Pav,out = Pdel,L = ··· = · |bS |2 , (1.39)
Γout =Γ∗L |1 − S11 ΓS |2 · (1 − |Γout |2 )

where Γout = Γ∗L has also been used to get rid of ΓL and Γin in the expression for aL .

1.4.2 Power gain

The power gain, GP , is defined as the ratio of the power delivered to the load (port 2) and the
power delivered to the amplifier input (port 1).

Pdel,L 1 − |Γin |2
   
GP = = |aL |2 (1 − |ΓL |2 ) |bS |2 (1.40)

Pdel,in |1 − ΓS Γin |2
|S21 |2 · (1 − |ΓL |2 )
= . (1.41)
(1 − |Γin |2 ) · |1 − ΓL S22 |2

Since Γin depends on [S] and ΓL , the power gain is a function GP ([S], ΓL ).

1.4.3 Available gain

The available gain, GA , is defined as the ratio of the power available from the amplifier output
(port 2) and the power available from the generator (port 1).

Pav,out |S21 |2 1
   
GA = = |bS |2 |bS |2 (1.42)

Pav,S |1 − S11 ΓS | · (1 − |Γout | )
2 2 1 − |ΓS |2
|S21 |2 · (1 − |ΓS |2 )
= . (1.43)
|1 − S11 ΓS |2 · (1 − |Γout |2 )

Since Γout depends on [S] and ΓS , the available gain is a function GA ([S], ΓS ).

1.4.4 Transducer gain

The transducer gain, GT , is defined as the ratio of the power delivered to the load (port 2) and
the power available from the generator (port 1).

Pdel,L 1
   
GT = = |aL |2 (1 − |ΓL |2 ) |bS |2 (1.44)

Pav,S 1 − |ΓS |2
|S21 |2 · (1 − |ΓS |2 ) · (1 − |ΓL |2 )
= . (1.45)
|1 − ΓS Γin |2 · |1 − S22 ΓL |2

The transducer gain is a function GT ([S], ΓS , ΓL ).

Chapter 1. Basics 13

1.4.5 Achieving maximum gain

The maximum available gain, Gmax , can be achieved when both ports are conjugately matched.
This simultaneous matching condition is a solution of the following system of complex equations:
∗ S ∗ Γ∗
S12
ΓS = Γ∗in = S11
∗
+ 21 L
(1.46)
1 − S11
∗ Γ∗
L
∗ S ∗ Γ∗
S12
ΓL = Γ∗out = S22 +
∗ 21 S
∗ Γ∗ . (1.47)
1 − S22 S

A solution exists only if the two-port is absolutely stable. In this case, one has GP = GA =
GT = Gmax , and

S21 
Gmax = K − K2 − 1 , (1.48)
p 
S12

where K is the Rollet (or Linvill) stability factor. It can be seen that Gmax is only real when
K ≥ 1. For conditionally stable devices Gmax is not meaningful, and one instead defines the
maximum stable gain, Gms as
S21
Gms = . (1.49)
S12
Trying to get more gain than Gms from a conditionally stable device is not advisable.

1.4.6 Unilateral gain

When active devices are employed at high frequencies, one is often interested in the maximum
frequency up to which a device can provide gain (and hence, excite oscillations). However, the
previously defined gains, including Gmax , all vary with feedback: Internal feedback (S12 ) is often
negative and results in diminished Gmax . External feedback may change this for the better or
the worse.
This problem is overcome by the maximum unilateral power gain, GU , also called Mason’s
invariant, which is an important figure of merit for the evaluation of active devices, or transistor
technologies in general. Its main advantage is that it is invariant with respect to any lossless,
reciprocal, four-port network in which the device may be embedded. Here, only the resulting
expression for GU is given. A derivation and detailed discussion can be found in [6]:

|S12 − S21 |2
GU = . (1.50)
det [1 − SS ∗ ]

As depicted in Fig. 1.11, such four-port embedding networks may provide external feedback
(Fig. 1.11c) or even result in different circuit configurations (Figs. 1.11a,b). Yet, their calculated
GU is the same. Because of its universal nature, GU is used for the definition of the maximum
frequency of oscillation, fmax :

f = fmax where GU (f ) = 1. (1.51)

def

The physical meaning of GU is the maximum gain of a two-port that is unilateralized, i.e., the
two-port’s internal feedback is canceled by proper external circuitry. In principle, a two-port
can be unilateralized with the embedding network in Fig. 1.11c. Here, the goal is that a voltage
V 02 at the output does not cause a voltage V 01 at the input:
14 1.4. Gain definitions

(a) common-source (b) common-drain (c) embedding network

configuration configuration for unilateralization

Figure 1.11: Examples of reciprocal, lossless four-port embedding networks.

• The reactance jX changes the angle ∠(V 02 , V 1 ) to zero,

• and the transformer ratio n adjusts the amplitude to |V 02 |/|n| = |V 1 |, so that the trans-
former’s secondary voltage cancels the device’s internal feedback (V 01 = −V 1 + V 1 = 0).

• Lastly, if the input and output are both conjugately matched, the outer network (device
and embedding network) has Gmax = GU , hence, the circuit operates at its maximum
unilateral power gain.

For a unilateral device (where S12 = 0 or is small enough), one obtains Γin = S11 and Γout = S22 ,
and the transducer gain is simplified to three separate gain factors

1 − |ΓS |2 1 − |ΓL |2
GTU = GT = · |S |2
· , (1.52)
|1 − S11 ΓS |2 | {z } |1 − S22 ΓL |2
21
S12 =0 | {z } G0 | {z }
GS GL

where GS and GL are the gain factors of input and output matching networks, respectively, and

input matching output matching

network active device network

Figure 1.12: Amplifier and matching networks with unilateral approximation.

G0 is the intrinsic gain of the two-port (Fig. 1.12). In practice, the non-reciprocity of a properly
biased transistor often leads to nearly unilateral behavior. Then, conjugate matching is achieved
by setting
ΓS = S11
∗
and ΓL = S22 ∗
, (1.53)
and the maximum power gain of a unilateral device is
1 1
GTUmax = max GT = (1.54)
n o
· |S21 |2 · .
S12 =0 1 − |S11 |2 1 − |S22 |2
Note that for a nearly unilateral device, simply setting S12 to zero has not the same effect
as the unilateralization technique described above, and that in this case (1.52) and (1.54) are
Chapter 1. Basics 15

approximations. The error associated with using GTUmax (S12 = 0) instead of Gmax (S12 6= 0)
has then upper and lower bounds which are determined by
1 Gmax 1
≤ ≤ , where (1.55)
(1 + δ)2 GTUmax (1 − δ)2
S11 S22 S12 S21
δ= . (1.56)
(1 − |S11 |2 ) · (1 − |S22 |2 )
Chapter 2

Microwave transistors

2.1 Bipolar-junction transistor (BJT)

References: [7, 5, 8, 9]
Before dealing with high-frequency phenomena, let us first briefly review the current gain in
bipolar transistors. Fig. 2.1 shows the one-dimensional model of an npn-BJT: The base–emitter
voltage VBE , the collector–base voltage VCB as well as the emitter-, base-, and collector currents
(IE , IB , and IC ) are drawn for the forward active mode, and the gray shaded areas represent the
resulting space-charge regions.

n++-emitter p+-base n-collector n++-substrate

electrons

holes recombination

Figure 2.1: Electron (blue) and hole (red) carrier flows within a BJT.

The common-base current gain α = β · γ is governed by the emitter efficiency γ and the base
transport factor β. For a high common-emitter current gain, α should be close to unity.
The emitter efficiency is the ratio of the electron diffusion current InE injected at the emitter–base
junction and the total emitter current IE ,
InE InE
γ= = . (2.1)
IE InE + IpE
The emitter–base doping contrast determines the magnitudes of electron and hole currents at
the junction; a high contrast NE  NB results in InE  IpE and hence, in emitter efficiencies
close to unity. γ is assumed to be frequency-independent.
The base transport factor is the ratio between the electron diffusion current InC arriving at the
collector and the injected electron diffusion current InE ,
InC 1 1 d2B 1 d2B
β= = ≈1− = 1 − , (2.2)
InE cosh(dB /LnB ) 2 L2nB 2 DnB · τ

16
Chapter 2. Microwave transistors 17

√
where dB is the neutral base width, and LnB = DnB · τ is the diffusion length (the distance at
which the injected minority carrier concentration has decayed to a fraction of e−1 ). Furthermore,
the approximation dB  LnB is justified because a small base width is needed to achieve a high
transport factor. For low frequencies the DC value of β can be used, but at high frequencies
the RF component of the minority carriers has to be considered in the semiconductor boundary
conditions. Then, the resulting base transport factor β ∗ has the same form as above if one
replaces the real carrier lifetime τ with a complex carrier lifetime τ ∗ = τ /(1 + jωτ ) [7, 9]:

1 d2B 1 d2B
β∗ ≈ 1 − = 1 − (1 + jωτ ). (2.3)
2 DnB · τ ∗ 2 DnB · τ

The frequency-dependent common-base current gain is then expressed as

α 2DnB
α∗ = β ∗ · γ ≈ with ωα = , (2.4)
1 + j(ω/ωα ) d2B

where ωα is the 3-dB corner frequency of the common-base current gain, and DnB is the diffusion
constant of the base.

2.1.1 Maximum frequency of oscillation

The maximum available gain Gmax of a bipolar transistor is determined by the α-cutoff frequency,
the base resistance rB , and the depletion-layer capacitance cBC at the base–collector junction. It
is frequency-dependent with 1/ω 2 :
αωα ωα
Gmax = ≈ . (2.5)
4ω 2 · rB · cBC 4ω · rB · cBC
2

The maximum frequency of oscillation, ωmax , is defined as the frequency where Gmax = 1. This
represents the boundary case where the active device becomes passive. Thus, beyond ωmax it is
impossible for the device to excite oscillations (wanted or unwanted) even if the output power is
completely fed back into the input. Setting Eq. 2.5 to unity and rearranging for ω = ωmax yields

1 ωα
r
ωmax = . (2.6)
2 rB cBC

Bipolar transistors for higher frequencies must have a sufficiently high ωmax . From Eq. 2.5 we
can now derive requirements for the high-frequency operation of BJTs:

• Small base width dB is required for high ωα

• Large cross-section and short path for base current (low rB )

• Wide collector depletion layer or small collector area (low cBC )

Some of these requirements are conflicting, for example, the cross-section for the base current
may be enlarged by dB , thereby violating the first requirement. It is therefore wise to review
the above requirements in the context of a particular transistor topology. The strip geometry
is depicted in Fig. 2.2: base and emitter are connected to the active region via contact strips
(“fingers”) the collector strip is placed next to the active region, and connected via heavily
doped ohmic contacts. The base resistance rB and the collector capacitance cBC depend on the
18 2.1. Bipolar-junction transistor (BJT)

p-base n+-emitter B
W SiO2
C
L

n Si epitaxial layer
E
n+ Si substrate
W
(a) (b)

Figure 2.2: BJT strip geometry — (a) cross-sectional view; (b) top view.

material properties and the transistor geometry:

W 1 W
rB = r0 · = · (2.7)
L σdB L
cBC = c0 · W L (2.8)
where c0 is the collector capacitance per area, and σ is the conductivity of the substrate. Plugging
these results into Eq. 2.6 one arrives at
1
ωmax ∼ √ . (2.9)
W dB
High performance in planar BJTs can thus be achieved for small W and dB . The finger width
W is limited by the photolithographic resolution and the mask alignment accuracy of the
semiconductor process. Nowadays, finger widths of 0.25 um or 0.13 um are possible. The base
width can be made very thin. The drawback, however, is base punch-through: For higher
base-collector voltages, the base–collector space charge region (inside the base) extends more
towards the base–emitter space charge region. At some point, both space-charge regions meet,
and the base is rendered ineffective (the transistor has no gain). The thinner the base, the
sooner base punch-through happens.
From here until end of BJT section: material from old lecture notes

Many strips: same voltage, larger current ⇒ power %, impedance &.

L
W

emitter
Ib
ie
db
base
collector W
(a) (b)

Figure 2.3: Field- and current distribution beneath emitter strip.

The base–emitter voltage at the edges of the emitter stripe is greater than in the middle due to
voltage drop over rB ,→ emitter current density is greater at the edges (Fig. 2.3).
Chapter 2. Microwave transistors 19

Emitter configuration with large ratio of length / area → three common geometries (Fig. 2.4).

E ++
E n
B

+
p
B
E B
(a) (b) (c)

Figure 2.4: RF bipolar transistor topologies: (a) interdigital geometry, (b) overlay geometry, (c)
mesh geometry.

• n+ pnn+ transistor (Fig. 2.5):

-3
cm
22
10

++
n -emitter +
n -substrate
18 +
10 p -base
|ND-NA|

n-collector

14
10
0 0.5 1 5 6 µm
x
Figure 2.5: Doping profile and layers.

– critical dimensions: dB < 0, 1µm, w < 1µm

– practical limit for dB : base-collector junction must not reach the emitter.
1. Upper limit for NE = 1021 cm−3 , otherwise: degeneration, only partly ionized.
→ µ ↓, rB ↑,
2. γ high: NB (0)  NE ,
3. doping / area:
Z db
N̄B = NB (x)dx > NC · dC
0

Z dB
NE max
NB (0)  ,→ dB from N̄B = NB (x)dx > NC · dC
100 0

• Equivalent circuit and cutoff frequencies: (Fig. 2.6)

20 2.1. Bipolar-junction transistor (BJT)

Ccb

0,02 pF
c0

0,2 pF α*ie

Rba Lba Lbi Rbi rbs rb 0 cC 0,02 pF rcs Lc Rc

C
0,2 Ω 1 nH 1 nH 0,2 Ω 5Ω 10 Ω rc 50 Ω 1 nH 0,2 Ω
ces
1 pF re 5 Ω 8 kΩ

res 1 Ω

ie
Rei 0,2 Ω

Lei 0,5 nH
Ceb Cec
0,2 pF 0,2 pF
Lea 0,5 nH

Rea 0,2 Ω

Figure 2.6: Equivalent circuit of a RF bipolar transistor.

– RF transistor: operation at higher frequencies ⇒ additional parasitic effects occur

(delay times; R; C; L)
– dashed: inner transistor supplemented by rbs , rcs , res (contact resistance)
– c0 : base-collector junction capacitance not beneath the emitter.

e−jmωτb 1 d2b
αb∗ = α · ; τb = =
1 + jωτb ωα 2ηDn,b

m,η: function (doping profile of base)

– gradient of doping profile within base > accelerating drift field τb ≈ 3 . . . 10ps

• Three additional effects alter α∗ :

1. Charge delay by ces (flows through re )

1 kT
,→ αe = τe = re · ces re =
1 + jωτe q · Ie
Ie A
τe small for Ie large; limit: = 1000 2
A cm
majority carrier = minority current in base → increases base width → τb too high
typical: τe ≈ 5 . . . 10ps
2. Transit time through collector junction
Chapter 2. Microwave transistors 21

– dc depends on Nc and VBC (reverse voltage), should extent over whole epitaxial
layer (otherwise rcs too high)
– strong electric field E → charge carriers accelerated to vs (saturated drift velocity)
– transit time: τc = dc /vs
– charge carriers in collector junction like charge q between two plates (capacitor),
q influences charge on each plate:

-q

Figure 2.7: Illustration of transit time effects in collector region.

dc − x x
q1 = q ; q2 = q
dc dc

q moves at vs → influences current during transit

dq2 q · vs q
i= = =
dt dc τc
– alternating current with phasor i0 at base
– current distribution across junction:
ω
ix = i0 · e−j vs ·x

– charge at x within dx: dq = ix /vs · dx ,→ contribution to collector current:

dq dx
di = = ix ·
τc vs · τc
– total collector current
1 − e−jωτc
Z dc
ic = di = i0
0 jωτc

at ωτc = 2π → ic = 0 : positive and negative half-wave cancel each other;

otherwise attenuation and delay; at

ω · τc  1 : ic ≈ i0 · exp(−jωτc /2)
1
,→ ic = αc · i0 , αc ≈
1 + jω τ2c
22 2.2. Heterojunction bipolar transistor (HBT)

– e.g. dc = 2µm; vs = 107 cm

s ,→ τc
2 = 10ps

3. Collector charge delay

– delay and attenuation from rcs and cc
– first charge cc , then ”overcome” rcs
– estimation:
1
αcs = ; τs = rcs · cc : typical 1 .. 5 ps
1 + jωτs
• Effective current gain of inner transistor:
∗
αef f = α·e−jmωτb
(1+jωτb )(1+jωτe )(1+jω τ2c )(1+jωτs )

For ω  1 1 1 1
τb , τ e , τc , τs :
∗
αef f ≈ α·e−jmωτb
1+jωτec , τec = τb + τe + τc
2 + τs
√
∗ | = α/ 2 :
effective α-cutoff frequency for |αef f

1
,→ ωec =
τec

• Common emitter configuration:

– favorable input- and output impedance

– higher current gain at RF than common base configuration

• Current gain from αe∗ = α∗

1−α∗

∗ α · e−jmωτb
αef ⇒ αe∗ ≈
f
1 − α · e−jmωτb + j ωωec

– transit frequency: |αe∗ (ωT )| = 1

– low frequency transistor: ωT = ωα
– high frequency transistor: ωT = ωec
1+mωec τb (α ≈ 1)

2.2 Heterojunction bipolar transistor (HBT)

HBT section: Figures were updated. Text is from old lecture notes.

The cutoff frequency of the bipolar transistor was limited by

• cutoff frequencies of bipolar transistor: limited by rb and db

• opposite requirements: db small for high ωα , db large for low rb

• Solution: bipolar transistor featuring emitter-base heterojunction (Figs. 2.8 and 2.9)

– E: n-AlGaAs, B: p-GaAs, C: n-GaAs

– bandgap E-B → holes in B do not reach E ,→ γ ≈ 1
Chapter 2. Microwave transistors 23

Emitter Base Collector Emitter Base Collector

E E

n-AlGaAs p-GaAs n-GaAs (barrier)

x x

Figure 2.8: Band diagram of an HBT under (a) flat-band condition, (b) forward-active bias.

– base current mainly from recombination current

– base width db very small (< 0, 1µm) ,→ τec &, fT %
→ large Nb to reduce rb (∼ 1019 cm−3 )

• Other materials:

– E: n - Si
– B: p - Si/Ge
– C: n - Si
– Ge shapes band diagram and has high µp

• Additional advantages of HBT:

n-AlGaAs (emitter)
Implanted n+-GaAs
E p+-GaAs (base)
isolation
B B
n-GaAs (collector)
C

n+-GaAs

Semi-insulating GaAs substrate

Figure 2.9: Typical HBT structure.

– vertical structure → fewer surface states compared to planar devices (e.g. FET)
→ lower 1/f-noise (important for nonlinear applications like oscillators). HBT has
similar performance to bipolar transistors, ca. 10 dB better than FET devices.
– high power since whole emitter cross section can be used.

• Performance:

– AlGaAs/GaAs-HBT: fT = 105 GHz, fmax = 175 GHz,

24 2.3. Metal semiconductor field-effect transistor (MESFET)

– Pout |10GHz ≈ 4W per mm emitter stripe length; (=

ˆ 0.4 W CW at ltotal = 100µm).

Pout − Pin
→ ηadd = 48%
PDC

(”power added efficiency”)

fmax may be reduced to yield higher Pout :

fmax ↑→ Pout ↓
fmax ↓→ Pout ↑

– under development:
∗ Si/SiGe: npn and pnp (complementary) for fast logic.
∗ InAlAs or InP emitter with InGaAs base on InP substrate. Expectation: higher
µn in base, higher thermal conductivity of substrate material. → αe ≈ 10000
has been realized.

• Applications: digital IC, high power microwave devices, monolithic optoelectronic switches.

2.3 Metal semiconductor field-effect transistor (MESFET)

MESFET section: all material from old lecture notes.

[7, 5, 9, 10]

S G D

Rg

C gs Cgd
Rs rg
gm
CD gd

Figure 2.10: Elements of equivalent circuit model.

Fig. 2.10 illustrates the physical model of the FET and Fig. 2.11 the resulting equivalent circuit
model.

• RF operation: drain conductance gd unchanged

• Transconductance gm
gm,static
gm ≈ for ωτs  1
1 + jωτs

τs : transit time
Chapter 2. Microwave transistors 25

S*Vgs

Rg C gd innerer FET
G D
Cgs
V gs gd
CG rg gm*Vgs
CD
RG Rs

S
Figure 2.11: RF equivalent circuit model for MESFET.

• Estimation of τs :
l VDS
τs ≈ , v̄D = µn
v̄D l
l
τs ≈ , if Es (saturation) over whole channel
vs
• Additional RF elements:

– gate-source capacitance:
Z l
dx
Cgs = ε · a ·
0 w
a - gate width, w = w(x) - depletion layer width, l - gate length

ε·a·l
Cgs ≈ (2 . . . 3) · C0 , C0 = depletion-layer capacitance for w = d
d

– rg : equivalent resistance (channel =

ˆ RC lowpass network)
 2
→ from rg , Cgs : time constant, estimation τg = ε
σ · l
d
– gate-drain:
Cgd : fringe capacitance at gate edge

Cgd 1pF
Cgd ≈ ε · a for εr = 12 : =
a cm
Outer elements: path- and contact resistances, fringe capacitances.
– source resistance: Rs path resistance between source and channel. Rs divides input
voltage
→ reduced drive level, feedback.
Rs small → S - contact near channel.
– gate - resistance Rg
0

series resistance per unit length of gate contact

gate contact small → Rg high (limited electrolytic reinforcement)
0
26 2.3. Metal semiconductor field-effect transistor (MESFET)

e.g. Rg = 10kΩ /cm

0

Gate contact =
ˆ transmission line with Rg0 , Cg0 and length a (G0 , L0 ≈ 0).
v
u Rg0
u
chracteristic impedance: Zg = t 0
jωCg

propagation constant: γ =
q
jωCg0 · Rg0
input impedance (open circuit): ZE = Zg · coth(γ · a)
1
0
Rg · a
for |γ · a|  1 → ZE ≈ +
0
jωCg · a 3
0
Rg · a
Rg = Re{ZE } =
3
input voltage divided between Rg and inner FET (like Re {ZE }/ Im {ZE }) ,→
division ∼ a2 ⇒ gate width small → several gate contacts in parallel.
e.g.: two gate contacts in parallel:
0
Rg · a
,→ Rg = ≈ 30Ω a: total gate width
12
at a/2 = 200 µm
– fringe capacitances CG , CD at gate and drain, feed lines and contacts:
e.g.: GaAs-FET, l = 1 µm, a = 300 µm

gm = 30 mS Cgs = 0, 4 pF rg = 3Ω Rg = 2Ω
gd = 2mS Cgd = 0, 01 pF τs = 3 ps Rs = 5Ω

• Frequency limits

– transit time delay and attenuation:

gm,static 1
|gm | = √ at ωs =
2 τs

– transit frequency fT and short-circuit current gain αSC :

with simplified equivalent circuit

i2
G D
i1
Vgs gm*Vgs gd V2

S S
Figure 2.12: Simplified equivalent circuit model of FET.

i2 gm
αSC = =
i1 V2 =0 jωCgs
Chapter 2. Microwave transistors 27

1 gm
|αSC | = 1 at f = fT = ·
2π Cgs
gm gm gd
ωT = 2πfT = = · =µ·B
Cgs gd Cgs
– µ: open-circuit voltage gain, B: bandwidth.
µn
with ωT 0 = 2πfT 0 = 2 · Vp · 2 Vp : pinchoff voltage, µn : mobility
l
and gm , Cgs from detailed physical model (Fig. 2.13):

l or Vs large: ,→ fT ∼ l−2
l or Vs small: ,→ fT ∼ l−1

Vs = Es · l
Es : saturated electric field
vg : normalized gate voltage

1
2
fT ~ 1/l
0.5
ug=
0.1
fT ~ 1/l
0.6

0.1
fT/fTO
0.05
0.9

0.01
0.1 0.5 1 5 10
VS/VP

Figure 2.13: transit frequency

FET - noise

[11]

• Source:

– thermal noise of channel resistance and parasitic resistances.

– shot noise of residual currents within schottky contact.
reverse current small → negligible.
– recombination noise: generation- and recombination noise at defects and surface
states. Spectrum ∼ 1/f : not important in linear applications
in nonlinear applications 1/f-noise is upconverted to RF by nonlinear Cgs and can
severly impair performance.
28 2.3. Metal semiconductor field-effect transistor (MESFET)

gd- dV gm,+ (dV-dI/g d-)

S

dI
(a) (b)

Figure 2.14: Thermal noise of FET: (a) geometry and (b) equivalent circuit model

• Thermal noise of FET channel (Fig. 2.14)

– static case
– thermal noise in dx:

dx dx
dV 2 = 4kT BdR, dR = = √
σa(d − w) σad(1 − v)
VD + Vsg + V (x)
v(x) = ;
Vp
VD : diffusion potential,
Vsg : source-gate voltage,
V (x): forward voltage drop over channel
dV between FET (0 . . . x), unsaturated:

ad q
gd− = σ
[1 − v(x)]
x
gate: short-circuit → no current source
and FET (x . . . l), saturated:

gd = 0
ad q
gm,+ = σ [1 − v(x)] for Vs → ∞
l−x
– gm,+ is controlled by voltage dV − gd− .
dI

dI dV
 
→ dI = gm,+ dV − → dI = 
+

gd− 1 1
gm,+ gd−
q dx
⇒ dI 2 = 4kT BG0 [1 − v(x)]
l
ad
=
ˆ noise current of dx , G0 = σ conductance of open channel
l
– For integration: dx = G0 lVp (1 − v(x))dv/I transformed to dv
p

1 1
Z q
I 2 = 4k T B G20 Vp (1 − v(x))2 dv = 4kT Bgm P (vg ), vg = v(0)
I vg

0.5 ≤ P (vg ) ≤ 0.67, typical P (vg ) ≈ 0.65

Chapter 2. Microwave transistors 29

2.4 High electron mobility transistor (HEMT)

[12, 13, 14, 15]

In GaAs-MESFET technology, a shorter gate length l enables operation towards higher frequen-
cies. The channel depth, however, must be reduced accordingly, so that the gate still controls
the whole channel. This, in turn, leads to a smaller device cross-section, which increases the
channel resistance and limits the current/power capability of the device. Commonly, the current
capability is restored by means of higher doping ND , which lowers the channel resistance. This
method reaches a limit when the donor concentration ND is so high that Coulomb-scattering of
electrons at the donor atoms severely affects electron mobility µn and saturated drift velocity vs .
At this point the scaling limit of MESFET technology is reached.
In the late 1970s, researchers observed carrier transport parallel to an AlGaAs/GaAs hetero-
junction with improved electron mobility and saturated drift velocity. Subsequently, in the early
1980s, the first field-effect transistors based on this effect were demonstrated. The resulting
device is now commonly known as high electron mobility transistor (HEMT). Other names, such
as HFET1 or MODFET2 are also sometimes used. HEMT technology matured in the 1990s and
is now state of the art for microwave and mm-wave field-effect transistors in both low-noise and
power applications.

n-AlGaAs i-GaAs
y

W 2-D electron gas

Figure 2.15: Energy band diagram of an AlGaAs/GaAs heterojunction at thermal equilibrium.

Let us first consider an AlGaAs/GaAs heterojunction (Fig. 2.15): A heavily doped (ND ≈
1017 cm−3 ) n-AlGaAs-layer is brought in contact with an undoped GaAs-layer. AlGaAs has a
wider bandgap than GaAs, which leads to discontinuities ∆WC and ∆WV in the conduction- and
valence band, respectively. In the conduction band, a quantum well is formed in the GaAs-layer
close to the heterojunction. The donors in the AlGaAs-layer provide free electrons, which move
to the lowest possible energy state, and hence, accumulate in the quantum well, where they
are confined. The carriers form a charge sheet, which is only a few nanometers thin and also
called two-dimensional electron gas (2DEG). Carrier transport is only possible lateral to the
heterojunction.
The main benefit is that the AlGaAs-layer can be heavily doped so that enough carriers are
available to the channel, which enables high transconductance gm and high saturated drain current
IDSS in a transistor. Here, the low electron mobility in the AlGaAs-layer (µn ≈ 4500 cm2 /Vs
in the above example) is tolerated because the electrons are mainly located in the undoped
GaAs-layer, where the impurity concentration is low, resulting in higher vs and µn . With this
structure, higher fT and fmax (compared to a GaAs MESFET with same gate length) are
achieved.
1
Heterojunction FET
2
Modulation-doped FET
30 2.4. High electron mobility transistor (HEMT)

The equivalent circuit model of the HEMT is essentially the same as for the MESFET so that
the transit frequency is written as
gm gm
fT = ≈
2π(Cgs + Cgd ) 2πCgs
Thus, for a given gate length, the higher transconductance of the HEMT leads to the improved
transit frequency.

