a

r
X
i
v
:
p
h
y
s
i
c
s
/
0
6
0
8
2
1
1
v
1


[
p
h
y
s
i
c
s
.
i
n
s
-
d
e
t
]


2
1

A
u
g

2
0
0
6
Tutorial on the double balanced mixer
Enrico Rubiola
web page http://rubiola.org
FEMTO-ST Institute
CNRS and Universit´e de Franche Comt´e, Besan¸ con, France
2nd February 2008
Abstract
Smart use of mixers is a relevant issue in radio engineering and in
instrumentation design, and of paramount importance in phase noise
metrology. However simple the mixer seems, every time I try to explain
to a colleague what it does, something goes wrong. One difficulty
is that actual mixers operate in a wide range of power (150 dB or
more) and frequency (up to 3 decades). Another difficulty is that the
mixer works as a multiplier in the time-domain, which is necessary to
convert frequencies. A further difficulty is the interaction with external
circuits, the input sources and the load. Yet far the biggest difficulty is
that designing with mixers requires a deep comprehension of the whole
circuit at system level and at a component level. As the electronic-
component approach is well explained in a number of references, this
tutorial emphasizes the system approach, aiming to provide wisdom
and insight on mixes.
1
2 E. Rubiola Tutorial on mixers February 2, 2008
Most used symbols
A(t) slow-varying (baseband) amplitude
h
lp
, h
bp
impulse response of lowpass and bandpass filters
h, k, n, p, q integer numbers
i(t), I current
I (goes with Q) in-phase in/out (of a two-phase mixer/modulator)
IF intermediate frequency
j imaginary unit, j
2
= −1
ℓ mixer voltage loss, 1/ℓ
2
= P
i
/P
o
LO local oscillator
P power
P
i
, P
o
power, input and output power
P
p
, P
S
LO (pump) power and internal LO saturation power
Q (goes with i) quadrature in/out (of a two-phase mixer/modulator)
R resistance
R
0
characteristic resistance (by default, R
0
= 50 Ω)
R
G
source resistance (Th´evenin or Norton model)
U dimensional constant, U = 1 V
v(t), V voltage
v

, v
′′
real and imaginary, or in-phase and quadrature part
v
i
(t), v
o
(t) input (RF) voltage, and output (IF) voltage
v
p
(t) LO (pump) signal
v
l
(t), V
L
internal LO signal
V
O
saturated output voltage
V
S
satureted level of the internal LO signal v
l
(t)
x(t) real (in-phase) part of a RF signal
y(t) imaginary (quadrature) part of a RF signal
φ, φ(t) static (or quasistatic) phase
ϕ(t) random phase
ω, f angular frequency, frequency
ω
i
, ω
l
input (RF) and pump (LO) angular frequency
ω
b
, ω
s
beat and sideband angular frequency
note: ω is used as a shorthand for 2πf
Most used subscripts
b beat, as in |ω
s
−ω
i
| = ω
b
i, I input
l, L local oscillator (internal signal)
o, O output
p, P pump, local oscillator (at the input port)
s sideband, as in |ω
s
−ω
i
| = ω
b
S saturated
note: in reverse modes, i is still the input, and o the output
February 2, 2008 E. Rubiola Tutorial on mixers 3
Contents
Most used symbols 2
1 Basics 5
2 Signal representations 8
3 Linear modes 10
4 Mixer loss 21
5 Saturated Modes 26
6 Reversed Modes 36
7 Special Mixers and I-Q Mixers 40
8 Non-ideal behavior 47
9 Mixer Noise 48
10 Where to learn more 49
References 50
4 E. Rubiola Tutorial on mixers February 2, 2008
February 2, 2008 E. Rubiola Tutorial on mixers 5
1 Basics
It is first to be understood that the mixer is mainly intended, and mainly doc-
umented, as the frequency converter of the superheterodyne receiver (Fig. 1).
The port names, LO (local oscillator, or pump), RF (radio-frequency), and
IF (intermediate frequency) are clearly inspired to this application.
knob
preselector
IF amplifier
local
oscillator
RF IF
LO
detector
tuning
Figure 1: Superheterodyne receiver.
The basic scheme of a mixer is shown in Fig. 2. At microwave frequencies
a star configuration is often used, instead the diode ring. Under the basic
v
o
(t)
IF out
LO input
RF input
mixer
D3 D4
D1 D2
input
LO
R
G
LO source
v
p
(t)
input
RF
R
G
RF source
R
L
IF load
v
i
(t)
IF
out
Figure 2: Double balanced mixer and its switch-network equivalent.
assumptions that v
p
(t) is large as compared to the diode threshold, and that
v
i
(t) is small, the ring acts a switch. During the positive half-period of v
p
(t)
two diodes are reverse biased and the other two diodes are forward biased
to saturation. During the negative half-period the roles are interchanged.
For the small RF signal, the diodes are open circuit when reverse biased,
and small resistances when forward biased. As a result, the IF signal v
o
(t)
switches between +v
i
(t) and −v
i
(t) depending on the sign of v
p
(t). This is
equivalent to multiplying v
i
(t) by a square wave of amplitude ±1 that takes
6 E. Rubiola Tutorial on mixers February 2, 2008
the sign from v
p
(t). In most practical cases, it is sufficient to describe the
frequency conversion mechanism as the product between v
i
(t) and the first
term of the Fourier expansion of the square wave. More accurate models
account for the higher-order Fourier terms, and for the dynamic resistance
and capacitance of the diodes.
At the RF and LO sides, a balun is necessary in order to convert the
unbalanced inputs into the balanced signals required for the ring to operate
as a switch. Conversely, no adapter is needed at the IF output, which is
already unbalanced. In low-frequency mixers (from a few kHz to 2–3 GHz)
the baluns are implemented with power iron tore transformers. At higher
frequencies, up to some tens of GHz, transformers are not available, for
microstrip networks are the preferred balun types. The typical LO power is
of 5–10 mW (7–10 dBm), whereas in some cases a power up to 1 W (30 dBm)
is used for highest linearity. The RF power should be at least 10 dB lower
than the LO power. The diodes are of the Schottky types, because of the low
forward threshold and of the fast switching capability. The characteristic
impedance to which all ports should be terminated is R
0
= 50 Ω, with rare
exceptions.
The mixer can be used in a variety of modes, each with its “personality”
and peculiarities, listed in Table 2, and detailed in the next Sections. In short
summary, the mixer is (almost) always used with the LO input saturated at
the nominal power. Then, the main parameters governing the behavior are:
Input power. The input (RF) power is usually well below the saturation
level, as in Figures 1–2. Yet, the input can be intentionally saturated.
Frequency degeneracy. When the input (RF) and LO frequency overlap,
the conversion product also overlap.
Interchanging the RF and IF ports. The difference is that the RF port
is coupled in ac, while the IF port is often coupled in dc.
Additionally, the mixer is sometimes used in a strange mode, with both
LO and RF inputs not saturated.
1.1 Golden rules
1. First and foremost, check upon saturation at the LO port and identify
the operating mode (Table 2).
2. Generally, all ports should be reasonably impedance matched, other-
wise reflected waves result in unpredictable behavior.
3. When reflected waves can be tolerated, for example at low frequencies
or because of some external circuit, impedance plays another role. In
fact, the appropriate current flow is necessary for the diodes to switch.
4. In all cases, read carefully Sections 3.1 to 3.3.
February 2, 2008 E. Rubiola Tutorial on mixers 7
Table 2: Operating modes of the double balanced mixer.
mode condition note
frequency P or I
LC ν
i
= ν
l
P
i
≪P
S
Linear frequency Converter. Typical
of the superetherodyne radio receiver.
SD ν
i
= ν
l
P
i
≪P
S
Synchronous Detector. Used the lock-
in amplifiers, in coherent receivers,
and in bridge noise measurements.
SC ν
i
= ν
l
P
i
≥ P
S
Saturated frequency Converter.
Mainly used in frequency synthesis.
DC
ν
l
=pν
0
ν
i
=qν
0
p, q small integers
P
i
≥ P
S
Degenerated frequency Converter.
Only used in some cases of metrology
and frequency synthesis.
N
o
r
m
a
l
M
o
d
e
s
PD ν
i
= ν
l
P
i
≥ P
S
Phase Detector. RF and LO signals
are to be in quadrature.
LM ν
i
≈ 0 I
i
≪I
S
Linear Modulator, driven with a near-
dc input current I
i
(t).
RLC ν
i
≫0 P
i
≪P
S
Reverse Linear Converter, driven
with a narrowband signal at ν
i
.
DM ν
i
≈ 0 P
i
≥ P
S
Digital Modulator. Information is lo-
cated close to dc.
RSC ν
i
≫0 P
i
≥ P
S
Reverse Saturated Converter. Some
cases of in frequency synthesis.
R
e
v
e
r
s
e
M
o
d
e
s
RDC
ν
l
=pν
0
ν
i
=qν
0
p, q small integers
P
i
≥ P
S
Reverse Degenerated Converter. Sim-
ilar to the DC mode, and only used in
some odd cases.
S
t
r
a
n
g
e
AD ν
i
= ν
l
P
i
<P
S
P
l
<P
S
Amplitude-modulation detector.
Scarce information. Used at NIST for
the measurement of AM noise.
1.2 Avoid damage
However trivial, avoid damage deserves a few words because the device can
be pushed in a variety of non-standard operation modes, which increases
the risk.
1. Damage results from excessive power. Some confusion between maxi-
mum power for linear operation and the absolute maximum power to
prevent damage is common in data sheets.
8 E. Rubiola Tutorial on mixers February 2, 2008
2. The nominal LO power (or range) refers to best performance in the
linear conversion mode. This value can be exceeded, while the absolute
maximum power can not.
3. The maximum RF power is specified as the maximum power for linear
operation. When linearity is not needed this value can be exceeded,
while the absolute maximum power can not.
4. Voltage driving may result in the destruction of the mixer for two
reasons. The diode i = i(v) characteristics is exponential in v, for the
current tend to exceed the maximum when the diode is driven by a
voltage source. The thin wires of the miniature transformers tend to
blow up as a fuse if the current is excessive.
5. In the absence of more detailed information, the absolute maximum
power specified for the LO port can be used as the total dissipated
power, regardless of where power enters.
6. The absolute maximum LO power can also be used to guess the max-
imum current through one diode. This may be useful in dc or degen-
erated modes, where power is not equally split between the diodes.
Better than general rules, a misfortunate case occurred to me suggests to be
careful about subtle details. A $ 3000 mixer used as a phase detector died
unexpectedly, without being overloaded with microwave power. Further
analysis showed that one rail of a dc supply failed, and because of this the
bipolar operational amplifier (LT-1028) connected to the IF port sank a
current from the input (20 mA?).
2 Signal representations
The simple sinusoidal signal takes the form
v(t) = A
0
cos(ω
0
t + φ) . (1)
This signal has rms value A
0
/

2 and phase φ. An alternate form often
encountered is
v(t) = V
rms

2 cos(ω
0
t + φ) (2)
= V


2 cos(ω
0
t) −V
′′

2 sin(ω
0
t) , (3)
with
V

= V
rms
cos φ (4)
V
′′
= V
rms
sin φ (5)
V
rms
=

(V

)
2
+ (V
′′
)
2
(6)
φ = arctan(V
′′
/V

) . (7)
February 2, 2008 E. Rubiola Tutorial on mixers 9
The form (2)-(3) relates to the phasor representation
1
V = V

+ jV
′′
= |V |e

, (8)
which is obtained by freezing the ω
0
oscillation, and by turning the amplitude
into a complex quantity of modulus
|V | =

(V

)
2
+ (V
′′
)
2
= V
rms
(9)
equal to the rms value of the time-domain sinusoid, and of argument
φ = arctan
V
′′
V

(10)
equal to the phase φ of the time-domain sinusoid. The “sin ω
0
t” term in
Eq. (3) has a sign “−” for consistency with Eq. (8).
Another form frequently used is the analytic (complex) signal
v(t) = V e

0
t
, (11)
where the complex voltage V = V

+ jV
′′
is consistent with Eq. (8). The
analytic signal has zero energy at negative frequencies, and double energy
at positive frequencies.
The product of two signals can only be described in the time domain
[Eq. (1), (2), (3)]. In fact, the phasor representation (8) is useless, and the
analytic signal (11) hides the down-conversion mechanism. This occurs be-
cause e
jωat
e

b
t
= e
j(ωa+ω
b
)t
, while the product of two sinusoids is governed
by
cos(ω
a
t) cos(ω
b
t) =
1
2
cos

ω
a
−ω
b

t +
1
2
cos

ω
a
+ ω
b

t (12)
sin(ω
a
t) cos(ω
b
t) =
1
2
sin

ω
a
−ω
b

t +
1
2
sin

ω
a
+ ω
b

t (13)
sin(ω
a
t) sin(ω
b
t) =
1
2
cos

ω
a
−ω
b

t −
1
2
cos

ω
a
+ ω
b

t . (14)
Thus, the product of two sinusoids yields the sum and the difference of the
two input frequencies (Fig. 3). A pure sinusoidal signal is represented as a
pair of Dirac delta function δ(ω −ω
0
) and δ(ω + ω
0
) in the spectrum, or as
a single δ(ω − ω
0
) in the case of the analytic signal. All the forms (1), (2),
(3), (8), and (11) are also suitable to represent (slow-varying) modulated
signals. A modulated signal can be represented
2
as
v(t) = A

(t) cos(ω
0
t) −A
′′
(t) sin(ω
0
t) . (15)
1
This is also known as the complex representation, or as the Fresnel vector represen-
tation.
2
The factor

2 is dropped, for A is a peak amplitude. Thus, A

(t) and A
′′
(t) are the
time-varying counterpart of V


2 and V
′′

2.
10 E. Rubiola Tutorial on mixers February 2, 2008
ω
a
RF
IF
ω
b
ω
a
−ω
b
ω
a
+ ω
b
LO
S
p
e
c
t
r
a
ω
ω
ω
Figure 3: Frequency conversion. Negative frequencies are not shown.
A

(t) and A
′′
(t) are the low-pass signals that contain information. They
may include a dc term, which accounts for the carrier, like in the old AM
and PM. Strictly, it is not necessary that A

(t) and A
′′
(t) are narrow-band.
The time-depencence of A

(t) and A
′′
(t) spreads the power around ω
0
. The
spectrum of the modulated signal is a copy of the two-side spectrum of A

(t)
and A
′′
(t) translated to ±ω
0
. Thus, the bandwidth of the modulated signal
(15) is twice the bandwidth of A

(t) and A
′′
(t). Not knowing the real shape,
the spectrum can be conventionally represented as a rectangle centered at
the carrier frequency, which occupies the bandwidth of A

and A
′′
on each
side of ±ω
0
(Fig. 4).
Of course, Equations (12)–(14) also apply to the product of modulated
signals, with their time-dependent coefficients A

(t) and A
′′
(t). Using mix-
ers, we often encounter the product of a pure sinusoid [Eq. (1)] multiplied
by a modulated signal [Eq. (15)]. The spectrum of such product consists of
two replicas of the modulated input, translated to the frequency sum and
to the frequency difference (IF signal Fig. 4).
3 Linear modes
For the mixer to operate in any of the linear modes, it is necessary that
• the LO port is saturated by a suitable sinusoidal signal,
• a small (narrowband) signal is present at the RF input.
The reader should refer to Sec. 3.3 for more details about linearity.
3.1 Linear frequency converter (LC) mode
The additional condition for the mixer to operate as a linear frequency
converter is that the LO and the RF signals are separated in the frequency
domain (Fig. 4).
It is often convenient to describe the mixer as a system (Fig. 5), in which
the behavior is modeled with functional blocks. The clipper at the LO input
limits the signal to the saturation level V
S
, while the clipper at the RF port
February 2, 2008 E. Rubiola Tutorial on mixers 11
ω
l
RF
IF
ω
i
ω
i
−ω
l
ω
i
+ ω
l
LO
LSB USB
S
p
e
c
t
r
a
ω
ω
ω
Figure 4: Frequency domain representation of the linear converter mode.
Negative frequencies are not shown.
(idle) 50 Ω
50 Ω
50 Ω
IF out
v
o
(t)
v
l
(t) = ±V
S
internal LO signal
multiplier
saturated
P ≥ P
S
LO input
v
p
(t)
not saturated
P ≪P
S
RF input
v
i
(t)
clipper
clipper
(active)
Figure 5: Model of the double balanced mixer operated as a linear converter.
is idle because this port is not saturated. The overall effect is that the inter-
nal LO voltage v
l
(t) is approximately a trapezoidal waveform that switches
between the saturated levels ±V
S
. The value of V
S
is a characteristic pa-
rameter of the specific mixer. The effect of higher LO power is to shrink
the fraction of period taken by the slanted edges, rather than increasing V
S
.
The asymptotic expression of v
l
(t) for strong saturation is
v
l
(t) =
4
π
V
S
¸
odd k≥1

−1

k−1
2
1
k
cos(kω
l
t) (16)
=
4
π
V
S
¸
cos ω
l
t −
1
3
cos 3ω
l
t +
1
5
cos 5ω
l
t −. . . + . . .