2.4.1 Classic device structure

G
S D W

n+-GaAs n+-GaAs
50...100 nm n-AlGaAs
n+ n+
10...30 nm i-AlGaAs
2-D electron gas
1 µm 2DEG
i-GaAs
GaAs
300 µm
S.I.-GaAs substrate
y

Figure 2.16: Device structure of a ‘classic’ HEMT.

In principle, a HEMT can be fabricated by adding gate-, drain-, and source-contacts on top of
an AlGaAs-layer which is located above a GaAs substrate. In practice, a few more steps are
added for better performance, as is shown in Fig. 2.16:

• A buffer layer of undoped GaAs is epitaxially grown onto the semi-isolating GaAs substrate.
The buffer has very high (monocrystalline) quality and ensures that substrate impurities
cannot diffuse into the channel during the fabrication.

• At the heterojunction, a thin spacer layer of undoped AlGaAs is used to further separate
the impurity atoms of the doped AlGaAs from the channel.

• The n-AlGaAs-layer is highly doped for high transconductance gm and high saturated
drain current IDSS .

• Drain and source need a low resistance path down to the channel, which is achieved by
highly doped regions. Ohmic source and drain contacts are fabricated with gold-germanium
(Au/Ge) alloys.

• At the gate, a Schottky contact is fabricated with Ti/Pt/Au alloys. The parasitic gate
resistance which results from small gate lengths can be lowered when ‘mushroom-’ or
T-shaped gate electrodes are employed.

2.4.2 Pseudomorphic HEMT (PHEMT)

One modification of the HEMT structure uses an AlGaAs/InGaAs heterojunction, where

the charge carriers accumulate in the undoped InGaAs-layer. The indium fraction x in the
Inx Ga1−x As compound can be varied; higher x leads to better µn and vs and also lowers the
Chapter 2. Microwave transistors 31

bandgap of the InGaAs compound. It can be seen in the energy band diagram of Fig. 2.17 that
there are now two energy band discontinuities, and that the quantum well is deeper and more
pronounced, which leads to higher carrier concentration in the 2DEG compared to the classic
AlGaAs/GaAs HEMT.

G
S D W

n+-GaAs n+-GaAs
50...100 nm n-AlGaAs
n+ n+
10...30 nm i-AlGaAs
5...10 nm

1 µm 2DEG
i-GaAs
GaAs
300 µm
S.I.-GaAs substrate
y
i-InGaAs

Figure 2.17: Device structure of a pseudomorphic HEMT (PHEMT)

However, the indium fraction cannot be varied arbitrarily: the lattice constant of Inx Ga1−x As
increases with x and is mismatched to the lattice constants of GaAs-buffer or AlGaAs-spacer
(which are similar). For example, with x = 0.22, there is 1.5 % mismatch between InGaAs and
GaAs/AlGaAs.
In general, the interface of two lattice-mismatched crystals exhibits defects. These appear as
additional energy states and can lead to degraded noise figure or memory effects. In the PHEMT,
they are avoided by making the InGaAs-layer very thin (about 10 nm), so that its lattice is
squeezed to fit onto the GaAs or AlGaAs lattice. Hence, the name ‘pseudomorphic’3 HEMT.
The induced strain can only be compensated up to the critical thickness, where defects start to
occur.
The PHEMT with GaAs buffer layer depicted in Fig. 2.17 is a called single-heterojunction
HEMT. In contrast, the double-heterojunction HEMT has a doped AlGaAs-layer also beneath
the undoped InGaAs-layer. This has two advantages: firstly, the bandgap discontinuity between
InGaAs and AlGaAs is greater than between InGaAs and GaAs (i.e., better carrier confinement),
and secondly, the doped lower AlGaAs-layer may also provide free electrons (i.e., higher carrier
concentration in the 2DEG).

2.4.3 High performance materials

Nowadays, the GaAs PHEMT is widely used in low-noise or power applications up to mm-wave
frequencies. However, when performance is critical, or when the operating frequencies are beyond
100 GHz other material systems, such as the InP HEMT, maybe used.
In Fig. 2.18a, the InP HEMT structure is shown. The device is grown on an InP substrate, with
InAlAs buffer, spacer, and donor layers. When the InGaAs channel is grown on the InAlAs
buffer, the indium fraction can be increased significantly (up to x = 0.80), leading to higher µn
and vs than possible with the AlGaAs/GaAs PHEMT. The advantages and disadvantages can
be summarized as follows:

+ Exhibits very low noise figure.

3
Pseudomorphic can be loosely translated from Greek as “having false shape”
32 2.4. High electron mobility transistor (HEMT)

G SiN passivation G
S D S D
n+-InGaAs n+-InGaAs n+-GaN n+-GaN

n-GaN
n+ n+ n+ n+
n-AlGaN

2DEG 2DEG
i-GaN

S.I.-InP substrate S.I.-SiC or -Si substrate

(a) InP HEMT (b) GaN HEMT

Figure 2.18: Device structures on high-performance materials.

+ Can be operated up to the THz region. InP HEMTs with fmax > 1 THz were demonstrated
recently. A 0.48 THz amplifier using this particular technology was also reported.
- The breakdown voltage is low (low power).
- InP wafers are expensive and small (3 or 4 inches), which means high initial cost.
- InP is a brittle material which makes processing difficult. This further adds to the cost
and results in low yield.

InP HEMTs are used in high performance ‘niche’ applications where the high fabrication cost is
justified.
HEMT technology continues to evolve with the GaN HEMT as a promising candidate for
high power applications. The structure is depicted in Fig. 2.18b and uses an AlGaN/GaN
heterojunction on silicon-carbide (SiC) or silicon (Si) substrate. GaN has a very wide bandgap of
Wg = 3.4 eV, and Alx Ga1−x N has an even wider bandgap (for example, Wg = 4.1 eV for x = 0.3).
Similar to the InP HEMT, the µn and vs increase with the Al fraction.
The advantages and disadvantages can be summarized as follows:

+ The maximum power density is about one order of magnitude higher than that of GaAs
PHEMTs.
+ Because of the wide bandgap, GaN HEMTs can be operated at much higher temperatures
than GaAs, SiGe, or silicon devices. Furthermore, the SiC substrate has very good thermal
conductivity and can be used as heat sink.
+ The wide bandgap also ensures good carrier confinement in the quantum well.
+ The low ionization rate leads to high breakdown voltages, commonly more than 100 V in
GaN HEMTs. Supply voltages of >30 V are possible, a huge improvement compared to
8–12 V for GaAs. In power amplifiers, higher voltage amplitudes mean higher impedance
levels. With an impedance level close to the system impedance, a broadband match is
obtained more easily.
− The reliability is not yet comparable to GaAs PHEMT processes.
− GaN HEMT technology is still expensive.
Chapter 3

Circuits

3.1 Matching Networks

Impedance matching in linear amplifiers often refers to conjugate matching, which maximizes
the power transfer from generator to load. In contrast, the noise match and the power match
(treated in later sections) are cases where a non-conjugate match is desired. Yet, regardless of
these differences, the design method is the same, and the problem can be differentiated into:

Input or output match, which refers to the impedance match between the amplifier input
or output and the external circuitry. It requires the transformation of the (usually real)
system impedance to the (generally complex) desired source or load impedance.

Inter-stage match, which refers to the impedance match between subsequent amplifier stages
in a multi-stage amplifier and requires the transformation of the (generally complex) output
impedance of one stage to the (generally complex) source impedance of the next stage.

The required impedance transformation is done by the matching network. Depending on the
operating frequency and the available technology, their realization differs:

• Lumped components are commonly used at low frequencies (i.e., up to several hundred
MHz). Standard surface-mount components are usually not designed for use at higher
frequencies where they may become lossy or exhibit resonances. While specially designed
components or standard components with small form factor are sometimes usable up to
several GHz, they require more care during design and fabrication.

• Transmission lines are employed in the GHz range because of their low loss and small size
(their dimensions become smaller at higher frequencies).

• Hybrid technology refers to the use of surface-mount components (packaged chips, bare
dies, lumped components) and transmission lines such as microstrips or striplines on the
same planar circuit. Thin-film- or thick-film technology are advanced variants of this where
lumped components are fabricated by deposition (thin-film) or screen printing (thick-film)
of resistive, dielectric, and metallic layers.

• Monolithic integration refers to the integration of active and passive components on a

single chip located on a large semiconductor wafer. The semiconductor process usually
provides lumped components (thin-film resistors and capacitors, spiral inductors) but one

33
34 3.1. Matching Networks

-j

Figure 3.1: Narrowband matching (a) and design (b)

may also use transmission lines. Chip area or Q-factor often determine whether a lumped,
distributed, or hybrid implementation is preferred.

3.1.1 Narrowband Matching

Section 1.4 gives formulas for optimum source and load reflection coefficients in Eqs. 1.46 and
1.47. A simple matching procedure for unilateral (S12 = 0) or approximately unilateral (S12 ≈ 0)
devices is also given in Eq. 1.53. From the reflection coefficients one can then determine the
desired source and load impedances.
A simple narrowband matching network may be implemented by a lumped LC-section, or by a
distributed single-stub L-section (“L” describes the shape of the microstrip lines).
From here on: old text.

Design of matching network with admittance diagram. Transformation of admittances using the
shortest path → small ω-dependence
error by approximation S12 = 0, Gm instead of G0m :
for example

|S11 | = 0.38 |S12 | = 0.13 δ = 0.12

)

|S21 | = 2.5 |S22 | = 0.57 0.8 < GG0m

m
< 1.3

3.1.2 Broadband Matching

networks between input, output and amplifier stages for:

1. impedance transformation
2. gain flatness

power gain = f(ω) ∼ 1

ω2
for ω ≥ (1 − α) ωe,c , decreases by 6 dB/octave.
matching network with attenuation for compensation of amplifier gain response
1st method: transformation network using reactances (L,C) (Fig. 3.2)
→ attenuation by ω-dependent mismatch
Chapter 3. Circuits 35

18.15 nH 3.2 pF 3.8 pF 80 pF

1 nH 2.5 nH

18.25 nH 3.13 pF 6.8 nH 5 pF

Transistor input

Figure 3.2: Matching network of lossless elements.

transistor input
attenuation

frequency

(a) (b)

Figure 3.3: Matching network of lossy elements (a) equivalent circuit model and attenuation
(b) impedance.

• frequency dependence of reflection is given

• approximation by rational function of jω

• continued fraction expansion

• synthesis of network, equivalent circuit model of transistor input must be taken into
account.

2nd method: transformation network using R, L, C (Fig. 3.3)

• → matching + equalization of gain

• no synthesis procedure → empirical design (CAD).

From here on: edited material.

36 3.1. Matching Networks

Bode-Fano Limit

Real-to-real impedance transformations are often easy to achieve. Narrowband matching is

achieved with a single quarter-wave line, and for broadband matching several cascaded quarter-
wave lines or a line which gradually changes its impedance along the structure (“tapered line”)
may be used.
The actual challenge in broadband matching are real-to-complex or complex-to-complex trans-
formations because complex impedances are naturally frequency-dependent. This makes it
challenging to provide a good match over a broad frequency band. Thus, the following text
refers to matching of complex loads only.
We have seen that narrowband matching results in a perfect match at one single frequency. It is
also possible to achieve a perfect match at several frequencies but one observes that between
these frequencies reflections become large. A broadband match cannot be achieved in this way,
and, as it turns out, it is impossible to get a perfect match over any bandwidth greater than
zero.
As a consequence, the reflection coefficient must be traded for bandwidth: matching over a
certain bandwidth comes at the cost of a minimum reflection coefficient |Γmin | which must be
tolerated. When the matching network is at the limit of what is physically realizable, any
attempt to get less reflections than |Γmin | at one frequency leads to reflections greater than
|Γmin | at other frequencies. The ideal limiting case is therefore characterized by total reflection
(|Γ| = 1) outside of the desired bandwidth and by a constant reflection of |Γmin | within the
desired bandwidth. The resulting rectangular response is depicted in Fig. 3.4. It is equivalent to
the response of a lossless Chebyshev matching network with infinite order, and can thus only be
approximated in practice.
Regardless of the actual response, Bode and Fano proved that a network which matches a
shunt-RC or series-RL load always fulfills
1
Z ∞
π
ln dω ≤ , (3.1)
0 |Γ(ω)| τ
where τ is the time constant of the load (τ = RC, or τ = L/R, respectively. Expressions for
shunt-RL or series-RC networks have also been derived.

|Γ|
1
ln(1/|Γ|)

area =
Γmin
0

Figure 3.4: Limiting case of the reflection coefficient of a broadband matching network.

When the idealized response is plugged into the above equation, the relation becomes
1 π
ln · ∆ω ≤ , (3.2)
|Γmin | τ
leading to
π
 
|Γmin | ≥ exp − . (3.3)
∆ω · τ
Chapter 3. Circuits 37

Therefore, for simple shunt-RC or series-RL loads the limit of the reflection coefficient can be
determined from the desired bandwidth.

3.2 Balanced Amplifiers

As discussed in the previous section, broadband amplifiers with flat gain may have significant
input mismatch at those frequencies where the gain is too high and thus reduced by lossless (i.e.,
reflective) input matching networks. Furthermore, in low-noise amplifiers or power amplifiers
the input or output return loss is commonly traded for other design parameters (more about
that in later sections). The balanced amplifier first introduced in 1965 [16] is a solution to the
problem of input/output mismatch but also has other important advantages.

Amp1
1 2 4' 3'
Input
Termination

Amp2
Termination Output
3 4 2' 1'
Quadrature couplers

Figure 3.5: Balanced amplifier circuit topology using a 90degree hybrid in microstrip technology.

The circuit topology is shown in Fig. 3.5. It consists of two amplifiers which are operated in
phase-quadrature, which means that the coupler at the input splits the input signal at port 1
into two equal parts with each −3 dB lower power and with a phase shift of 90degree between
port 2 and port 4. The split signals are then amplified by the individual amplifiers. Since by
reciprocity, a power splitter is always also a power combiner, the same type of coupler attached
to the output is used to combine the two amplified signals with 90degree phase shift at port 2’
and 4’ into one output signal at port 1’.
The illustrated coupler is called branchline coupler but any coupler with 90degree phase shift
between two ports will work. Such couplers are ideally characterized by the following S-
parameters:
0 −j 0 −1
 

1  −j 0 −1 0 
S=√  . (3.4)
 
2  0 −1 0 −j 
−1 0 −j 0

Assuming that port 1 of the input coupler √ is excited by a wave

√ a1 = 1, the wave appears at
the amplifiers’ input ports as b2 = −ja1 / 2 and b4 = −a1 / 2. Each amplifier then reflects a
portion of its input signal back to the coupler according to its input reflection coefficient (Γ2 or
Γ4 ). Finally, port 3 is matched, so that all incident waves read

1
   
a1
Γ b  −jΓ /√2
a =  2 2 =  2
. (3.5)
   
 0   0√ 
Γ4 b4 −Γ4 / 2
38 3.2. Balanced Amplifiers

So the reflected waves b2 and b4 cause the incident waves a2 = Γ2 b2 and a4 = Γ4 b4 which
simultaneously travel back through the coupler. The reflected waves resulting at port 1 and port
3 may be computed by superposition:
√ √  
Γ2 b2 · (−j/ 2) + Γ4 b4 · (−1/ 2) (Γ4 − Γ )/2
 
2
b2   −j/√2 
b= = (3.6)

 √ √  
.

Γ2 b2 · (−1/ 2) + Γ4 b4 · (−j/ 2) j(Γ2 + Γ4 )/2
√
b4 −1/ 2

The result is that at port 1, the reflected wave is half of the difference of the amplifiers’ individual
reflection coefficients. Ideally, the transistors are equal, and from Γ2 = Γ4 follows b1 /a1 = 0. In
practice, the two amplifiers are not equal but quite similar, so that the resulting input reflection
is small (b1 /a1 ≈ 0). On the contrary, the reflected wave at port 3 is related to the sum of Γ2
and Γ4 . When both transistors are equal, all reflected power is absorbed in the termination of
port 3.
Without loss of generality, let amplifier 1 have gain G = V and amplifier 2 have gain G = 1.
Then, the incident waves at the output coupler are

Γ01 b01
 
 −1/√2 
a0 =  , (3.7)
 
 0√ 
−jV / 2
√ √
where Γ01 is the load reflection coefficient, and b01 = (−jV / 2)S14 + (−1/ 2)S12 = 2j (V + 1) is
the wave traveling towards the load. Then, the reflected waves at the output coupler are

j(V + 1)
 

1  (V + 1)Γ0 /√2 
b0 =  1
(3.8)
 
2 (1 − V ) √ 


−j(V + 1)Γ01 / 2.

As a result, in the ideal case where the load is matched (Γ01 = 0) and both amplifiers are
equal (V = 1), the complete output power appears at port 1’. With two unequal but matched
amplifiers, the worst case is when one of them fails so that V = 0. The output power is then
divided equally between port 1’ and 3’. Thus, balanced amplifiers are somewhat robust against
device tolerances or failure, provided that the termination at port 3’ can absorb that much
power. With two equal but mismatched amplifiers (Γ01 = 6 0) the reflected power at port 1’ is
split and half of it arrives at each amplifier output.
The main advantages of balanced amplifiers may be summarized as follows:

• Due to the isolation of the couplers (S24 = 0), the amplifiers are decoupled at the input
and output. This reduces the risk of instabilities due to signals traveling between the two
parallel amplifiers.

• Multiple stages of balanced amplifiers may be cascaded. Since ideal balanced amplifiers
are fully matched, there are no reflected signals traveling between stages, which reduces
the risk of instabilities due to load mismatch.

• As described above, the balanced amplifier is robust against device tolerances. Yet, each
individual amplifier may be trimmed to make the circuit more symmetric.
Chapter 3. Circuits 39

• The balanced amplifier is also a power-combining technique. The power which can be
handled by a certain amplifier (or transistor) is often limited but this limit can be doubled
by using two amplifiers (transistors) in parallel.

While the drawback is greater circuit complexity, especially the area required by the couplers,
the advantages often justify the additional effort.

3.3 Low-noise amplifiers (LNAs)

Amplifiers employed at the input of receiver front-ends often handle very small signals. In order
to achieve such high sensitivity, the received signal must be distiguishable from the noise power
present in the same frequency band. Thus, the amplifier itself should generate as little noise as
possible—hence the name low-noise amplifier (LNA).
The noise properties of two-ports are described by the noise figure, which, in the case of bipolar
transistors in common-emitter topology, and at medium frequencies, may be written as [7]

rb + 12 re Pn,av
F =1+ + (3.9)
RS kT BG
where rb and re are the intrinsic base and emitter resistances, respectively, and RS is the
generator impedance. The second term in Eq. 3.9 describes the contributions due to thermal
noise of rb and due to shot-noise of the emitter–base junction. The third term describes the
noise contribution due to base recombination. The factor Pn,av ,
GL
Pn,av = q(1 − β)Ic Brc2 , (3.10)
2
is the available noise power, which is delivered within the bandwidth B when the two-port is
matched to the load admittance GL . Then, G is the power gain under this condition, Ic is
the collector current at the operating point, and β is the DC base transport factor. Note that
Eq. 3.9, particularly the last term, is still valid under mismatched conditions because the noise
figure of a two-port is independent of the load.
Yet, to keep it simple, we consider the case of a two-port matched to GL for which we use
Eq. 3.10 along with the formula for the associated power gain,

4α2 rc2 RS GL
G= , (3.11)
|(RS + rb + re )(1 + rc GL ) − αrc GL (RS + rb )|2
and the expression for F may be rewritten as

rb + r2e (RS + rb + re )2 (1 − β)qIc

F =1+ + , (3.12)
RS RS 2α2 kT
from which the it can be seen that the noise figure depends on RS . For transistors at higher
frequencies, the equation essentially stays the same, albeit with the following substitutions:

rb → rb + rbs ,
RS → Re{ZS },
GL → Re{YL },
G → GU ,
40 3.4. Nonlinear distortions

where rbs is the base contact resistance, ZS is the complex generator impedance, YL is the
complex load admittance (matched to the noise source Pn,av ), and GU is the unilateral gain
when the output is matched. The noise figure then becomes

rb + rbs + r2e q(1 − β)Ic

F =1+ + . (3.13)
Re{ZS } 2kT GU Re{YL }

For low-noise operation, the base resistances must be minimized, and the bias should be
chosen such that (1 − β)Ic is small. The noise figure also depends on GU and the real part of
the matched load admittance Re{YL }, which are intrinsic parameters of the transistor. The
frequency-dependence of F mainly stems from the unilateral gain GU (ω), which, above the
α-cutoff frequency, ω > (1 − α)ωec , falls off with GU (ω) ∼ 1/ω 2 . Consequently, the noise figure
increases with F (ω) ∼ ω 2 , i.e., with 20 dB per decade.
As already pointed out, the noise figure depends on RS = Re{ZS }: for small Re{ZS } the noise
contributions of thermal and shot noise dominate, whereas for large Re{ZS } recombination is the
major noise source. In between, the noise figure is minimal with F = Fmin . Choosing Re{ZS }
such that F = Fmin is called noise match. Due to GU ∼ 1/ω 2 , the generator resistance decreases
with increasing frequency. In general, the generator resistance for the noise match is different
from the generator resistance for the conjugate (gain-) match, resulting in a trade-off between
gain and noise figure. The lost gain may be compensated by using more amplifier stages but the
noise match also leads to higher input reflections. This problem can be solved by designing the
LNA as balanced amplifier, or by using an isolator or circulator at the input.

3.4 Nonlinear distortions

Semiconductor devices such as diodes and transistors are essentially nonlinear. Under small-
signal conditions their characteristics may be linearized by taking the slope at the operating
point. For example, the nonlinear Vgs –Ids relationship of a FET is often approximated by the
linear transconductance gm = ∂Ids /∂Vgs . This linearization is useful as it allows us to apply
the methods of linear circuit theory (e.g., S-parameters or stability factors). However, under
large-signal conditions, i.e. with increasing signal level, larger portions of the nonlinear device
characteristic are driven. The regime is no longer linear, and the principle of superposition, which
is inherent to linearity, is not applicable any more. This means that one can not characterize a
nonlinear circuit with a simple frequency-domain transfer function. The analysis of nonlinear
circuits must be conducted in the time domain.
In the following, some theoretical and practical aspects of nonlinear circuits are introduced. Our
emphasis is on weakly nonlinear effects, which are discussed in more detail in [17].

3.4.1 Weakly Nonlinear Effects

Obviously, the transition between a linear and a nonlinear regime is gradual. In this context,
weakly nonlinear effects originate either from mildly nonlinear device characteristics or from
moderate drive levels. Without loss of generality, the example depicted in Fig. 3.6 is considered
in the following.
This simple circuit consists of a voltage source VS , a series resistance R, and a device, for
instance a diode, with a nonlinear current–voltage characteristic I = f (V ). The function f
may be expanded into a power series about the DC operating point and truncated to yield the
Chapter 3. Circuits 41

R I

VS V

Figure 3.6: Example nonlinear circuit.

approximation:
I = f (V ) ≈ aV + bV 2 + cV 3 . (3.14)

The truncation—here after the cubic term—determines the order of the approximation. In the
following the generator voltage is assumed to be a two-tone signal

VS (t) = V1 cos(ω1 t) + V2 cos(ω2 t). (3.15)

We first consider the case R = 0, in which the voltage drop across the nonlinearity equals the
generator voltage, V (t) = VS (t), and the various terms of the expansion are

ia (t) = aVS (t)

= aV1 cos(ω1 t) + aV2 cos(ω2 t), (3.16)
ib (t) = bVS2 (t)
b

= (V12 + V22 ) + V12 cos(2ω1 t) + V22 cos(2ω2 t) + 2V1 V2 cos((ω1 + ω2 )t)
2

+ 2V1 V2 cos((ω1 − ω2 )t) , (3.17)

and

ic (t) = cVS3 (t)

c 3

= V cos(3ω1 t) + V23 cos(3ω2 t)
4 1

+ 3V12 V2 [cos((2ω1 + ω2 )t) + cos((2ω1 − ω2 )t)]

+ 3V22 V1 [cos((2ω2 + ω1 )t) + cos((2ω2 − ω1 )t)]


+ 3(V13 + 2V22 V1 ) cos(ω1 t) + 3(V23 + 2V12 V2 ) cos(ω2 t) , (3.18)

where ia (t), ib (t), and ic (t) are, respectively, the linear, quadratic, and cubic contributions to
the total device current I(t):
I(t) ≈ ia (t) + ib (t) + ic (t). (3.19)

If R = 0, I(t) is called the short-circuit current. The two-tone voltage leads to a large number
of spectral components in the short-circuit current, and this number obviously increases with
the model order.
If R 6= 0 the generator voltage does not drop directly across the nonlinearity. A part of the
nonlinear device current drops across R, and the voltage drop across the nonlinearity is then
42 3.4. Nonlinear distortions

V = VS − RI, from which one concludes that V and I must have the same spectral components.
The resistor introduces feedback from I to V , so that the resulting spectra have, at least in
principle, an infinite number of components and must be calculated iteratively. In this case, the
power series approach discussed here often becomes impracticable.
The resulting spectral components are commonly called mixing products and may only appear
at frequencies
ωm,n = mω1 + nω2 with {m, n} ∈ Z2 . (3.20)

The mixing products ωm,n are of order N = |m| + |n|. A closer look at this definition may lead
to some confusion as the first-order terms of Eq. 3.16 and the third-order terms of Eq. 3.18 both
have spectral components at ω1 , which, by the above rule, has order N = 1. More precisely
speaking, mixing products which are caused by the N th-order power series term are called
N th-order mixing products, so at ω1 there are actually first- and third-order contributions.
However, the less precise rule is often used because in practice, these subtleties rarely matter:

• The cause of a certain mixing product is often not clear. This is particularly true for R 6= 0
because feedback from device current I to device voltage V makes it impossible to clearly
distinguish cause and effect.

• For small voltages V1 and V2 (i.e., moderate drive), higher-order contributions have
less power than lower-order contributions. Thus, in our example circuit, the first-order
contribution at ω1 is dominant.

3.4.2 Classes of nonlinear phenomena

A closer look at the current spectrum caused by the two-tone excitation allows us to identify
different classes of nonlinear phenomena.

Harmonics

Harmonics, i.e., the mixing products at mω1 and nω2 , may appear under single- and multi-tone
excitation and are often unwanted signals. However, in narrowband systems, they are out-of-
band and easily suppressed with adequate filtering. In some applications, such as in frequency
multipliers or power amplifiers, they are desired or play an important role in the design.

Saturation

The DC power which can be supplied to a nonlinear system is limited. As a result of power
conservation, the output power is limited as well and saturates at higher input power levels.
Consider a single-tone input signal, for instance at ω1 . The fundamental frequency component
of the current is:
3
 
iω1 (t) = aV1 + cV13 cos(ω1 t). (3.21)
4
For the truncated power series c < 0 ensures physical behavior (i.e., the current saturates at
higher amplitudes of V1 ). This is also known as gain compression or AM/AM conversion. An
important figure of merit is the 1 dB compression point, P1dB . It is defined as the output power
at which the gain is 1 dB smaller than under small-signal operation.
Chapter 3. Circuits 43

In the case of a two-tone input signal with V1  V2 , the ω2 -component of the current reads:

3
 
iω2 (t) = aV2 + c(V23 + 2V12 V2 ) cos(ω2 t) (3.22)
4
3 2
 
≈ a + cV1 V2 cos(ω2 t). (3.23)
2

It can be seen from the term 32 cV12 that a strong interfering signal at ω1 may even compress the
gain of a small signal at ω2 . This is also known as desensitization.

Intermodulation

Intermodulation (IM) is caused by mixing products of two- or multi-tone signals. The current
spectrum of Eqs. 3.16–3.18 is essentially an intermodulation spectrum. In general IM-products
have a low power level compared to the signals at ω1 and ω2 .
Even-order IM-products (e.g., of second order at ω2 − ω1 or at ω1 + ω1 ) are commonly out-
of-band and thus easily filtered at the output. Filtering at the input is equally important
because two strong out-of-band interferers could generate significant in-band intermodulation
products. Second-order IM-products matter in broadband systems (an octave or more), as
in-band signals/products alone may cause the above problems, which means that filtering is not
an option.
Odd-order IM-products (e.g., of third order at 2ω2 − ω1 and 2ω1 − ω2 ) are often in-band.
Therefore, they can generally not be suppressed by filtering. Among all odd-order products, the
third-order products have the highest power level and are thus commonly considered in amplifier
design.

Cross modulation

Cross modulation describes how the modulation of one signal affects other signals. To illustrate
the effect, consider the case of a small unmodulated signal V1 in the presence of a strong
amplitude-modulated signal V2 with modulation index |m(t)| < 1:

VS = V1 cos(ω1 t) + (1 + m(t)) · V2 cos(ω2 t). (3.24)

With V1  V2 , the ω1 -component of the cubic current term ic (t) becomes

3
ic,ω1 (t) ≈ cV22 (1 + 2m(t) + m2 (t)) · V1 cos(ω1 t), (3.25)
2
which is equivalent to an AM with m0 (t) = 2m(t) + m2 (t), i.e., a distorted replica of m(t). For
large values of cV22 , there can be significant cross modulation.

AM/PM conversion

AM/PM conversion refers to a situation in which the phase shift between input and output
changes with the amplitude of the input signal. The power series of Eq. 3.14 cannot explain this
behavior as all resulting terms may only have 0degree or 180degree phase. This is correct for
memoryless nonlinearities such as nonlinear resistances or transconductances but not generally
valid for reactive nonlinearities such as nonlinear capacitors.
44 3.4. Nonlinear distortions

In the presence of a nonlinear capacitor (e.g., the space-charge capacitance of a diode), there
may be a phase shift Θ between the first-order and third-order terms caused by a single-tone
signal at ω1 . The contributions are then written as phasors and the resulting current phasor at
ω1 becomes
3
iω1 = aV1 + cV13 exp(jΘ). (3.26)
4
The phase shift Θ generelly depends on V1 but even if Θ is independent of V1 , the phase of iω1
depends on the amplitude of the input signal.
AM/PM conversion is most critical in systems which employ phase- or frequency-modulation.

3.4.3 Intercept point

When plotting the fundamental output power versus the (available) input power in a double-
logarithmic scale (e.g., both powers in dBm), the slope in the small-signal region is one. Due
to saturation, the slope decreases towards zero for higher input powers. The higher-order
mixing products show similar behavior: their slope is characterized by their order (slope = n for
nth order), and they, too, saturate at higher input powers. This is shown in Fig. 3.7 for the
fundamental output power, and second- and third-order IM-products.

Pout [dBm]

OIP3
OIP2 Fundamental
2nd order
3rd order

IIP2 IIP3 Pin [dBm]

Figure 3.7: Definition of the intercept point.

The intercept point is defined as the intersection of the extrapolated IM-product line with the
extrapolated fundamental line. Second- and third-order intercept points are called IP2 and IP3 ,
respectively. When explicitly referring its value to the abscissa, it may be called IIPn , and when
referring to the ordinate, it may be called OIPn (input-referred or output-referred intercept
points).
Intercept points are an important figure of merit as their knowledge allows to easily calculate
the intermodulation ratio of a circuit. In general, in a logarithmic scale, the powers obey the
relation:
PIMn = nPin + P0 , (3.27)
where P0 is a device-dependent offset. With the gain G this can be reformulated as:

PIMn = n(Pout − G) + P0 = nPout + P00 (3.28)

At the intercept point, the condition

PIMn = Pout = OIPn (3.29)

!
Chapter 3. Circuits 45

is met. Eliminating P00 by means of

P00 = (1 − n)OIPn (3.30)

one finally obtains the desired expression:

PIMn = nPout − (n − 1)OIPn . (3.31)

From here on: material from old lecture notes.

With the (linear) gain, which reads

G = 10 · log(4a21 | H(ω) |2 RRL )

Pin can be eliminated. This yields:

9a23
!
PIM 3 = 10 log + 3Plin − 60
4a61 RL
2

and
2 a31
!
IP3 = 10 log RL + 30 .
3 a3

An example shall illustrate these results. Fig. 3.8 shows the power characteristics of a device
with a third order intercept point of IP3 = 20 dBm and a linear gain of G = 10 dB.

Pout / dBm

20
IP3

10
r
ea
lin

0
-10 0 10 Pin / dBm
r
orde
3rd

-10

Figure 3.8: Examplary power characteristics and 3rd order intercept point.

What is the maximum input power level Pin which still garantees intermodulation ratio of 20
dB?
46 3.5. Power Amplifiers

With the above one has

PIM 3 = 3Plin − 2IP3 .

This allows to rewrite the intermodulation ratio A = Plin − PIM 3 as

A = −2Plin + 2IP3 .

The relation Plin = Pin + G for the linear output power finally leads to the allowable input
power:
1
Pin = IP3 − G − A = 0 dBm
2

40

20
LINEAR
0

Pout
dBm -20 IM
3rd order

-40 f = 8 GHz
Df = 5 MHz
VDS = 10 V
-60

-80
-20 -10 0 10 20 30

Pin dBm

Figure 3.9: Measured intermodulation of GaAs-MESFET.

The analysis becomes more complicated for complex transfer nonlinearity or in presence of nonlin-
ear reactances. Fig. 3.9 shows the results of a more detailed investigation. The intermodulation
graphs are no longer straight lines. This is attributable to several causes:

• The FET exhibits several elements with higher order nonlinearity,

• The impedance for lowest distortion is a function of the input power,
• Nonlinear effects of different elements partly cancel each other which leads to the observed
fluctuations.

3.5 Power Amplifiers

The power amplifier (PA) is an important component in the RF front-end of transmitters. As a

performance bottleneck it largely determines many critical system parameters:

Output power requirements may vary from a few milliwatts to several hundred watts, de-
pending on the application. For high output power, large transistors or power-combining
techniques are used. Large power transistors are expensive (“$ per watt”), so that one
tries to get as much power from a device as possible.
Chapter 3. Circuits 47

Efficiency is important in systems which are operated from a battery or an otherwise limited
power supply (e.g., cell phones, satellites). But even in fixed transmitter sites (base stations,
TV broadcasting) efficiency is crucial because power consumption significantly contributes
to the operating cost.

Bandwidth requirements are commonly high in radar and military applications. Civil mobile
communication systems are traditionally narrowband but also there the trend towards
increasing data rates, or towards coverage of multiple wireless standards within one
front-end, has lead to higher bandwidth requirements.

Linearity is especially important when spectrum is expensive1 . In such scenarios out-of-band

signals due to intermodulation are unacceptable and must be kept at very low levels.
Furthermore, one often employs spectrally efficient modulation schemes (e.g., 64-QAM)
which require a linear amplifier. A moderately nonlinear amplifier may be linearized with
some sort of predistortion.

Most of this section’s material is covered in detail in [18].

3.5.1 Loadline Match Vs. Conjugate Match

We have seen that for low-noise amplifiers a conjugate match at the input does not generally lead
to a minimum noise figure. Similarly, for power amplifiers, a conjugate match at the output does
not generally lead to good performance in terms of efficiency or output power. This may seem
contradictory to fundamental principles of linear circuit theory, by which a conjugate match
always results in maximum power transfer from the generator to the load. However, linear
circuit theory relies on assumptions which must be questioned in the context of a nonlinear
design problem.
This becomes obvious when recalling that power transistors are often modeled as a voltage-
controlled current source (Fig. 3.10a). The conjugate match of such a generator is an infinite
resistance, and the power transfer to such a load would be unbounded. Even in the presence
of a large but finite internal resistance, the conjugate match still results in excessively large
voltage swings across the current source. In a real device, voltage clipping, current saturation, or
even device failure due to thermal stress or electrical breakdown may occur. So the problem of
maximum power transfer is in fact a constrained optimization problem, which can be stated as

maximize Pout while V (θ) ∈ [0 ; Vmax ] and I(θ) ∈ [0 ; Imax ], (3.32)

where V (θ) and I(θ) are, respectively, the voltage and current during one RF cycle, and Vmax
and Imax are the corresponding maximum ratings of the device. The loadline resistance is
the solution to the above problem. For a given power amplifier mode, it will maximize the
output power while staying within the voltage and current limits of the device. Recall the case
of a Class-A power amplifier, where one tries to stay within the linear region of the transfer
characteristic. The optimum loadline resistance is then determined by RL = Vmax /Imax and (as
usual), it depends only on the device maximum ratings. If the amplifier is not driven too hard,
both current and voltage are sinusoidal and swing from zero to the maximum rated value, which
ensures high output power.
1
One example of expensive spectrum is the revenue of about 50 billion Euros the German government made in
August 2000 by auctioning off the UMTS licenses. This amounts to about 400 Euros per Hz bandwidth.
48 3.5. Power Amplifiers

Lbw I '
ds
Ids Ids
Vgs gmVgs Vds Vgs Cds V'ds

(a) (b)

Figure 3.10: Models for field-effect power devices — (a) simple unilateral current source model;
(b) simple unilateral current source model with internal and external parasitics.

3.5.2 Passive Load-pull Measurement

In practice, finding the correct load impedance is more complicated than simply calculating a
loadline resistance because firstly, the device may not be unilateral, and secondly, the device has
parasitic resistances and reactances. For instance, Fig. 3.10b shows the inclusion of a (generally
nonlinear) output capacitance Cds , and a bondwire inductance Lbw in the device model. The
optimum loadline resistance at the current generator reference plane is then transformed to
a (generally) complex impedance at the package reference plane. Often the optimum load is
determined experimentally by a large-signal measurement technique called “load-pull”. Small-
signal (i.e., linear) measurement techniques such as S-parameters are not suitable for power
amplifiers because high power or high efficiency are only obtained when the amplifier is driven
at or beyond its 1 dB compression point at which the voltage and current waveforms are not
sinusoidal any more.

Pavs Pref Pout

Bias Control
Generator Load

Isolator Couplers Tuner1 DUT Tuner2 Coupler

Figure 3.11: Typical load-pull measurement setup.

A typical load-pull measurement setup is shown in Fig. 3.11.

• The device under test (DUT) could be a transistor or even a complete amplifier. Its
bias voltages/currents are provided by the bias control, from which one also obtains the
consumed DC power PDC .

• In their simplest form, the load tuners at the input and output are made of mechanically
tunable/movable pieces of transmission line. They can transform the 50 Ω system impedance
to almost any impedance on the Smith Chart, which the DUT then ‘sees’ at its input and
output.

• The couplers and corresponding power meters are used to measure the available generator
power, Pavs , the power reflected at the DUT’s input, Pref , and the delivered output power,
Pout . From these values, power gain, transducer gain, and input reflection coefficient may
be calculated.
Chapter 3. Circuits 49

• The generator provides the input signal. An isolator is used to protect the generator from
power which may be reflected at the DUT’s input or at the input tuner.

For the load-pull analysis, the Smith Chart is discretized into many points, each of which
represents a certain load impedance ZL . For each ZL , the following procedure is repeated:

1. Adjust the output tuner so that the DUT sees the load impedance ZL at its output.

2. At the input, a conjugate match is often desired: adjust the input tuner so that the reflected
power Pref is minimized. The delivered input power should be the same throughout the
whole procedure.

3. Record the output power, Pout , the delivered input power, Pavs − Pref , and the consumed
DC power, PDC .

Manually covering all required points on the Smith Chart can be quite time-consuming: for
each point, a network analyzer measurement of the tuner at the output is performed to change
its impedance to the desired value. However, automatic load-pull equipment exists, which uses
stepper motors and communicates with the network analyzer to calibrate or measure without
user interaction.
Once the Smith Chart is covered, points with equal output power or efficiency may be connected
to closed contours. The contours usually have a characteristic ellipsoid shape. An example
load-pull contour plot for output power is shown in Fig. 3.14.

3.5.3 Active Load-pull Measurement

The passive load tuner at the DUT output of the setup in Fig. 3.11 presents a reflection coefficient
ΓL which creates a wave that travels back towards the DUT output. The corresponding voltage
wave is shown as V2 in Fig. 3.12.

Input A
V3
Tuner DUT
Pin V2

Y0 Vin Vout Y0

Figure 3.12: Voltages and impedances at the DUT during load-pull.

Active load-pull is a measurement technique which does not need a passive output tuner. It is
based on the idea [19] that the DUT does not know whether V2 is caused by a mechanical tuner
or whether it is a signal which is generated (hence, ‘active’ load-pull) and then injected into the
DUT output. However, it is important that V2 is physically correlated to the original excitation
of the DUT, so that V2 may be derived from V3 or from the input voltage, Vin .
In the following, we look towards the DUT output and thus refer to
1 1
P2 = Y0 V22 and P3 = Y0 V32 (3.33)
2 2
50 3.5. Power Amplifiers

as injected and reflected power, respectively. Likewise, V2 and V3 are called, respectively, injected
and reflected voltage. The delivered output power is then written as

1
Pout = P3 − P2 = Re{YL }Vout
2
, (3.34)
2
where Vout = V2 + V3 is the voltage at the DUT output, and YL is the load admittance presented
to the DUT at the fundamental frequency. To determine YL , we define two reflection coefficients
in plane A of Fig. 3.12: Γ2 towards the DUT output and, towards the opposite direction, ΓL .
These are related by

Y0 − Y2 V3 Y0 − YL V2 1 Y0 + Y2
Γ2 = = and ΓL = = = = , (3.35)
Y0 + Y2 V2 Y0 + YL V3 Γ2 Y0 − Y2

from which we gather YL = −Y2 , i.e., the DUT’s dynamic output admittance Y2 and the load
admittance YL have opposite phase. This relation shows that when the presented load is passive,
the DUT is active (delivering net power), and vice versa. Further, the load impedance or
admittance is easily obtained from the voltage ratio V2 /V3 .

Pavs Pref VNA P3 P2

Bias Control
CV3

CV2
Generator
Tuner DUT Low-pass
filter Reflection
Splitter bridge

Attenuator or Phase shifter

amplifier

Figure 3.13: Active load-pull measurement setup.

Fig. 3.13 shows a simplified measurement setup for active load-pull: at the input, the setup is
similar to passive load-pull, only that half of the available input power is routed to a second
branch: after adjusting the amplitude (amplification or attenuation) and phase, the resulting
signal V2 is injected into the DUT output.
The two couplers at the output are used to detect injected and reflected power, from which (by
Eq. 3.34), the output power is measured. The reflection bridge (essentially a directional coupler)
couples a fraction C of the injected and reflected voltages to the VNA ports, so that CV3 is
fed into the reference receiver and CV2 into the measurement receiver. Thus, the VNA directly
measures the ratio ΓL = CV2 /CV3 .
Again, from the locations ΓL on the Smith Chart and the recorded output powers, contours
may be drawn. This method can also be made automatic by using electronically controlled
components in the lower signal branch (variable gain amplifier, adjustable attenuator and phase
shifter) and some form of optimization loop for ΓL .
Active load-pull systems are of course more complex but have the following advantages:
Chapter 3. Circuits 51

• Passive tuners are lossy, which makes it impossible to present loads with |ΓL | ≈ 1 to the
DUT. This is however needed in high power amplifiers, or for testing robustness against
highly mismatched loads.

• Mechanical tuners are often difficult to fabricate at high frequencies because of the smaller
dimensions.

• An active load-pull setup may also present active loads |ΓL | > 1 to the DUT.

Fig. 3.14 shows the output power contours obtained by active load-pull of a medium power
MESFET at 6 GHz.

x
300

250
200
150

Pout = 100 mW
f = 6 Ghz
Pin= 95.5mW

Figure 3.14: Typical output power contours of a medium power MESFET.

3.5.4 Performance Metrics

The remaining part of this chapter is devoted to high-efficiency power amplifier modes (namely,
Class-E and Class-F). Before going into that, let us briefly review metrics which are commonly
used to judge the theoretical performance of such modes.
The drain efficiency (or collector efficiency in bipolar devices) is defined as the ratio of funda-
mental RF output power and consumed DC power:

Pout
η= (3.36)
PDC

and can be calculated directly from the Fourier coefficients of voltage and current waveforms
at the device output. Using η, one can calculate the efficiency of output waveforms without
regarding their generation in the nonlinear device.
The above definition does not account for the fundamental RF power supplied to the input of
the amplifier. However, high-efficiency PAs often require high levels of gain compression, and are
thus driven with substantial input power. The power-added efficiency, PAE, relates the power
52 3.5. Power Amplifiers

added by the amplification process to the consumed DC power:

Pout − Pin 1
 
PAE = = 1− η. (3.37)
PDC G

From Eq. 3.37 one sees that for high gain, PAE approaches η. For low gain (especially G < 10 dB),
the PAE is a more ‘honest’ efficiency metric. However, computations involving large-signal gain
require a detailed device model, so that in the following derivations η will be used.
The power utilization factor, PUF, is a performance metric which relates the fundamental RF
output power to the peak voltage and current:

Pout
PUF = . (3.38)
max V (θ) · max I(θ)

It is well-known that a Class-C power amplifier exhibits 100% efficiency only in conjunction with
zero power utilization (i.e., no output power). So it is important that high-efficiency modes are
also compared in terms of output power. A physical interpretation of the PUF is the output
power (in Watt) one can get with a particular mode when the device current/voltage limits are
Imax = 1 A and Vmax = 1 V, respectively.

3.5.5 Conditions for 100% Efficiency

Pdiss Pout,3f0
Pin Pout,2f0
Pout,f0

PDC1 PDC2

Figure 3.15: Power balance in an amplifier.

The power balance for a typical PA is shown in Fig. 3.15. We neglect the DC power contribution
at the input, which is ideally zero for field-effect devices and sufficiently small in bipolar devices.
In order to achieve 100% drain efficiency (DC to fundamental RF), one must have

Pout,f0 = PDC , (3.39)

which leads to the conditions

Pout,nf0 = 0 for n ≥ 2, and (3.40)

Z 2π
Pdiss = 0 ⇔ V (θ)I(θ)dθ = 0. (3.41)
0

The former condition requires that there is no power generated at higher harmonics, and the
latter requires that the overlap integral for device voltage and current is zero, which means that
during the whole RF cycle, either the current or the voltage must be zero. These two conditions
together are sufficient and necessary for 100% drain efficiency.
Chapter 3. Circuits 53

3.5.6 Switch-Mode Power Amplifiers

There are different methods for achieving η > 80%. The classic “ABC” modes use reduced
conduction angles. In addition to the penalty on output power mentioned above, the classic
modes require a purely sinusoidal voltage waveform, so that all harmonics must be short-circuited.
The device output capacitance will only take care of that at very high frequencies. Otherwise,
the (harmonic) matching network has to do the job.
The Class-E power amplifier also uses the concept of a conduction angle (or duty cycle), but
as will be shown, it has the advantage of a simple circuit topology, and it exhibits 100% drain
efficiency at all conduction angles.

Switch-Mode Operation

Ids Ids
Transistor +
pulse-shaped drive

Vgs t

Cds Modulated switch

t

Figure 3.16: Switch-mode operation with 180degree conduction angle in a field-effect device.

Fig. 3.16 shows that by picking a bias point in the middle of the transconductive region and
driving the input very hard (blue curve), one may get an almost rectangular current with
180degree conduction angle at the output (red curve). Other conduction angles may be obtained
with different bias points and drive levels.
The transistor is either on or off and sweeps fast through its linear transconductive region. Thus,
the device may be modeled as an ideal switch which is modulated according to the conduction
angle. While the transistor’s output current is limited by saturation, such limits do not exist for
an ideal switch. The current through the switch must therefore be limited to realistic values by
appropriate choice of the load resistance. Furthermore, an output capacitance may be added to
capture the essential device parasitics.

Simple switch-mode PA

Vdd
Bias-T
LF C Ids Vds
B

Imax Vmax
RL

Figure 3.17: Simple switch-mode amplifier and waveforms for 180degree conduction angle.
54 3.5. Power Amplifiers

Let us now analyze the switch model inside an amplifier. Fig. 3.17 presents a simple switch-mode
amplifier comprising only of a bias-T (feed inductor LF and blocking capacitor CB , for decoupling
DC and RF signal paths) and the load resistance RL .
Assuming 180degree conduction angle, the voltage swings between zero and twice the supply,
Vmax = 2Vdd , and the current is limited by the load resistance, Imax = Vmax /RL = 2Vdd /RL . At
any time, either current or voltage is zero, hence, the ideal switch does not dissipate power.
The rectangular waveforms of this amplifier can be written as Fourier series

4 X sin kθ 4 X sin kθ
! !
Imax Vmax
I(θ) = 1+ , and V (θ) = 1− (3.42)
2 π k odd k 2 π k odd k

for the device current and voltage, respectively. From the Fourier coefficients one can compute
DC and fundamental RF output powers,

1 1 4
 2
Vmax Imax Vmax Imax
PDC = V0 · I0 = and Pout = V1 I1∗ = , (3.43)
4 2 2 π 4

from which the drain efficiency, η = 8/π 2 = 81%, is obtained. This is not a huge improvement
over Class-B (η = 78%), and one may wonder about the missing 19 percentage points. The
answer is that only condition 3.41 is fulfilled, while condition 3.40 is not because a significant
amount of power is contained in harmonics and thus lost in RL .

Class-E PA

The previous example illustrates that an ideal switch does not automatically result in better
efficiency, and that harmonics must be blocked at the output. A quite popular type of switch-
mode power amplifier, known as Class-E, uses a series resonator to block harmonics. The
topology is depicted in Fig. 3.18, where it can be seen that our simple switch-mode amplifier
has been complemented by a few additional elements:

Vdd
LF
IDC
isw itot iRF Cs Ls
iC
Cp RL

Figure 3.18: Class-E power amplifier circuit topology.

The shunt capacitor Cp is needed for the circuit functionality and also includes the transistor
output capacitance Cds , so that Cp ≥ Cds is required. The series inductor Ls forms a resonator
with Cs and RL . To let fundamental RF pass through and block higher harmonics a sufficiently
high Q is required, and the circuit elements must be related to the fundamental frequency f0 by

1
f0 = √ . (3.44)
2π Ls Cs
Chapter 3. Circuits 55

The large feed inductor LF lets only DC currents pass, and the series resonator lets only
fundamental RF currents pass. Therefore, the total current, itot (θ) is must be a fundamental
RF current with DC offset:
itot (θ) = IDC + IRF sin(θ). (3.45)
The total current then splits at the switch–capacitor node: when the switch is closed, itot (θ) flows
through the switch, and when the switch is open, itot (θ) flows through the capacitor. Fig. 3.19
shows how itot (θ) is decomposed into the switch current isw (θ) and the capacitor current iC (θ)
and also illustrates the used nomenclature. Since our sine has a DC offset, the phase angle at
which the switch turns on (zero-crossing, positive slope) is given by −θ1 and the phase angle at
which the switch turns off is given by θ2 . The conduction angle φ is related to these angles by
itot isw iC

IRF
IDC
0 0 0

(a) (b) (c)

Figure 3.19: Current waveforms — (a) total current; (b) switch current; (c) capacitor current.

φ = θ1 + θ2 . (3.46)
Secondly, the DC and RF amplitudes of the total current are related to the turn-on phase angle:
IDC
itot (−θ1 ) = IDC + IRF sin(−θ1 ) = 0 ⇔ sin θ1 = . (3.47)
IRF
Finally, a physically meaningful capacitor current must not have a DC-term:
1 1
Z 2π Z 2π−θ1
iC (θ)dθ = IDC + IRF sin θdθ = 0. (3.48)
2π 0 2π θ2

These three conditions are used to eliminate most of the above parameters. After some algebraic
manipulations, one obtains

1 2π − φ + sin φ
s 2
IRF

= = 1+ , (3.49)
IDC sin θ1 1 − cos φ
so that the conduction angle φ alone determines characteristic phase angles θ1 and θ2 as well
as the ratio between fundamental RF and DC currents. The resulting current waveforms are
depicted in Fig. 3.20a for a fixed DC current of IDC = 1. For a fixed DC current, lower conduction
angles lead to higher peak currents but the absolute value of the peak current is given by
1
 
max itot (θ) = IDC + IRF = 1+ IDC (3.50)
sin θ1
and may be adjusted later on by using IDC as a current scaling factor.
The next step is to calculate the voltage waveform at the switch–capacitor node. When the
switch is closed, one has vsw (θ) = 0, and when the switch is open, the voltage is that of the
charging capacitor with zero initial charge, vC (θ) = ωC
1 R
p
iC (θ)dθ. In summary:

for θ2 < θ < 2π − θ1
 
 IDC θ − θ − cos θ−cos θ2
vsw (θ) = ωCp (3.51)
2 sin θ1
0 otherwise.
56 3.5. Power Amplifiers

The peak voltage at the capacitor occurs when iC (θ) = 0, so that

IDC 2
 
max vC (θ) = vC (π + θ1 ) = 2θ1 − π + . (3.52)
ωCp tan θ1

isw vsw isw,vsw

= 60°
= 60° = 90°
10 20 10

= 180°
= 180°
1 1
0 0 = 180° 0
(a) (b) (c)

Figure 3.20: Class-E waveforms — (a) current waveforms; (b) voltage waveforms; (c) current
and voltage waveforms on same scale for 90degree and 180degree conduction angle.

The voltage waveform contains an additional scaling parameter, ωCp . In Fig. 3.20b these
waveforms are plotted for different conduction angles and for a voltage scaling parameter
ωCp = 1. The voltage scaling parameter allows to draw some conclusions regarding the limits of
Class-E operation:

• Assuming the current scaling parameter IDC is fixed, the power may be increased by scaling
up the voltage, which leads to lower Cp . The limit may be reached when the peak voltage
equals the device breakdown voltage, or when Cp equals the device output capacitance.
Whether Class-E operation is breakdown-limited or capacitance-limited depends on the
particular device technology.

• For Class-E operation at higher frequencies, the same waveforms can only be maintained
when Cp is scaled down, which is ultimately limited by the device capacitance.

Using conditions 3.40 and 3.41 it is now easy to prove that the derived current and voltage
waveforms have 100% drain efficiency:

1. The series resonator is an open circuit at all higher harmonics, which are thus terminated
by Cp and effectively blocked. The load resistance is visible to the device only at the
fundamental frequency, which is terminated with (RL || Cp ).

2. Voltage and current waveforms do not overlap (cf. Fig. 3.20c), so no power is dissipated.

In contrast to Class-C, Class-E exhibits 100% efficiency for all conduction angles φ. It is now
interesting to compare different conduction angles in terms of power utilization.
In order to analyze the power utilization, the peak current and peak voltage of Eqs. 3.50 and 3.52
are used. The output power may be computed from Pout = PDC = VDC IDC . The DC voltage
must still be determined. It reads

1 sin θ2
Z 2π 2
IDC

VDC = vC (θ)dθ = · 1+ . (3.53)
2π 0 4πωCp sin θ1
Chapter 3. Circuits 57

Following that, the power utilization factor may be obtained from

VDC IDC
PUF = . (3.54)
max vC (θ) · max itot (θ)

PUF
B AB A E
0.125 100
0.098 C
C B
E 50 AB A

0 360 0 360
(a) (b)

Figure 3.21: Comparison of Class-E and -A to -C — (a) power utilization; (b) drain efficiency.

A comparison of Class-E with other common PA modes in terms of efficiency and power utilization
is shown in Fig. 3.21. The power utilization of Class-E has maximum value of PUF = 0.098 at
φ ≈ 176 degree, which is about 1 dB less power than one gets from a Class-A or Class-B design
(PUF = 0.125) with the same device.