The filters account for the bandwidth limitations of the actual mixer. The
IF output is often coupled in dc. As an example, Table 3 gives the main
characteristics of two typical mixers.
A simplified description of the mixer is obtained by approximating the
12 E. Rubiola Tutorial on mixers February 2, 2008
Table 3: Example of double balanced mixers.
port HF-UHF mixer microwave mixer
LO 1–500 MHz 8.4–18 GHz
7 dBm ±1 dB 8–11 dBm
swr < 1.8 swr < 2
RF 1–500 MHz 8.4–18 GHz
0 dBm max 0 dBm max
swr < 1.5 swr < 2
IF dc – 500 MHz dc – 2 GHz
0 dBm max 0 dBm max
swr < 1.5 swr < 2
ssb loss 5.5 dB max ssb loss 7.5 dB max
all ports terminated to 50 Ω
internal LO waveform v
l
(t) with the first term of its Fourier expansion
v
l
(t) = V
L
cos(ω
l
t) . (17)
The input signal takes the form
v
i
(t) = A
i
(t) cos [ω
i
t + φ
i
(t)] , (18)
where A
i
(t) and φ
i
(t) are the slow-varying signals in which information is
coded. They may contain a dc term. The output signal is
v
o
(t) =
1
U
v
i
(t) v
l
(t) (19)
=
1
U
A
i
(t) cos

ω
i
t + φ
i
(t)

V
L
cos(ω
l
t) (20)
=
1
2U
V
L
A
i
(t)

cos


l
−ω
i
)t −φ
i
(t)

+ cos


l
+ ω
i
)t + φ
i
(t)

¸
.
(21)
The trivial term U = 1 V is introduced for the result to have the physical
dimension of voltage.
An optional bandpass filter, not shown in Fig. 5, may select the upper
sideband (USB) or the lower sideband (LSB). If it is present, the output
signal is
v
o
(t) =
1
2U
V
L
A
i
(t) cos


l
−ω
i
)t −φ
i
(t)

LSB (22)
v
o
(t) =
1
2U
V
L
A
i
(t) cos


l
+ ω
i
)t + φ
i
(t)

USB . (23)
February 2, 2008 E. Rubiola Tutorial on mixers 13
LO
ω
l
ω
RF
S
p
e
c
t
r
a
ω
IF
ω
ω

i
= ω
l
−ω
b
ω
′′
i
= ω
l
+ ω
b
ω
b
ω
b
ω
b
LSB USB
RF
LO
S
p
e
c
t
r
a
ω
l
ω
IF
ω
ω/ω
0
ω
b
2 1 3 4 etc.
ω
b
ω
b
filter mask
IF RF
LO
v
o
(t)
not saturated
v
i
(t)
P ≪P
S saturated
v
l
(t) ω
b
P = P
S
Figure 6: Image frequency in a conversion circuit.
Image frequency. Let us now consider the inverse problem, that is, the
identification of the input signal by observing the output of a mixer followed
by a band-pass filter (Fig. 6 top). In a typical case, the output is a band-pass
signal
v
o
(t) = A
o
(t) cos

ω
b
t + φ
o
(t)

, (24)
centered at ω
b
, close to the filter center frequency. It is easily proved that
there exist two input signals
v
L
(t) = A
L
(t) cos


l
−ω
b
)t + φ
L
(t)

LSB (25)
v
U
(t) = A
U
(t) cos


l
+ ω
b
)t + φ
U
(t)

USB , (26)
that produce a signal that passes through the output filter, thus contribute
to v
o
(t). It is therefore impossible to ascribe a given v
o
(t) to v
L
(t) or to
its image v
U
(t) if no a-priori information is given. Fig. 6 (middle) gives
the explanation in terms of spectra. The USB and the LSB are image
of one another with respect to ω
l
. In most practical cases, one wants to
detect one signal, so the presence of some energy around the image frequency
14 E. Rubiola Tutorial on mixers February 2, 2008
is a nuisance. In the case of the superheterodyne receiver, there results
ambiguity in the frequency at which the receivers is tuned. Even worse, a
signal at the image frequency interferes with the desired signal. The obvious
cure is a preselector filter preceding the mixer input.
More generally, the input signal can be written as
v
i
(t) =
¸
n
A

n
(t) cos(nω
0
t) −A
′′
n
(t) sin(nω
0
t) , (27)
which is a series of contiguous bandpass processes of bandwidth ω
0
, centered
around nω
0
, and spaced by ω
0
. The output is
v
o
(t) =
1
U

v
l
(t) v
i
(t)

∗ h
bp
(t) , (28)
where “∗” is the convolution operator, and h
bp
(t) the impulse response of the
bandpass IF filter. The convolution ∗ h
bp
(t) defines the pass-band filtering.
Accordingly, the terms of v
i
(t) for which |nω
0
− ω
l
| is in the pass-band of
the filter contribute to the output signal v
o
(t). Fig. 6 (bottom) shows the
complete conversion process.
Multi-harmonic conversion. In usual conditions, the LO port is well
saturated. Hence it makes sense to account for several terms of the Fourier
expansion (16) of the LO signal. Each term of Eq. (16) is a sinusoid of
frequency kω
l
that converts the portions of spectrum centered at |kω
l

b
|
and |kω
l
−ω
b
| into ω
b
(Fig. 7), thus
v
o
(t) =
1
U
v
i
(t) v
l
(t) (29)
=
1
U
A
i
(t) cos

ω
i
t + φ
i
(t)

4
π
V
S
¸
odd k≥1

−1

k−1
2
1
k
cos(kω
l
t) (30)
=
1
2U
4
π
V
S
A
i
(t)
¸
odd k≥1

−1

k−1
2
1
k

cos

(kω
l
−ω
i
)t −φ
i
(t)

+
+ cos

(kω
l
+ ω
i
)t + φ
i
(t)

¸
. (31)
With k = 1, one term can be regarded as the signal to be detected, and
the other one as the image. All the terms with k > 1, thus 3ω
0
, 5ω
0
, etc.,
as stray signals taken in because of distortion. Of course, the mixer can
be intentionally used to convert some frequency slot through multiplication
by one harmonic of the LO, at the cost of lower conversion efficiency. A
bandpass filter at the RF input is often necessary to stop unwanted signals.
Sampling mixers are designed for this specific operation. Yet their internal
structure differs from that of the common double-balanced mixer.
February 2, 2008 E. Rubiola Tutorial on mixers 15
ω
b
ω
b
ω
b
ω
b
stray
ω
b
ω
b
IF
ω ω
b

l
stray
LO
Spectra
RF
ω

1
ω

3
ω
′′
3
ω

5
ω
′′
5
ω
ω
signal & image
ω
′′
1
ω
l

l
Figure 7: Multi-harmonic conversion.
In real mixers the Fourier series expansion of v
l
(t) can be written as
v
l
(t) =
¸
odd k≥1

−1

k−1
2
V
L,k
cos(kω
l
t + φ
k
) , (32)
for Eq. (31) becomes
v
o
(t) =
1
2U
A
i
(t)
¸
odd k≥1

−1

k−1
2
V
L,k

cos

(kω
l
−ω
i
)t −φ
i
(t)

+
+ cos

(kω
l
+ ω
i
)t + φ
i
(t)

¸
. (33)
The first term of Eq. (32) is equivalent to (17), thus V
L,1
= V
L
. Equation
(32) differs from Eq. (16) in the presence of the phase terms φ
k
, and in that
the coefficient V
L,k
decrease more rapidely than 1/k. This due to non-perfect
saturation and to bandwidth limitation. In weak saturation conditions the
coefficient V
L,k
decrease even faster.
Looking at Eq. (16), one should recall that frequency multiplication re-
sults in phase noise multiplication. If the LO signal contains a (random)
phase ϕ(t), the phase kϕ(t) is present in the k-th term.
For a more accurate analysis, the diode can no longer be modeled as a
switch. The diode forward current i
F
is governed by the exponential law
i
F
= I
s

e
v
F
ηV
T
−1

(34)
where V
F
is the forward voltage, I
s
the inverse saturation current, η ∈
[1 . . . 2] a technical parameter of the junction, and V
T
= kT/q the thermal
voltage at the junction temperature. At room temperature, it holds that
V
T
= kT/q ≃ 25.6 mV. The term “−1” is negligible in our case. In the
presence of a sinusoidal pump signal, the exponential diode current can be
16 E. Rubiola Tutorial on mixers February 2, 2008
expanded using the identity
e
z cos φ
= I
0
(z) + 2

¸
k=1
I
k
(z) cos(kφ) , (35)
where I
k
(·) is the modified Bessel function of order k. As a consequence of
the mixer symmetry, the even harmonics are canceled and the odd harmonics
reinforced. Ogawa [OMK80] gives an expression of the IF output current
i
o
(t) = 4I
s
V
rf
ηV
T
¸
odd k≥1
I
k

V
lo
ηV
T

cos(kω
l
+ ω
i
)t + cos(kω
l
−ω
i
)t

. (36)
Equation (36) is valuable for design purposes. Yet, it is of limited usefulness
in analysis because some parameters, like I
s
and η are hardly available. In
addition, Eq. (36) holds in quasistatic conditions and does not account for a
number of known effects, like stray inductances and capacitances, varactor
effect in diodes, bulk resistance of the semiconductors, and other losses.
Nonetheless, Eq. (36) provides insight in the nature of the coefficients V
L,k
.
Rules for the load impedance at the IF port. The product of two
sinusoids at frequency ω
i
and ω
l
, inherently, contains the frequencies ω
i
±ω
l
.
At the IF port, current flow must be allowed at both these frequencies,
otherwise the diodes can not switch. The problem arises when IF selection
filter shows high impedance in the stop band. Conversely, low impedance
Z ≪ R
0
is usually allowed. Figure 8 shows three typical cases in which a
filter is used to select the |ω
i
− ω
l
| signal at the IF output, and to reject
the image at the frequency |ω
i
+ ω
l
|. The scheme A is correct because the
image-frequency current can flow through the diodes (low impedance). The
scheme B will not work because the filter is nearly open circuit at the image
frequency. The scheme C is a patched version of B, in which an additional
RC cell provides the current path for the image frequency. The efficient use
of a mixer as a multi-harmonic converter may require a specific analysis of
the filter.
In microwave mixers, the problem of providing a current path to the
image frequency may not be visible, having been fixed inside the mixer.
This may be necessary when the image frequency is out of the bandwidth,
for the external load can not provide the appropriate impedance.
Rules are different in the case of the phase detector because the current
path is necessary at the 2ω
l
frequency, not at dc.
Can the LO and RF ports be interchanged? With an ideal mixer yes,
in practice often better not. Looking at Fig. 2, the center point of the LO
transformer is grounded, which helps isolation. In the design of microwave
February 2, 2008 E. Rubiola Tutorial on mixers 17
IF
LO
RF
IF
LO
RF
IF
LO
RF
50 Ω
filter
ω
i
ω
l
50 Ω
Z
i
load
ω

i
−ω
l
| |ω
i
+ ω
l
|
|Z
i
|
50 Ω
(short)
low Z
A: correct
filter
ω
i
ω
l
50 Ω
Z
i
load
50 Ω
ω

i
−ω
l
| |ω
i
+ ω
l
|
high Z
(open)
|Z
i
|
B: incorrect
filter
ω
i
ω
l
50 Ω
Z
i
load
ω

i
−ω
l
| |ω
i
+ ω
l
|
|Z
i
|
C: patched
Figure 8: The mixer is followed by a filter that selects the |ω
i
−ω
l
| frequency.
mixers, where the transformers are replaced with microstrip baluns, opti-
mization may privilege isolation from the LO pump, and low loss in the RF
circuit. This is implied in the general rule that the mixer is designed and
documented for the superheterodyne receiver. Nonetheless, interchanging
RF and LO can be useful in some cases, for example to take benefit from
the difference in the input bandwidth.
3.2 Linear Synchronous Detector (SD) Mode
The general conditions for the linear modes are that the LO port is saturated
by a suitable sinusoidal signal, and that a small (narrowband) signal is
present at the RF input. The additional conditions for the mixer to operate
in the SD mode are: (1) the LO frequency ω
l
is tuned at the center of
the spectrum of the (narrowband) RF signal, and (2) the IF output is low-
passed.
The basic mixer operation is the same of the frequency conversion mode,
with the diode ring used as a switch that inverts or not the input polarity
18 E. Rubiola Tutorial on mixers February 2, 2008
loss ℓ
2
/2
power
power power
loss ℓ
2
loss ℓ
2
ω ω
0
−ω
0
RF
ω ω
0
Spectra
LO
1/ℓ
2
1/ℓ
2
ω
IF
−ω
0
−2ω
0

0
Figure 9: Frequency-domain sketch of the linear synchronous detection.
Spectra

l
ω
l

l
ω
ω
ω
u
n
w
a
n
te
d
u
n
w
a
n
te
d
LO
RF
IF
Figure 10: Signals are converted to IF by the harmonics at frequency mul-
tiple than the LO frequency.
dependig on the sign of the LO. The model of Fig. 5 is also suitable to
the SD mode. Yet, the frequency conversion mechanism is slightly different.
Figure 9 shows the SD mode in the frequency domain, making use of two-
sided spectra. Using one-sided spectra, the conversion products of negative
frequency are folded to positive frequencies. Of course, the multi-harmonic
frequency conversion mechanism, due to the harmonics multiple of the LO
frequency still works (Figure 10).
The simplest way to understand the synchronous conversion is to rep-
resent the input and the internal LO signal v

l
(t) = V
L
cos(ω
0
t + φ
L
) in
Cartesian coordinates
3
v
i
(t) = x(t) cos ω
0
t −y(t) sin ω
0
t (37)
v

l
(t) = V
L
[cos φ
L
cos ω
0
t −sin φ
L
sin ω
0
t] (38)
3
In this Section we use x and y in order to emphasize some properties of the synchronous
detection tightly connected to Cartesian-coordinate representation. Here, x and y are the
same thing of A

and A
′′
of Eq. (15).
February 2, 2008 E. Rubiola Tutorial on mixers 19
v
l
= V
L
[cos φ
L
cos ω
0
t −sin φ
L
sin ω
0
t]
RF
internal
LO signal
IF
equivalent to
v
i
= xcos ω
0
t −y sin ω
0
t X =
V
L
2U
[xcos φ
L
+ y sin φ
L
]
v
l
= V
L
cos(ω
0
t + φ
L
)
Figure 11: Linear synchronous detection.
The signal at the output of the low-pass filter is
4
(Fig. 11)
X(t) =
1
U
v
i
(t) v

l
(t) ∗ h
lp
(39)
=
1
U

xcos ω
0
t −y sin ω
0
t

V
L

cos φ
L
cos ω
0
t −sin φ
L
sin ω
0
t

∗ h
lp
(40)
=
1
2U
V
L

xcos φ
L
+ y sin φ
L
+ (2ω terms)