3.5.7 Class-F PA

The Class-F mode uses shaping of the voltage waveform at the output. The interesting feature
is that one can increase efficiency and output power at the same time. To understand how this
is done, consider the typical Class-B waveforms, normalized to their DC Fourier coefficient:

v(θ) = 1 + sin θ, and (3.55)

π 2
i(θ) = 1 − sin θ − cos kθ, (3.56)
X
2 k even
k 2−1

for the voltage and current, respectively. For the voltage coefficients, we have v0 = v1 = 1, so
that v(θ) swings between zero and two (cf. Fig. 3.22a). We further assume that the device can
only sustain voltages between these limits (v ∈ [0 ; 2]), and that the DC Fourier coefficients are
fixed so that in any case, the (normalized) DC power consumption is pDC = 1.
The fundamental RF output power can be increased if the voltage waveform is modified such that
it contains more fundamental amplitude while staying within the above limits. With pDC = 1
fixed, a higher fundamental RF output power automatically leads to a higher drain efficiency.
This may be accomplished by an additional third-harmonic voltage coefficient:

v(θ) = 1 + v1 sin θ + v3 sin 3θ. (3.57)

The fundamental term has a positive peak at θ = π/2, where the third harmonic has a negative
peak (cf. Fig. 3.22b). By adding the third-harmonic voltage one can flatten the voltage waveform,
thereby ‘hiding’ a higher fundamental amplitude inside. However, v1 cannot be arbitrarily high,
and the limit shall be explored in the following.
For small deviations, i.e. v1 & 1, one can stay exactly within the limits by choosing v3 = 1 − v1 ,
and the resulting waveform still has one positive peak. This changes when the waveform becomes
58 3.5. Power Amplifiers

3.5 3.5 3.5

Normalized Voltage

Normalized Voltage

Normalized Voltage
3 Vds 3 Vds 3 Vds
2.5 2.5 3rd harm. 2.5 3rd harm.
2 2 2
1.5 1.5 1.5
1 1 1
0.5 0.5 0.5
0 0 0
-0.5 -0.5 -0.5
0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6
Phase Angle (rad) Phase Angle (rad) Phase Angle (rad)
(a) v1 = 1.000 (b) v1 = 1.125 (c) v1 = 1.154

Figure 3.22: Evolution towards Class-F voltage waveforms: (a) original Class-B voltage; (b)
maximally-flat case; (c) maximum-efficiency case. Any included third-harmonic
components are also plotted separately (dashed).

maximally-flat, i.e., when also the second derivative vanishes at the peak (θ = π/2):
π π 3π v1
 
v 00
θ= = −v1 sin − 9v3 sin = −v1 + 9v3 = 0 ⇔ v3 = . (3.58)
2 2 2 9
Thus, under maximally-flat conditions (shown in Fig. 3.22b), the voltage waveform becomes
9 1
v(θ) = 1 + sin θ + sin 3θ. (3.59)
8 8
When further increasing the fundamental amplitude, a waveform which is within the limits will
exhibit two positive peaks, and the analysis becomes more cumbersome. By defining x = cos θ
and exploiting multiple-angle formulas, one can write v 0 (θ) as a polynomial in x:

v 0 (x) = (v1 − 9v3 )x + 12v3 x3 , (3.60)

and from v 0 (x? ) = 0 the phase angle θ? of the first positive peak is gathered:

3
s
v1
θ = arccos(x ),
? ?
with x =
?
− . (3.61)
4 12v3
√
Likewise, by using y = sin θ = 1 − x2 , the voltage may be written as a polynomial in y

v(y) = 1 + y v1 + 3v3 − 4v3 y 2 , (3.62)

 

√
and by plugging in y ? = 1 − x? and equating this to the upper voltage limit, v(y ? ) = 2, one
arrives at a relation between the fundamental and third-harmonic voltage coefficient:
3
v1 √

v3 + = v3 ⇔ v1 = 3( 3 v3 − v3 ) (3.63)
3
The limiting case at which we cannot further increase the fundamental voltage is characterized
by ∂v1 /∂v3 = 0 and yields the maximum-efficiency voltage waveform of Fig. 3.22c:
2 1
v(θ) = 1 + √ sin θ + √ sin 3θ. (3.64)
3 3 3
The above case represents the optimum with additional third harmonic. The fundamental
voltage can only be further increased by adding other odd harmonics, such as the fifth or seventh.
Chapter 3. Circuits 59

While the mathematical analysis of these cases is outside the scope of this text, it can be easily
guessed that with infinitely many odd harmonics, the voltage waveform approaches a rectangle.
Of all periodic functions, the rectangular square-wave is the waveform which contains the most
fundamental amplitude between a given upper and lower boundary. This limiting case is depicted
in Fig. 3.23a and known as “Ideal Class-F”. It can be shown that the corresponding waveforms
exhibit 100% drain efficiency:

1. The higher odd harmonics are only contained in the voltage waveform, whereas the higher
second harmonics are only contained in the current waveform (refer to Eq. 3.56). So none
of the higher harmonics (n ≥ 2) contains power because Re{vn i∗n } = 0.

2. The “ideal” rectangular voltage and the half-wave sinusoidal current do not overlap, so
that there is no power dissipated in the device.

In practice, however, one usually takes the third-harmonic version (Class-F3 , see Fig. 3.23b)
because for the “ideal” version, one needs to design a matching network which takes infinitely
many harmonics into account. Such a network is hard to design and its complexity would result
in losses which degrade efficiency, thus rendering such effort pointless.
Normalized Waveforms

Normalized Waveforms

3.5 3.5 3.5

Normalized Current
3 Vds 3 Vds 3 Ideal
2.5 Ids 2.5 Ids 2.5 Class-F3
2 2 2
1.5 1.5 1.5
1 1 1
0.5 0.5 0.5
0 0 0
-0.5 -0.5 -0.5
0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0.5 1 1.5 2
Phase Angle (rad) Phase Angle (rad) Normalized Voltage
(a) Ideal Class-F (b) Class-F3 (third-harmonic version) (c) Corresponding loadlines

Figure 3.23: Class-F waveforms.

At this point, it is interesting to assess the efficiency and power utilization of the (sub-optimum)
Class-F3 . Clearly, its waveforms show some overlap, which is also visible when Ids is plotted
versus Vds (Fig. 3.23c, where the light green loadline goes through states of finite dissipated
power). The associated efficiency is computed as

pout 1
v1 i1 1 2 π
η= = 2 = · √ · = 90.7%. (3.65)
pDC 1·1 2 3 2

and the power utilization is

π
1
2 v1 i 1
√
1
PUF = = 2 3
= √ = 0.144, (3.66)
max v(θ) · max i(θ) 2·π 4 3

which is about 15% (0.6 dB) higher than a comparable Class-A or Class-B design.
The last important point is the circuit realization. While in the Class-E section, a certain
circuit topology was presented and then analyzed, we have started with the waveform analysis of
Class-F and will now derive requirements for a (harmonic) matching network. From the voltage
60 3.5. Power Amplifiers

and current Fourier coefficients, we infer

 I1 = 3π · IDC for n = 1 (fundamental)


 V1 √2 VDC

Zn = 0
I2 =0 for n = 2 (second harmonic) (3.67)
=∞ for n = 3 (third harmonic).

 V3

0

This means that the fundamental impedance is resistive, whereas the second- and third-harmonic
terminations are a short and open circuit, respectively. Harmonics of order n ≥ 4 must also
be short-circuited but this happens often automatically due to the device output capacitance,
which makes any external load at such high harmonic frequencies negligible.
Until now, the analysis is based on the addition of a third-harmonic voltage coefficient and
proper harmonic impedance termination at the output. The origin of the third-harmonic voltage
and how its amplitude or phase may be adjusted has not yet been discussed. The appearance
of a third-harmonic voltage may seem like a mystery, given that there is no third-harmonic
current at the output (i3 = 0). Together with the required third-harmonic impedance (Z3 = ∞),
we have a physically indeterminate case, where every finite value of v3 satifies the condition
v3 = Z3 · i3 = ∞ · 0.

Ids Vgs Ids Vgs Ids Vgs

Vds Vds Vds

Vk
(a) (b) (c)

Figure 3.24: Models for transistor output I–V characteristics — (a) Ids = f (Vgs ) without
dependence on Vds ; (b) dependence on Vds by output conductance λ; (c) dependence
on Vds by output conductance λ and knee-voltage Vk .

A physical solution for v3 may be found [20] by using a more realistic transistor model, where the
device current is also a function of the drain voltage. For example, with I˜ds (Vgs ) as the ‘baseline’
drain current due to the input voltage, one may use Ids = f (Vgs , Vds ) = I˜ds (Vgs ) · (1 + λVds ) to
model a finite output conductance, λ, as shown in Fig. 3.24b. In addition, one may include a
“knee-voltage”, Vk , i.e., a boundary between linear and saturation regions of a FET (Fig. 3.24c).
When introducing λ > 0, the drain current becomes
π 2 2 2
 
i(θ) = 1 − sin θ − sin 2θ − sin 4θ − sin 6θ − · · · × (1 + λ + λv1 sin θ + λv3 sin 3θ),
2 3 15 35
(3.68)
and the third-harmonic current component of this product is
36 4
 
i3 = λ v3 − v1 . (3.69)
35 15
Setting i3 to zero as is required for Class-F operation, we obtain
7
v3 = v1 ≈ 0.26v1 . (3.70)
27
The result shows two things: firstly, it shows that although λ > 0 was introduced to solve the
indeterminate problem, the solution of v3 does not depend on λ and is also valid for λ = 0.
Chapter 3. Circuits 61

Secondly, the obtained ratio v3 /v1 ≈ 0.26 depends only on the coefficients of the baseline
current I˜ds (here, a half-wave sine) and does not equal the optimum ratio of Eq. 3.64, which is
v3 /v1 = 1/6.
The conclusion is that the baseline current waveform must also be manipulated in order to adjust
the ratio v3 /v1 . This is usually done by careful selection of bias and drive level at the input. In
addition, harmonic matching networks at the input may be used to shape the driving waveform.
It is usually difficult to generate waveforms which precisely resemble those of Fig. 3.23b.

3.5.8 Conclusions

Let us conclude the discussion of high-efficiency power amplifiers with the following remarks:

• The need for high drive levels at the input in Class-E and Class-F PAs leads to gain
compression, thereby decreasing the PAE. At less than 10 dB of “linear” gain, the described
concepts are often not useful any more.

• Both concepts are highly nonlinear because they work only at high drive levels. This is
inevitable; efficient amplifiers are always nonlinear. They can, however, be linearized by
employing predistortion techniques.

• Class-E is an alternative to Class-C as it provides more output power at 100% efficiency

and has a simple topology. Yet, its power utilization is not optimal because of high voltage
and current peaks.

• Class-F with three harmonics is a good compromise between efficiency and circuit com-
plexity. Furthermore, its advantage is improved power utilization.
Part II: Mixers

The second part of this lecture deals with mixers, which are three-port devices. The ideal
mixer produces an output at the sum and at the difference frequencies of the two input signals
and thus enables frequency up- and down-conversion. The principle of operation is based on a
nonlinearity which also generates harmonics and other products of the two input frequencies.
As a consequence sufficient filtering is important.
This part is structured into three main chapters. In the first chapter, the fundamentals of
mixers are presented by application of energy-conservation relations and the conversion matrix
analysis. The second section deals with three mixer devices: the pn-diode, the Schottky diode
and the field-effect transistor. After laying these foundations, the third section discusses the
Schottky diode downconverter and the field-effect transistor downconverter as examples of circuit
topologies.

62
Chapter 4

Mixer Basics

4.1 Energy-conservation relations in nonlinear reactances

The energy-conservation relations in nonlinear reactances were first derived by Manley and Rowe
[21] and can be formulated as

X Pi ∂fi
∀k ∈ {1, 2, . . . , nind } : · = 0. (4.1)
i
fi ∂find,k

As illustrated in Fig. 4.1a by example of a nonlinear capacitance,

• find,k with k ∈ {1, 2, . . . , nind } are nind independent frequencies with which the nonlinear
reactance is excited,

• fi with i ∈ {1, 2, . . . , ndep } are ndep dependent frequencies resulting from mixing between
the independent frequencies. Recall that excitation of a nonlinear system with nind
frequencies causes mixing products at dependent frequencies f = ni=1 ki · find,i with
P ind

ki ∈ Z. As is shown in Fig. 4.1a, filters may be used to select a finite number of frequencies
which contribute to the power balance.

• Pi is the power at the frequency fi , where negative values refer to powers generated by
the nonlinear reactance, and positive values refer to powers supplied to the nonlinear
reactance.

... ... Mixer

find,1 find,2 find,k f1 f2 fk f1 f3

...
...
... ... f2
(a) (b)

Figure 4.1: (a) Illustration of independent and dependent frequencies in a nonlinear capacitance;
(b) Block diagram of an ideal mixer.

63
64 4.1. Energy-conservation relations in nonlinear reactances

An idealized mixer is shown as block diagram in Fig. 4.1b. It is a special case of Fig. 4.1a, as
there are two independent frequencies at f1 and f2 and one dependent frequency at f3 . The
excitation at f1 is converted to a signal at f3 by means of the excitation at f2 . The source
at f2 is called local oscillator (LO) and provides a strong signal. The signals at f1 and f3 are
usually weak. Assuming that the above mixer is based on a nonlinear reactance (e.g., an ideal
varactor), the Manley-Rowe relations can be applied to analyze its power spectrum relations.
Such an analysis shows the applications for which varactor-based mixers are suited and indicates
potential problems. Note that the above relations cannot be applied to varistor-based mixers
because varistors are nonlinear resistances. Chapter 5 treats varactors and varistors in more
detail.
In the following, four mixer configurations are analyzed by applying the above energy-conservation
relations. In these derivations, one key assumption is that the power P3 at the converted frequency
is generated by the varactor, and thus negative. This is illustrated in Fig. 4.1b by the direction
of the arrow for f3 . The arrows for f1 and f2 are the result of the analysis; there may be one or
two surprises.

4.1.1 Upconverter

In an upconverter, the source signal f1 is a low frequency called intermediate frequency (IF)
because it is often located between baseband and carrier frequency. The IF is converted to a
high frequency f3 , called radio frequency (RF). There are two possibilites, depending on the
location of RF with respect to LO: in the upper sideband case (Fig. 4.2a) one has f3 > f2 , and
in the lower sideband case (Fig. 4.2b) one has f3 < f2 .

P LO P LO

RF RF
IF IF

Figure 4.2: Upconverter power spectrum — left: upper sideband upconverter; right: lower
sideband upconverter.

In the upper sideband case, the dependent frequency f3 is related to the independent frequencies
by f3 = f2 + f1 . The Manley-Rowe relations yield two equations:
Chapter 4. Mixer Basics 65

3
Pi ∂fi
For f1 : =0 (4.2)
X
·
i=1
fi ∂f1
P1 ∂f1 P2 ∂f2 P3 ∂f3 P1 P3
⇔ · + · + · = + = 0, and (4.3)
f1 ∂f1 f2 ∂f1 f3 ∂f1 f1 f3
|{z} |{z} |{z}
=1 =0 =1
3
Pi ∂fi
For f2 : =0 (4.4)
X
·
i=1
fi ∂f2
P1 ∂f1 P2 ∂f2 P3 ∂f3 P2 P3
⇔ · + · + · = + = 0. (4.5)
f1 ∂f2 f2 ∂f2 f3 ∂f2 f2 f3
|{z} |{z} |{z}
=0 =1 =1

Solving for the power P3 of the upconverted signal, one obtains

f3
P3 = − P1 , (4.6)
f1

meaning that P3 < 0 is generated by the varactor if P1 is positive (i.e., supplied to the varactor).
Also, there is a conversion gain of G = P3 /P1 = f3 /f1 . The necessary additional power is taken
from the power supplied by the LO at f2 .
In the lower sideband case, the frequency relation is f3 = f2 − f1 . The energy-conservation law
yields two equations,

P1 P3
− = 0, and (4.7)
f1 f3
P2 P3
+ = 0, (4.8)
f2 f3

from which the RF power P3 is calculated as

f3
P3 = P1 , or, equivalently, (4.9)
f1
f3
P3 = − P2 . (4.10)
f2

If the LO power P2 is positive (supplied to the varactor), the powers P1 and P3 are both
generated (same sign). This means that a signal at f1 is reflected at its input with |Γ| > 1,
which may lead to instabilities. On the other hand, the power gain at f1 may be useful: this
effect is exploited in so-called parametric amplifiers.

4.1.2 Downconverter

In a downconverter, the source signal f1 is the high frequency RF signal, which is converted to
f3 , in this case the low frequency IF signal. Like in the upconverter, there are two possibilites:
in the upper sideband case (Fig. 4.3a) one has f1 > f2 , and in the lower sideband case (Fig. 4.2b)
one has f1 < f2 .
66 4.2. Conversion matrix analysis

P LO P LO

RF RF
IF IF

Figure 4.3: Downconverter power spectrum — left: upper sideband downconverter; right: lower
sideband downconverter.

In the upper sideband case, one has f3 = f1 − f2 , and the two energy-conservation equations are
P1 P3
+ = 0, and (4.11)
f1 f3
P2 P3
− = 0, (4.12)
f2 f3
which leads to an IF power of
f3
P3 = − P1 , or, equivalently, (4.13)
f1
f3
P3 = P2 . (4.14)
f2

Note that now f3 /f1 is a small ratio, which leads to attenuation of the downconverted signal.
Also, from the relation of their signs, only the RF power P1 is supplied. Both LO and IF power
are generated, which may lead to instability at the LO port.
In the lower sideband case, the frequency relation is f3 = f2 − f1 . The Manley-Rowe relation
yields
P1 P3
− = 0, and (4.15)
f1 f3
P2 P3
+ = 0, (4.16)
f2 f3
which results in the following expressions for the IF power:
f3
P3 = P1 , or, equivalently, (4.17)
f1
f3
P3 = − P2 . (4.18)
f2
Like in the upper sideband downconverter, one has attenuation due to f3 /f1 , and potential
instability (here, at the RF port). One important conclusion is that nonlinear reactances
(varactors) are not useful for downconversion. Instead, one may use varistors.

4.2 Conversion matrix analysis

The conversion matrix analysis (or large-signal–small-signal analysis) is useful for calculation of
nonlinear circuits in which the nonlinear device is excited with one periodic, large signal while
one or multiple signals with much smaller amplitude are also present.
Chapter 4. Mixer Basics 67

The large signal (called “carrier”, “pump”, or “local oscillator”) periodically changes the operating
point. Small-signal circuit elements, such as the differential resistance rD of a diode, depend
on the operating point and thus fluctuate with the same periodicity. The small signals then
see linear, but time-varying circuit elements, which leads to the frequency conversion. It is
important to note that the small signals are still linearly related—even if the pump is a nonlinear
excitation. The conversion matrix then describes the frequency mixing between the small signals.
Conversion matrices are not only used in conjunction with mixers and modulators but also for
analyzing noise or stability in nonlinear circuits.

4.2.1 Derivation

large signal
(pump) nonlinear
V
device
small signals

Figure 4.4: Nonlinear diode circuit.

Let us derive the conversion matrix by example of the diode mixer in Fig. 4.4.
The large-signal local oscillator voltage VLO (t) is given as

VLO (t) = Vdc + V̂LO · cos(ωLO t), (4.19)

and applied to the diode’s terminals, it results in a large-signal diode current

ID (t) = f (VLO (t)), (4.20)

where ID = f (VD ) is the diode’s nonlinear current–voltage relationship. If the small-signal

voltages are taken into account, they must be superimposed with VLO (t). For simplicity, all
small-signal voltages are gathered in the term ∆v(t), and the voltage at the diode terminals
becomes VLO (t) + ∆v(t). This leads to a diode current

ID (t) = f (VLO (t) + ∆v(t)) = ILO (t) + ∆i(t), (4.21)

where ID has already been split into a large- and small-signal term. We are mainly interested in
the interaction between the small signal voltages and currents, ∆v(t) and ∆i(t), and the latter
can be obtained by expanding the nonlinearity f into a Taylor series around VLO
(∆v)2
f (VLO + ∆v) = f (VLO ) + f 0 (VLO ) · ∆v + f 00 (VLO ) · + ··· (4.22)
2
Because ∆v is very small compared to VLO (t), the series can be truncated after the first-order
terms, and the small-signal currents are computed as

∆i ≈ f 0 (VLO ) · ∆v. (4.23)

The local oscillator voltage is a periodic function, and this also holds for the derivative term
f 0 (VLO ). A periodic function can be expressed by a Fourier series:
∞
f 0 (VLO ) = (4.24)
X
Gn ejnωLO t ,
n=−∞
68 4.2. Conversion matrix analysis

where the Fourier coefficients Gn (here, they take the form of conductances) are determined as
1
Z π
Gn = f 0 (VLO ) exp(−jnωLO t) d(ωLO t). (4.25)
2π −π

Recall from the theory of LTI-systems, that functions which are real in time-domain are hermitian
in frequency-domain, that is, G−n = G∗n . Knowledge of either the right or left half of a spectrum
is sufficient because the other half is easily obtained.

4.2.2 Example 1: conversion matrix with a single small signal

Assume that the small-signal voltage ∆v(t) has only one spectral component at ωIF :
1 
∆v(t) = √ v IF ejωIF t + v ∗IF e−jωIF t , (4.26)

2
where v IF is the effective voltage phasor at ωIF . We obtain the small-signal current from (4.23)
and plug the Fourier series representation of f 0 (4.24):

1 ∞
∆i(t) = √ · Gn ejnωLO t · v IF ejωIF t + v ∗IF e−jωIF t (4.27)
X  
2 n=−∞
G0
 
= √ · v IF ejωIF t + v ∗IF e−jωIF t
2
G−1
 
+ √ · v IF e j(ωIF −ωLO )t
+ v IF e
∗ −j(ωIF +ωLO )t
2
G1
 
+ √ · v IF ej(ωIF +ωLO )t + v ∗IF e−j(ωIF −ωLO )t
2
G−2
 
+ √ · v IF e j(ωIF −2ωLO )t
+ v IF e
∗ −j(ωIF +2ωLO )t
+ ···
2
The small-signal current has infinitely many spectral components: the voltage at ωIF leads to
currents at all mixing frequencies nωLO ± ωIF . It can be seen that the exponential terms of the
Fourier series lead to the frequency conversion, where the nth Fourier coefficient Gn is always
associated with a shift of nωLO , and positive n cause the spectrum to be shifted right (negative
n: shifted left).

iIF = G0 · v IF , (4.28)
iIF+nLO = Gn · v IF , and (4.29)
iIF−nLO = G−n · v IF = G∗n · v IF . (4.30)

In (4.27), terms that belong to the same frequency have been marked in red (ωIF − ωLO ) and
blue (ωIF + ωLO ). The spectral components of ∆i(t) are hermitian, and more importantly, the
phasors of ∆i(t) depend linearly on the phasors of ∆v(t). The linear relationship is also valid if
∆v(t) consists of multiple frequencies, as in the following example.

4.2.3 Example 2: conversion matrix with multiple small signals

In the previous example, only voltages at ωLO and ωIF were considered, and all other voltages
were short-circuited. While it is true that some voltages at the device terminals may be short-
circuited (e.g., due to parasitic capacitances), voltages at other mixing frequencies may see a
Chapter 4. Mixer Basics 69

termination resistance or parasitic resistances (such as the diode’s path resistance), which leads
to a voltage drop. It is then useful to allow these voltages in the analysis by including them in
∆v(t).

V
I

resonance at

Figure 4.5: Diode upconverter with resonant circuit.

In this example, ∆v(t) has two spectral components, at ωIF and ωRF = ωLO + ωIF :

1  1 
∆v(t) = √ v IF ejωIF t + v ∗IF e−jωIF t + √ v RF ejωRF t + v ∗RF e−jωRF t . (4.31)
 
2 2

For the calculation of the power spectrum the related currents are needed:

G0
 
∆i(t) = √ · v IF ejωIF t + v ∗IF e−jωIF t + v RF ejωRF t + v ∗RF e−jωRF t
2
G−1
 
+ √ · v IF ej(ωIF −ωLO )t + v ∗IF e−j(ωIF +ωLO )t + v RF ej(ωRF −ωLO )t + v ∗RF e−j(ωRF +ωLO )t
2
G1
 
+ √ · v IF ej(ωIF +ωLO )t
+ v IF e
∗ −j(ωIF −ωLO )t
+ v RF e j(ωRF +ωLO )t
+ v RF e
∗ −j(ωRF −ωLO )t
+ ···
2

Here, the current contributions at ωIF are marked with blue color, and the contributions at
ωRF are marked in green. The higher-order (|n| ≥ 2) Fourier terms do not lead to spectral
components at ωIF or ωRF and thus do not contribute to the power spectrum.
The above contributions can be grouped and expressed in matrix form. The current phasors iIF
and iRF are hermitian, so only the right half of the spectrum is taken into account. This leads
to the conversion matrix: ! ! !
iRF G0 G1 v RF
= . (4.32)
iIF G∗1 G0 v IF

The conversion matrix describes the linear relationships between the small signals at a non-
linear device which is driven by one large signal. The number of considered small-signal
voltages/currents determines the matrix dimension. The large signal does not appear in the
voltage/current vectors but it determines the complex-valued matrix elements Gn . In the matrix,
the main diagonal elements relate voltages and currents of the same frequency. The off-diagonal
elements describe the frequency conversion.
The n-dimensional matrix can be used like the n-port matrix known from classic circuit theory.
The difference is that now different ‘ports’ may refer to the same physical terminal, but at a
different frequency.
Assume that the resonant circuit presents an admittance YRF to the signal at ωRF . In terms of
the voltages and currents in Fig. 4.5, we have iRF = −v RF · YRF . The admittance presented at
70 4.2. Conversion matrix analysis

ωIF can be calculated by using the first matrix row, which leads to
G1 YRF
iRF = v , (4.33)
G0 + YRF IF
and the second matrix row, which gives
G1 G∗1
 
iIF = G0 − v IF . (4.34)
G0 + YRF
| {z }
YIF

The admittance at ωIF thus depends on the admittance at ωRF . This is similar to classic circuit
theory, where the admittance at one port influences the input admittance at the other port.
With the above matrix, it is also possible to compute a voltage gain, for example the ratio
v RF /v IF :
G1
v RF /v IF = − . (4.35)
G0 + YRF
The conversion matrix analysis can be adapted for nonlinear resistances, capacitances, induc-
tances, transconductances and other nonlinearities. In section 6.1, the conversion matrix is used
to calculate the conversion gain of a nonlinear resistance downconverter.
Chapter 5

Mixer Devices

5.1 PN-diode (varactor)

The ideal varactor is a lossless, nonlinear device, for example, a nonlinear capacitance. The
depletion layer inside a reverse-biased pn-diode is such a nonlinear capacitance: the depletion
layer width, and hence, the junction capacitance varies with the applied reverse-bias voltage.

Figure 5.1: Equivalent circuit model of a reverse-biased pn-diode.

The equivalent circuit model of a reverse-biased pn-diode is depicted in Fig. 5.1. It consists of
an ideal nonlinear capacitance c(V ) plus additional parasitic elements:

• losses due to bulk- and contact resistances are summarized in rb ,

• the feed inductance Lp and the package capacitance Cp depend on the diode layout and
package dimensions.