∗ h
lp
, (41)
thus,
X(t) =
1
2U
V
L

x(t) cos φ
L
+ y(t) sin φ
L

. (42)
Eq. (42) can be interpreted as the scalar product
X =
1
2U
V
L
(x, y) · (cos φ
L
, sin φ
L
) , (43)
plus a trivial factor
1
2U
V
L
that accounts for losses.
Let us now replace the LO signal v

l
(t) with
v
′′
l
(t) = −V
L
sin(ω
0
t + φ
L
) = −V
L
[sinφ
L
cos ω
0
t −cos φ
L
sin ω
0
t] . (44)
In this conditions, the output signal is
Y (t) =
1
U
v
i
(t) v
′′
l
(t) ∗ h
lp
(45)
=
1
U

xcos ω
0
t −y sin ω
0
t

V
L

−sin φ
L
cos ω
0
t −cos φ
L
sin ω
0
t

∗ h
lp
(46)
=
1
2U
V
L

−xsin φ
L
+ y cos φ
L
+ (2ω terms)

∗ h
lp
, (47)
4
Once again, we emphasize the properties connected with the Cartesian-coordinate
representation. X(t) is the same thing of vo(t) of other sections.
20 E. Rubiola Tutorial on mixers February 2, 2008
φ
L
X
x
(x, y)
(
X
,
Y
)
y
Y
Figure 12: Cartesian-frame rotation. The coefficient
1
2U
V
L
is implied.
RF IF
LO
RF IF
LO v
′′
l
= −V
L
sin(ω
0
t + φ
L
)
90

v
i
= xcos ω
0
t −y sin ω
0
t
(in-phase)
(quadrature)
pump
X =
V
L
2U
[xcos φ
L
+ y sin φ
L
]
v

l
= V
L
cos(ω
0
t + φ
L
)
Y =
V
L
2U
[−xsin φ
L
+ y cos φ
L
]
Figure 13: Basic I-Q detector.
thus,
Y (t) =
1
2U
V
L

−x(t) sin φ
L
+ y(t) cos φ
L

. (48)
Finally, by joining Equations (42) and (48), we find
¸
X(t)
Y (t)

=
1
2U
V
L
¸
cos φ
L
sin φ
L
−sin φ
L
cos φ
L
¸
x(t)
y(t)

. (49)
Equation (49) is the common form of a frame rotation by the angle φ
L
in
Cartesian coordinates (Fig. 12).
The simultaneous detection of the input signal with two mixers pumped
in quadrature is common in telecommunications, where QAM modulations
are widely used
5
. The theory of coherent communication is analyzed in
[Vit66]. Devices like that of Fig. 13, known as I-Q detectors, are com-
mercially available from numerous manufacturers. Section 7 provide more
details on these devices.
5
For example, the well known wireless standard 811g (WiFi) is a 64 QAM. The trans-
mitted signal is of the form (37), with x and y quantized in 8 level (3 bits) each.
February 2, 2008 E. Rubiola Tutorial on mixers 21
3.3 Linearity
A function f(·) is said linear [Rud76] if it has the following two properties
f(ax) = af(x) (50)
f(x + y) = f(x) + f(y) . (51)
The same definition applies to operators. When a sinusoidal signal of ap-
propriate power and frequency is sent to the LO port, the mixer is linear,
that is, the output signal v
o
(t) is a linear function of the input v
i
(t). This
can be easily proved for the case of simple conversion [Eq. (21)]
v
o
(t) =
1
2U
V
L
A
i
(t)

cos


l
−ω
i
)t −φ
i
(t)

+ cos


l
+ ω
i
)t + φ
i
(t)

¸
The linearity of v
o
(t) vs. v
i
(t) can also be demonstrated in the case of the
multi-harmonic conversion, either by taking a square wave as the LO internal
signal [Eq. (31)], or by using the internal LO signal of real mixers [Eq. (33)].
In fact, the Fourier series is a linear superposition of sinusoids, each of which
treated as above. In practice, the double balanced mixer can be used in a
wide range of frequency (up to 10
4
), where it is linear in a wide range of
power, which may exceed 10
16
(160 dB).
In large-signal conditions, the mixer output signal can be expanded as
the polynomial
v
o
(v
i
) = a
0
+ a
1
v
i
+ a
2
v
2
i
+ a
3
v
3
i
+ . . . . (52)
The symmetric topology cancels the even powers of v
i
, for the above poly-
nomial can not be truncated at the second order. Yet, the coefficient a
2
is
nonzero because of the residual asymmetry in the diodes and in the baluns.
Another reason to keep the third-order term is the adjacent channel inter-
ference. In principle, transformer nonlinearity should also be analyzed. In
practice, this problem is absent in microwave mixers, and a minor concern
with ferrite cores. The coefficient a
1
is the invese loss ℓ. The coefficients
a
2
and a
3
are never given explicitely. Instead, the intercept power (IP2 and
IP3) is given, that is, the power at which the nonlinear term (a
2
v
2
i
and a
3
v
3
i
)
is equal to the linear term.
4 Mixer loss
The conversion efficiency of the mixer is operationally defined via the two-
tone measurement shown in Fig. 14. This is the case of a superheterodyne
receiver in which the incoming signal is an unmodulated sinusoid v
i
(t) =
V
i
cos ω
i
t, well below saturation. The LO sinusoid is set to the nominal
saturation power. In this condition, and neglecting the harmonic terms
higher than the first, the output signal consists of a pair of sinusoids of
22 E. Rubiola Tutorial on mixers February 2, 2008
P
l
≃ P
S
loss
LO
RF
IF
ω
i

i
−ω
l
| |ω
i
+ ω
l
|
saturated
loss
linear
P
i
P
o
= P
l ω
l
P
o
= P
i
P
o
log-log
S
p
e
c
t
r
a
(P
o
) P
o
P
l
P
i
ω
ω
ω

2

2
scale
saturated
ω
i
not saturated

i
−ω
l
|
P
o
P
i
≪P
S
IF
RF LO
ω
l
Figure 14: Definition of the SSB loss ℓ.
frequency ω
o
= |ω
l
±ω
i
|. One of these sinusoids, usually |ω
l
−ω
i
| is selected.
The SSB power loss ℓ
2
of the mixer is defined
6
as
1

2
=
P
o
P
i
SSB loss ℓ (53)
where P
i
is the power of the RF input, and P
o
is the power of the IF output
at the selected freqency. The specifications of virtually all mixes resort to
this definition.
The loss is about constant in a wide range of power and frequency. The
upper limit of the RF power range is the saturation power, specified as the
compression power P
1 dB
at which the loss increases by 1 dB.
Intrinsic SSB loss. The lowest loss refers to the ideal case of the zero-
threshold diode, free from resistive dissipation. The LO power is entirely
wasted in switching the diodes. Under this assumptions, the ring of Figure 2
works as a loss-free switch that inverts or not the polarity of the RF, v
o
(t) =
±v
i
(t), according to the sign of v
l
(t). Of course, the instantaneous output
power is conserved
1
R
0
v
2
i
(t) =
1
R
0
v
2
o
(t) . (54)
Nonetheless, the mixer splits the input power into the conversion products
at frequency |ω
i
±ω
l
| and higher harmonics, for only a fraction of the input
power is converted into the desired frequency. There result a loss inherent
in the frequency conversion process, found with the definition (53).
6
In our previous articles we took ℓ = Pi/Po instead of ℓ
2
= Pi/Po. The practical use
is unchanged because ℓ is always given in dB.
February 2, 2008 E. Rubiola Tutorial on mixers 23
In the described conditions, the internal LO signal is a unit square wave
(V
S
= 1 V), whose Fourier series expansion is
v
l
(t) =
4
π
¸
cos ω
l
t −
1
3
cos 3ω
l
t +
1
5
cos 5ω
l
t −. . . + . . .

. (55)
Only the first term of the above contributes to the down-converted signal at
the frequency ω
b
= |ω
i
−ω
l
|. The peak amplitude of this term is V
L
=
4
π
V.
Hence,
v
o
(t) =
1
U
v
l
(t) v
i
(t) (56)
=
4
π
cos(ω
l
t) V
i
cos(ω
i
t) (57)
=
4
π
V
i
1
2

cos[(ω
i
−ω
l
)t] + cos[(ω
i
+ ω
l
)t]
¸
(58)
=
2
π
V
i
cos[ω
b
t] rubbing out the USB (59)
The RF and IF power are
P
i
=
V
2
i
2R
0
and P
o
=
1
2R
0
4V
2
i
π
2
(60)
from which the minimum loss ℓ =

P
i
/P
o
is
ℓ =
π
2
≃ 1.57 (3.92 dB) minimum SSB loss. (61)
SSB loss of actual mixers. The loss of microwave mixer is usually be-
tween 6 dB for the 1-octave devices, and 9 dB for 3-octave units. The dif-
ference is due to the microstrip baluns that match the nonlinear impedance
of the diodes to the 50 Ω input over the device bandwidth. In the case of a
narrow-band mixer optimized for conversion efficiency, the SSB loss can be
of 4.5 dB [OMK80]. The loss of most HF/UHF mixers is of about 5–6 dB
in a band up to three decades. This is due to the low loss and to the large
bandwidth of the tranmission-line transformers. Generally, the LO satura-
tion power is between 5 and 10 mW (7–10 dBm). Some mixers, optimized
for best linearity make use of two or three diodes in series, or of two diode
rings (see Fig. 26), and need larger LO power (up to 1 W). The advantage
of these mixers is high intercept power, at the cost of larger loss (2–3 dB
more). When the frequencies multiple of the LO frequency are exploited to
convert the input signal, it may be necessary to measure the conversion loss.
A scheme is proposed in Fig. 15.
24 E. Rubiola Tutorial on mixers February 2, 2008

l
ω
b
reference
common
synthesizer
synthesizer
IF RF
LO
power
meter
analyzer
spectrum
RF
ω ω
i
ω
b
ω
IF
loss

2
3
power
bandwidth
filter
P ≪P
S
not saturated
P = P
S
ω
l
ω
i
saturated
odd k
ω
b
= ω
i
−kω
l
LO
Spectra
ω ω
l

l
Figure 15: Measurement of the mixer loss in harmonic coversion.
Derivation of the internal LO voltage from the loss. For the purpose
of analytical calculus, the amplitude V
L
of the internal LO signal is often
needed. With real (lossy) mixers, it holds that V
L
<
4
π
V. V
L
can be derived
by equating the output power P
i
/ℓ
2
to the power of the output product. The
usefulness of this approach is in that ℓ is always specified. Let
v
i
(t) = V
i
cos [ω
i
(t) + φ
i
] (62)
the RF input, and select the lower
7
output frequency ω
b
= |ω
i
− ω
l
|. The
internal LO signal is
v
l
(t) = V
L
cos(ω
l
t + φ
l
) . (63)
Measuring the output power, we can drop the phases φ and φ
l
. Hence, the
output signal is
v
o
(t) =
1
U
V
i
V
L

cos ω
i
t + cos ω
l
t

∗ h
bp
(t) (64)
=
1
U
V
i
V
L
1
2
cos(ω
i
−ω
l
)t (65)
The output power is
P
o
=
1
2R
0
1
4U
2
V
2
i
V
2
L
(66)
when the input power is
P
i
=
1
2R
0
V
2
i
. (67)
7
Some experimental advantages arise from taking ω
b
= |ωi−ω
l
| instead of ω
b
= |ωi+ω
l
|.
February 2, 2008 E. Rubiola Tutorial on mixers 25
Parameter: LO input power
Mixer MCL TFM−10514
+7 dBm
+1 dBm
−2 dBm
−5 dBm
−8 dBm
−11 dBm
−14 dBm
−20 dBm
−17 dBm
+10 dBm
+4 dBm
RF input power, dBm
−80
−70
−60
−50
−40
−30
−20
−10
0
I
F
o
u
t
p
u
t
p
o
w
e
r
,
d
B
m
−90
−50 −40 −30 −20 −10 0 +10
Figure 16: Conversion loss measured at various LO power levels (1990 p. 12).
Combining the two above Equations with the definition of ℓ [Eq. (66)], we
obtain
1

2
1
2R
0
A
2
i
=
1
2R
0
1
4U
2
A
2
i
V
2
L
, (68)
hence
V
L
=
2U

Internal LO peak amplitude. (69)
Interestingly, the loss of most mixers is close to 6 dB, for V
L
≃ 1 V, while
the intrinsic loss ℓ = π/2 yields V
L
= 4/π ≃ 1.27 V.
What if the LO power differs from the nominal power? When the
LO input is saturated, the LO power has little or no effect on the output
signal. This fact is often referred as power desensitization (also LO desen-
sitization, or pump desensitization). In a narrow power range, say ±2 dB
26 E. Rubiola Tutorial on mixers February 2, 2008
from the nominal power, the conversion loss changes slightly, and noise also
varies. The internal Schottky diodes exhibit exponential i = i(v) character-
istics, hence lower LO power is not sufficient to saturate the diodes, and the
the ring is unable to switch. The conversion efficiency 1/ℓ is reduced, and
drops abruptly some 10 dB below the nominal LO power. As a side effect of
loss, white noise increases. Figure 16 shows an example of output power as
a function of the RF power, for various LO power levels. Below the nominal
LO power, flicker noise increases. Whereas this phenomenon is still unclear,
we guess that this is due to the increased fraction of period in which the
diodes are neither open circuit or saturated, and that up conversion of the
near-dc flickering of the junction takes place during this transition time .
Insufficient LO power may also impair symmetry, and in turn the cancel-
lation of even hamonics. The physical explanation is that saturated current
is limited by the diode bulk resistance, which is more reproducible than
the exponential law of the forward current. Increasing the fraction of time
in which the exponential law dominates emphasizes the asymmetry of the
diodes.
Too high LO power may increase noise, and damage the mixer. Special
care is recommanded with high-level mixers, in which the nominal LO power
of of 50 mW or more, and in the miniaturized mixers, where the small size
limits the heat evacuation.
According to the model of Fig. 5, the LO clipper limits the internal
voltage to ±V
S
, which turns the input sinusoid into a trapezoidal waveform.
Hence, the input power affects the duration of the wavefronts, and in turns
the harmonic contents. As a result, a circuit may be sensitive to the LO
power if stray input signals are not filtered out properly.
Finally, changing the LO power affects the dc voltage at the IF output.
This can be a serious problem when the mixer is used as a synchronous
converter or as a phase detector.
5 Saturated Modes
When both RF and LO inputs are saturated, the mixer behavior changes
radically. The mixer can no longer be described as a simple switch that
invert or not the RF signal, depending on the LO sign. Instead, at each
instant the largest signal controls the switch, and sets the polarity of the
other one. Of course, the roles are interchanged continuously. Strong odd-
order harmonics of the two input frequencies are present, while even-order
harmonics are attenuated or cancelled by symmetry. Saturation means that
amplitude has little effect on the output, for saturated modes are useful in
phase detectors or in frequency synthesis, where amplitudes are constant.
A further consequence of saturation is phase noise multiplication, which is
inherent in harmonic generation. In the case of saturated modes, phase noise
February 2, 2008 E. Rubiola Tutorial on mixers 27
multiplication takes place in both LO and RF.
In saturated modes the specified maximum power at the RF port is
always exceeded. When this maximum power is exceeded, the mixer leaves
the “normal” linear operation, still remaining in a safe operating range until
the “absolute maximum ratings” are approached. Read page 7.
The model of Fig. 5 describes some characteristics, as it emphasizes
the internally clipped waveforms, and the cancellation of even harmonics.
Yet, the model fails in predicting amplitude because the ring is no longer a
multiplier. The output amplitude is lower than expected.
5.1 Saturated Frequency Converter (SC) Mode
The conditions for the mixer to operate in SC mode are
• the LO and the RF ports are saturated by sinusoidal signals,
• the input frequencies are not equal, and the ratio ω
l

i
is not too close
to the the ratio of two small integers (say, 5–7),
• the output is band-passed.
Let the input signals
v
i
(t) = V

P
cos ω
i
t (70)
v
p
(t) = V
′′
P
cos ω
l
t . (71)
If possible, the saturated amplitudes V

P
and V
′′
P
should be equal. The main
output signal consists of the pair of sinusoids
v
o
(t) = V
O
cos(ω
l
−ω
i
)t + V
O
cos(ω
l
+ ω
i
)t (72)
that derives from the product v
i
(t) v
l
(t). Yet, the output amplitude V
O
is
chiefly due to the internal structure of the mixer, and only partially in-
fluenced by V

P
and V
′′
P
. A bandpass filter selects the upper or the lower
frequency of (72).
The unsuitability of the model of Fig. 5 to predict amplitude can be seen
in the following example.
Example 1 Replacing V

P
and V
′′
P
with V
L
yields V
O
=
1
2
UV
2
L
. Let us
consider typical mixer that has a loss of 6 dB when the LO has the nominal
power of 5 mW (7 dBm). From Eq. (69) we get V
L
≃ 1 V, thus we expect
V
O
= 250 mV, and an output power V
2
O
/2R
0
= 2.5 mW (+4 dBm) with
R
0
= 50 Ω. Yet, the actual power is hardly higher than 1.25 mW (+1 dBm).

28 E. Rubiola Tutorial on mixers February 2, 2008

i
Spectra
ω
LO

l

l
ω
o
= hω
l
+ kω
i
ω
IF
ω ω
i
RF

i
ω
l
3 3
1
1
3
−1
3
5
−1
5
1
1
5 5
−1 3
−3
3
3 −1
−1
1 1
1
−3
5
5
−3 h
k
Figure 17: Frequency conversion with a saturated mixer.
Accounting for the harmoncs, the output signal is
v
o
(t) =
¸
odd h,k
V
hk
cos(hω
l
+ kω
i
)t
positive frequencies
ω
hk
= hω
l
+ kω
i
> 0
, (73)
where the sum is extended to the positive output frequencies, i.e., hω
l
+kω
i
>
0. V
hk
decreases more rapidely than the product |hk|, and drops abruptly
outside the bandwidth. Figure 17 shows an example of spectra involving
harmonics.
The contition on the ratio ω
l

i
two output frequencies ω
h

k
′ and ω
h
′′
k
′′
do not degenerate in a single spectral line, at least for small h and k. This
problem is explained in Section 5.2.
Other authors write the output frequencies as |±hω
l
±kω
i
|, with positive
h and k. We recommend to keep the sign of h and k. One reason is that the
positive and negative subscripts of V
hk
make the spectrum measurements
unambiguously identifiable. Another reason is that input phase fluctuations
are multiplied by h and k, and wrong results may be obtained discarding
the sign.
5.2 Degenerated Frequency Converter (DC) Mode
The conditions for the mixer to operate in DC mode are the following
• the LO and the RF ports are saturated by sinusoidal signals,
• the input frequencies are not equal, and the ratio ω
l

i
is equal or
close to the the ratio of two small integers (say, 5–7 max.),
• the output is band-passed.
When ω
l
and ω
i
are multiple of a common frequency ω
0
, thus
ω
l
= pω
0
and ω
i
= qω
0
integer p>0, q>0, p=q , (74)
February 2, 2008 E. Rubiola Tutorial on mixers 29
the sum (73) degenerates, and groups of terms collapse into fewer terms of
frequency nω
0
, integer n. The combined effect of saturation and symmetry
produces strong odd-order harmonics hω
l
and kω
i
ω
l
: v
l1
= V
1
cos(pω
0
t + φ
l
) ω
i
: v
i1
= V
1
cos(qω
0
t + φ
i
)

l
: v
l3
= V
3
cos(3pω
0
t + 3φ
l
) 3ω
i
: v
i3
= V
3
cos(3qω
0
t + 3φ
i
)
· · · · · · · · · · · · · · · · · ·

l
: v
lh
= V
h
cos(hpω
0
t + hφ
l
) kω
i
: v
ik
= V
k
cos(kqω
0
t + kφ
i
)
· · · · · · · · · · · · · · · · · ·
inside the mixer. After time-domain multiplication, all the cross products
appear, with amplitude V
hk
, frequency (hp + kq)ω
0
, and phase hφ
l
+ kφ
i
.
The generic output term of frequency nω
0
derives from the vector sum of
all the terms for which
hp + kq = n , (75)
thus
v
n
(t) =
¸
h,k pair :
hp+kq=n
V
hk
cos(nω
0
t + hφ
l
+ kφ
i
) (76)
Reality is even more complex than (76) because
• some asymmetry is always present, thus even-order harmonics,
• each term of (76) may contain an additional constant phase φ
hk
,
• for a given ω
l
ω
i
pair, several output frequencies nω
0
exist, each one
described by (76). Due to nonlinearity, the v
n
(t) interact with one
another.
Fortunately, the amplitudes V
hk
decrease rapidly with |hk|, therefore the
sum (76) can be accurately estimated from a small number of terms, while
almost all the difficulty resides in parameter measurement. For this reason,
there is no point in devlopping a sophisticated theory, and the few cases of
interest can be anlyzed individually. The following example is representative
of the reality.
Example 2 The input frequencies are f
l
= 5 MHz and f
i
= 10 MHz, and
we select the output frequency f
o
= 5 MHz with an appropriate bnd-pas
filter. Thus f
0
= 5 MHz, p = 1, q = 2, and n = 1. The output signal (76)
results from the following terms
hf
l
+ kf
i
= nf
0
hp + kq = n v
n
(t)
−1×5 + 1×10 = 5 −1×1+1×2 = 1 V
−1 1
cos(ω
0
t−φ
l

i
)
+3×5 −1×10 = 5 +3×1−1×2 = 1 V
3 −1
cos(ω
0
t+3φ
l
−φ
i
)
−5×5 + 3×10 = 5 −5×1+3×2 = 1 V
−5 3
cos(ω
0
t−5φ
l
+3φ
i
)
+7×5 −3×10 = 5 +7×1−1×2 = 1 V
7 −3
cos(ω
0
t+7φ
l
−3φ
i
)
−9×5 + 5×10 = 5 −9×1+5×2 = 1 V
−9 5
cos(ω
0
t+7φ
l
−3φ
i
)
· · · · · · · · ·
30 E. Rubiola Tutorial on mixers February 2, 2008
–2
–1
0
1
2
–3 –2 –1 0 1 2 3
Phi
φ
i
= 0
V
3−1
V
−1 1

l
−ϕ
l
V
n
V
−1 1

l
V
n
V
3−1
ℑ ℑ
ℜ ℜ
−ϕ
l
f
l
= 5 a.u. f
i
= 10 a.u., f
n
= 5 a.u.
V
−1 1
= 1 a.u., V
3 −1
= 0.2 a.u. (−14 dB)
phase φ
l
output voltage, a.u.
phase gain
Figure 18: Simplified picture of degenerated frequency conversion. Only
V
−1 1
and V
3−1
are taken into account, with φ
i
= 0. Top: phasor rep-
resentation. Bottom: output voltage, and phase gain as a function of the
static phase φ
l
.

5.3 Phase Amplification Mechanism
Introducing the phasor (Fresnel vector) representation
8
Eq. (76) becomes
V
n
=
¸
V
hk
, thus
1

2
V
n
e
jφn
=
¸
h,k pair :
hp+kq=n
1

2
V
hk
e

hk
with φ
hk
= hφ
l
+ kφ
i
. (77)
Both V
n
and φ
n
are function of φ
l
and φ
i
, thus function of the phase rela-
tionship between the two inputs. Let ϕ the fluctuation of the static phase
8
In this section we use uppercase boldface for phase vectors, as in V = V e

. V is the
rms voltage.
February 2, 2008 E. Rubiola Tutorial on mixers 31
φ. The output phase fluctuation is
ϕ
n
=
∂φ
n
∂φ
l
ϕ
l
+
∂φ
n
∂φ
i
ϕ
i
, (78)
where the derivatives are evaluated in the static working point. There follows
that the input phase fluctuations φ
l
and φ
i
are amplified or attenuated
(gain lower than one) by the mixer. The phase gain/attenuation mechanism
is a consequence of degeneracy. The effect on phase noise was discovered
studying the regenerative frequency dividers [ROG92].
Figure 18 shows a simplified example in which a 5 MHz signal is obtained
by mixing a 5 MHz and a 10 MHz, accounting only for two modes (10 − 5
and 3×5 − 10). For φ
l
= 0, the vectors are in phase, and the amplitude is
at its maximum. A small negative φ
n
results from V
−11
and V
3 −1
pulling
in opposite directions. A phase fluctuation is therefore attenuated. For
φ
l
= π/4 ≃ 0.785, the vectors are opposite, and the amplitude is at its
minimum. The combined effect of V
−11
and V
3 −1
yields a large negative
φ
n
. With V
3 −1
/V
−1 1
= 0.2 (−14 dB), the phase gain ∂φ
n
/∂φ
l
spans from
−0.33 and 2, while it would be −1 (constant) if only the −1, 1 mode was
present.
The experimentalist not aware of degeneracy may obtain disappointing
results when low-order harmonics are present, as in the above example. The
deliberate exploitation of degeneracy to manage phase noise is one of the
most exhotic uses of the mixer.
Parameter Measurement. There are two simple ways to measure the
parameters of a degenerated frequency converter (Fig. 19).
The first method is the separate measurement of the coefficients V
hk
of
Eq. (76) by means of a spectrum analyzer. One input signal is set at a fre-
quency δ off the nominal frequency ω
l
(or ω
i
). In this condition degeneracy
is broken, and all the terms of Eq. (76) are visible as separate frequencies.
The offset δ must be large enough to enable the accurate measurement of all
the spectral lines with a spectrum analyzer, but small enough not to affect
the mixer operation. Values of 10–50 kHz are useful in the HF/UHF bands,
and up to 1 MHz at higher frequencies. Figure 20 provides an example.
This method is simple and provides insight. On the other hand, it is not
very accurate because it hides the phase errors φ
hk
that may be present in
each term.
The second method consists of the direct measurement of V
n
[Eq. (77)]
as a function of the input phase, φ
l
or φ
i
, by means of a vector voltmeter.
This gives amplitude and phase, from which the phase gain is derived. For
the measurement to be possible, the three signals must be converted to the
same frequency ω
0
with approprate dividers. Of course, the mixer must be
measured in the same conditions (RF and LO power) of the final applica-
tion. While one vector voltmter is sufficient, it is better to use two vector
32 E. Rubiola Tutorial on mixers February 2, 2008
phase
reference
common
synthesizer
synthesizer
÷n
divider
÷q
divider
reference
common
synthesizer
synthesizer
IF RF
LO
analyzer
spectrum
saturated
P = P
S
saturated
P = P
S
ω
i
= pω
0
ω
l
= qω
0
+ δ

0
÷p
divider
vector
voltmeter
vector
voltmeter
saturated
P = P
S
saturated
P = P
S
ω
i
= pω
0
ω
l
= qω
0

0
ω
0
adjustable
Figure 19: Parameter masurement of a degenerated frequency converter.
voltmters because the measurement accounts for the reflected waves in the
specific circuit. In some cases good results are obtained with resistive power
splitters located close to the mixer because these splitters are not directional.
Interestingly, most frequency synthesizers can be adjusted in phase even if
this feature is not explicitely provided. The trick consists of misaligning
the internal quartz oscillator when the instrument is locked to an external
frequency reference. If the internal phase locked loop does not contain an
integrator, the misalignamet turns into a phase shift, to be determined a
posteriori. The drawback of the direct measurement method is that it re-
quires up to two vector voltmeters, two frequency synthesizers and three
frequency dividers. In the general case, the dividers can not be replaced
with commercial synthesizers because a synthesizer generally accepts only a
small set of round input frequencies (5 MHz or 10 MHz). Figure 21 shows
an example of direct measurement, compared to the calculated values, based
on the first method.
5.4 Phase Detector (PD) Mode
The mixer works as a phase detector in the following conditions
• the LO and the RF ports are saturated by sinusoidal signals of the
same frequency ω
0
, about in quadrature,
February 2, 2008 E. Rubiola Tutorial on mixers 33
0 250
+10
0
−10
−20
−30
−40
−50
−60
−70
−250
I
F
o
o
t
p
u
t
p
o
w
e
r
,
d
B
m
center 239.4 MHz
span 500 kHz
res bw 1 kHz
video filter 3 kHz
frequency
kHz
−38

R
−7ν
L

R
−9ν
L

R
−5ν
L

R
−3ν
L
ν
R
−ν
L

L
−3ν
R

L
−2ν
R

L
−ν
R
ν
L
11ν
L
−5ν
R

L
−4ν
R
+1
−8
−32
−35
−40
−45
−49 −48
Figure 20: Amplitude, phase, and phase gain in a degenerated frequency
converter.
• the output is low-passed.
The product of such input signals is
cos

ω
0
t + φ

cos

ω
0
t −
π
2

=
1
2
sin


0
t + φ


1
2
sinφ , (79)
from which one obtains a sinusoid of frequency 2ω
0
, and a dc term −
1
2
sinφ
that is equal to −
1
2
φ for small φ. The output signal of an actual mixer is a
distorted sinusoid of frequency 2ω
0
plus a dc term, which can be approxi-
mated by
v
o
(t) = V
2
sin