The expression for the depletion layer capacitance of a symmetrically doped pn-junction shall be
derived. The doping profile is given as a power of x, which is a general description including
abrupt and linearly graded junctions. Assuming complete ionization of the impurity atoms and
negligible minority carrier concentrations, the net charge distribution can be expressed as

gxn x ≥ 0
(
ρ(x) = q(ND (x) − NA (x)) = q · (5.1)
−g(−x)n x < 0

By integrating Poisson’s equation, dE(x)

dx =  ,
ρ(x)
the electric field is found:
n+1 !
qg w w w

E(x) = |x|n+1 − , with − ≤x≤ , (5.2)
(n + 1) 2 2 2

71
72 5.1. PN-diode (varactor)

neutral region depletion layer neutral region

p-doped n-doped

Figure 5.2: Symmetric pn-junction with linearly graded doping profile in thermal equilibrium.

where the width Rof the (symmetric) depletion layer is w. One more integration yields the
potential Ψ = − E(x)dx. We are interested in the potential difference across the depletion
layer:
2qg
 n+2
w w w w
     
Ψ −Ψ − = 2Ψ = VD − V = , (5.3)
2 2 2 (n + 2) 2
where VD is the diffusion voltage and V is an externally applied voltage. The charge distribution,
electric field, and potential are plotted in Fig. 5.2 for a linearly graded pn-junction at thermal
equilibrium.
The above expression for VD can be solved for w, and one gets
1
(n + 2)VD
1
V
   
n+2 n+2
w = w0 1− with the zero-bias width w0 = 2 · . (5.4)
VD 2qg

The depletion layer capacitance is determined from the formula of the parallel-plate capacitor:
A Cj0
Cj = = γ , (5.5)
w 1 − VVD

where Cj0 = wA

0
is the zero-bias capacitance, and γ = n+2
1
is called the characteristic exponent.
By choosing the doping profile exponent n, one can adjust the characteristic exponent γ and
thus the voltage dependence.

Example:

One important application where the doping profile of a varactor is chosen to result in a certain
voltage dependence is in tuning varactors for voltage-controlled oscillators (VCOs). Here, the
varactor and an inductor form an LC-resonator which determines the VCO’s operating frequency.
Naturally, a linear tuning relation between externally applied voltage and resonance frequency
ω0 is desired.
1 V V −2
   
ω0 = p ∼ 1− ⇒ Cj ∼ 1 − .
LCj VD VD
The characteristic exponent for linear frequency tuning is γ = 2, or equivalently, n = − 32 . This
type of doping profile is also called hyperabrupt junction (Fig. 5.3). In practice, the impurity
Chapter 5. Mixer Devices 73

concentration cannot be infinitely increased, and close to the junction the actual profile deviates
from x−3/2 . However, further away from the junction, the doping profile must resemble the
desired function as closely as possible.

Figure 5.3: Ideal (dotted) and practical (solid) charge distribution of the hyperabrupt junction.

5.1.1 Normalized charge–voltage relationship

The depletion layer size is limited by two effects: first of all, at V = VD , the diode becomes
forward-biased, and does not operate as a varactor any more. Secondly, at very negative
(reverse-bias) voltages breakdown occurs. The voltages and charges can be normalized to these
limits. For the normalized voltage,
VD − V
v= , with 0 ≤ v ≤ 1. (5.6)
VD − VB

The charge–voltage relationship can be calculated from Cj (V ) = dV .

dQ
Integration yields

Cj0 VDγ
Z V
Q = QD + Cj (V )dV = QD − (VD − V )1−γ , (5.7)
VD 1−γ
where QD is the stored charge at VD . The above expression for the charge QB at the breakdown
voltage VB then becomes

Cj0 VDγ
Z VB
QB = QD + Cj (V )dV = QD − (VD − VB )1−γ , (5.8)
VD 1−γ
and the normalized charge–voltage relationship can be written as
1−γ
QD − Q VD − V

1
q= = = v 1−γ ⇔ v = q 1−γ , 0≤q≤1 (5.9)
QD − QB VD − VB

5.1.2 Cutoff frequency

The parasitic elements in Fig. 5.1 have lowpass behavior (bulk resistance rb in series with
capacitance c), which limits the usable frequency range of the varactor diode. Neglecting the
influence of packaging (Cp , Lp ), the cutoff frequency is determined by
1
fc = (5.10)
2πrb c
The bulk resistance rb is in fact dependent on the drive voltage: with greater reverse-bias V ,
the depletion layer grows. Hence, the neutral bulk region, which contributes to rb , becomes
smaller. The periodic variation of rb commonly has little influence on the operation of the
74 5.2. Schottky diode (varactor / varistor)

varactor [22], so that one may use a constant value for rb . Under small-signal conditions, the
bulk resistance at the operating point, and under large-signal conditions, the average value
during one RF cycle may be used. Ideally, the device length d equals the depletion layer width
at the maximum voltage V = VB because then, rb,min = 0. The maximum bulk resistance is
reached at V = VD , where the depletion layer is negligibly small, so that rb,max = σA
d
, where σ is
the bulk conductivity, and A is the diode cross-section. The average value of the bulk resistance
at full drive is thus
1 1 d
rb = (rb,min + rb,max ) = rb,max = . (5.11)
2 2 2σA
In order to average the capacitance over one RF cycle, we use its reciprocal, the elastance
S = 1/C = A w
(for a meaningful average, the variable term should be in the numerator). In a
similar fashion, the minimum elastance, Smin = 0, occurs at V = VD , where the w is negligibly
small. The maximum elastance, Smax = A d
is reached at V = VB . Thus, the average elastance is

1 1 d
S= (Smax + Smin ) = Smax = . (5.12)
2 2 2A
The cutoff frequency is then computed as

1 S σ
fc = = = . (5.13)
2πrb c 2πrb 2π

It can be understood that for a high cutoff frequency, high bulk conductivity (which, for a
semiconductor, means high electron mobility) and a low dielectric constant are advantageous.
Nowadays, n-GaAs is commonly used due to its high mobility.
Physically, fc is the dielectric relaxation frequency, at which the displacement current ωE
equals the conduction current σE. Above fc , the displacement current becomes dominant, so
that ultimately, the pn-junction acts like a linear capacitance.
For applications, in which pn-diodes are operated under periodic drive, one often defines a diode
quality factor Q = fc /f . From Eq. 5.13 it can be readily obtained for the ideal diode with
optimum device length d:
S σ
Q= = . (5.14)
ωrb ω

5.2 Schottky diode (varactor / varistor)

The Schottky diode is not only the oldest semiconductor device, but, as it was in place before
vacuum tubes, it is even the oldest electronic device in general. The effect of a nonlinear
current–voltage relationship at a metal-semiconductor (MES) junction was discovered in 1874
by Ferdinand Braun, but a theoretical description based on energy band diagrams was provided
only in 1938 by Walter Schottky. The first applications were as point-contact rectifiers.
Schottky diodes only involve majority carriers (normally electrons). Thus, in contrast to pn-
diodes, which are slowed down by minority carrier storage effects, they are very fast. They
also exhibit lower threshold voltages and smaller resistances under forward bias, while under
reverse-bias conditions they behave like an ideal abrupt pn-junction. Their noise performance is
also favorable because there is no charge recombination (as in pn-diodes).
The device is mostly employed as varistor (variable resistor) in rectifiers, mixers, modulators,
switches, limiters, and frequency multipliers but may also be used as varactor, e.g., in frequency
multipliers and parametric amplifiers.
Chapter 5. Mixer Devices 75

vacuum level vacuum level

WC
WF WB qVD
WC
WFm WFm WF
WV

WV
metal semiconductor metal semiconductor
(a) (b)

Figure 5.4: Energy-band diagrams — (a) for separated metal and semiconductor; (b) for metal
and semiconductor in contact without bias (V = 0) and at thermal equilibrium.

Current flow

In order to qualitatively analyze the current flow of a Schottky diode under forward and reverse
bias, the energy-band diagram of the Schottky contact is helpful. In Fig. 5.4a the band diagrams
of (separate) metal and semiconductor bulk materials are shown. Their Fermi levels WFm and
WF are related by their distance to the vacuum level, WF − WFm = q(φm − φs ).
When the two materials are brought in contact, and after the decay of any transient processes,
the contact reaches thermal equilibrium, where the Fermi levels of the two materials must be flat
and equal (illustrated in Fig. 5.4b). Deep inside the semiconductor, the influence of the contact
is negligible, so that the relation between valence band, WV , conduction band, WC , and Fermi
level, WF , is like in the bulk material. However, the alignment of the Fermi levels causes the
conduction and valence bands as well as the vacuum level to bend near the contact inside the
semiconductor.

WC
WB WB
WF
WFm WFm
WC
WV WF

metal semiconductor
WV
(a) (b)

Figure 5.5: Energy-band diagrams — (a) Schottky contact under forward bias; (b) Schottky
contact under reverse bias.

The Schottky contact is characterized by an energy barrier, WB , which allows only few electrons
to flow from the metal to the semiconductor. The corresponding current −I0 is called saturation
current. In the other direction the current flow depends on the biasing conditions:

• At zero bias (V = 0, Fig. 5.4b), only the built-in potential (or diffusion voltage) VD
causes the bands to bend. The Fermi level is flat, and the macroscopic current is zero:
I = I0 + (−I0 ) = 0. The currents in both directions compensate each other.
76 5.2. Schottky diode (varactor / varistor)

• In forward bias, (V > 0, Fig. 5.5a), the energy level is raised in the semiconductor, thus
lowering the barrier and increasing the current flow towards the metal. For the electrons
in the metal, the barrier and, in turn, the current (−I0 ) remains unchanged.

• In reverse bias, (V < 0, Fig. 5.5b), the current flowing towards the metal vanishes almost
completely because of the high energy barrier. Again, the current flowing towards the
semiconductor remains the same.

Despite the simple structure and technology of the device, the theory of the Schottky diode is
quite complicated. As a consequence, there is no closed-form expression which describes the
current–voltage relationship. In practice, the following heuristic formula is often used:
 V

qV
I = I0 e nkT − 1 = I0 e VT − 1 , (5.15)
 

where VT = nkT /q is the thermal voltage, and n is called the ideality factor. Its value is between
n = 1 . . . 2 and has to be determined by measuring the device (or a few samples of a batch).

Lp
contacts (Au)

rb

Cp
c

semiconductor
(a) (b)

Figure 5.6: (a) Equivalent circuit model; (b) Schottky diode in beam-lead technology.

Varistor equivalent circuit model

The equivalent circuit model of a Schottky diode (Fig. 5.6a) comprises the following elements:

• The small-signal resistance of the junction depends on the quiescent current I (or equiva-
lently, on the bias voltage V ) and is determined as

∂V 1 VT V VT
 
r= = ∂I = exp − = . (5.16)
∂I ∂V
I0 VT I + I0

• The depletion layer exhibits a small capacitance c in parallel to r.

• The model further includes a bulk resistance rb .

• Package parasitics may be modeled by Cp and Lp . They are kept small in bare dies or
beam-lead packages (Fig. 5.6b).

When operated exclusively in reverse-bias, the junction resistance r is very high, and the
equivalent circuit model of a reverse-biased pn-diode (Fig. 5.1) may be used. In this case, the
variation of c is the dominant effect, and the Schottky diode acts as a varactor.
Chapter 5. Mixer Devices 77

However, we focus on the Schottky diode’s operation as varistor, which is achieved when the
bias and driving voltage swing the diode between reverse- and forward-biased region. In this
case, the variation of r(V ) becomes the dominant effect, and the equivalent circuit of Fig. 5.6a
should be used. c and rb also vary with the driving voltage but do not determine the overall
nonlinear operation.

GaAs
Si

V
-3V -2V -1V U

0V 1V 2V

Figure 5.7: Schottky diode small-signal resistance versus bias voltage.

It can be seen in Fig. 5.7 that the junction resistance varies in a very wide range of several orders
of magnitude (1 TΩ down to about 1 Ω) when the bias voltage is varied by less than 1 V. In
an environment with Z0 = 50 Ω system impedance, these values range from r  Z0 to r  Z0 ,
which is almost like a short or an open circuit. In this way, the diode can easily operate as a
bias-controlled switch or modulator.
The comparison of Si and GaAs technology shows that the higher electron mobility µn of GaAs
leads to a lower forward-bias resistance, and that the higher bandgap energy of GaAs leads to
a higher reverse-bias resistance and to a shift of the forward-bias region towards higher bias
voltages.
We conclude this section with a discussion of the Schottky diode’s cutoff frequency, which mainly
depends on the shunt-RC lowpass formed by r and c. Since r decreases with the forward-bias
voltage, the cutoff frequency increases. A common definition of the cutoff frequency uses r = rb ,
which is obtained at V ≈ VD − 0.1 V and leads to
1
fc = . (5.17)
4πrb c
Typical values of fc exceed several hundreds of GHz and may even reach 5 THz. For the
depletion layer capacitance, c ∼ A, where A is the area of the contact. The resistance rb is
mainly determined by the substrate resistance. Because of the small contact area and the
relatively thick substrate, the current narrows near the contact and flows into the contact from
all directions of the half-space. This leads to the (unusual) relation rb ∼ √1A . For the cutoff
frequency one obtains
1
fc ∼ √ , (5.18)
A
so that a small contact area A is favorable.

5.3 Field-effect transistor

Since transistor technology is discussed in previous chapters, we simply review the most important
nonlinear effects which may be exploited in field-effect devices to achieve frequency conversion:
78 5.3. Field-effect transistor

• The input nonlinearity of the transistor, IGS = f (VGS ), may be driven like a varistor. The
input then behaves like a Schottky diode, and the nonlinear gate resistance may be used
for frequency mixing.

• The transconductive nonlinearity, IDS = f (VGS ), may be used to operate the transistor
as “gate mixer”: the local oscillator and the RF signal are both applied to the gate, so
that the LO drive modulates the transconductance gm . The desired mixing products are
obtained and filtered at the drain.

• The output nonlinearity, IDS = f (VDS ), can cause two effects: the modulation of the
transconductance with the drain voltage (especially below the ‘knee’-voltage), and the
modulation of the channel resistance with the drain voltage. As “drain mixer”, the LO is
attached to the drain, the RF signal is fed into the gate, and the desired mixing products
appear at the drain.

The nonlinear input capacitance, Cgs and the parasitic gate resistance, rg usually do not
contribute to the mixing process. Also, the drain conductance gd = 1/R0 has only negligible
VGS -dependence and is thus not modulated when the LO drives the gate.
Chapter 6

Mixer Circuits

6.1 Schottky diode downconverter

Downconverters are mainly used in so-called superheterodyne receivers to translate the received
RF signal to a fixed intermediate frequency where it can be easily amplified, filtered, and
demodulated. The desired RF-band is selected by changing the LO frequency. Mixers are
linear for the small RF signal and thus more sensitive than rectifiers, which exhibit a square-law
characteristic.
As discussed in section 4.1 varactors are not suited for downconversion because of their high
conversion loss, which increases with the frequency ratio ωRF /ωIF . Instead, varistors (e.g.,
Schottky diodes) are preferred. In this section, the typical conversion gain and noise figure of a
Schottky diode downconverter is derived.

6.1.1 Conversion gain

The conversion gain analysis is carried out using the circuit topology of Fig. 6.1. The DC
bias and the local oscillator are not explicitly shown but the resulting large-signal voltage,
VLO (t) = VDC + V̂LO cos(ωLO t), drives the Schottky diode and appears in the circuit as periodic
modulation of the nonlinear diode resistance R(t) = R(VLO (t)). The received RF signal, here
supplied by the voltage source VS at frequency ωRF , shall be downconverted to ωIF = ωRF − ωLO
(upper sideband case).

RS RIM RIF

VS

Figure 6.1: Circuit topology (current drive) of a Schottky diode downconverter.

The topology employs current drive, i.e., only a finite number of current components pass
through the (ideal) filters and, hence, through the diode. By restricting the current spectrum,

79
80 6.1. Schottky diode downconverter

lower losses occur due to the diode’s bulk resistance. On the other hand, the voltage spectrum
at the diode is generally infinite and may consist of components at nωLO ± ωRF , (n ∈ Z). A
dual (voltage drive) topology with a finite number of voltage components and infinite current
spectrum is also possible but would have higher losses.

RF* IM* IM RF
IF* IF

Figure 6.2: General current spectrum of the downconverter.

In Fig. 6.2 the general (unfiltered) current spectrum is shown. The LO and its harmonics
are grey, and the components at ωRF and ωIF , which are desired for the mixer operation, are
highlighted in red. Other, undesired components may be converted to ωIF , (or ωRF ) and, thus,
must be suppressed. In practice, however, the so-called image frequency at ωIM = 2ωLO − ωRF is
often quite close to ωLO and ωRF and, therefore, difficult to reject with a filter.
When (as in the topology of Fig. 6.1) small-signal currents at ωIF , ωIM , and ωRF are considered,
the conversion matrix for effective voltage- and current phasors becomes
    
v RF R0 R1 R2 iRF
 v IF  = R1∗ R0 R1   iIF  . (6.1)
    
∗ ∗ ∗ ∗
v IM R2 R1 R0 iIM

The phasors at the image frequency are hermitian because they are taken from the negative
half of the spectrum (IM∗ ). It can be understood from Fig. 6.2 that among each other, the red
spectral components are related by frequency shifts of nωLO . Likewise, the blue components are
only related to each other. Consequently, the conversion matrix can only relate IM∗ , IF, and
RF (or IM, IF∗ , and RF∗ ). Recall from Section 4.2 that the hermitian matrix elements Rn∗ are
caused by frequency shifts with negative n.
Before computing the matrix elements, the periodic modulation of the diode’s small-signal
resistance R(t) must be obtained. We plug the LO voltage into Eq. 5.16:

VLO (t) VDC + V̂LO · cos(ωLO t)

!
VT VT
 
R(t) = exp − = exp − , (6.2)
I0 VT I0 VT

and the matrix elements are the Fourier coefficients Rn of R(t):

1
Z π Z π
VT −VDC /VT
Rn = R(t)e−jnθ dθ = e e−V̂LO ·cos(θ)/VT e−jnθ dθ, (6.3)
2π −π 2πI0 −π

where the abbreviation θ = ωLO t has been used.

Chapter 6. Mixer Circuits 81

10

0
0 1 2 3 4 5
x

Figure 6.3: Modified Bessel function of the first kind for different orders n.

The cosine in an even function, so R(t) must as well be an even function in time. Hence, the
Fourier coefficients Rn are real. The coefficients can be computed analytically and take the form
of modified Bessel functions In :
!
V̂LO
Rn = RT0 · (−1)n · In , with the auxiliary, bias-dependent resistance (6.4)
VT
VT −VDC /VT
RT0 = e . (6.5)
I0

In a real Schottky diode, the bulk resistance rb leads to a voltage drop that degrades the local
oscillator drive and changes the matrix elements Rm .
To simplify the following calculation of the conversion gain, we assume that the image frequency
is far enough from ωLO and ωRF so that it can be filtered. The filter presents an open circuit at
ωIM , so that the current iIM is zero, and the conversion matrix becomes two-dimensional:
! ! !
v RF R0 R1 iRF
= . (6.6)
v IF R1 R0 iIF

If the intermediate frequency is terminated with a load RIF , one can use this constraint and
rearrange the matrix equation to obtain the input resistance Rin,RF at ωRF :

R02 + R0 RIF − R12

Rin,RF = . (6.7)
R0 + RIF

After some manipulations, the transducer power gain GT ([R], RS , RIF ) becomes

Pdel,IF |v |2 |v S |2 4RS RIF R12


GT = = IF = ··· = . (6.8)
Pav,S RIF 4RS (R02 + R0 RIF − R12 + R0 RS + RIF RS )2

Trying to maximize the power gain by evaluating the above expression is quite cumbersome.
Instead, we use the following line of reasoning:
82 6.1. Schottky diode downconverter

RF IF

LO

Figure 6.4: Two-port interpretation of the conversion matrix of the downconverter.

1. Recall from Section 1.4 that the maximum power gain is obtained in a two-port for
simultaneously matched input and output ports. The conversion matrix describes the
two-port of Fig. 6.4, where the RF and IF signals are at the input and output, respectively.

2. The conversion matrix is symmetric, so the impedance terminations required for a simulta-
neous match must follow this symmetry: ZS = ZIF .

3. The matrix elements are real, so only resistances must be considered (ZS = RS , etc.)

4. From the above conditions follows: RS = Rin,RF = RIF = Rin,IF = R? .

q
The solution R? may be used in Eq. 6.7 to arrive at R? = R02 − R12 . Plugging R? into RS and
RIF in Eq. 6.8 yields the maximum gain
R12
R12 R02
Gmax =  q 2 =  r 2 , (6.9)
R0 + R02 − R12 1+ 1−
R12
R02

which is a only function of the ratio R1 /R0 . One can show that for all positive, even functions
(R(t) > 0), the Fourier coefficients are related by R1 < R0 . A high ratio R1 /R0 for good Gmax
is achieved when the diode is driven hard so that R(t) has sharp peaks (similar to a Class-C
current waveform), as is illustrated in Fig. 6.5.

V t
VDC

Figure 6.5: LO driving voltage V (t) leading to peak-shaped modulation of R(t)

From Eq. 6.4 one can see that R1 /R0 and, thus, the gain, depends only on V̂LO /VT . The gain
usually improves with the LO amplitude, and for large arguments V̂LO /VT , one can use the
Chapter 6. Mixer Circuits 83

asymptotic approximations of the modified Bessel functions for large arguments x, which read:
ex 1 ex 3
   
I0 (x) ≈ √ 1+ + ··· , and I1 (x) ≈ √ 1− + ··· . (6.10)
2πx 8x 2πx 8x
These approximations yield:
 q −2
Gmax ≈ 1 + 1/x , with x = V̂LO /VT . (6.11)

Gmax
1

0.5

0 5 10 15

Figure 6.6: Maximum gain of the downconverter versus LO drive.

While the gain is independent of the DC bias, the RF/IF impedance level R? varies with the LO
drive level and the DC bias voltage. For large x = V̂LO /VT , the impedance level reads:
ex VT −VDC /VT ex
R? ≈ RT0 √ = e √ . (6.12)
2πx I0 2πx
In order to maintain the same impedance level, the DC voltage must be increased for higher LO
drive levels x. Assuming the DC voltage is set for a certain impedance at x = V̂LO /VT = 1. Then,
at higher drive levels, the DC bias must be increased by ∆VDC , which, for large x, becomes:

∆VDC ≈ VT (x − ln x). (6.13)

Practical values of x are V̂LO /VT ≤ 10, by which the achievable conversion gain becomes
Gmax = 0.52 ≈ −3 dB (Fig. 6.6). When the LO drive is larger than the above limit, example
V̂LO > 0.25 V, out formulation of the LO voltage is not valid as there will be significant voltage
drop over the bulk resistance. Further, the semi-logarithmic plot of the diode resistance of
Fig. 5.7 shows that for larger LO amplitudes the upper and lower limits of R(V ) are reached; at
this point the conversion gain does not improve.

Conversion gain with image frequency considered

This subsection: old material

RIM = RS (image frequency close to RF signal)

For V̂LO  VT : Gmax ≤ 0.5
,→ PS divided evenly between RIF and RIM
Influence of rB and c rendered by average RB and C
84 6.1. Schottky diode downconverter

wS 1 2 2w-wS 3 wS-w
Rp

jXS jXIM RIF

RS RIM
C R(V)
VG ~
~

Figure 6.7: Mixer equivalent circuit model with realistic Schottky-diode model.

(Fig. 6.7):
no current drive of R(V ) due to capacitive shunt circuit at ωLO → compensation by external
inductance → residual conductance small for ωLO < 2CR 1
B
; current drive holds approximately
C present only at ωRF not at lower ωIF (Fig. 6.8).
wS wS-w
L Rp Rp
1 2

RS
C R(V) RIF
VG ~
~

Figure 6.8: Equivalent circuit model (RIM → ∞).

For ωLO → ωc = 1
CRB
IF-match increasingly difficult ⇒ mismatch ⇒ gain reduction
From here on: revised material

6.1.2 Noise figure

Downconverters inside receiver front-ends have to be very sensitive. The sensitivity is usually
not limited by the conversion gain but rather by noise. The noise power at the output (the IF
port) comes from thermal noise and from the shot noise of the Schottky contact.
The thermal noise comes directly from the bulk resistance at the IF, and indirectly, from the
bulk and external resistances at the RF and, unless filtered, the image frequency. The Schottky
noise, however, is the dominant contribution and shall be considered first.

Shot noise

Within the bandwidth B, the Schottky contact may be modeled as a noise current source with
mean squared amplitude
i2 = 2qIB, (6.14)
Chapter 6. Mixer Circuits 85

and an associated internal impedance R(V ), which is determined by Eq. 5.16. Equivalently, it
may be modeled as a noise voltage source with mean squared amplitude
v 2 = 2qIR2 (V )B ≈ 2nkT B · R(V ), (6.15)
where the commoly valid relation I  I0 is assumed.

RS

Figure 6.9: Noise equivalent circuit model for blocked image frequency and open IF port.

An equivalent circuit of the downconversion mixer with blocked image frequency (RIM → ∞) is
shown in Fig. 6.9. Since the noise figure does not depend on the output termination, the IF port
may be open-circuited (RIF → ∞), so that only noise voltages must be considered. The noise
power, v 2 /2R(V ) = nkT B, is constant, i.e., the noise spectrum is white. Again, there is not
only a direct contribution at ωIF but also indirect contributions due to mixing or modulation at
the periodically oscillating varistor R(V ).
The IF noise voltage is caused by frequency components at all combinations of mωLO √ ± ωIF .
For a sufficiently small (non-overlapping) RF-band, and using the abbreviation p = 2 nkT B,
at each frequency component, a noise voltage source with the following root-mean-squared
amplitude may be defined:
√
1. : p R cos(ωIF t + α0 ),
√
2. : p R cos[(ωLO + ωIF )t + α1 ],
√
3. : p R cos[(ωLO − ωIF )t + β1 ], (6.16)
√
4. : p R cos[(2ωLO + ωIF )t + α2 ],
√
5. : p R cos[(2ωLO − ωIF )t + β2 ], ...
The phase angles αi , βi are random and uncorrelated. Hence, the above voltages may be treated
like separate, uncorrelated voltage sources without internal impedance.
√
Since R(t) is periodic, so is R. We start with the Fourier series of R(t), which, without loss of
generality, may be an even function. Exploiting R−k = Rk , its Fourier series is written as
R(t) = R0 + 2R1 cos(ωLO t) + 2R2 cos(2ωLO t) + · · · . (6.17)
√
The function R is then also even:
√
R = r0 + 2r1 cos(ωLO t) + 2r2 cos(2ωLO t) + · · · , (6.18)
and its coefficients, rk , are related to the Fourier coefficients Rk of R(t):
R0 = r02 + 2r12 + 2r22 + · · · (6.19)
R1 = 2r0 r1 + 2r1 r2 + 2r2 r3 + · · · , etc. (6.20)

Now,
√ for each of the uncorrelated noise voltages of Eq. 6.16, there is a primary translation by
R to the IF, and a secondary translation, which is summarized by these steps:
86 6.1. Schottky diode downconverter

√
• conversion by R to a noise voltage at ωRF ,
• this noise voltage causes a noise current at ωRF to flow through the RF branch, R0 + RS ,
• and the current is subsequently converted by R to a noise voltage at the IF.
√
Notice that the translation due to the factor R in Eq. 6.16 (modulation of the white noise
source) is described by the Fourier coefficients rk , whereas the usual translation at the varistor
R is described by the Fourier coefficients Rk .

1. The first source of Eq. 6.16 yields a primary noise voltage at the IF:
pr0 cos(ωIF t + α0 ). (6.21)
The secondary translation starts with conversion of the white noise to an RF noise voltage:
pr1 cos(ωRF t + α0 ), (6.22)
which then causes a noise current to flow through the RF branch:
−pr1
cos(ωRF t + α0 ). (6.23)
R0 + RS
At the varistor, the noise current is converted to an IF noise voltage:
R1
− pr1 cos(ωIF t + α0 ). (6.24)
R0 + RS
Because the primary and secondary translations come from the same origin, they are
completely correlated (note how the phase α0 stays the same). Thus, the primary and
secondary IF noise voltages must be added:
R1
q  
2
vIFnoise,1 = pr0 − pr1 cos(ωIF t + α0 ). (6.25)
R0 + RS
2. The second source of Eq. 6.16 similarly yields a noise voltage at the IF:
R1
q  
2
vIFnoise,2 = pr1 − pr0 cos(ωIF t + α1 ). (6.26)
R0 + RS
3. Likewise, the third source yields:
R1
q  
2
vIFnoise,3 = pr1 − pr2 cos(ωIF t − β1 ), (6.27)
R0 + RS
and so on.