0
t + φ

−V
0
sinφ . (80)
V
2
and V
0
are experimental parameters that depend on the specific mixer and
on power. Due to saturation, the maximum of |v
o
(t)| is about independent
of ϕ, hence V
2
decreases as the absolute value of the dc term increases.
Using the 2ω
0
output signal to double the input frequency is a poor
choice because (i) the quadrature condition can only be obtained in a limited
bandwidth, (ii) the IF circuit is usually designed for frequencies lower than
the RF and LO. A better choice is to use a reversed mode.
When the PD mode is used close to the quadrature conditions, the devi-
ation of dc response from sinφ can be ignored. After low-pass filtering, the
34 E. Rubiola Tutorial on mixers February 2, 2008
measured phase
∆ measured phase gain ∇ calculated phase gain
I
F
a
m
p
l
i
t
u
d
e
,
m
V
φ
o
,
d
e
g
r
e
e
s
p
h
a
s
e
g
a
i
n
φ
i
−φ
l
, degrees
⋄ calculated phase
+ measured amplitude
× calculated amplitude
Figure 21: Amplitude, phase, and phase gain in a degenerated frequency
converter.
output signal is
9
v
o
= −k
φ
φ + V
os
, (81)
where k
φ
is the phase-to-voltage gain [the same as V
0
in Eq. (80)], and V
os
is the dc offset that derives from asymmetry. Figure 22 shows an example
of phase detector charactaristics. The IF output can be loaded to a high
resistance in order to increase the gain k
φ
.
It is often convenient to set the input phase for zero dc output, which
compensate for V
os
. This condition occurs at some random—yet constant—
phase a few degrees off the quadrature conditions, in a range where the
mixer characteristics are virtually unaffected.
Due to diode asymmetry, the input power affects V
os
. Exploiting the
asymmetry of the entire v(i) law of the diodes, it is often possible to null
the output response to the fluctuation of the input power, therefore to make
the mixer insensitive to amplitude modulation. This involves setting the
phase between the inputs to an appropriate value, to be determined exper-
imentally. In our experience, the major problem is that there are distinct
9
The phase-to-voltage gain is also written as kϕ (with the alternate shape of ϕ) because
it is used with the small fluctuations ϕ.
February 2, 2008 E. Rubiola Tutorial on mixers 35
O
u
t
p
u
t

v
o
l
t
a
g
e
,

V
Ω 1 k
100 Ω
50 Ω
0 45 90 135 360
0
0.5
1.0
−0.5
−1.0
Phase difference, degrees
P
h
a
s
e
-
t
o
-
v
o
l
t
a
g
e
g
a
i
n
,
m
V
/
r
a
d
Input power, dBm
M14A term. to 50 Ω
DB0218LW2 term. to 10 kΩ
−10 −5 0 +10 +5
0
−15
100
200
300
400
500
Figure 22: Example of phase detector characteristics: output voltage as a
function of ϕ (data are from a handbook Macom) and phase-to-voltage gain
as a function of power (measured).
AM sensitivities
dv
o
dP
l
,
dv
o
dP
i
,
dv
o
d(P
l
+ P
i
)
, (82)
and that nulling one of them is not beneficial to the other two. In some
cases the nulls occurr within some 5

from the quadrarure, in other cases
farther, where the side effects of the offset are detrimental.
36 E. Rubiola Tutorial on mixers February 2, 2008
R
L
mixer
D3 D4
D1 D2
input
LO
R
G
v
p
(t)
RF
IF
out
v
o
(t)
input
LO source
IF
R
G
source
i
i
(t)
load
RF
Figure 23: Reversed-mode modulator.
6 Reversed Modes
The mixer can be reversed taking the IF port as the input and the RF port
as the output (Fig. 23). The LO signal makes the diodes switch, exactly
as in the normal modes. The major difference versus the normal modes is
the coupling bandwidth: the output is now ac-coupled via the RF balun,
while the input is in most cases dc-coupled. When impedance-matching is
not needed, the IF input can be driven with a current source.
6.1 Linear Modulator (LM)
The mixer works as a LM in the following conditions
• the LO port is saturated by a sinusoidal signal,
• a near-dc signal is present at the IF input,
• the IF input current is lower than the saturation current
10
I
S
.
As usual, the LO pump forces the diodes to switch. At zero input cur-
rent, due to symmetry, no signal is present at the RF output. When a
positive current i
i
is present, the resistance of D2 and D4 averaged over
the period decreases, and the conduction angle of D2 and D4 increases.
The average resistance of D1 and D3 increases, and their conduction angle
decreases. Therefore, a small voltage v
o
(t) appears at the RF output, of
amplitude proportional to i
i
, in phase with v
p
(t). Similarly, a negative i
i
produces an output voltage proportional to i
i
, of phase opposite to v
p
(t).
The mixer can be represented as the system of Fig. 24, which is similar to
the LC model (Fig. 5) but for the input-output filters. The internal sat-
10
The mixer saturation current, which can be of some mA, should not be mistaken for
the diode reverse saturation current. The latter can be in the range from 10
−15
A to 10
−15
A.
February 2, 2008 E. Rubiola Tutorial on mixers 37
signal
saturated
P ≥ P
S
LO input
v
p
(t)
50 Ω
50 Ω
50 Ω
not saturated
i
i
≪I
S
IF input
i
i
(t) v
o
(t)
idle
clipper
clipper
active
v
l
(t) = ±V
S
internal LO signal
multiplier
RF out
dc or sinus.
Figure 24: Reverse-mode model of a mixer.
urated LO signal can be approximated with a sinusoid v
l
(t) = V
L
cos ω
l
t,
[Eq. (17)], or expanded as Eq. (32). Strictly, V
L
can not be derived from
the reverse loss, which is not documented. Reciprocity should not given for
granted. Nonetheless, measuring some mixers we found that the ‘conven-
tional’ (forward) SSB loss ℓ and Eq. (69) provide useful approximation of
reverse behavior. Thus, the mixer operates as a linear modulator described
by
v
o
(t) =
1
U
v
i
(t) v
l
(t) (83)
=
1
U
v
i
(t) V
L
cos ω
l
t . (84)
Example 3 The LO signal of a mixer (Mini-Circuits ZFM-2) is a sinusoid of
frequency f
l
= 100 MHz and power P = 5 mW (7 dBm). In such conditions
the nominal SSB loss is ℓ = 2 (6 dB). By virtue of Eq. (69), V
L
= 1 V. When
the input current is i
i
= 2 mA dc, the input voltage is v
i
= R
0
i
i
= 100 mV
with R
0
= 50 Ω. After Eq. (84), we expect an output signal of 100 mV peak,
thus 71 mV rms. This is close to the measured value of 75 mV. The latter
is obtained fitting the the low-current experimental data of Fig. 25. Beyond
i
i
= 3 mA, the mixer lives gradually the linear behavior, and saturates at
some 230 mV rms of output signal, when i
i
≈ 12 mA dc. Similar results
were obtained testing other mixers.
6.2 Reverse Linear Converter (RLC)
The mixer works as a RLC in the following conditions
• the LO port is saturated by a sinusoidal signal,
38 E. Rubiola Tutorial on mixers February 2, 2008
200
0
20
40
60
80
100
120
140
160
0 2 4 6 8 10
IF input current, mA
R
F
v
o
l
t
a
g
e
,
m
V
r
m
s
12
180
RF output term. to 50 Ω
Mini Circuits ZFM-2, I input
(vol. 5 p. 52)
Figure 25: Gain of a mixer used ad a modulator. Input is driven with a
current source. Output is terminated to 50 Ω.
• a small narrowband signal is present at the IF input, which is not
saturated,
• LO and the IF separated in the frequency domain,
• an optional filter selects one of the beat products.
This mode is similar to the LM mode. Letting v
i
(t) = A
i
(t) cos[ω
i
(t)+φ
i
(t)]
the input, the output signal is
v
o
(t) =
1
U
v
i
(t) v
l
(t) (85)
=
1
U
A
i
(t) cos

ω
i
t + φ
i
(t)

V
L
cos(ω
l
t) (86)
=
1
2U
V
L
A
i
(t)

cos


l
−ω
i
)t −φ
i
(t)

+ cos


l
+ ω
i
)t + φ
i
(t)

¸
.
(87)
The model of Fig. 24 still holds, and the internal LO amplitude V
L
can be
estimated using Eq. (69) and the ‘conventional’ SSB loss ℓ.
If an external bandpass filter, not shown in Fig. 24, is present, the output
signal is
v
o
(t) =
1
2U
V
L
A
i
(t) cos


l
−ω
i
)t −φ
i
(t)

LSB, or (88)
v
o
(t) =
1
2U
V
L
A
i
(t) cos


l
+ ω
i
)t + φ
i
(t)

USB , (89)
February 2, 2008 E. Rubiola Tutorial on mixers 39
under the obvious condition that the signal bandwidth fits into the filter
passband.
6.3 Digital Modulator (DM) Mode
The mixer works as a DM in the following conditions
• the LO port is saturated by a sinusoidal signal,
• a large near-dc current is present at the IF input, which is saturated,
• the RF output is bandpassed.
Let v
p
= V
P
cos ω
l
t the LO input signal, i
i
= ±I
i
the IF input current, and
V
O
the saturated output amplitude. The output signal is
v
o
(t) = sgn(i
i
) V
O
cos ω
l
t , (90)
where sgn(·) is the signum function. Equation (90) represents a BPSK
(binary phase shift keying) modulation driven by the input current i
i
.
6.4 Reverse Saturated Converter (RSC) Mode
The mixer works in the RSC mode under the following conditions
• the LO and the IF ports are saturated by sinusoidal signals,
• the input frequencies are not equal, and the ratio ω
l

i
is not too close
to the the ratio of two small integers (say, 5-7 max.),
• the output is band-passed.
The RSC mode is similar to the SC mode, for the explanations given in
Section 5.1 also apply to the RSC mode. The only difference between SC
and RSC is the input and output bandwidth, because IF and RF are inter-
changed.
6.5 Reverse Degenerated Converter (RDC) Mode
The mixer works in the RDC mode when
• the LO and the IF ports are saturated by sinusoidal signals,
• the input frequencies are equal, or the ratio ω
l

i
is equal or close to
the the ratio of two small integers (say, no more than 5–7),
• the output is band-passed.
40 E. Rubiola Tutorial on mixers February 2, 2008
The RDC mode is similar to the DC mode (Section 5.1) but for the trivial
difference in the input and output bandwidth, as the roles of IF and RF are
interchanged. The output signal results from the vector addition of several
beat signals, each one with its own phase and amplitude.
It is to be made clear that when two equal input frequencies (ω
i
= ω
l
=
ω
0
) are sent to the input, the reverse mode differs significantly from the
normal mode. In the DC mode, this condition would turn the degenerated
converter mode into the phase-detector mode. But in the reversed modes
no dc output is permitted because the RF port is ac coupled. Of course,
a large 2ω
0
signal is always present at the RF output, resulting from the
vector addition of several signals, which makes the RDC mode an efficient
frequency doubler.
Example 4 The input frequencies are f
l
= f
i
= 5 MHz, and we select the
output f
o
= 10 MHz. Thus f
0
= 5 MHz, p = 1, q = 1, and n = 2. The
output signal [Eq. (76)] results from the follwoing terms
hf
l
+ kf
i
= nf
0
hp + kq = n v
n
(t)
+1×5 + 1×5 = 10 +1×1+1×1 = 2 V
1 1
cos(ω
0
t+φ
l

i
)
+3×5 −1×5 = 10 +3×1−1×1 = 2 V
3 −1
cos(ω
0
t+3φ
l
−φ
i
)
−1×5 + 3×5 = 10 −1×1+3×1 = 2 V
−1 3
cos(ω
0
t−φ
l
+3φ
i
)
+5×5 −3×5 = 10 +5×1−3×1 = 2 V
5 −3
cos(ω
0
t+5φ
l
−3φ
i
)
−3×5 + 5×5 = 10 −3×1+5×1 = 2 V
−3 5
cos(ω
0
t−3φ
l
+5φ
i
)
· · · · · · · · ·

7 Special Mixers and I-Q Mixers
Phase Detector. Some mixers are explicitely designed to operate in the
phase detector mode. In some cases such devices are actually general-
purpose mixers documented for phase detector operation. Often the IF
output impedance is larger than 50 Ω, typically 500 Ω. The main advan-
tage of this higher impedance is a lower residual white noise of the system.
In fact, the output preamplifier can hardly be noise-matched to an input
resistance lower than a few hundreds Ohms. The IF bandwidth reduction
that results from the increased output impedance is not relevant in practice.
The residual flicker, which is the most relevant parameter for a number of
measurements, is usually not documented
11
.
Analog Modulator / Variable Attenuator. A mixer can be designed
and documented to be used in a reverse mode as an analog modulator (See
Sec. 6.1). The fancy name “variable attenuator” is sometimes used. Yet, the
11
I never come across a phase detector whose residual flicker is documented.
February 2, 2008 E. Rubiola Tutorial on mixers 41
20 to 30 dBm
7 to 13 dBm Class I
Class III
20 to 30 dBm
Class III
20 to 30 dBm
Class III
Type 1
13 to 24 dBm
Class II
Type 1
13 to 24 dBm
Type 1
Class II
Type 2
Type 3
Diode circuit LO power
Figure 26: Diode assemblies of high linearity mixers.
mixer operation is more general than that of a simple attenuator because
the mixer input current can be either positive or negative, and the output
signal changes sign when the input current is negative.
BPSK Modulator. The BPSK modulator differs from the analog modu-
lator in that the IF input is saturated (See Sec. 6.3). Once again, the device
may differ from a general-purpose mixer mostly in the documentation.
High Linearity Mixers. In some cases low intermodulation performance
must be achieved at any cost. Special mixers are used, based on a ring in
which the diodes are replaced with the more complex elements shown in
Fig. 26 (classes I-III). High linearity is achieved by forcing the diodes to
switch abruptly in the presence of a large pump signal. These mixers, as
compared to the single-diode ones, need large LO power, up to 1 W, and
show higher loss.
Improved Impedance-Matching Mixers. The 90

hybrid junction,
used as a power splitter, has the useful property that the input (output)
is always impedance matched when the isolation port is correctly termi-
nated and the two outputs (inputs) are loaded with equal impedances. This
property is exploited joining two equal double-balanced mixers to form the
improved mixer of Fig. 27 (Class IV mixer). Other schemes are possible,
based on the same idea.
42 E. Rubiola Tutorial on mixers February 2, 2008

9
0
°

9
0
° 0
° 0
°

0
°

1
8
0
°
−90°
−90°
0
°
0
°
LO
LO
IF
IF
RF
RF
+
1
2

2U
V
I
V
L
sin(a −b)
R
0
LO
input
V
L
sin b
V
L
cos b
V
S
cos b
V
S
>

2V
L
1
2

2U
V
I
V
L
sin(a + b)
+
1
2

U
V
I
V
L
sin(a −b)
1
2

2U
V
I
V
L
sin(a + b)
out
IF
R
0
1
2U
V
I
V
L
sin(a + b)
+
1
2U
V
I
V
L
sin(a −b)
Class IV
input
RF
R
0
V
I
cos a
1

2
V
I
sin a
1

2
V
I
cos a
Figure 27: Improved impedance-matching mixer.
out
RF
input
LO
input
IF
Figure 28: Double-double-balanced mixer.
Double-Double-Balanced Mixers. The double-double-balanced mixer
(Figure 28) shows high 1 dB compression point, thus high dynamic range
and low distortion, and high isolation. This device is sometimes called triple
balanced mixer because it is balanced at the three ports. Other schemes are
possible.
Image-Rejection Mixer. Let us go back to the frequency conversion
system of Fig. 6, in which the LSB and the USB are converted into the
same IF frequency ω
b
. The scheme of Fig. 29 divides the IF components,
enabling the selection of the LSB or the USB input (RF) signal.
Let us for short a = ω
i
t and b = ω
l
t the instantaneous phase of the RF
and LO signal. The converted signals, at the IF output of the mixers are
v
1
=
1

2U
V
I
V
L
sin a cos b
v
2
=
1

2U
V
I
V
L
cos a cos b ,
February 2, 2008 E. Rubiola Tutorial on mixers 43
1
8
0
° 0
°
0
°
0
°
−90°
−90°
0
°
0
°
−90°
−90°
0
°
0
°
LO
LO
IF
IF
RF
RF
LO
input
R
0
V
S
cos b
V
S
>

2V
L
USB
LSB
IF out
input
RF
R
0
v
′′
1
+ v

2
v

1
+ v
′′
2
RF
LO
USB
ω
ω
ω
ω
ω
a
ω
b
ω
a
+ ω
b
LSB
RF
LO
USB
ω
ω
ω
ω
ω
b
ω
a
−ω
b
ω
a
−ω
b
LSB
ω
a
ω
a
+ ω
b
ω
a
+ ω
b
ω
a
−ω
b
out
out out
out
ω
a
+ ω
b
rejected rejected
rejected rejected
ω
b
−ω
a
ω
a
< ω
b
ω
a
> ω
b
V
I
cos a
1

2
V
I
sin a
1

2
V
I
cos a
v
1
=
1

2U
V
I
V
L
sin a cos b
v
2
=
1

2U
V
I
V
L
cos a cos b
V
L
cos b
V
L
cos b
Figure 29: Image-rejection mixer.
thus
v
1
=
1
2

2U
V
I
V
L

sin(a −b) + sin(a + b)

v
2
=
1
2

2U
V
I
V
L

cos(a −b) + cos(a + b)

.
The path of the hybrid junction labeled ‘−90

’ turns the phase of the
positive-frequency signals by −90

, and the phase of the negative-frequencies
signal by +90

. The rotated signals are
v
′′
1
=

1
4U
V
I
V
L

−cos(a −b) −cos(a + b)

a>b
1
4U
V
I
V
L

+cos(a −b) + cos(a + b)

a<b
v
′′
2
=

1
4U
V
I
V
L

+sin(a −b) + sin(a + b)

a>b
1
4U
V
I
V
L

−sin(a −b) −sin(a + b)

a<b
which also account for a factor 1/

2 due to energy conservation. The non-
44 E. Rubiola Tutorial on mixers February 2, 2008
rotated signals are
v

1
=
1
4U
V
I
V
L

sin(a −b) + sin(a + b)

v

2
=
1
4U
V
I
V
L

cos(a −b) + cos(a + b)

.
The output signals are
v
USB
= v
′′
1
+ v

2
=

1
4U
V
I
V
L

sin(a −b) + sin(a + b)

a>b (USB taken in)
0 a<b (LSB rejected)
(91)
v
LSB
= v

1
+ v
′′
2
=

0 a>b (USB rejected)
1
4U
V
I
V
L

cos(a −b) + cos(a + b)

a<b (LSB taken in)
(92)
The unwanted sideband is never cancelled completely. A rejection of 20
dB is common in practice. The main reason to prefer the image-rejection
mixer to a (simple) mixer is noise. Let us assume that the LO frequency ω
l
and the IF center frequency ω
IF
are given. The mixer converts both |ω
l
−ω
IF
|
and |ω
l
+ ω
IF
| to ω
IF
, while the image-rejection mixer converts only one of
these channels. Yet, the noise of the electronic circuits is present at both
frequencies.
Example 5 The IF filter of a FM receiver has a bandwidth of 300 kHz
centered at 10.7 MHz. In order to receive a channel at 91 MHz, we tune the
local oscillator to 101.7 MHz (101.7 − 10.7 = 91). A mixer down-convert
to IF two channels, the desired one (91 MHz) and the image frequency at
122.4 MHz (101.7 +10.7 = 122.4). In the best case, only noise is present at
the image frequency (122.4 MHz), which is taken in by the mixer, yet not
by the image-rejection mixer.
SSB Modulator. The SSB modulator (Fig. 30) is a different arrangement
of the same blocks used in the image-rejection mixer. The main purpose of
this device is to modulate a carrier by adding only one sideband, either LSB
or USB. All explanations are given on the scheme, in Fig. 30.
I-Q Detectors and Modulators. The two-axis synchronous detector in-
troduced in Section 3.2 is commercially available in (at least) two practical
implementations, shown in Fig. 32. Of course, the conversion loss is in-
creased by the loss of the input power splitter, which is of 3–4 dB. For
the same reason, the required LO power is increased by 3–4 dB. The I-Q
mixer can be reversed, operating as a modulator, as the simple mixer did
February 2, 2008 E. Rubiola Tutorial on mixers 45

9
0
°

9
0
° 0
° 0
°
0
°
0
°

180°
−90°
−90°
0
°
0
°
LO
LO
RF
IF RF
IF
LSB
LO
input
R
0
RF out
USB
V
S
cos b
V
S
>

2V
L
V
L
cos b
V
L
sin b

1
2

2U
V
I
V
L
sin(b −a)
+
1
2

2U
V
I
V
L
sin(b + a)
1
2

2U
V
I
V
L
sin(b −a)
+
1
2

2U
V
I
V
L
sin(b + a)
1
2U
V
I
V
L
sin(b + a)

1
2U
V
I
V
L
sin(b −a)
input
R
0
IF
V
I
cos a
1

2
V
I
cos a
1

2
V
I
sin a
Figure 30: SSB modulator.
1
8
0
° 0
°
0
°
0
°
0
° 0
°

9
0
°

9
0
°
−90°
−90°
0
°
0
°

0
°

1
8
0
°
LO
LO
IF
IF
RF
RF
LO
LO
IF
IF
RF
RF
1

2U
VL [xcos a + y sin a]
R
0
R
0
Type 1 modulator
y
Q input
I input
x
R
0 input
LO
VS cos a
VS >

2VL
VL cos a
VL cos a
Type 2 modulator
y
I input
Q input
x
R
0 input
LO
VS cos a
VS >

2VL
VL cos a
VL sin a
out
RF
out
RF
VLxcos a
VLy sin a
VLxcos a
VLy cos a
1

2U
VL [xcos a + y sin a]
Figure 31: I-Q modulators.
(Sec. 6.1). A number of I-Q modulators are available off the shelf, shown in
Fig. 31. Other configurations of I-Q detector/modulator are possible, with
similar characteristics.
The Type-2 detector seems to work better than the Type-1 because the
180

junction exhibit higher symmetry and lower loss than the 90

junction.
Some power loss and asymmetry is more tolerated at the LO port, which is
saturated. Figure 33 gives an idea of actual loss asymmetry. In addition,
there can be a phase error, that is a deviation from quadrature, of a few
degrees.
46 E. Rubiola Tutorial on mixers February 2, 2008
1
8
0
° 0
°
0
°
0
°
−90°
−90°
0
°
0
°
0
° 0
°

9
0
°

9
0
°


0
°
1
8
0
°
LO
LO
IF
IF RF
RF
LO
LO
IF
IF
RF
RF
VS cos a
R
0
xcos a −y sin a
input
RF
R
0
xcos a −y sin a
input
RF
+
1
2

2U
VL x
+ 2a terms
+
1
2

2U
VL y
+ 2a terms
+
1
2

2U
VL x
+ 2a terms

1
2

2U
VL y
+ 2a terms
R
0
I out
Q out
input
LO
VL cos a
VL cos a
1

2
[xcos a −y sin a]
1

2
[xsin a + y cos a]
Type 1 detector
VS >

2VL
VS cos a
Type 2 detector
R
0
Q out
I out
input
LO
VL cos a
VL sin a
1

2
[xcos a −y sin a]
1

2
[xcos a −y sin a]
VS >

2VL
VS cos a
Figure 32: I-Q detectors.
14
8.2
6.0
6.2
6.4
6.6
6.8 D
S
B

l
o
s
s
,

d
B
LO power, dBm
7.0
7.2
7.4
7.6
7.8
8.0
3 4 5 6 7 8 9 10 11 12 13 15
ER−67 I−Q detector Vol.6 p.9, Mar 2002
Figure 33: DSB loss of a home-made VHF I-Q detector, based on Mini
Circuits mixers and power splitters.
Finally, it is worth pointing out that the phase relationships shown in
Figures 32–31 result from a technical choice, for they should not be given
February 2, 2008 E. Rubiola Tutorial on mixers 47
F
F
T


a
n
a
l
y
z
e
r
d
u
a
l

c
h
a
n
n
e
l
Q
I
LO
RF
d
e
t
e
c
t
o
r
I

Q
ω
b
= |ω
s
−ω
l
|
reference
common
synthesizer
synthesizer
ω
l
P = P
S
ω
s
P ≪P
S
(ref)
(signal)
(linear)
(saturated)
Figure 34: Understanding the phase relationships inside an I-Q detector.
1 dB compression point
primary response
two-tone
intermodulation
noise level
I
F
o
u
t
p
u
t
p
o
w
e
r
,
d
B
m
RF input power, dBm
third-order intercept
Figure 35: .
for granted. Letting the phase of the LO arbitrary, there are two possible
choices, Q leads I or Q lags I. The experimentalist may come across unclear
or ambiguous documentation, hence inspection is recommended. Figure 34
shows a possible method. The FFT analyzer is used to measure the phase
of the signal Q versus the reference signal I. I have some preference for
ω
s
> ω
l
, and for a beat note
1

ω
b
=
1


s
−ω
b
| of some 1–5 kHz. A phase-
meter, a vector voltmeter, or a lock-in amplifier can be used instead of the
dual-channel FFT analyzer.
8 Non-ideal behavior
Most of the issues discussed here resort to the general background on radio-
frequency and microwave background, for they are listed quickly only for
the sake of completeness. The book [Raz98] is a good reference.
Impedance matching. Inputs and output of the mixer only approximate
the nominal impedance, for reflection are present in the circuit. In
practice, the impedance mismatching depends on frequency and power.
48 E. Rubiola Tutorial on mixers February 2, 2008
Isolation and crosstalk. A fraction of the input power leaks to the out-
put, and to the other input as well. Often, isolating the LO port is
relevant because of power.
1 dB compression point. At high input power, of about 10 dB below the
LO power, the mixer starts saturating, hence the SSB loss increases.
The 1 dB compression power is defined as the compression power at
which the loss increases by 1 dB (Figure 14).
Non-linearity. The mixer behavior deviates from the ideal linear model
of Section 3.3, for the input-output relationship is of the form v
o
(v
i
) =
a
0
+ a
1
v
i
+ a
2
v
2
i
+ a
3
v
3
i
+ . . . [Eq. (52), here repeated]. In radio
engineering the cubic term, a
3
v
3
i
, is often the main concern. This is
due to the fact that, when two strong adjacent-channel signals are
present at ∆ω and 2∆ω off the received frequency ω
i
, a conversion
product falls exactly at ω
i
, which causes interference. Being ∆ω ≪ω
i
,
a preselector filter can not fix the problem.
Offset. In ‘synchronous detector’ mode, the output differs from the ex-
pected value by a dc offset, which depends on the LO power and of
frequency. The same problem is present in the in ‘phase detector’
mode, where also the RF power affects the offset. This occurs because
of saturation.
Internal phase shift. The presence of a small phase lag at each port
inside the mixer has no effect in most application. Of course, in the
case of I-Q devices the quadrature accuracy is relevant.
9 Mixer Noise
The mixer noise were studied since the early time of radars [TW48, Ber58].
Significantly lower noise was later obtained with the Schottky diode [Bar67,
Gew71], and afterwards with the double balanced mixer. More recent and
complete analysis of the mixer noise is available in [HK78a, HK78b, Ker79c,
Ker79a, Ker79b]. Nonetheless in the design electronics, and even in low-
noise electronics, the mixer noise is often a second-order issue because:
1. Nowadays mixers exhibit low noise figure, of the order of 1 dB.
2. The mixer is almost always preceded by an amplifier.
3. The mixer picks up noise from a number of frequency slots sometimes
difficult to predict.
Noise pick-ups from various frequency slots is probably the major practical
issue. The presence of the USB/LSB pair makes the image-rejection mixer
February 2, 2008 E. Rubiola Tutorial on mixers 49
(Fig. 29, p. 43) appealing. Two phenomena deserve attention. The first one
is the multi-harmonic frequency conversion (Fig. 7 p. 15 and Fig. 10 p. 18),
by which noise is converted to the IF band from the sidebands of frequencies
multiple of the LO frequency. The second phenomenon is a step in the
output noise spectrum at the LO frequency, in the presence of white noise
at the RF port (Fig. 36). Only a graphical proof is given here. The output
slots IF1, IF2, and IF3 are down-converted from the input slots RF3+RF4,
RF2+RF5, and RF1+RF6, respectively. Thus, the conversion power loss
is ℓ
2
/2. At higher frequencies, the output slots IF4, IF5, . . . , come from
RF7, RF8, . . . , for the loss is ℓ
2
. The analytical proof follows exactly the
graphical proof, after increasing to infinity the number of frequency slots so
that their width is dω.
RF8 RF7 RF6 RF5 RF4 RF3 RF2 RF1
N/ℓ
ω
N
2N/ℓ
Spectra
ω
0