Since the IF noise voltages, vIFnoise,k , are uncorrelated to each other, the total IF noise power is
gathered by adding the powers (or, with open-circuited IF port, by adding the squared voltages):

=
X
2
vIFnoise 2
vIFnoise,k
k 
 " # 
p2 h R12
 
R1
= r02 + 2r12 + 2r22 + · · · 1+ − 4 [r0 r1 + r1 r2 + r2 r3 + · · · ]
i 
2 
| {z } (R0 + RS )2 | {z } R0 + RS 

2R1
 
R0
" #
R12 R0 R12
= 2nkT B R0 − 2 + (6.28)
R0 + RS (R0 + RS )2
Chapter 6. Mixer Circuits 87

√
For the above identity, the relation between the Fourier series of R and R of Eqs. 6.19 and
6.20 were used.
The noise figure may now be derived. It is defined as the ratio of the total noise power at the
output (IF port) and the output noise power caused by the thermal noise at the input (RF port).

2
vIFnoise
F =1+ , (6.29)
2
vIFnoise,th

where vIFnoise,th
2 is the IF noise power due only to the thermal noise of the load RS . The thermal
noise at the RF load may be modeled by a voltage noise source with mean squared amplitude

2 = 4kT BR .
vth (6.30)
S

This noise source has a white spectrum and results in a current flowing through the RF branch:

1
i2RFnoise,th = 4kT BRS , (6.31)
(R0 + RS )2

which is converted at the varistor to an IF noise voltage:

R12
2
vIFnoise,th = 4kT BRS (6.32)
(R0 + RS )2

In summary, the noise figure is

" ! #
n R0 R02 R0 RS

F =1+ 2+ −1 + , (6.33)
2 RS R12 R12

and it depends on the input (RF) load, RS , as well as the LO drive (by means of R0 and R1 ).
The noise figure is infinite for an RF short circuit (RS → 0) or open circuit (RS → ∞), which
makes sense because the input will be completely mismatched.
The minimum noise figure, Fm , is obtained by setting ∂F/∂RS = 0 and solving for RS :
s s !
R2 R2 R2
Fm = 1 + n 02 1 − 12 1+ 1 − 12 . (6.34)
R1 R0 R0
q
The minimum is at RS = R02 − R12 , which means that the condition for noise matching is
equivalent to matching for maximum conversion gain. The minimum noise figure may thus be
written as a function of the maximum conversion gain, Gmax :

n 1
 
Fm = 1 + −1 . (6.35)
2 Gmax

The result shows that a diode ideality factor of n = 2 leads to the same behavior as a passive
two-port, which has Fm = 1/Gmax . Thus, an ideality factor of n = 1 is favorable because then,
the mixer performs better than a passive two-port.
The best possible noise figure, Fm = 1 (no noise contribution) may be achieved regardless of the
ideality factor when Gmax = 1. However, due to the losses in practical circuits, this result is
only a theoretical limit.
88 6.1. Schottky diode downconverter

Thermal noise

Apart from shot noise of the Schottky contact, and the thermal noise at the mixers RF port,
which was already included in the above derivation, another contribution to the noise figure
is due to the thermal noise of the bulk resistance rb at ωRF and ωIF . The thermal noise at
ωIM = 2ωLO − ωRF must only be taken into account when the image frequency is not blocked.
However, we assume current drive and suppression of the image frequency, which leads to the
equivalent circuit in Fig. 6.10:

L rb rb

RS
c RIF

GS Gmax GIF

Figure 6.10: Equivalent circuit for inclusion of thermal noise under current drive and with
blocked image frequency (RIM → ∞).

The bulk resistance is taken out of the mixer and placed separately in the IF and RF branches,
where it incurs the same losses. The diode’s parasitic capacitance c is also included and may be
compensated (matched) by an external inductor. As indicated by the boxes, the circuit may
now be subdivided logically into three two-ports with gain and noise figure of, respectively,

(GS , FS ), (Gmax , Fm ), and (GIF , FIF ).

The second two-port is a downcoverter that is only characterized by shot noise, i.e., with the
properties derived in the last section. The first and third two-ports are cascaded with the mixer
and yield a total noise figure of

Fm − 1 FIF − 1
F = FS + + . (6.36)
GS GS Gmax

If (as is usually desired), the input and output of the first and third two-port are matched, their
noise figure takes the form F = 1/G, so that one gets

1 1
FS = , and FIF = . (6.37)
GS GIF

It thus follows for the overall noise figure of the mixer including thermal noise:

1 n(1 − Gmax ) 1 − GIF

F = + + . (6.38)
GS 2Gmax GS GIF Gmax GS

When the bulk resistance is zero, GS = GIF = 1 and one has F = Fm . For rb = 6 0, one has
F ≥ Fm , and the difference is especially pronounced when approaching ωc = 1/rb c: Near and
beyond the cutoff frequency, the shunt capacitance c lowers the impedance level required for IF
and RF matching, so that in comparison to RIF and RS , rb becomes significant.
Chapter 6. Mixer Circuits 89

Local oscillator noise

In addition to the thermal and shot noise caused by the varistor, the noise generated by the
local oscillator may degrade the mixer’s sensitivity. While an ideal oscillator generates a discrete
frequency and thus, exhibits a Dirac-shaped spectrum, real oscillators generally have a fluctuating
instantaneous amplitude and frequency, which results in a finite spectral width. The envelope of
such a spectrum usually decays with the offset to the center frequency, and its slope depends
on the oscillator Q. Spectral components of the LO at ωLO ± ωIF are also converted to the IF
where they interfere with the desired signal. Especially when a low intermediate frequency ωIF
is chosen, the corresponding offset to the LO center frequency is small. The LO noise may then
be quite significant because the noise envelope has not yet sufficiently decayed.
The LO noise contribution is reduced when high-Q filters are employed at the oscillator output
but sufficiently high Q is often only available with waveguide cavity resonators. These are heavy,
bulky, and expensive and thus, not always an attractive solution. Another option is the frequency
stabilization of the oscillator by a phase-locked loop (PLL). Still, such a solution increases the
complexity of the system.

i0 IF
RIF

(a) (b)

Figure 6.11: Single balanced mixer — (a) basic topology; (b) topology with RF isolation.

An effective solution at circuit-level is the suppression of LO AM-noise by use of an (anti-)

symmetric circuit. Consider the so-called push-pull- or single balanced mixer, which is depicted
in Fig. 6.11a and represents the most simple symmetric topology. In the figure, we assume
voltage drive with only the RF and LO voltages present.
A varistor with current–voltage relationship i(v) will, after reversing its polarity, exhibit a
relationship −i(−v). Similar to a push-pull amplifier, the first varistor is driven by the sum of
the RF voltage, VS , and the LO voltage, VLO , and the second varistor is driven by the difference.
For simplicity, the IF port is grounded, and we calculate the current i0 flowing towards ground
as
i0 = i(VLO + VS ) − i(VLO − VS ). (6.39)

Without an RF input signal (VS = 0), there is no current i0 . Even when a resistor (such as an
IF load) is placed between the diodes and ground, there will be no voltage drop. So any IF load
is isolated from both desired and undesired (i.e., noise) components of the LO signal.
From here on: old material

• but: i0 component at nωLO ± ωRF

• from: i0 = 2i0 (V )VS (conversion matrix analysis)

90 6.1. Schottky diode downconverter

1 Z/√2 2 Z

λ/4
λ/4
Z Z

3 Z/√2 4 Z

Figure 6.12: 3 dB-coupler in stripline technology.

Example: single balanced mixer with RF isolation Fig. 6.11b

voltage over RIF at ωIF
C short circuit for |nωLO ± ωRF | > ωIF )
Advantages of symmetrical and anti-symmetrical mixers: isolation of different signal paths
At RF: 3 dB-coupler instead of differential transformer (Fig. 6.12), simple and easy to realize.
Principle:
incident wave travels from 1 → 2
and is coupled via line 1 - 3 to line 3-4
and from 2 - 4 to line 3-4
forward travelling waves on 3 - 4 : in phase
backward travelling waves on 3 - 4 : 180◦ phase difference;
waves cancel each other for correct load impedance at 2 and 4.

power is divided equally between 2 and 4 (see Fig. 6.12).

proof: short circuit at 3 →
load at 4: Z
load at 2: Z2
→ is transformed to Z at 1.
V1 at 1 generates:
V1 V1
V2 = −j √ , V4 = − √ ⇒ 3dB-division
2 2
short circuit current at 3:
V1 √ V4
 
I3 = j + 2 =0
Z Z
isolation of port 1 and 3

0 0 1
 
j
1  j 0 1 0
S-matrix of coupler :

−√ 
 
2 0 1 0 j


1 0 j 0
Chapter 6. Mixer Circuits 91

j (V +V)

RF-short circuit (lowpass)

S
-VS Ö2

-jV 1 (V -V)
S
Ö2

Figure 6.13: Push-pull mixer in stripline technology.

In a push-pull mixer, the absolute phase of VS and V is not relevant, only their phase differences
between port 2 and 4 at IF.

advantages

1. separate LO and RF terminals (V , VS ), isolated except for varistor reflections

2. an additional 90◦ -phase shifter ( λ4 -transmission line) at one varistor can be used to isolate
the LO and RF terminals completely (assuming identical varistor reflections)

3. suppression of LO-AM-noise (amplitude variations of the signal)

push-pull mixer with 90◦ -hybrid in thin film technology (cm-waves)

(fs = 8 GHz, fIF = 500 M Hz) (see Fig. 6.13)
1
Fmin = = 4dB at 80% IF-bandwidth
Gm

6.2 FET downconverter

here: gate mixer (Fig. 6.14)

mixing of signal at ω1 and ω0 (ω1 < ω0 ) at nonlinearity µ(t) = gm (t) · R0 (t)
⇒ IF: ω3 = ω0 − ω1 image at ω2 = 2ω0 − ω1
assumptions:

• fixed LO (µ → µ(t)) with VG0

• Fi , Fi0 : ideal bandpass filters

∗ : conjugate-complex phasor of signal voltage
• VG1

with local oscillator (→ ω0 ) : gm → gm (t)

∞
gm (t) =
X
gm,k ejkω0 t
k=−∞
92 6.2. FET downconverter

I0 I0'
Z0 F0 Rg Rd F0' Z0'
VG0 ~ w0 (LO)
Vgs
w0
I1
*
Cgs R0 I4
*

Z1* F1 * * F1' Z4*

VG1* ~ w1 (signal)
V1
rg µ(T)Vgs ~ V4
w1
I2 I5
Z2 F2 Vin Vout F2' Z5
V2 V5
w2 (image) RS w2
I3 I6
Z3 F3 V3 V6 F3' Z6
w3 (IF) w3
Figure 6.14: Equivalent circuit model of FET mixer.

Z2π
1
gm,k = gm (t)e−jkω0 t d(ω0 t)
2π
0

R0 (t) ≈ R0

In practice:

⇒ µ(t) ≈ R0 gm (t) open-circuit voltage gain

loop voltage (outer mesh):

Vk = VGk − Ik Zk k = 1, 2 . . . , 6
VGk = 0 for k 6= 1

loop voltage (inner mesh, incl. conversion matrix elements): [V ] = [Zm ][I]:

V1∗ ∗ 0 0 Z14∗ 0 0 I1∗

    
Z11
 V2   0 Z22 0 0 Z25 0  I2 
0 0 Z33 0 0 Z36
    
V3 I3
=
    
    
V4∗ Z41 0
∗ Z43 Z44 0
∗ 0 I4∗
  
    
0 Z52 Z53 0 Z55 0
    
 V5    I5 
V6 ∗
Z61 Z62 Z63 0 0 Z66 I6
Chapter 6. Mixer Circuits 93

Realistic assumption: mixing products caused by gm,2 negligible (mixing of signal and image)
1
Zkk (ωk ) = Rg + rg + RS + , k = 1, 2, 3
jωk Cgs
Zkk (ωk ) = Rd + R0 + RS , k = 4, 5, 6
Z14 = Z25 = Z36 = RS
g0,k R0 gm,1 R0
Z41 = − + RS Z61 = −
jω1 Cgs jω1 Cgs
gm,0 R0 gm,1 R0
Z52 = − + RS Z62 = −
jω2 Cgs jω2 Cgs
gm,0 R0 gm,1 R0
Z63 = − + RS Z43 = Z53 = −
jω3 Cgs jω3 Cgs

With load matrix

Z1∗ 0 0 0 0 0
 
 0 Z2 0 0 0 0 
0 0 Z3 0 0 0
 
[Zt ] = 
 
 
0 0 0 Z4∗ 0 0

 
0 0 0 0 Z5 0
 
 
0 0 0 0 0 Z6
one obtains [VG ] = ([Zm ] + [Zt ]) [I], [VG ]T = [VG1
∗
, 0, 0, 0, 0, 0]
Conversion gain between 1 and 6, (RFin → IFout )

| I6 |2 Re {Z6 } I6 2
Gm = 2
= 4RG RL
| VG1 | /[4Re {Z1 }] VG1

Z1 = RG + jXG and Z6 = RL + jXL

Gm : complicated function of terminations at different frequencies.

simplifications for:
)
R0  RS , Rd
for typical microwave FET
gm,0 RS  1

ω3  ω1 : normal operation
!2
2gm,1 R0 RG RL
⇒ Gm ≈ · 2 ·
ω1 Cgs (RG + Rin )2 + XG − (R0 + RL )2 + XL2

1
ω1 Cgs

Rin = Rg + rg + RS
conjugate complex match at input/output terminals:

RG = Rin , XG = ω1 C
1
gs
RL ≈ R0 , XL = 0
2
gm,1 R0
⇒ max. Gain: Gmax ≈
4ω12 Cgs
2 R
in
94 6.2. FET downconverter

gm,1
mS
gm
10
Pinch-off

Vgs0

-3.3 0 0.5 V -4 -3 -2 -1 V 0
Vgs0

Vgs

Figure 6.15: Nonlinear drive of FET.

comparable to amplifier gain:

2
gm R0
Gmax, amp =
4ω1 Cgs Rin
2 2

local oscillator:
VG0 (t) = Vgs0 + V̂gs cos(ω0 t)

gm,1 large for Vgs0 ≈ VP 0 (Fig. 6.15)

Vgs also high: gm,1 ≈ gm,max /3 (cf. ideal step function: gm,1 = gm,max /π) (Fig. 6.16)
⇒ LO drive as large as possible

mS
10

Vgs0 » Vp0
gm,1

0
0 1 2 3 V
V

Figure 6.16: Local oscillator drive.

Chapter 6. Mixer Circuits 95

L.O.

RF trap lowpass filter

IF output
50W

l/4
D.C. block
D
G
signal S drain
coupler bias

lowpass

short circuit for RF IF-matching circuit

(transformation from 1.5-2kW to
50W using lumped elements)

gate bias

low impedance at IF

Figure 6.17: Mixer realized in microstrip technology

microstrip realization (Fig. 6.17)

FET:

• l = 1.4µm

• a = 500µm

Mixer:

• f ≈ 8GHz

• fIF = 30MHz

features, performance

– coupler for signal+LO → additional loss (noise %)

+ FET: mixing with gain (Fig. 6.18)

+ high output power (∼ 5dBm at 1 dB compression)

+ PLO for maximum gain &, if VP O &

+ excellent intermodulation: IP3 ≈ 20dBm at PLO ∼ 10mW

– high noise figure caused by 1/f- noise.

noise &
+ gate length & ⇒ (Fig. 6.19)
gain %
96 6.2. FET downconverter

10 FET: l = 1.4 µm
a = 500 µm
Gm
dB 5

0 f » 8 GHz
fIF= 30 MHz

-5
0 5 10 15 20 mW 25
LO power (PLO)
Figure 6.18: Gain of FET-mixer.

0.5 µm FET
1 µm FET
Gain 10
10
8
Gain
Noise N.F.
Figure 8 6
dB
dB 6 4
N.F.
Gain
4 2

-5 0 5 dBm 10
PLO
Figure 6.19: Noise of FET-mixer.
Chapter 6. Mixer Circuits 97

low pass

IF circuitry IF
G2
LO match
D
high pass
S
signal match G1

Figure 6.20: Equivalent circuit model of dual-gate-mixer.

mA

+0.5 70
0.22
60
0
VGS2/V -0.5 50

-1.0 40 IDS
30
-1.5
20
-2.0
10
0
V -3 -2 -1 0
VGS1
Figure 6.21: Drain-source current versus gate voltage.

Dual-gate-mixer (Fig. 6.20)

LO at G2 modulates transconductance of G1 , signal at G1 → IF at drain
transconductance of G1 (=∆I
ˆ DS /∆VGS1 ) depends strongly on VGS2 (Fig. 6.21), if 0.5V ≥
VGS2 ≥ 0V → high gain for low PLO (Fig. 6.22).
noise figures ∼ 6 − 8 dB
Dual-gate-mixer with image rejection

• low IF
• suppresses noise at image frequency

dB
30 VGS2 +0.22 V

20 -0.5 V
-1.0 V
10

0
Gm
-10
VDS = 5 V
-20 VGS1=-0.5 V
-30
-20 -16 -12 -8 -4 0 4 8 dBm
PLO
Figure 6.22: Gain versus LO drive.
98 6.2. FET downconverter

low pass
D

G2
0
90 IF
high pass hybrid
termination G1 A S A

G1 B S B
signal 0
3dB 90 high pass
hybrid G2

D
0
0 -power divider low pass

LO

Figure 6.23: Balanced configuration of dual-gate-mixer.

Balanced configuration (Fig. 6.23):

ω = ωLO + ωIF , ω 0 = ωLO − ωIF

phase upper lower

sideband sideband
FET A, G1
π 0 π
ej(ωt− 2 ) ej(ω t− 2 )
FET B, G1
0
ejωt ejω t
FET A, D j(ωIF t− π2 ) π
e ej(ωIF t+ 2 )
FET B, D ejωIF t ejωIF t
terminal A
A, D
π π
ej(ωIF t− 2 ) ej(ωIF t+ 2 )
B, D +ej(ωIF t− 2 ) +ej(ωIF t− 2 )
π π

terminal B
A, D ej(ωIF t−π) ejωIF t
B, D +ej(ωIF t) +ejωIF t

measured characteristics:

• gain: ∼ 10 dB

• image rejection: ∼ 20 dB

• VSWR < 2.0, f =8..12 GHz

• noise figure: ∼ 8.5 dB at fIF =30 MHz

Part III: Oscillators

After treating amplifiers and mixers in Part I and II of this lecture, the oscillator is introduced
in this chapter as the third major RF/microwave system component. Oscillators are needed in
almost any receiver or transmitter architecture; as local oscillators, they provide the large-signal
(pump) power which is necessary for the frequency translation in mixers. In radar or measurement
systems, they are used as a signal source.
Oscillators are fundamentally different from forced systems such as amplifiers or mixers, which
are driven by external signals. Oscillators are autonomous systems, which convert DC power to
RF power at their frequency of oscillation. Like all frequency conversion processes, oscillation
due to self-excitation depends on a nonlinear active device.
This part is structured into three main chapters. In the first chapter, the fundamentals of
oscillators are presented with conditions for oscillation start-up, determination of the oscillator’s
stationary state and its stability. The second section deals with oscillator devices: the IMPATT
diode, the Gunn-Element and the field-effect transistor. After that, the third section discusses
circuit topologies for stabilization of oscillators by application of resonators.

99
Chapter 7

Oscillator Basics [1]

The main ingredients for an oscillator are shown in the block diagram of Fig. 7.1: an active
element (commonly a diode or transistor) is employed because of its nonlinear behavior, a
resonator is needed to select the desired oscillation frequency with sufficient spectral purity, and
a (generally complex) load impedance ZL is present at the oscillator output. At low frequencies
the resonator is often implemented as lumped element LC-tank, whereas at high frequencies
distributed elements, such as transmission-line resonators, are generally used. An impedance
transformation network can be used to transform the (usually real) system impedance Z0 of the
outside world to any complex impedance at the marked reference plane. This matching network
is implemented with lumped elements at low frequencies and with transmission lines at high
frequencies. Consequently, at high frequencies, all transmission line elements in the circuit have
an impact on the performance, and thus, their effects must be considered already at the design
stage.

impedance
transformation
active device network
& resonator
reference plane

Figure 7.1: Block diagram of a one-port oscillator.

Two main types of oscillators can be distinguished: one-port oscillators, which are based on
two-terminal devices (e.g., diodes), and two-port oscillators, which rely on three-terminal devices
(e.g., FETs). This lecture focuses on one-port oscillators because they allow for a straight-forward
analysis. However, at the load impedance’s reference plane, it is irrelevant whether, looking away
from the load, the remaining circuit is based on a diode or a transistor. Hence, many of the
results gained from the analysis of one-port oscillators can be applied or extended to two-port
oscillators.
One-port oscillators can often be modeled as a negative differential resistance1 or conductance
with a series or parallel resonant circuit, respectively. These resistances/conductances are defined

1
Note that a negative resistor is unphysical since it generates power out of nowhere. A negative differential
resistance, however, generates RF power from its DC supply and is thus perfectly physical.

100
Chapter 7. Oscillator Basics [1] 101

as
∂V
−RD = and (7.1)
∂I
∂I
−GD = , respectively, (7.2)
∂V
and, as differential quantities, they are only valid in a small-signal equivalent circuit which has
been linearized about its operating point. Such an equivalent circuit is shown in Fig. 7.2 for
the case of a resistance with series resonant circuit, where −RD (with RD > 0) is the negative
small-signal resistance and XD the reactance of the active element. The resonant circuit is given
by a series LC tank, where Rloss accounts for the resonator loss. The circuit drives a real load
RL .

Figure 7.2: Equivalent circuit of a one-port oscillator with negative differential resistance.

The dual of Fig. 7.2, a negative differential conductance with parallel resonant circuit, is shown
in Fig. 7.3. All derivations in this chapter will be carried out for the resistance/series case but
can be easily adapted to the conductance/parallel case.

Figure 7.3: Equivalent circuit of a one-port oscillator with negative differential conductance.

Two-port oscillators can often be described as an amplifier with positive feedback. This is
illustrated in Fig. 7.4, where a coupling network is connected to the FET’s output. The coupling
network provides both load impedance matching and appropriate feedback (magnitude & phase)
to the FET’s input.
As already stated above, there is no fundamental difference between one-port and two-port
oscillators; when the oscillation reaches its steady state the load impedance RL absorbs the
power generated by the active device.
Coupling networks with minimum complexity can, for example, be implemented as T - or Π-
networks, which are shown in Fig. 7.5. While in Fig. 7.4 the load resistance RL is separated from
the lossless subnetwork to aid visualization of the power flow, it is regarded as part of the T - or
Π-network in Fig. 7.5: One impedance Zi , i ∈ {1, 2, 3}, contains the load resistance, such that
102 7.1. Conditions for oscillation start-up

D lossless
network
G S

Figure 7.4: Two-port oscillator [23].

(a) T -type coupling net- (b) Π-type coupling net-

work work

Figure 7.5: Coupling network topologies with minimum circuit complexity.

Zi = RL + jXL , and the other two are lossless reactances Zk = jXk , k ∈ {1, 2, 3}, with k =
6 i. In
total, four degrees of freedom are needed: two for load coupling and two for feedback coupling.

7.1 Conditions for oscillation start-up

Oscillations arise from small signals which are present in the circuit, and which grow exponentially
due to the circuit’s unstable poles. Noise is often the origin of oscillations because it is present
at virtually all frequencies and at every circuit node. Transients that are still present after
switch-on are another possible source. For noise, the linearized small-signal equivalent circuit
with RD and XD can be used, and as long as the current i(t) stays small enough, RD and XD
are independent of the current amplitude.

7.1.1 Case 1: One-port oscillator

Now we derive the conditions for oscillation start-up for the circuit in Fig. 7.2. To simplify the
following mathematical analysis, the resonator loss and the load resistance are combined as
R = Rloss + RL , and the diode’s reactance XD is regarded as part of the resonator’s reactance
ωL − ωC1
. Also, it is assumed that RD is frequency independent (at least, compared with the
resonator). These simplifications will be used throughout Chapter 7.
By applying Kirchhoff’s voltage law, one arrives at a second order ordinary differential equation
(ODE) for i(t):

di(t) 1
Z
L + (R − RD )i(t) + i(t)dt = 0 (7.3)
dt C
d2 i(t) R − RD di(t) 1
⇐⇒ + + ω02 i(t) = 0, with ω02 = . (7.4)
dt2 L dt LC
Chapter 7. Oscillator Basics [1] 103

The solutions of this ODE are usually2 of the form

Ak exp(pk t), with pk = σk + jωk , k ∈ {1, 2}. (7.5)

The eigenvalues, or poles p1 and p2 , of the resulting equation are

s
RD − R (RD − R)2 C
p1,2 = ± jω0 1 − . (7.6)
2L 4L

Oscillation start-up is only possible if σk > 0, which implies RD > R. Also, since we are only
interested in periodic solutions, p1 and p2 must be a complex-conjugate pair, which is only true
for under-damped systems, where one has

σ 2
(RD − R)2 C
= < 1. (7.7)
ω0 4L

7.1.2 Case 2: Two-port oscillator [2]

To fulfill the conditions for oscillation start-up in a two-port oscillator, one has to make the
active device unstable. Most transistors are designed for absolute stability over a broad range of
frequencies but by applying positive feedback, one can often bring the device into an unstable
operating point.

feedback
1 2
active
1 2
element 3
feedback active
element 3
(a) series feedback topology (b) parallel feedback topology

Figure 7.6: Feedback topologies for two-port oscillators.

Two common configurations for positive feedback are shown in Fig. 7.6. The first is series
feedback at the transistor’s emitter/source terminal, and the second is shunt feedback between
the transistor’s collector/drain and base/gate terminals. Although this description is based on
the common-emitter or common-source topology, similar feedback configurations can be applied
in other topologies as well.
For a transistor oscillator design, one could then follow this step-by-step procedure:

1. choose feedback type (series or shunt)

2. plot stability circles for the two-port including feedback networks for different bias volt-
ages/currents and feedback impedances to arrive at an unstable circuit

3. choose a termination in the unstable region of the Smith Chart

2
This assumption is not true for critically damped systems, in which two identical eigenvalues lead to a solution
of the form (A + Bt) exp(pt).
104 7.2. Stationary operation

4. adjust the phase of the waves that are reflected by the resonator, e.g., by varying the
length of the connecting transmission lines.

The last point is important: choosing a load outside of the stable region indicated by the stability
circles is only a necessary condition for oscillation start-up. The phase determines whether the
feedback is positive or negative.

7.2 Stationary operation

Now that the phenomenon of oscillation start-up has been investigated with linear models, the
next step is to determine the oscillator’s steady state; for example, we might be interested in the
stationary current amplitude. Note that the exponential current growth has to diminish because
unlimited growth is unphysical.
In stationary operation, currents are high enough that the small-signal equivalent circuit loses
its validity. Instead of the small-signal resistance RD we then adopt a large-signal resistance
RD , which is defined later. For now, it is only important to note that in physically realizable
systems, RD has to decrease if the current increases beyond a certain amplitude. In stationary
operation, the net series resistance of the circuit evolves to zero:

lim RD = R. (7.8)
t→∞

To calculate stationary current amplitudes, one needs a nonlinear model of the active device.
These models can grow quite complex; numerical techniques are often unavoidable. In this
example, however, the nonlinear I-V relationship is given by the (rather simple) van der Pol
model
V (t) = −RD I(t) + KI 3 (t), (7.9)

in which K determines the large-signal behavior. Since XD has previously been absorbed into LC,
the current dependence of the diode’s reactance is neglected (again, for further simplification):
ˆ → ∂XD = 0.
XD 6= f (I) ∂ Iˆ
For a quantitative analysis, we start with the observation that due to the resonant circuit, the
current only has one frequency component:

I(t) = Iˆ cos(ωt). (7.10)

Now the method of the describing function is applied. The method was originally invented for
nonlinear control systems and later also applied in nonlinear circuit analysis. In this method,
the current I(t) is regarded as an input to the nonlinear system. This input results in a voltage
V (t), which is understood as the output of the system. The describing function relates the
spectral components of V (t) and I(t) at the fundamental frequency ω and has the dimension of
a resistance.
Plugging the current I(t) into the van der Pol model results in

3 1
V (t) = −RD + K Iˆ2 Iˆ cos(ωt) + K Iˆ3 cos(3ωt).
 