0
LO
RF
ω
0
IF
ω
ω
. . . . . . . . .
. . . . . .

0
power loss ℓ
2
/2
ω
0
power loss ℓ
2

0

0

0

0
IF1 IF2 IF3 IF4 IF5 IF6 IF7 IF8
RF9
Figure 36: A step appears in the conversion of white noise.
Flicker (1/f) noise is generally not documented. All the references found
about the mixer noise are limited to classical white noise, that is, thermal
and shot noise, while the flicker noise is not considered. The flicker behavior
of mixer may depend on the operating mode, as listed in Table 2 (p. 7).
Yet, the general rule is that flicker noise is a near-dc phenomenon, powered
by the LO pump. Then, the near-dc flicker is up-converted by non-linearity
and brougt to the output; or available at the output, in the ‘synchronous
detector’ mode (Sec. 3.2) and in the ‘phase detector’ mode (Sec. 5.4), where
the dc signal is taken at the output.
10 Where to learn more
Our approach, which consists of identifying and analyzing the modes of
Table 2, is original. Thus, there are no specific references.
A lot can be learned from the data sheets of commercial mixers and from
50 E. Rubiola Tutorial on mixers February 2, 2008
the accompaining application notes. Unfortunately, learning in this way re-
quires patience because manufacturer tend to use their own notation, and
because of the commercial-oriented approach. Another problem is that the
analysis is often too simplified, which makes difficult to fit technical informa-
tion into theory. Watkins Johnson
12
application notes [Hena, Henb] provide
useful general description and invaluable understanding of intermodulation
[Che]. We also found useful the Anzac [Anzb, Anza], Macom [M/A] and
Mini-Circuits [Minb, Mina] application notes.
Reading books and book chapters on mixers, one may surprised by the
difference between standpoints. A book edited by E. L. Kollberg [Kol84]
collects a series of articles, most of which published in the IEEE Transac-
tions on Microwave Theory and Technology and other IEEE Journals. This
collection covers virtually all relevant topics. The non-specialist may be in-
terested at least in the first part, about basic mixer theory. The classical
book written by S. A. Maas [Maa93] is a must on the subject.
A few books about radio engineering contains a chapter on mixers. We
found useful chapter 3 (mixers) of McClaning & al. [MV00, pp. 261–344],
chapter 7 (Mixers) of Krauss & al. [KBR80, pp. 188–220], chapter 6 (Mixers)
of Rohde & al. [RWB96, pp. 277–318], and Chapter 7 (Microwave Mixer
Design), of Vendelin & al.[VPR90].
Some radio amateur handbooks provide experiment-oriented information
of great value, hard to find elsewere. Transmission-line transformers and
baluns are described in [Sev01]. Recent editions of the the ARRL Hand-
book [Str99] contain a chapter on mixers (chapter 15 in the 1999 edition),
written by D. Newkirk and R. Karlquist, full of practical information and
common sense.
References
[Anza] Anzac, Adams-Roussell Co., Inc, Burlington, MA, Biphase and
quadriphase digital modulators, In RF and Microwave Signal Pro-
cessing Components Handbook, 1990.
[Anzb] Anzac, Adams-Roussell Co., Inc, Burlington, MA, Double bal-
anced mixers, In RF and Microwave Signal Processing Compo-
nents Handbook, 1990.
[Bar67] Mark R. Barber, Noise figure and conversion loss of the schot-
tky barrier mixer diode, IEEE Trans. Microw. Theory Tech. 15
(1967), no. 11, 629–635.
[Ber58] Saul M Bergmann, One aspect of minimum noise figure microwave
mixer design, IRE Trans. Microw. Theory Tech. (1958), 324–326.
12
http://www.wj.com/technotes/
February 2, 2008 E. Rubiola Tutorial on mixers 51
[Che] Daniel Cheadle, Selecting mixers for best intermod performance,
part 1 and 2, Watkins Johnson Company, Palo Alto, CA, In RF
and Microwave Design Handbook, 1997–98.
[Gew71] J. W. Gewartowski, Noise figure for a mixer diode, IEEE Trans.
Microw. Theory Tech. 29 (1971), no. 5, 481.
[Hena] Bert C. Henderson, Mixers: Part 1. characteristics and perfor-
mance, Watkins Johnson Company, Palo Alto, CA, In RF and
Microwave Design Handbook, 1997–98.
[Henb] , Mixers: Part 2. theory and technology, Watkins Johnson
Company, Palo Alto, CA, In RF and Microwave Design Hand-
book, 1997–98.
[HK78a] Daniel N. Held and Anthony R. Kerr, Conversion loss and noise
of microwave and millimeter-wave mixers: Part 1—theory, IEEE
Trans. Microw. Theory Tech. 26 (1978), no. 2, 49–55.
[HK78b] , Conversion loss and noise of microwave and millimeter-
wave mixers: Part 2—experiment, IEEE Trans. Microw. Theory
Tech. 26 (1978), no. 2, 55–61.
[KBR80] Herbert L. Krauss, Charles W. Bostian, and Frederick H. Raab,
Solid state radio engineering, John Wiley & Sons, New York, 1980.
[Ker79a] Anthony R. Kerr, Noise and loss in balanced and subharmoni-
cally pumped mixers: Part i—theory, IEEE Trans. Microw. The-
ory Tech. 27 (1979), no. 12, 938–943.
[Ker79b] , Noise and loss in balanced and subharmonically pumped
mixers: Part ii—application, IEEE Trans. Microw. Theory Tech.
27 (1979), no. 12, 944–950.
[Ker79c] , Shot-noise in resistive-diode mixer and the attenuator
noise model, IEEE Trans. Microw. Theory Tech. 27 (1979), no. 2,
135–140.
[Kol84] Erik L. Kollberg (ed.), Microwave and millimeter-wave mixers,
IEEE, New York, 1984.
[M/A] M/A-Com, Ltd, Dunstable, UK, Mixers. application note m562,
In RF Microwave and Millimeter Wave Handbook, 1996.
[Maa93] S. A. Maas, Microwave mixers, Artech House, 1993.
[Mina] Mini-Circuits, Brooklyn, NY, Modern mixer terms defined, In
RF/IF Designer’s Handbook, 1997.
52 E. Rubiola Tutorial on mixers February 2, 2008
[Minb] Mini-Circuits, Brooklyn, NY, Understanding mixers, In RF/IF
Designer’s Handbook, 1997.
[MV00] Kevin McClaning and Tom Vito, Radio receiver design, Noble,
Atlanta, GA, 2000.
[OMK80] H. A. Ogawa, A. Masayoshi, and M. Kozo, K-band integrated
double-balanced mixer, IEEE Trans. Microw. Theory Tech. 28
(1980), no. 3, 180–185.
[Raz98] Behzad Razavi, RF microelectronics, Prentice Hall, Upper Saddle
River, NJ, 1998.
[ROG92] Enrico Rubiola, Marcel Olivier, and Jacques Groslambert, Phase
noise in the regenerative frequency dividers, IEEE Trans. Instrum.
Meas. 41 (1992), no. 3, 353–360.
[Rud76] Walter Rudin, Principles of mathematical analysis, McGraw Hill,
1976.
[RWB96] Ulrich L. Rohde, Jerry Whitaker, and T. T. N. Bucher, Com-
munications receivers: Principles and design, McGraw Hill, New
York, 1996.
[Sev01] Jerry Sevick, W2FMI, Transmission line transformers, Noble, At-
lanta, GA, 2001.
[Str99] Dean R Straw, N6BV (ed.), The ARRL handbook, American Ra-
dio Relay League, Newington CT, 1999, Published yearly.
[TW48] Henry C. Torrey and Charles A. Whitmer, Crystal rectifiers, Ra-
diation Laboratory Series, vol. 15, McGraw Hill, 1948.
[Vit66] Andrew J. Viterbi, Principles of coherent communication, Mc-
Graw Hill, New York, 1966.
[VPR90] George D. Vendelin, Anthony M. Pavio, and Ulrich L Rohde,
Microwave circuit design, John Wiley & Sons, New York, 1990.

2

E. Rubiola Tutorial on mixers

February 2, 2008

Most used symbols
A(t) hlp , hbp h, k, n, p, q i(t), I I (goes with Q) IF j ℓ LO P Pi , Po Pp , PS Q (goes with i) R R0 RG U v(t), V v ′ , v ′′ vi (t), vo (t) vp (t) vl (t), VL VO VS x(t) y(t) φ, φ(t) ϕ(t) ω, f ωi , ωl ωb , ωs note: ω is used slow-varying (baseband) amplitude impulse response of lowpass and bandpass filters integer numbers current in-phase in/out (of a two-phase mixer/modulator) intermediate frequency imaginary unit, j 2 = −1 mixer voltage loss, 1/ℓ2 = Pi /Po local oscillator power power, input and output power LO (pump) power and internal LO saturation power quadrature in/out (of a two-phase mixer/modulator) resistance characteristic resistance (by default, R0 = 50 Ω) source resistance (Th´venin or Norton model) e dimensional constant, U = 1 V voltage real and imaginary, or in-phase and quadrature part input (RF) voltage, and output (IF) voltage LO (pump) signal internal LO signal saturated output voltage satureted level of the internal LO signal vl (t) real (in-phase) part of a RF signal imaginary (quadrature) part of a RF signal static (or quasistatic) phase random phase angular frequency, frequency input (RF) and pump (LO) angular frequency beat and sideband angular frequency as a shorthand for 2πf

Most used subscripts b beat, as in |ωs − ωi | = ωb i, I input l, L local oscillator (internal signal) o, O output p, P pump, local oscillator (at the input port) s sideband, as in |ωs − ωi | = ωb S saturated note: in reverse modes, i is still the input, and o the output

February 2, 2008

E. Rubiola Tutorial on mixers

3

Contents
Most used symbols 1 Basics 2 Signal representations 3 Linear modes 4 Mixer loss 5 Saturated Modes 6 Reversed Modes 7 Special Mixers and I-Q Mixers 8 Non-ideal behavior 9 Mixer Noise 10 Where to learn more References 2 5 8 10 21 26 36 40 47 48 49 50

4

E. Rubiola Tutorial on mixers

February 2, 2008

February 2, 2008

E. Rubiola Tutorial on mixers

5

1

Basics

It is first to be understood that the mixer is mainly intended, and mainly documented, as the frequency converter of the superheterodyne receiver (Fig. 1). The port names, LO (local oscillator, or pump), RF (radio-frequency), and IF (intermediate frequency) are clearly inspired to this application.

preselector
RF LO IF

IF amplifier

detector

tuning knob

local oscillator

Figure 1: Superheterodyne receiver. The basic scheme of a mixer is shown in Fig. 2. At microwave frequencies a star configuration is often used, instead the diode ring. Under the basic
RF input IF out

LO input mixer LO source RG LO input vp (t)
D3 D4 D2 D1

RF input vi (t)

RF source RG

IF out

vo (t) IF load RL

Figure 2: Double balanced mixer and its switch-network equivalent. assumptions that vp (t) is large as compared to the diode threshold, and that vi (t) is small, the ring acts a switch. During the positive half-period of vp (t) two diodes are reverse biased and the other two diodes are forward biased to saturation. During the negative half-period the roles are interchanged. For the small RF signal, the diodes are open circuit when reverse biased, and small resistances when forward biased. As a result, the IF signal vo (t) switches between +vi (t) and −vi (t) depending on the sign of vp (t). This is equivalent to multiplying vi (t) by a square wave of amplitude ±1 that takes

2. whereas in some cases a power up to 1 W (30 dBm) is used for highest linearity. Yet. for microstrip networks are the preferred balun types. impedance plays another role. 3. Additionally. all ports should be reasonably impedance matched. First and foremost. Rubiola Tutorial on mixers February 2. the input can be intentionally saturated.1 Golden rules 1. The RF power should be at least 10 dB lower than the LO power. In fact. 1. the appropriate current flow is necessary for the diodes to switch. the mixer is (almost) always used with the LO input saturated at the nominal power. because of the low forward threshold and of the fast switching capability. and detailed in the next Sections. a balun is necessary in order to convert the unbalanced inputs into the balanced signals required for the ring to operate as a switch. The diodes are of the Schottky types. check upon saturation at the LO port and identify the operating mode (Table 2). In all cases. The difference is that the RF port is coupled in ac. When the input (RF) and LO frequency overlap.1 to 3. Then. When reflected waves can be tolerated. listed in Table 2. it is sufficient to describe the frequency conversion mechanism as the product between vi (t) and the first term of the Fourier expansion of the square wave. Frequency degeneracy. read carefully Sections 3. the mixer is sometimes used in a strange mode. The mixer can be used in a variety of modes. no adapter is needed at the IF output. up to some tens of GHz. More accurate models account for the higher-order Fourier terms. 2008 the sign from vp (t). each with its “personality” and peculiarities. The input (RF) power is usually well below the saturation level. the conversion product also overlap. as in Figures 1–2. In short summary. with rare exceptions. 4. Generally. Interchanging the RF and IF ports. the main parameters governing the behavior are: Input power. for example at low frequencies or because of some external circuit. otherwise reflected waves result in unpredictable behavior. . which is already unbalanced. transformers are not available. with both LO and RF inputs not saturated. In low-frequency mixers (from a few kHz to 2–3 GHz) the baluns are implemented with power iron tore transformers. At higher frequencies. Conversely.6 E. while the IF port is often coupled in dc. The typical LO power is of 5–10 mW (7–10 dBm). In most practical cases.3. and for the dynamic resistance and capacitance of the diodes. At the RF and LO sides. The characteristic impedance to which all ports should be terminated is R0 = 50 Ω.

Similar to the DC mode. Used the lockin amplifiers. Reverse Degenerated Converter. Reverse Linear Converter. Linear Modulator. Amplitude-modulation detector. q small integers νi = νl Pi <PS Pl <PS AD 1. Damage results from excessive power. Scarce information. . Information is located close to dc. driven with a narrowband signal at νi . in coherent receivers. avoid damage deserves a few words because the device can be pushed in a variety of non-standard operation modes. which increases the risk. 1. νl =pν0 Pi ≥ PS νi =qν0 p. Synchronous Detector. Digital Modulator. and in bridge noise measurements. Mainly used in frequency synthesis. Degenerated frequency Converter. Used at NIST for the measurement of AM noise. 2008 E. Reverse Saturated Converter. driven with a neardc input current Ii (t). and only used in some odd cases. Some cases of in frequency synthesis.2 Avoid damage However trivial. Rubiola Tutorial on mixers 7 Table 2: Operating modes of the double balanced mixer.February 2. RF and LO signals are to be in quadrature. Only used in some cases of metrology and frequency synthesis. Typical of the superetherodyne radio receiver. mode LC Normal Modes SD SC DC PD LM RLC DM RSC RDC Strange condition frequency P or I νi = νl νi = νl νi = νl Pi ≪ PS Pi ≪ PS Pi ≥ PS note Linear frequency Converter. Some confusion between maximum power for linear operation and the absolute maximum power to prevent damage is common in data sheets. q small integers νi = νl νi ≈ 0 νi ≫ 0 νi ≈ 0 νi ≫ 0 Pi ≥ PS Ii ≪ IS Pi ≪ PS Pi ≥ PS Pi ≥ PS Reverse Modes νl =pν0 Pi ≥ PS νi =qν0 p. Phase Detector. Saturated frequency Converter.

the absolute maximum power specified for the LO port can be used as the total dissipated power. while the absolute maximum power can not. with V ′ = Vrms cos φ V = Vrms sin φ Vrms = (V ′ )2 + (V ′′ )2 ′′ ′ ′′ (4) (5) (6) (7) φ = arctan(V /V ) . Further analysis showed that one rail of a dc supply failed. This may be useful in dc or degenerated modes. 6. . while the absolute maximum power can not.8 E. Rubiola Tutorial on mixers February 2. This value can be exceeded. and because of this the bipolar operational amplifier (LT-1028) connected to the IF port sank a current from the input (20 mA?). 4. without being overloaded with microwave power. An alternate form often encountered is √ (2) v(t) = Vrms 2 cos(ω0 t + φ) √ √ (3) = V ′ 2 cos(ω0 t) − V ′′ 2 sin(ω0 t) . a misfortunate case occurred to me suggests to be careful about subtle details. 5. 3. Voltage driving may result in the destruction of the mixer for two reasons. The nominal LO power (or range) refers to best performance in the linear conversion mode. When linearity is not needed this value can be exceeded. A $ 3000 mixer used as a phase detector died unexpectedly. where power is not equally split between the diodes. for the current tend to exceed the maximum when the diode is driven by a voltage source. 2 Signal representations The simple sinusoidal signal takes the form v(t) = A0 cos(ω0 t + φ) . Better than general rules. The thin wires of the miniature transformers tend to blow up as a fuse if the current is excessive. The diode i = i(v) characteristics is exponential in v. In the absence of more detailed information. (1) √ This signal has rms value A0 / 2 and phase φ. 2008 2. The maximum RF power is specified as the maximum power for linear operation. The absolute maximum LO power can also be used to guess the maximum current through one diode. regardless of where power enters.