(7.11)
4 4
| {z }
V̂
Chapter 7. Oscillator Basics [1] 105

The large-signal voltage V (t) at the diode has spectral components at ω and 3ω. Only the
ˆ and results from
component at ω is considered for the describing function, which is called RD (I)

ˆ = V̂ = −RD + 3 K Iˆ2 .
− RD (I) (7.12)
Iˆ 4

ˆ turns out to be the large-signal resistance introduced earlier, and

The describing function RD (I)
the method yields a quasi-linearization of the nonlinear system. When a time-harmonic input is
used (as with I(t) in our case), the method is also sometimes termed harmonic linearization.
RD decreases quadratically with Iˆ and is plotted in Fig. 7.7a.
Applying Kirchhoff’s voltage law as condition for stationary oscillation leads to

ˆ = 0.
I R + jXL − RD (I) (7.13)
 

Eq. 7.13 is a complex equation which yields two equations for two unknowns. The oscillation
frequency ωosc is obtained from the imaginary part

1
jXL = 0 ⇒ ωosc = √ , (7.14)
LC

and the amplitude of oscillation Iˆosc is obtained from the real part

4 1
r
− RD (Iˆosc ) + R = 0 ⇒ Iˆosc = (RD − R). (7.15)
3K

The solution of the complex equation system is illustrated by the loci in Fig. 7.7b, where the
intersection of both loci results in the stationary solution.

(a) The large-signal resistance RD is monotoni- (b) Loci of ZL (load line) and RD (device line)
cally decreasing with the current amplitude intersect at the stationary solution.
ˆ
I.

Figure 7.7: Graphical solution of oscillation frequency and amplitude for a series resonant circuit.
106 7.3. Stability of the stationary solution

7.2.1 Two-port oscillators

Two-port oscillators generally have more complicated loci[23]. However, the maximum output
power PL,max can be estimated from the saturation characteristics of the transistor amplifier
which is often at the heart of the two-port oscillator.
For the amplifier without feedback, one can plot the input/output power relationship, which
is shown as red curve in Fig. 7.8. The amplifier with feedback has a power flow relationship
PL = PD − PG , as was previously illustrated in Fig. 7.4 and plotted as blue curve in Fig. 7.8.

Figure 7.8: Output power level.

The maximum output power is attained where the partial derivative with respect to PG is zero:

∂PL
= 0 (7.16)
∂PG
∂PD
⇔ = 1 (7.17)
∂PG

From this condition, and with knowledge of the PD (PG ) characteristic, the maximum output
power can be computed.

7.3 Stability of the stationary solution

In the previous section, a stationary solution for the oscillator was found. However, we have
not yet addressed the question whether this stationary state is stable. The word “stable” is not
to be confused with the (in)stability during oscillation start-up. Here, “stable” means that the
stationary solution has to be robust against small changes.
For example, noise may perturb the stationary state and lead to small fluctuations ∆i of the
oscillation amplitude. In a stable (robust) system these fluctuations decay, and the system
returns back to its original state.
In order to assess robustness against perturbations, the frequency- or current-dependent terms
Chapter 7. Oscillator Basics [1] 107

in the oscillation condition are expanded into a first-order Taylor polynomial:

dRD dXL
− RD (Iˆosc ) − · ∆i + R + jXL (ωosc ) + j · (∆σ + j∆ω) = 0, (7.18)
dIˆ Iˆosc
d(jω) ωosc

where ∆i is the (noise) current fluctuation, ∆ω is the change of the oscillation frequency, and
∆σ accounts for the change of the damping factor (towards exponential increase or decrease).
Subtraction of the stationary oscillation condition (Eq. 7.13) leads to the simpler expression

dRD dXL
− · ∆i + · (∆σ + j∆ω) = 0. (7.19)
dIˆ Iˆosc
dω ωosc

Looking at the imaginary part of this complex equation, we get

dXL
· ∆ω = 0 ⇒ ∆ω = 0, (7.20)
dω ωosc

which means that current fluctuations ∆i do not result in a change of frequency.

Evaluating the real part we obtain

dRD dXL
− · ∆i + · ∆σ = 0 (7.21)
dIˆ Iˆosc
dω ωosc
 
dRD
∆σ dIˆ Iˆosc 
=  (7.22)

∆i dXL

dω ωosc

For a stable solution, an increase in current (∆i > 0) must be counteracted by a shift towards
exponential decrease (i.e., more damping, ∆σ < 0), and vice-versa. A stable stationary state
can thus be identified by  
dRD
∆σ  dIˆ Iˆosc 
= < 0, (7.23)
∆i dXL

dω ωosc

which is also known as Kurokawa’s criterion.

The underlying assumption for the above derivation is that of the simple series resonant circuit,
where RD and ZL intersect perpendicularly. However, in reality the matter is usually more
complicated. For instance, the load impedance may have a frequency-dependent real part R(ω),
or the diode’s reactance may be current-dependent. In the most general case, every part of the
ˆ
model may be frequency- or current-dependent, and these ω- or I-dependent terms must then
be considered in the Taylor polynomial of Eq. 7.18. Such modifications of the model lead to
different real and imaginary parts for the above complex equation, and the loci might be difficult
to visualize.
The locus ZL of a practical system where parasitic reactances and transmission line elements
lead to a frequency-dependent real part R(ω) is shown in Fig. 7.9. It is important to note that
Kurokawa’s criterion also holds in this more complicated case.
The load locus contains several loops, and three points fulfill the oscillation condition (P1 , P2 ,
and P3 ). However, not all of these solutions are stable:

• P1 is unstable because dRD

dIˆ
and dXL
dω have the same sign.
108 7.3. Stability of the stationary solution

Figure 7.9: Locus of oscillator with series- and additional parallel resonant circuit.

• P2 and P3 are stable.

The history of the system determines which operating point (P2 , P3 ) is reached. Detuning of the
resonator shifts the locus of ZL up- or downwards. When shifting the locus upwards, at some
point the only solution is P3 . Conversely, when shifting the locus downwards, at some point
only P2 is left as stable solution. The hysteresis and mode-jumping often observed in oscillators
can be the result of such loops in the resonator/load locus.
Chapter 8

Oscillator Devices

The nonlinear devices on which oscillators are based will be discussed in this section. Only
solid-state devices are treated here, and our focus is on the (probably) less known IMPATT diode
and Gunn element. It is assumed that knowledge of the transistor’s basic device physics has
been established by previous lectures on solid-state devices. Furthermore, advanced transistor
technologies for microwave and mm-wave frequencies have already been covered in Part 1 of
this lecture. Therefore, these things will not be repeated, and we only briefly highlight certain
transistor properties which are important for oscillator performance.
For a long time, IMPATT diodes and Gunn elements were the only solid-state devices that could
be effectively used for mm-wave signal generation up to and beyond 100 GHz, and they were only
rivaled by vacuum tubes, which are favorable in high power applications. With the advances in
transistor technology, the transit frequency fT continues to increase, and GaAs HEMTs have
been employed as fundamental oscillators even beyond 100 GHz. However, IMPATT diodes and
Gunn elements are still important because they offer better performance in certain areas, and
engineers can rely on decades of experience with these devices.

8.1 IMPATT Diode [7, 1, 24, 25]

The IMPATT (IMPact-ionization Avalanche Transit-Time) diode is named after the two main
effects that are exploited. The first effect is avalanche multiplication of charge carriers, which
leads to a delay due to the buildup time of the avalanche current. The second effect is a
transit-time delay due to the drift of carriers among the diode structure. These two delays add
up, and when the resulting phase lag between current and voltage exceeds 90degree the real part
of the diode impedance becomes negative. IMPATT diodes that are used as oscillators are often
called IMPATT oscillators.
The main benefits and drawbacks of IMPATT diodes are:

⊕ IMPATT diodes offer the possibility to generate high power at high frequencies and have
been used to generate or amplify microwave signals beyond 100 GHz. More than 1 W of
CW power at 40 GHz can be achieved.

⊕ Efficiencies between 10% and 20% can be obtained.

IMPATT oscillators exhibit more phase noise (approximately 10 dB more than Gunn
elements) due to the randomness of the avalanche multiplication process.

109
110 8.1. IMPATT diode

The device has a large reactance which also depends on the amplitude of oscillation and
may lead to detuning or even burnout of the device.

It may be difficult to integrate the diode because a heat sink is often needed.

8.1.1 Principle of Operation

IMPATT diodes can be realized in many different ways. One device realization is the p+ nin+ -
structure, which is an instructive example because it allows for a separate treatment of avalanche
effect and transit-time delay. By example of this structure the operation of IMPATT diodes is
explained.

avalanche region

drift region

operating
point

avalanche
breakdown

(a) IMPATT diode I-V curve. (b) E-field distribution in an IMPATT diode
with p+ nin+ -structure.

Figure 8.1: Properties of a p+ nin+ structure with abrupt junctions.

Essentially, IMPATT diodes are pn-diodes that are biased into the region of reverse avalanche
breakdown. The I-V curve of the p+ nin+ -structure is depicted in Fig. 8.1a along with its
operating point. Assuming abrupt junctions between regions of different doping, the E-field
distribution may look as in Fig. 8.1b. There is a field maximum at the p+ n-junction, and
in the region between x = 0 and x = wa the E-field is above breakdown: |E(x)| > EB ,
(EB ≈ 4 · 105 V/cm in Si). Adjacent to the avalanche region is the drift region where the electrons
travel at the saturation drift velocity vs (vs ≈ 107 cm/s in Si). The width w of the drift region is
approximately equal to the width of the i-region because the n-region is usually very small.
The avalanche effect starts when electron-hole pairs are accelerated by the electric field such that
they accumulate enough energy to knock off other charge carriers by impact ionization. The
electron-hole pairs that are generated in this way may in turn cause further impact ionization,
which leads to an avalanche-like increase of charge carriers.
One defines the ionization rate α(x) as
Z W !
α(x) = αn (x) exp αn (ξ) − αp (ξ)dξ , (8.1)
x

where αn and αp are the electron ionization rate and hole ionization rate, respectively. α(x) tells
how many electron-hole pairs are generated by one charge carrier per length, and in materials
where the electron- and hole ionization rates are equal, one has α = αn = αp .
Chapter 8. Oscillator Devices 111

At the onset of avalanche multiplication, each electron-hole pair generates exactly one additional
carrier pair on average while traversing the device. To determine the condition for avalanche
multiplication, the ionization rate must be integrated over the whole device length, and one
obtains: Z W Z wa
α(x)dx ≈ α(x)dx = 1. (8.2)
0 0
In the above formulation we have made use of the fact that α(x) depends on E(x), which results
in a highly localized avalanche region and a very steep function α(E(x)). For example, an
electrical field of E = 5 · 105 V/cm, which is above breakdown in Si, leads to an ionization rate
of α = 105 cm−1 . Outside of the avalanche region, the ionization rate is usually several orders of
magnitude lower.
In the case of static fields, the integral of Eq. 8.2 may not exceed unity, or otherwise the device
may be destroyed. For time-varying fields E(t), however, the above integral may exceed unity for
a fraction of the oscillation period. During this time, the current increases almost exponentially.
In Fig. 8.2, the voltage and current waveforms of the IMPATT diode are illustrated: The voltage
V (t) between the device terminals is depicted in Fig. 8.2a and can be written as
V (t) = V0 + V̂1 sin(ωt), (8.3)
where V0 is a DC bias close to the breakdown voltage, V0 ' VB , and V̂1 is the amplitude of a
superimposed RF voltage. All other voltage spectral components are short-circuited, for example,
by means of an external parallel-resonant circuit.
Considering first the avalanche region, during the positive half-wave, when |E| > EB , the
ionization rate α(x) increases rapidly in this region, leading to the generation of additional
charge carriers. Assuming that this is the dominating generation process, we can neglect
contributions due to drift or diffusion in the avalanche region. Further assuming that electrons
and holes have equal drift velocities vs and equal ionization rates, α = αn = αp , we arrive at the
same equations for the electron and hole generation rates:
∂n ∂p
= αvs (n + p), and = αvs (n + p). (8.4)
∂t ∂t
Due to symmetry, one can eliminate the hole density p in Eq. 8.4 and arrive at
∂n
= 2αvs · n (8.5)
∂t

(a) (b) (c)

Figure 8.2: Voltage and currents at IMPATT-diode.

112 8.1. IMPATT diode

which yields the important result that the electron generation rate is proportional to the
ionization rate α and to the carriers already present in the n-region.
The above differential equation for n can also be rewritten for the injected electron current
In = Jn · A = qvs nA, where Jn is the injected electron current density, A is the diode’s cross
section, and q is the elementary charge. Then, in a similar fashion, the increase of injected
electron current is proportional to α and the instantaneous injected electron current:
∂In
= 2αvs · In . (8.6)
∂t

The ionization rate α(x) is a monotonically increasing function of the instantaneous field. Thus,
it follows the externally applied voltage V (t) in phase. However, as can be seen in Fig. 8.2b the
injected electron current In (t) is not in phase with V (t); it still increases, even after the voltage
has passed beyond its peak value. This is because its increase depends on its instantaneous value.
The peak current is reached when the avalanche effect ceases, which is approximately when the
external voltage approaches its mean value V0 . The resulting injected electron current In (t) has
a pulse shape and its fundamental spectral component In,ω lags the external RF voltage V̂1 by
approximately 90◦ . In phasor notation the result is

I n,ω ∼ −jV 1 . (8.7)

In the previous description we have only considered the generation of electrons. This is because
the generated holes can quickly flow into the p+ region where they do not contribute significantly
to the diode current ID (t) at the device terminals. The generated electrons, on the other hand,
flow into the drift region where they are delayed by the transit time τd . In the drift region the
E-field is high enough that the electron can travel at their saturation drift velocity vs . The
transit-time delay τd is determined by the width wd of the drift region,
wd
τd = . (8.8)
vs

Now let us consider the drift region. The electron current In which is injected into the drift
region can be expressed as space-charge wave. Since the avalanche region is very small, we
neglect it for the electron drift process, which simplifies the resulting expressions. We can then
assume that the electrons drift from x = 0 (at t = 0) to x = wd (at t = τd ) and write the
space-charge wave as
x
 
In (x, t) = In t − . (8.9)
vs

Electrostatic induction1 caused by the drifting electrons leads to the diode current ID (t) at the
device terminals:
1 1 t−τd
Z wd Z
ID (t) = In (x, t)dx = − In (η)dη. (8.10)
wd 0 τd t

We are mainly interested in the fundamental spectral component of ID (t). The phasor of this
spectral component is called I D,ω and can be computed as a function of I n,ω :

I n,ω sin(Θ/2) −jΘ/2

I D,ω = (1 − e−jωτd ) = I n,ω e , with Θ = ωτd . (8.11)
jωτd Θ/2
1
German native speakers, please be aware that electrostatic induction is English for “Influenz” and not to be
confused with electromagnetic induction.
Chapter 8. Oscillator Devices 113

For a detailed derivation of Eqs. 8.10 and 8.11 you may revisit the section about the base-collector
junction delay in BJTs, which is covered in the lecture notes of Part 1.
The fundamental phasor of ID has a phase lag of Θ/2 with respect to the fundamental phasor of
In , see Fig. 8.2c. The total phase lag Θtot of I D,ω with respect to V 1 is the sum of the phase lag
due to the avalanche charge buildup (→≈ 90◦ ) and due to the transit-time delay (→ Θ/2):

π+Θ
Θtot ≈ . (8.12)
2

Figure 8.3: Phasor diagram.

This relation is also visualized in the phasor diagram of Fig. 8.3, where it can be seen that the
projection of the diode current phasor I D,ω onto the voltage phasor V 1 results in an negative
(opposite phase) component. This is observed as negative real part of the device impedance at ω,
and it leads to a negative real power (P < 0), i.e., power which is delivered to the device output:

V̂1 IˆD,ω π+Θ 1 ˆ Θ

   
P = Re(V 1 I D,ω )
∗
= √ √ cos = − V̂1 ID,ω sin . (8.13)
2 2 2 2 2

If the width of the drift region is

πvs vs
wd = = , (8.14)
ω 2f
the voltage V 1 and the current I D,ω are in opposite phase. For example, an IMPATT diode in
Si (where one has vs = 107 cm/s) operating at f = 10 GHz must have a drift region width of
wd = 5 µm to fulfill the above condition.

8.1.2 Efficiency

IMPATT diodes are used in high power applications, where efficiency is often an important
concern. For IMPATT oscillators the DC to RF conversion efficiency must be considered. It is
defined as the ratio of the RF power delivered to the output and the DC power consumed by
the diode:
PRF
ηosc = . (8.15)
PDC

Under the assumption of very sharp pulses for In (t), the resulting diode current ID (t) has a
rectangular shape (also cf. Fig. 8.2c):

Imax for ωt ∈ [π, π + Θ]

(
ID (t) = (8.16)
0 otherwise.
114 8.1. IMPATT diode

The DC component of such a rectangular function can be found in Fourier tables and is
Θ
ID,0 = Imax . (8.17)
2π

The fundamental spectral component amounts to

2 Θ
IˆD,ω = Imax sin
 
. (8.18)
π 2

Using Eq. 8.13 for the magnitude of PRF , and with PDC = V0 ID,0 , the DC to RF conversion
efficiency becomes
1/2 · V̂1 IˆD,ω sin(Θ/2) V̂1 sin2 (Θ/2)
ηosc = = (8.19)
V0 ID,0 V0 Θ/2

For Θ = 0, 2π, . . . , n · 2π the efficiency becomes ηosc = 0. The angle Θ that results in maximum
efficiency can be found numerically: It is Θ = 2.33 ≈ 43 π and results in a theoretical maximum
efficiency of
V̂1
max ηosc = 0.725 . (8.20)
V0

Assuming a ratio of V̂1 /V0 = 0.5 between RF amplitude and DC bias, the maximum efficiency is
ηosc = 36%. Practical values are Θ ≈ π2 . . . π and ηosc ≈ 10% . . . 20%.

8.1.3 Diode Impedance

The foregoing derivations have neglected capacitive currents and the drive-dependent variation
of the diode impedance, which is important under large-signal excitation.
A simplified model for the diode impedance when the drift angle is Θ = π is given as
 
1  2
ZD = RD + jXD = (8.21)

   − j
,
ωCD
π 1 − ω2 Φ(
ω2

V̂ ) a 1

where CD = A
wd is the diode capacitance (A is the diode cross-sectional area, and again because
wa  wd , the width of the avalanche region is neglected). The function Φ(V̂1 ) accounts for
nonlinear effects, i.e., the variation of RD with the voltage amplitude V̂1 .
Φ has the following properties:

• Φ(0) = 1, this determines the small-signal impedance of the IMPATT diode.

• Φ(V̂1 ) is monotonically decreasing with V̂1 .

• Φ(V̂1 ) is proportional to 1
V̂1
for large values of V̂1 .

ωa is called avalanche frequency and depends on the diode’s DC current. At the avalanche
frequency, the inductive current In is in resonance with the capacitive current in the avalanche
region, which leads to an infinite small-signal resistance. From Eq. 8.21 it is also obvious that a
negative small-signal resistance can only occur for frequencies ω > ωa . Only then, oscillation
start-up is possible.
Chapter 8. Oscillator Devices 115

For increasing oscillation amplitude V̂1 , −RD decreases as determined by the function Φ(V̂1 ).
This means that by application of Kurokawa’s stability criterion, an external series resonant
circuit, which must compensate CD and hence be inductive at the frequency of oscillation leads
to a stable stationary oscillation.

8.1.4 Output Power

The output power of IMPATT diodes is limited by two effects:

• Thermal effects play an important role up to approximately 10 GHz. This is because in

the lower range of frequencies IMPATT diodes can deliver several Watts of output power,
and the heat sink is the main limitation. For example, an output power of 10 W with an
efficiency of 20% means that 40 W of power are dissipated as heat.

• Electronic effects become the dominating limitation at frequencies above 10 GHz. If the
output power is too large, avalanche breakdown may also occur in the drift region, and
the IMPATT diode can no longer be safely operated.

For a constant diode reactance, the thermal limit leads to a power-frequency dependency of
P ∼ 1/f . To understand this, we assume that the diode is attached to a heat sink, and the
whole assembly has a thermal resistance RT
w
RT ≈ (8.22)
κA
where w ≈ wd is the width of the diode and thus the thickness of the heat path. κ is the thermal
conductivity, and A is the cross sectional area.
The relation between the increase of junction temperature ∆T and the power dissipated in the
diode is determined by the thermal resistance
∆T
P = (8.23)
RT
and usually, there is a maximum junction temperature, which limits ∆T and thus P .
In order to show P ∼ 1/f , we multiply power and frequency and expand with a factor of 2π:
∆T κA∆T 2π κA∆T
P ·f = ·f = ·f = · · f. (8.24)
RT w 2π w
After rearranging the term
κ∆T A
P ·f = · 2πf · = const. (8.25)
2π | {z w}
ωC

it is obvious that for constant diode reactance XD = − ωC

1
the maximum power is inversely
proportional to f .
The electronic limit leads to a power-frequency dependency of P ∼ 1/f 2 . Again, we assume
constant reactance and infer from the discussion of the phasor diagram in Fig. 8.3 that w ' 2f
vs

(or, at least w is proportional to f ). The breakdown condition

vs

vs
Vmax ≈ EB w ' EB (8.26)
2f
116 8.1. IMPATT diode

leads to a maximum voltage across the drift region.

The maximum injected electron current is also limited by breakdown due to the increased E-field.
Static fields in semiconductors are described by Poisson’s equation

∂E qn Jn
= − , with n = (8.27)
∂x  qvs
Jn In
= − =− . (8.28)
vs vs A

The maximum injected current is found by integrating, solving for In , and taking |E| < EB as
upper limit:
A
In,max = vs EB = Cvs EB . (8.29)
w
Although the diode current is the quantity that contributes to output power, it is sufficient to
look at the injected electron current because their fundamental components are proportional.
The maximum output power is then proportional to

vs2
P ∼ Vmax In,max = EB2 · C · , (8.30)
2f

and it can be seen that the output power is proportional to 1/f 2 :

vs2 v2
P · f 2 ∼ EB2 · f C · = EB2 · 2πf C · s = const. (8.31)
2 | {z } 4π
ωC

It may seem like a trick to get rid of an additional ω by assuming a constant reactance. However,
the practically used impedance levels are standardized (i.e., fixed) and thus device reactances
must also be within a reasonable range. In order to compare theoretical performance of devices
across several decades of ω in a fair way, the constant reactance assumption is helpful.
In Fig. 8.4 the experimentally measured performance of Si and GaAs IMPATT diodes is plotted.
One can see the thermal limit and that it can be overcome by pulsed operation. Another
observation is that the devices based on Si are the preferred technology beyond 100 GHz.

8.1.5 IMPATT Diode Designs

Throughout this section we have only considered the p+ nin+ -structure for IMPATT diodes and
used it to explain the functional principle. Another example is the p+ nn+ -structure, which is
grown by epitaxial layers, as shown in Fig. 8.5. Often, the diode is designed as mesa-type diode,
which means that it is mounted upside-down so that the avalanche region, where most of the
power is dissipated, is near the heat sink.
According to the structure shown in Fig. 8.6 the avalanche and drift regions of the p+ nn+ diode
can not be separated.
Another possible implementation is the ’double-drift’-diode p+ pnn+ :, where the breakdown
happens at the pn-junction in the middle of the structure. The advantage of this structure is
that both electrons and holes can drift, which results in a higher efficiency and output power.
The Schottky-nn+ -diode does not have a p+ contact region and instead uses a metal-semiconductor
Schottky barrier. The maximum field then occurs at the Schottky contact, which also serves as
Chapter 8. Oscillator Devices 117

50

Output Power [dBm] 40

30

20

10 Si, CW
Si, pulsed
GaAs, CW
GaAs, pulsed

0
1 10 100 1000
Frequency [GHz]

Figure 8.4: Output power of IMPATT-diodes, filled: pulsed operation, open: CW operation.

Au top metalization package top

Au bondwire ceramic

diode
solder

Au bottom metalization Cu heatsink

(heatsink contact) package bottom
(a) mesa-type diode ’upside-down’ (b) coaxial metal-ceramic housing

Figure 8.5: Fabricatoin and packaging of a mesa-type p+ nn+ IMPATT diode.

heat sink. The second advantage is that this type of IMPATT diode can be fabricated at low
temperatures, so that the quality of the epitaxial layer can be preserved.
The material that are mostly used for IMPATT diodes are Si and GaAs. Si-diodes exhibit higher
noise since ionization rates for electrons and holes are different. Recently, SiC (Silicon Carbide)
and GaN (Gallium Nitride) are under investigation as promising candidates for THz signal
118 8.2. Gunn-Element

avalanche region

drift region

Figure 8.6: pnn field distribution

sources with high output power.

8.2 Gunn-Element [7, 1, 24, 25]

The Gunn-element is a transferred electron device. This means that it uses an effect in the
semiconductor in which the electrons are scattered (transferred) from the main conduction band
minimum into neighboring satellite minima where they have lower mobility. Since the scattering
process increases gradually with an externally applied field, there is a region with negative
differential mobility. As will be explained later in more detail, the negative differential mobility
leads to the periodic formation of space-charges (so called dipole domains), which travel through
the device and can be used to extract RF power at their fundamental frequency. Gunn-elements
have been used in signal sources (oscillators) up to 300 GHz
The Gunn effect was first observed by J.B. Gunn in 1963: Gunn had applied a bias voltage
to small samples of GaAs when he observed current fluctuations as soon as the resulting E-
field across the sample exceeded a critical threshold value. The period of these oscillations is
proportional to the length of the sample, which is characteristic for transit-time effects. The
experimental setup and the observed current pulses at the device terminals are illustrated in
Figs. 8.7a and 8.7b, respectively.

n-GaAs

(a) Experimental setup for observa- (b) Periodic current fluctuations observed at
tion of the Gunn-effect. the device terminals.

Figure 8.7: Experimental setup and observation for the Gunn-effect.

After its initial observation in GaAs, the Gunn-effect was used with other III–V semiconductors
such as InP as well as with II–VI semiconductors (e.g., CdTe, ZnSe).
The start-up of oscillations in a Gunn-element is not based on a negative differential resistance
(except for the LSA mode of operation explained later) but relies on the negative differential
Chapter 8. Oscillator Devices 119

mobility which leads to space-charge instabilities.

Figure 8.8: Energy-band structure of GaAs.

Fig. 8.8 shows the energy-band structure of GaAs. Energy-band structure plots relate conduction-
and valence-band energy levels with momentum. Since momentum is a vector, the plot can
only be drawn for a one-dimensional cut plane of the momentum space. In our case this is the
momentum in direction of the <100> or <100> crystal planes.
The conduction band of GaAs consists of several subbands as can be seen from the minima
of WC . The bottom of the conduction band is located at <000>, the first higher subband at
<100>.
In thermal equilibrium and/or when only small E-fields are applied, the electrons are located at
<000>.
The effective mass of an electron is inversely proportional to the second derivative of the
condcution band curve:

d2 Wc
me ∼ , (8.32)
d(~k)2
so that small effective mass is obtained for the conduction band minimum.
The mobility of charge carriers is defined as

µe = const./me , (8.33)

which results in high electron mobility, thus low interaction with the crystal lattice at thermal
equilibrium. In GaAs, the electron mobility is µn1 ≈ 8000 cm
2
Vs
If now an external electric field is applied to the material, the electron energy increases, and
the momentum of electrons changes due to lattice interactions (which are also called phonons).
When a phonon provides enough additional momentum to an electron, the electron is scattered
into the <100> subband. In Fig. 8.8 one can see that the <100> subband has a soft curvature,
which means that the effective mass is higher and the lattice interactions are stronger than in
the <000> subband.
Above the critical electric field Ec (Ec ≈ 3.2kV/cm in GaAs), scattering into the <100> subband
becomes dominant, so that the electron mobility decreases.
120 8.2. Gunn-Element

n2
2 vD n1+n2 1
x10 cm
7
s
n2
n1+n2
vD
1 0.5

0 0
0 Ek 10 20 30 kV
cm
E
Figure 8.9: Drift velocity and relative carrier concentration in <100> subband.

The average drift velocity is defined as

n1 µn1 + n2 µn2
vD = · E, (8.34)
n1 + n2

where n1 is the carrier concentration in the <000> subband, n2 is the carrier concentration
in the <100> subband, and µn1 and µn2 are the electron mobilities in the <000> and <100>
subband, respectively. In Fig. 8.9, the average drift velocity is plotted along with the fraction of
electrons in the <100> subband.
The time constant of the scattering process is on the order of several ps, so that the vD –E curve
in Fig. 8.9 is valid up to > 100 GHz
negative differential mobility of E > Ec does not imply a negative differential conductivity σ
because n = n1 + n2 important: E > Ec ⇒ n = 6 n0 = ND (doping).
rather: space-charge instabilities, high-field domains.
σ > 0 : space-charges decrease exponentially

1 σ qnµn qn dvD
= = =
τd ε ε ε dE

τd : dielectric relaxation time constant

E > Ec ⇒ dvD /dE < 0 : space-charges increase exponentially + drift at vD

space-charge high if |τd | ≤ L/vD (transit time)

dvD
≈ −2000cm2 /V s (E > Ec ) → n0 L ≥ 1012 cm−2
dE
field inhomogeneity (=
ˆ space-charge)⇒ domain (Fig. 8.10)

• primary field inhomogeneity caused by field distribution at ohmic contact (‘cathode‘)

• secondary field inhomogeneity caused by doping or statistic fluctuations (noise).

Chapter 8. Oscillator Devices 121

cathode contact anode contact

- GaAs + V0

l
V0>|EC|•L

V(x)

primary secondary
field inhomogeneity
|E(x)|
EC

n(x)
n0,contact
accumulation

n0,GaAs
depletion

0 x L

Figure 8.10: (a) Gunn-element (schematic), (b) voltage-, (c) field- and (d) carrier distribution.
122 8.2. Gunn-Element

⇒ change of n (local accumulation or depletion). Secondary inhomogeneities are small in space

⇒ dipole domain.
rapid increase (→ vD (E)−curve)
high voltage drop over dipole domain

t3 t2
E(x)

t1
C x A
Figure 8.11: A dipole domain.

8.2.1 Stationary Operation

domain propagates at vD0 .

current pulses if domain reaches anode contact ⇒ again E > Ec → new domain
I/V-curve: due to dipole domain the I/V-curve splits into two parts:

• stationary curve =v
ˆ D (E) for E < Ec ,
• dynamic curve (after formation of a domain) for E > Ec .

constant drift velocity vD0 ⇒ diffusion must be taken into account (Fig. 8.12):
Overall, the domain propagates with constant drift velocity vd,0 .

1 ∂n
vd0 = vd (E) − D = const. (8.35)
n ∂x0
(assumption: D = const.)
Let us now derive Butcher’s equal-areas rule, which is a graphical tool that is helpful for
determining the Gunn element’s dynamic device properties. Again, we start out with Poisson’s
equation
∂E q(n − n0 )
∇·E = 0
= , (8.36)
∂x 
which yields a substitution for dx0 → dE. The substitution is used to prepare the velocity
relation (8.35)
1 ∂n 0
[vd (E) − vd0 ] dx0 = D dx , (8.37)
n ∂x0
q 1
[vd (E) − vd0 ] dE = (n − n0 )D dn, (8.38)
 n
so that it can be subsequently integrated. We integrate from a point outside of the domain, where
the electrical field and carrier concentration are constant at values Elow and n0 , respectively:
Z E Z n
qD n0

[vd (E) − vd0 ] dE = 1− dn. (8.39)
Elow  n0 n
Chapter 8. Oscillator Devices 123

Ehigh
vD0
UD0
E(x)
n(x), Elow
E(x)

accumulation
n>>n0
n=n0
n0

depletion
n=0
C A
x
Figure 8.12: Field- and carrier distribution of a dipole domain.

At the domain’s field maximum, Ehigh , the field derivative must be zero, ∂E/∂x0 = 0, and from
Poisson’s equation follows n = n0 for the carrier concentration. When the left integral’s upper
limit is chosen as E = Ehigh , then the right integral’s upper limit becomes n = n0 , resulting in
the cancellation of the right-hand side and, finally, Butcher’s equal-areas rule:
Z Ehigh
[vd (E) − vd0 ] dE = 0. (8.40)
Elow

vD vD0(Elow)

vD0
vmin

Elow Ehigh E'high

E'low |E|
Figure 8.13: Illustration of Butcher’s equal-areas rule.

As illustrated in Fig. 8.13, the rule enables a graphical solution which shows the relation between
the domain’s propagation velocity vd0 and the minimum and maximum electrical fields, Elow
and Ehigh . When picking the two min/max field values of a domain, Elow ≤ Ecrit ≤ Ehigh , the
propagation velocity can be uniquely determined as the level separating equal areas below or
above the vd –E curve. Similar min/max fields Elow . Ecrit . Ehigh result in fast domains with
124 8.2. Gunn-Element

vd0 . vd (Ecrit ) (a), whereas large differences between Elow and Ehigh lead to lower propagation
velocity (b). There is also a lowest Elow and a highest Ehigh ; these mark the range where
solutions for vd0 can be found.
The two min/max field values cannot be chosen arbitrarily: they are related to each other by the
domain shape and the externally applied dc voltage Vdc . The domain shape can be approximated
by a triangular field distribution, provided that the accumulation layer is very small (like a
Dirac-pulse) compared to the depletion layer (cf. Fig. 8.14).

Ehigh

E UD0 n
Elow n0
b

x x
Figure 8.14: Triangular approximation of a dipole domain.

The voltage drop across the device results from the voltage drop due to the constant ‘background’
E-field Elow and from the additional voltage drop Vdom due to the triangular dipole domain:

Vdc = Elow L + Vdom = Elow L + (Ehigh − Elow )2 (8.41)
2qn0

With fixed external dc voltage, the low field may be chosen freely, and the high field results from
(8.41). The domain velocity vd0 results from Butcher’s rule, and is now written as vd0 (Elow ).
The current density within the device resulting from the constant carrier drift is then

J = qn0 · vd0 (Elow ), (8.42)

and by taking the device cross-section A into account, the total current becomes I = J · A.
graphical solution: (n0 = 1015 cm−3 ): see Fig. 8.15
V0 (device line) and VD0 (Elow ) are given, Butcher: point of intersection yields Elow (,→ vD0 )
,→ J resp. I= J · A (area)
⇒ stationary (E < Ec ) and dynamic (domain) curves (Fig. 8.16)
domain formation and -extinction ⇒ transition from stationary to dynamic curve
domain extinction for (E < Es )

oscillator operation

• Gunn-oscillator suitable for broadband operation

• resonant circuit with medium or high quality factor (Q ≈ 100 - 10000) for low noise

• parallel circuit for voltage drive

Chapter 8. Oscillator Devices 125

10
V

VD0
6
ESL V0=10V, L=25µm

4
V0=10V,
L=20µm
2

0 kV
0 1 Elow 2 ES 3 EC 4 5 cm
Elow
Figure 8.15: Determination of operating point.

Ek

2
7cm
x10 s 11 -2
n0L=3x10 cm
Es 12
10

1
vD0 10
14
10
13

0 kV
0 2 4 6 8 10 12 cm
V0/L
Figure 8.16: Stationary and dynamic curves for different products of doping and length.

1. transit-time operation (Fig. 8.17)

• bias voltage V0 and load impedance chosen that at voltage minimum (V0 − V̂ ), E
greater than Ec
• domain → anode, extinction
126 8.2. Gunn-Element

Ec
Î
I i(t)
form-
extinction
ation
-p/2 p/2 3p/2 wt

V V0

0
V(t)
p/2
p

3p/2
2p
wt

Figure 8.17: Illustration of transit-time operation.

• ,→ current → stationary curve

• formation of new domain, current → dynamic curve

short formation- and extinction time → current-pulses

periodic time = transit time L/vD0
→ for V and I in opposite phase (fundamental harmonic)
⇒ resonant frequency f ≈ vD0 /L
example:
L = 10µm vD0 ≈ 107 cm/s → f ≈ 10 GHz

2. domain delay operation (Fig. 8.18)

• operation allows tuning (≈ 1 octave) using a varactor or YIG-resonator

• V0 and V̂ chosen that at voltage minimum:
(V0 − V̂ ) ⇒ (E < Ec )
• domain → anode, extinction
• current → stationary curve
• new domain if E > Ec
vD0 vD0
≤f ≤ (transit-time operation)
2L L

efficiency . 10%

3. domain quenching operation (Fig. 8.19)

• oscillation frequency > transit-time frequency

• ,→ at V0 − V̂ , En < Es (high amplitude)
Chapter 8. Oscillator Devices 127

Ec
wL
I i(t) vD0

wt

V V0

V(t)
^
wL/vD0 V

wt

Figure 8.18: Illustration of domain delay operation.

I I
ve

dyn. curve
ur
.c
at
st

T
t

V t1 t2 t3 wt
V0
t1 V(t)-V0

^
V
t2 T t

t3

wt

Figure 8.19: Illustration of domain quenching operation.

• domain extinction (t = t2 )
• new domain if E > Ec (t = t3 )
• broadband operation
• low efficiency (≤ 5%), decreases with f

4. LSA operation (Fig. 8.20)

LSA: Limited Space charge Accumulation
• utilization of whole vD (E)-curve (including decreasing part)
128 8.2. Gunn-Element

I
i(t)

0 wt

V0 V
V(t)

wt

Figure 8.20: Illustration of LSA operation.

• ,→ no domain formation
• bias voltage chosen that E > Ec , f high enough that during positive half-wave
domains are not fully formed
• amplitude high enough that signal drive is on stationary curve (⇒ compensation of
space-charge)

→ high efficiency (∼ 20%) possible

this mode of operation is difficult to obtain (→ high amplitude, demands on doping
homogeneity)
,→ in most cases no true LSA operation

device design, limits

• n-layer (thickness = L) on highly doped layer (n+ )

• n+ -contact layer

• mesa-type structure
Chapter 8. Oscillator Devices 129

+
Au/Ge Au/Ge
n -GaAs
+
n-GaAs L n -GaAs-substrate
+
n -GaAs n-GaAs L
+
n -GaAs
Au/Ge Au/Ge
(a) normal (b) upside-down

Figure 8.21: Different device designs.

• upside-down → better heat removal

transit-time effects 1
)
⇒ Pmax ∼
breakdown f2

(except LSA operation)

1000 GaAs
fundamental mode
mW

100 InP

P
GaAs harmonic
10 mode

1 10 100 GHz
f
Figure 8.22: Typical output power.

• for f ≤ 10GHz: Pmax limited by thermal effects

(higher L → thermal resistance %)

• alternative for GaAs: InP

– EcInP >> EcGaAs

→ InP requires higher voltage, ⇒ due to heat removal only useful for higher frequen-
cies
130 8.3. Field-effect transistor

vD max vD max
– vD min InP > vD min GaAs
,→ current amplitude higher
→ higher efficiency.

operation at higher harmonics: overcomes P-limit at ω-operation

circuit at ω (reactive load only)
,→ high amplitude → distortions → power at 2ω.

8.3 Field-effect transistor

nonlinearity of transconductance and drain conductance

• cause saturation of FET

• determined by static set of characteristic curves

nonlinearity of Cgs (Gate-Source-capacitance) causes upconversion of 1/f-noise

troublesome due to:

• strong 1/f -noise in FET

• upconversion very effective (→ Manley-Rowe)

result from simple conversion matrix analysis [26]

phase noise of oscillator ∼ Cgs1
Cgs1 : 1. fourier component of Cgs (Vgs ) at periodic, nonlinear drive.
Chapter 9

Oscillator Circuits

9.1 Resonator stabilization [27]

also: ”cavity”- stabilization, ”Q”- stabilization

long term stability (→ temperature)
short term stability (→ noise)
important design criteria for oscillators

• Q-stab.:

+ simple, economical, geometrical dimensions

– Ploss , several modes of operation (”mode-jumping”), tuning losses

• three basic circuits of one-port oscillators (Fig. 9.1)

a Q0 transmission resonator

Q0
b reaction resonator

c Q0 reflection resonator

Q0
oscillator resonator load

Figure 9.1: Three basic circuits.

• common equivalent circuit model

• three functional blocks (=

ˆ three resonant circuits)

– original (unstabilized) oscillator

– stabilization resonator
– coupling line (attenuated) (⇒ Q &)

• example: rectangular waveguide oscillator

131
132 9.1. Resonator stabilization

I I

λ/2 D

I H

Figure 9.2: Diode postcoupled to rectangular waveguide.

– diode is postcoupled to waveguide (Fig. 9.2).

– post behaves like λ/4 - TEM - transmission line
– parallel resonance (oscillator is ”short-circuit stable”: ⇒ no oscillation for short-
circuit);

λ/2

G0 J-Inverter J-Inverter G0
gD gP

Figure 9.3: Equivalent circuit model of waveguide mounting structure.

equivalent circuit model of waveguide mounting structure Fig. 9.3

– gd : conductance of diode, negative real
– here (assumption): depends only on amplitude, independent of frequency
– post with air gap =λ/4-transformator
ˆ + resonant circuit (losses =g
ˆ p)
– stabilization by cavity resonator
– cylindrical high-Q resonator
– modes of operation: H01n (indices: 1. circumference- (0), 2. radius- (1) and 3.
length-dependency (n) [28])
– equivalent circuit model near natural frequency: R L C - circuit, series / parallel
depending on resonator coupling
– resonator coupling in a way that oscillation is unambiguous
– reflection-resonator =
ˆ series resonant circuit,
– reaction-resonator =
ˆ parallel resonant circuit in series with coupling line,
– transmission-resonator =
ˆ series resonant circuit in series with coupling line.

• assumption: coupling reactance negligible → resonator is described by:

– unloaded Q-factor Q0
Chapter 9. Oscillator Circuits 133

– resonant frequency ωc

– coupling coefficient β resp. β1 /β2

• common equivalent circuit model for all three types of stabilization (Fig. 9.4):

A QH X Qt gDA QC

gD gH gC

X
A
yL

Figure 9.4: Common equivalent circuit model.

• three resonant circuits:

– diode-mounting structure

– coupling line

– cavity resonator

• length of coupling line that oscillator can be stabilized

reflection-resonator: → λ/2 − line

,→ reaction-resonator: → λ/4 − line
transmission-resonator: → λ/2 − line

• equivalent circuit model elements:

reaction- reflection- transmission-

resonator resonator resonator
gH gp gLoad + gp gp
gDA gLoad gDA gDA
1 1 (1+β2 )
gc β β β1
Qc Q0 Q0 Q0 /(1 + β2 )
134 9.1. Resonator stabilization

transmission:
”H” =
ˆ resonant circuit of unstabilized oscillator
”t” =
ˆ coupling line
”C” =
ˆ stabilization resonator (incl. transformed load)

reflection:
”H” : incl. load
”C” : excl. load

Reaction:
gDA =
ˆ load, no additional dampening necessary
,→ Qt ≈ 0

• condition of oscillation from locus in complex plane. here: gD real, |gD | decreases with
signal drive (→ short-circuit stable)

– important: yL (ω) shows 1 - 2 loops depending on quality-factor and resonant frequency

of resonant circuits.
– typical:
stabilization resonator : Q0 ≥ 104
coupling line : Qt ≈ 103
diode-mounting structure : QH ≈ 10 . . . 103
– ωc tunable, ωH = ωt = const.
– ωC = ωH = two loops within one another
– inner loop =
ˆ stabilization resonator

• operating point on outer loop ,→ no stabilization

⇒ dampening of series resonance until additional loop vanishes.

1. stabilization resonator detuned (Fig. 9.5)

imaginary

gDA << gcrit high damping

gDA = gcrit critical damping

gDA > gcrit no damping

real

yL(w)

Figure 9.5: Loci of resonator.

– design: (b =
ˆ susceptance)
Chapter 9. Oscillator Circuits 135

dbH dbt
+ ≥0
dω ω=ωH dω ω=ωH

due to short-circuit stability

– equivalent circuit model (Taylor series expansion):

ω − ωH
bH ≈ 2QH gH
ωH
s
QH
⇒ gDA ≤ gt gH = gcrit.
Qt

2. incl. stabilization-resonator: (ωc = ωH )

imaginary

imaginary

gDA = gcrit gDA << gcrit

A A
real real

yL(w)
yL(w)
(a) (b)

Figure 9.6: Approximation of locus of stabilization-resonators.

– A=ˆ operating point, gDA small that series resonant circuit can be neglected.
– equivalent circuit model:

2
gDA
yL (ω) = gH + gDA −
gDA +gc
2
1+

gc ·Ωc
gDA +gc
 2 
Ω

gDA
g c c
+j gH ΩH +
 gDA +gc
2
1+
 
gc Ωc
gDA +gc

ω − ωH,C
ΩH,C ≈ 2QH,C
ωH,C
– natural frequency from Im{yL (ω0 )} = 0
from yL (ω) : tuning behavior ω0 = f ct(ωC )
136 9.1. Resonator stabilization

wS-wH -yD( v )
2gHQH fC
wH C
A
imaginary

real
C

B
gDA gC
gH +
gDA+gC
yL(w)

gH+gDA

Figure 9.7: Loci of device and load.

– locus representation:
here: approximation of loop by circle (Fig. 9.7)
A: stable operating point at ω0 = ωC = ωH = ωt
tunable to point B, subsequently mode jumping to unstabilized operating point
(linear branch of yL )
geometry: (ωs = ω0 at point B i.e. detuning of ωC ):

ωs − ωH 1 gC gDA
 
2gH QH ≈ gDA −
ωH 2 gC + gDA
∆ωs ωs − ωH 1 2
gDA
,→ =2 ≈ ·
ωH ωH 2gH QH gc + gDA
– hold-in range

• oscillation frequency is tuned from linear branch of yL :

question: at which point does stabilization occur?

or : at which point is the resonator locked
⇒ lock-in range

– intersection point at C ⇒ mode jumping to resonator loop

– lock-in range can be calculated from:

d [Im (yL (ω))]

=0
dω ω=ωf
Chapter 9. Oscillator Circuits 137

– lock-in range smaller than hold-in range

C
w
Pout

0=
w0 b = reflection- and

w
reaction-resonator

a = transmission-
resonator
lock-in range

hold-in range

wC wC
(a) (b)

Figure 9.8: Typical behavior of natural frequency, oscillation frequency and output power.

• effect of stabilization is characterized best by loaded Q-factor QL of total passive circuit

(correct for one-port; for two-port oscillator more complicated due to feedback).

• choice useful since QL mostly determines phase noise.

hW i
QL = ω0 D E h...i time average
dW
− dt

1 d [Im {yL (ω0 )}] 2

hW i = û û : voltage over diode
4 dω0

dW 1
 
− ≡ PGen = Re [yL (ω0 )] û2
dt 2

hence:

2
QH + QC ggH

gDA
ω0 d[Im{ydωL0(ω0 )}] C

QL =
gDA +gC
≈
2Re {yL (ω0 )} 1+ gC
gH · gDA
gDA +gC

• for complete characterization:

⇒ determine various power losses

• large-signal model for diodes necessary

– device conductance is described by gD (û) = −gD (0) + kû2

(→ van-der-Pol-characteristic).
138 9.1. Resonator stabilization

– condition of oscillation

gD (û) = −Re {yL (ω0 )}

gD (0) − Re {yL (ω0 )}
,→ û2 =
k

– generated power:

1 1 gD (0) − Re {yL (ω0 )}

PGen = Re {yL (ω0 } û2 = Re {yL (ω0 )}
2 2 k
but:

1
Pgen = − gD (û)û2
2
– maximum power for:

dPgen gD (0)
= 0 ⇒ ûmax =
dû 2k
– ⇒ load admittance for maximum power:

2Re {yL (ω0 )} max

= gD (0)

– delivered power to yL (ω0 ):

1 2Re {yL (ω0 )}|max − Re {yL (ω0 )}

Pgen = Re {yL (ω0 )}
2 k
– distribution of generated power to different circuits by analysis of equivalent circuit.
– from this follows e.g. Pout

• typical results for

Q0 = 20000
gDA = gcrit , gp = 0.1
QH = 2000 (QH = 30 for reflection-oscillator)

– ⇒ output power can be traded for higher Q-factor depending on resonator coupling.
– here:

β1
βT =
1 + β2

– hold-in range vs. Pout

⇒ distinguishes properties of three stabilization techniques
choice is dictated by application.
Chapter 9. Oscillator Circuits 139

bT

12 transmission

10
b2=0.1
8 reaction
QL b2=0.5
3
10 6 b
b2=1.0
reflection
4 b2=2.0

b2=5.0 b
2
bT=2
bT=1
0
0 0.2 0.4 0.6 0.8 1
Pout
Figure 9.9: Loaded Q-factor vs. Pout .

reflection-
transmission-
6 resonator
resonator
(QH=30)
(QH=2000)

bT
4
3 Dw
10 wHS
0.1 0.5 1 2 5
b b
2

reaction-
resonator
b (QH=2000)
0
0 0.2 0.4 0.6 0.8 1.0
Pout

Figure 9.10: Hold-in range vs. output power.

9.2 YIG oscillator [29]

=
ˆ oscillator feat. YIG resonator

9.2.1 YIG: single-crystal sphere of Y3 F e5 O12

,→ yttrium-iron-garnet
140 9.2. YIG oscillator

• static magnetic field H0 → resonance in ferrite material, tunable.

H0 aligns the electron spins in parallel to H0 .

• RF-magnetic field (HRF ) perpendicular to H0

precession movement ,→ amplitude high if RF-frequency equals precession frequency.

• pronounced resonance if all spins are aligned in same direction and oscillate with the same
phase.
requirements:

– RF-field homogeneous
– perfect crystal
– YIG shape spherical
– smooth surface

• resonant frequency f0 of YIG-ellipsoid

(H0 on principal axis (z-axis), rotational symmetric with respect to H0 )

f0 = γ [H0 + Ha + (Nt − Nz ) MS ]

γ = 35.2 MHz/(kA/m): gyromagnetic ratio

Ha : anisotropic field (→ crystal orientation); YIG ≈ 0
Nt /Nz : transversal / axial demagnetization factor
MS : saturation magnetization; YIG: MS = 142 kA/m

• sphere: Nt = Nz = 1/3; Q-factor up to 104

⇒ no influence of MS (→ strong temperature dependence)

• lower resonant frequency fg if inner field=0:

Hiz = H0 − Nz MS = 0

,→ fg = γNz MS = 1666MHz

can be reduced by material composition (→ MS ).

• upper resonant frequency set by H0 , limited to 40 - 50 GHz

• tuning highly linear!

deviation from linearity ≈ per mill range

• advantageous for magnetic tunable oscillators

• example:
technical data (circuit Fig. 9.11):

– f0 = 2.0 − 6.2 GHz

– Pout ≈ 20 mW
Chapter 9. Oscillator Circuits 141

+V

+V
R2
pole
YIG sphere shoe
R2 C4
Tr1 A A’
-U0 R1 C2
matching Tr2

C1 pole R3
shoe
L1 C3 R5

-V
+V

R7
C6
matching output

A’ matching Tr3

R6
C5 R8

-V

Figure 9.11: YIG oscillator.

– oscillator =T
ˆ r1 + L1 (feedback) + YIG
– Tr2 and Tr3 : buffer amplifier stage for load decoupling

• application: e.g., swept frequency signal generator

Bibliography

[1] E. Voges, Hochfrequenztechnik, Band 1. 2. Auflage, Huethig-Verlag, Heidelberg, 1991.

[2] R. S. Pengelly, Microwave Field-Effect Transistors - Theory, Design and Applications.

Research Studies Press, Chichester, 1982.

[3] D. M. Pozar, Microwave Engineering, 3rd ed. Hoboken, NJ: Wiley, 2005.

[4] R. E. Collin, Foundations for Microwave Engineering, 2nd ed. IEEE Press, 2001, originally
published: New York, McGraw Hill, 1992.

[5] E. Voges, Hochfrequenztechnik: Bauelemente, Schaltungen, Anwendungen, 3rd ed. Bonn:

Hüthig-Verlag, 2004.

[6] M. S. Gupta, “Power gain in feedback amplifiers, a classic revisited,” IEEE Trans. Microw.
Theory Tech., vol. 40, no. 5, pp. 864–879, 1992.

[7] H.-G. Unger, W. Harth, Hochfrequenz-Halbleiterelektronik. S. Hirzel Verlag, Stuttgart,

1972.

[8] H.-G. Unger, W. Schultz, and G. Weinhausen, Elektronische Bauelemente und Netzwerke I.
Brauschweig: Vieweg, 1981.

[9] S. M. Sze, Physics of Semiconductor Devices, 3rd ed. Hoboken, NJ: Wiley-Interscience,
2007.

[10] R. S. Pengelly, Microwave Field-Effect Transistors: Theory, Design and Applications, 2nd ed.
Letchworth: Research Studies Press, 1986.

[11] R. S. Cobbold, Theory and Applications of Field-Effect Transistors. New York: Wiley-
Interscience, 1970.

[12] C. A. Liechti, “High speed transistors: Directions for the 1990s (state of the art reference
supplement),” Microwave Journal, pp. 165–177, Sep. 1989.

[13] T. Drummond, W. Masselink, and H. Morkoc, “Modulation-doped GaAs/(Al,Ga)As hetero-

junction field-effect transistors: MODFETs,” Proceedings of the IEEE, vol. 74, no. 6, pp.
773–822, 1986.

[14] D. Delagebeaudeuf and N. Linh, “Metal-(n) AlGaAs-GaAs two-dimensional electron gas

FET,” IEEE Trans. Electron Devices, vol. 29, no. 6, pp. 955–960, 1982.

[15] P. M. Smith and A. W. Swanson, “HEMTs–low noise and power transistors for 1 to 100
GHz,” Applied Microwave, pp. 63–72, 1989.

142
Bibliography 143

[16] R. S. Engelbrecht and K. Kurokawa, “A wide-band low noise L-band balanced transistor
amplifier,” Proc. IEEE, vol. 53, no. 3, pp. 237–247, Mar. 1965.

[17] S. A. Maas, Nonlinear Microwave and RF Circuits, 2nd ed. Boston, MA: Artech House,
2003.

[18] S. C. Cripps, RF power amplifiers for wireless communications, 2nd ed. Boston, MA:
Artech House, 2006.

[19] Y. Takayama, “A new load-pull characterization method for microwave power transistors,”
in IEEE MTT-S Int. Microw. Symp. Digest, Cherry Hill, 1976, pp. 218–220.

[20] S. C. Cripps, “Indeterminate inversion,” IEEE Microwave Magazine, vol. 12, no. 7, pp.
32–129, Dec. 2011.

[21] J. Manley and H. Rowe, “Some general properties of nonlinear elements—part i. general
energy relations,” Proc. IRE, vol. 44, no. 7, pp. 904–913, 1956.

[22] C. Burckhardt, “Analysis of varactor frequency multipliers for arbitrary capacitance variation
and drive level,” Bell Syst. tech. J., no. 44, pp. 675–692, 1970.

[23] A. F. Jacob, Stabilitaet und Rauschen von MESFET-Oszillatoren. Dissertation, TU

Braunschweig, 1986.

[24] S. M. Sze, Physics of Semiconductor Devices, 2nd Ed. John Wiley & Sons, New York,
1981.

[25] Y. C. Shik, H. J. Kuno, Solid State Sources from 1 to 100 GHz. Microwave Journal, State
of the Art Reference, S. 145 - 161, 1989.

[26] H. J. Siweris, B. Schiek, Analysis of Noise Upconversion in Microwave FET Osciallators.

IEEE Trans. Microwave Theory and Techniques, Vol. MTT-33, Nr. 3, S. 233 - 242, March
1985.

[27] R. Knoechel, K. Schuenemann, J.-D. Buechs, Theory and Performance of Cavity Stabilised
Microwave Oscillators. IEE Microwaves, Optics and Acoustics, Vol. 1, Nr. 4, S. 148 - 155,
1979.

[28] H.-G. Unger, Elektromagnetische Theorie für die Hochfrequenztechnik, Teil 1. Heidelberg:
Hüthig Verlag, 1988.

[29] B. Schiek, Grundlagen der Hochfrequenz-Messtechnik. Berlin: Springer, 1999.