Thus. while the product of two sinusoids is governed by cos(ωa t) cos(ωb t) = 1 1 cos ωa − ωb t + cos ωa + ωb t 2 2 1 1 sin(ωa t) cos(ωb t) = sin ωa − ωb t + sin ωa + ωb t 2 2 1 1 sin(ωa t) sin(ωb t) = cos ωa − ωb t − cos ωa + ωb t . and (11) are also suitable to represent (slow-varying) modulated signals. In fact. (8). 2008 E. and of argument φ = arctan V ′′ V′ (10) equal to the phase φ of the time-domain sinusoid. √ A is a peak amplitude. 2 2 (12) (13) (14) Thus. (3). 1 (15) This is also known as the complex representation. All the forms (1). and double energy at positive frequencies. (2). Another form frequently used is the analytic (complex) signal v(t) = V ejω0 t . A modulated signal can be represented2 as v(t) = A′ (t) cos(ω0 t) − A′′ (t) sin(ω0 t) . and the analytic signal (11) hides the down-conversion mechanism. The product of two signals can only be described in the time domain [Eq. √ 2 for The factor 2 is dropped. . Rubiola Tutorial on mixers 9 The form (2)-(3) relates to the phasor representation1 V = V ′ + jV ′′ = |V |ejφ . and by turning the amplitude into a complex quantity of modulus |V | = (V ′ )2 + (V ′′ )2 = Vrms (9) equal to the rms value of the time-domain sinusoid. 3). A′ (t) and A′′ (t) are the √ time-varying counterpart of V ′ 2 and V ′′ 2. (1). (3) has a sign “−” for consistency with Eq. This occurs because ejωa t ejωb t = ej(ωa +ωb )t . the product of two sinusoids yields the sum and the difference of the two input frequencies (Fig. (8). (8). or as a single δ(ω − ω0 ) in the case of the analytic signal. (3)]. A pure sinusoidal signal is represented as a pair of Dirac delta function δ(ω − ω0 ) and δ(ω + ω0 ) in the spectrum. (11) where the complex voltage V = V ′ + jV ′′ is consistent with Eq. The “sin ω0 t” term in Eq. or as the Fresnel vector representation. (8) which is obtained by freezing the ω0 oscillation. (2). The analytic signal has zero energy at negative frequencies.February 2. the phasor representation (8) is useless.

Negative frequencies are not shown. 2008 Figure 3: Frequency conversion. The spectrum of such product consists of two replicas of the modulated input. it is not necessary that A′ (t) and A′′ (t) are narrow-band.10 E. The time-depencence of A′ (t) and A′′ (t) spreads the power around ω0 . (1)] multiplied by a modulated signal [Eq. 3. The clipper at the LO input limits the signal to the saturation level VS . Equations (12)–(14) also apply to the product of modulated signals. The spectrum of the modulated signal is a copy of the two-side spectrum of A′ (t) and A′′ (t) translated to ±ω0 . Not knowing the real shape. 3 Linear modes • the LO port is saturated by a suitable sinusoidal signal. 5). Strictly. • a small (narrowband) signal is present at the RF input. which accounts for the carrier. Rubiola Tutorial on mixers LO ωa Spectra RF ωb IF ωa − ωb ωa + ωb ω ω ω February 2. 4). A′ (t) and A′′ (t) are the low-pass signals that contain information. 3. while the clipper at the RF port . They may include a dc term. like in the old AM and PM. the spectrum can be conventionally represented as a rectangle centered at the carrier frequency. with their time-dependent coefficients A′ (t) and A′′ (t). It is often convenient to describe the mixer as a system (Fig. we often encounter the product of a pure sinusoid [Eq.3 for more details about linearity. the bandwidth of the modulated signal (15) is twice the bandwidth of A′ (t) and A′′ (t). (15)]. Of course.1 Linear frequency converter (LC) mode The additional condition for the mixer to operate as a linear frequency converter is that the LO and the RF signals are separated in the frequency domain (Fig. For the mixer to operate in any of the linear modes. Thus. in which the behavior is modeled with functional blocks. Using mixers. translated to the frequency sum and to the frequency difference (IF signal Fig. 4). 4). which occupies the bandwidth of A′ and A′′ on each side of ±ω0 (Fig. it is necessary that The reader should refer to Sec.

rather than increasing VS . As an example. The IF output is often coupled in dc. . A simplified description of the mixer is obtained by approximating the . The effect of higher LO power is to shrink the fraction of period taken by the slanted edges. . Rubiola Tutorial on mixers 11 ωl Spectra RF ωi IF ωi − ωl LSB ωi + ωl USB ω ω ω Figure 4: Frequency domain representation of the linear converter mode. . The asymptotic expression of vl (t) for strong saturation is 4 vl (t) = VS π = −1 k−1 2 odd k≥1 1 cos(kωl t) k (16) 4 1 1 VS cos ωl t − cos 3ωl t + cos 5ωl t − . Table 3 gives the main characteristics of two typical mixers. + . Negative frequencies are not shown.February 2. . π 3 5 The filters account for the bandwidth limitations of the actual mixer. is idle because this port is not saturated. The value of VS is a characteristic parameter of the specific mixer. RF input vi (t) not saturated P ≪ PS LO input vp (t) saturated P ≥ PS clipper (idle) multiplier IF out vo (t) 50 Ω 50 Ω internal LO signal vl (t) = ±VS clipper (active) 50 Ω Figure 5: Model of the double balanced mixer operated as a linear converter. The overall effect is that the internal LO voltage vl (t) is approximately a trapezoidal waveform that switches between the saturated levels ±VS . 2008 LO E.

port LO HF-UHF mixer microwave mixer 1–500 MHz 8. may select the upper sideband (USB) or the lower sideband (LSB). If it is present. 2008 Table 3: Example of double balanced mixers.5 swr < 2 ssb loss 5.4–18 GHz 7 dBm ±1 dB 8–11 dBm swr < 1.5 dB max ssb loss 7.8 swr < 2 1–500 MHz 8. 5.5 swr < 2 dc – 500 MHz dc – 2 GHz 0 dBm max 0 dBm max swr < 1. (22) (23) . They may contain a dc term. VL Ai (t) cos (ωl − ωi )t − φi (t) + cos (ωl + ωi )t + φi (t) 2U (21) The trivial term U = 1 V is introduced for the result to have the physical dimension of voltage. An optional bandpass filter. Rubiola Tutorial on mixers February 2.12 E. The output signal is vo (t) = 1 vi (t) vl (t) (19) U 1 = Ai (t) cos ωi t + φi (t) VL cos(ωl t) (20) U 1 = . the output signal is vo (t) = 1 VL Ai (t) cos (ωl − ωi )t − φi (t) 2U 1 vo (t) = VL Ai (t) cos (ωl + ωi )t + φi (t) 2U LSB USB . (18) (17) where Ai (t) and φi (t) are the slow-varying signals in which information is coded. not shown in Fig.4–18 GHz 0 dBm max 0 dBm max swr < 1. The input signal takes the form vi (t) = Ai (t) cos [ωi t + φi (t)] .5 dB max all ports terminated to 50 Ω RF IF internal LO waveform vl (t) with the first term of its Fourier expansion vl (t) = VL cos(ωl t) .

thus contribute to vo (t). so the presence of some energy around the image frequency . 2008 vi (t) not saturated P ≪ PS E. Fig. one wants to detect one signal. the identification of the input signal by observing the output of a mixer followed by a band-pass filter (Fig. It is easily proved that there exist two input signals vL (t) = AL (t) cos (ωl − ωb )t + φL (t) LSB USB . In a typical case. 6 (middle) gives the explanation in terms of spectra. Rubiola Tutorial on mixers RF LO saturated IF vl (t) P = PS ωb vo (t) 13 LO ωl Spectra RF LSB USB ω ′ ′′ ωi = ωl − ωb ωi = ωl + ωb ω IF ωb ωb ωb ω LO ωl Spectra RF 1 2 3 4 etc. It is therefore impossible to ascribe a given vo (t) to vL (t) or to its image vU (t) if no a-priori information is given. The USB and the LSB are image of one another with respect to ωl . IF ωb ωb ωb ω/ω0 ω filter mask ω Figure 6: Image frequency in a conversion circuit. Image frequency. Let us now consider the inverse problem.February 2. that is. 6 top). the output is a band-pass signal vo (t) = Ao (t) cos ωb t + φo (t) . close to the filter center frequency. (24) centered at ωb . (25) (26) vU (t) = AU (t) cos (ωl + ωb )t + φU (t) that produce a signal that passes through the output filter. In most practical cases.

Accordingly. In the case of the superheterodyne receiver. centered around nω0 . . etc. Sampling mixers are designed for this specific operation. 2008 is a nuisance. n (27) which is a series of contiguous bandpass processes of bandwidth ω0 . the mixer can be intentionally used to convert some frequency slot through multiplication by one harmonic of the LO. 7). The output is vo (t) = 1 vl (t) vi (t) ∗ hbp (t) . Of course. the terms of vi (t) for which |nω0 − ωl | is in the pass-band of the filter contribute to the output signal vo (t).14 E. U (28) where “∗” is the convolution operator. thus vo (t) = 1 vi (t) vl (t) U 4 VS π k−1 2 (29) −1 k−1 2 1 = Ai (t) cos ωi t + φi (t) U = 1 4 VS Ai (t) 2U π −1 odd k≥1 1 cos(kωl t) k (30) odd k≥1 1 cos (kωl − ωi )t − φi (t) + k . Even worse. a signal at the image frequency interferes with the desired signal. thus 3ω0 . and spaced by ω0 . (16) is a sinusoid of frequency kωl that converts the portions of spectrum centered at |kωl + ωb | and |kωl − ωb | into ωb (Fig. there results ambiguity in the frequency at which the receivers is tuned. one term can be regarded as the signal to be detected. the LO port is well saturated. All the terms with k > 1. the input signal can be written as vi (t) = n ′′ A′ (t) cos(nω0 t) − An (t) sin(nω0 t) .. Rubiola Tutorial on mixers February 2. The convolution ∗ hbp (t) defines the pass-band filtering. More generally. and hbp (t) the impulse response of the bandpass IF filter. at the cost of lower conversion efficiency. 5ω0 . Yet their internal structure differs from that of the common double-balanced mixer. In usual conditions. Fig. Each term of Eq. and the other one as the image. A bandpass filter at the RF input is often necessary to stop unwanted signals. Hence it makes sense to account for several terms of the Fourier expansion (16) of the LO signal. (31) + cos (kωl + ωi )t + φi (t) With k = 1. 6 (bottom) shows the complete conversion process. as stray signals taken in because of distortion. The obvious cure is a preselector filter preceding the mixer input. Multi-harmonic conversion.

and in that the coefficient VL. Is the inverse saturation current. (16). Equation (32) differs from Eq. If the LO signal contains a (random) phase ϕ(t). thus VL.k decrease more rapidely than 1/k. . Rubiola Tutorial on mixers 15 3ωl stray ′ ω3 ′′ ω3 ωb ωb ′ ω5 5ωl stray ′′ ω5 ωb ωb ω ω IF ωb ω Figure 7: Multi-harmonic conversion.6 mV. (33) The first term of Eq. For a more accurate analysis. the exponential diode current can be . 2] a technical parameter of the junction. Looking at Eq. and VT = kT /q the thermal voltage at the junction temperature. In the presence of a sinusoidal pump signal. one should recall that frequency multiplication results in phase noise multiplication. At room temperature. the phase kϕ(t) is present in the k-th term. The diode forward current iF is governed by the exponential law vF iF = Is e ηVT − 1 (34) where VF is the forward voltage.1 = VL . This due to non-perfect saturation and to bandwidth limitation.k cos(kωl t + φk ) . In real mixers the Fourier series expansion of vl (t) can be written as vl (t) = odd k≥1 −1 k−1 2 VL. . The term “−1” is negligible in our case. it holds that VT = kT /q ≃ 25.k cos (kωl − ωi )t − φi (t) + + cos (kωl + ωi )t + φi (t) . (16) in the presence of the phase terms φk . 2008 Spectra LO ωl RF signal & image ′ ω1 ′′ ω1 ωb ωb E. the diode can no longer be modeled as a switch. In weak saturation conditions the coefficient VL.February 2. (32) for Eq.k decrease even faster. η ∈ [1 . (32) is equivalent to (17). (31) becomes vo (t) = 1 Ai (t) 2U −1 k−1 2 odd k≥1 VL.

for the external load can not provide the appropriate impedance. (36) holds in quasistatic conditions and does not account for a number of known effects. In the design of microwave . 2. The product of two sinusoids at frequency ωi and ωl . (36) provides insight in the nature of the coefficients VL. otherwise the diodes can not switch. Ogawa [OMK80] gives an expression of the IF output current io (t) = 4Is Vrf ηVT Ik odd k≥1 Vlo ηVT cos(kωl + ωi )t + cos(kωl − ωi )t . inherently. like Is and η are hardly available. 2008 expanded using the identity ∞ e z cos φ = I0 (z) + 2 k=1 Ik (z) cos(kφ) . Rules for the load impedance at the IF port. varactor effect in diodes. not at dc. the center point of the LO transformer is grounded. contains the frequencies ωi ± ωl . At the IF port. the even harmonics are canceled and the odd harmonics reinforced. As a consequence of the mixer symmetry. The scheme C is a patched version of B. in which an additional RC cell provides the current path for the image frequency. Eq. The scheme B will not work because the filter is nearly open circuit at the image frequency. This may be necessary when the image frequency is out of the bandwidth. The scheme A is correct because the image-frequency current can flow through the diodes (low impedance). Can the LO and RF ports be interchanged? With an ideal mixer yes.k . Rubiola Tutorial on mixers February 2. and to reject the image at the frequency |ωi + ωl |. low impedance Z ≪ R0 is usually allowed. current flow must be allowed at both these frequencies. having been fixed inside the mixer. in practice often better not.16 E. the problem of providing a current path to the image frequency may not be visible. Figure 8 shows three typical cases in which a filter is used to select the |ωi − ωl | signal at the IF output. it is of limited usefulness in analysis because some parameters. In addition. Looking at Fig. (35) where Ik (·) is the modified Bessel function of order k. Nonetheless. bulk resistance of the semiconductors. like stray inductances and capacitances. The efficient use of a mixer as a multi-harmonic converter may require a specific analysis of the filter. Conversely. Rules are different in the case of the phase detector because the current path is necessary at the 2ωl frequency. and other losses. which helps isolation. Yet. In microwave mixers. (36) Equation (36) is valuable for design purposes. The problem arises when IF selection filter shows high impedance in the stop band. Eq.

and low loss in the RF circuit. Rubiola Tutorial on mixers A: correct 17 ωi RF IF Zi filter load |Zi | 50 Ω low Z (short) ω |ωi − ωl | B: incorrect |ωi + ωl | ωl LO 50 Ω ωi RF IF Zi filter load |Zi | high Z (open) ωl LO 50 Ω 50 Ω |ωi − ωl | |ωi + ωl | ω C: patched ωi RF IF Zi filter load |Zi | 50 Ω 50 Ω |ωi − ωl | |ωi + ωl | ω ωl LO Figure 8: The mixer is followed by a filter that selects the |ωi −ωl | frequency. and that a small (narrowband) signal is present at the RF input. This is implied in the general rule that the mixer is designed and documented for the superheterodyne receiver. 3. The additional conditions for the mixer to operate in the SD mode are: (1) the LO frequency ωl is tuned at the center of the spectrum of the (narrowband) RF signal. Nonetheless. and (2) the IF output is lowpassed.February 2. The basic mixer operation is the same of the frequency conversion mode.2 Linear Synchronous Detector (SD) Mode The general conditions for the linear modes are that the LO port is saturated by a suitable sinusoidal signal. optimization may privilege isolation from the LO pump. where the transformers are replaced with microstrip baluns. with the diode ring used as a switch that inverts or not the input polarity . for example to take benefit from the difference in the input bandwidth. interchanging RF and LO can be useful in some cases. mixers. 2008 E.

due to the harmonics multiple of the LO frequency still works (Figure 10). 5 is also suitable to the SD mode. The simplest way to understand the synchronous conversion is to represent the input and the internal LO signal vl′ (t) = VL cos(ω0 t + φL ) in Cartesian coordinates3 vi (t) = x(t) cos ω0 t − y(t) sin ω0 t vl′ (t) 3 (37) (38) = VL [cos φL cos ω0 t − sin φL sin ω0 t] In this Section we use x and y in order to emphasize some properties of the synchronous detection tightly connected to Cartesian-coordinate representation.18 Spectra LO E. Figure 9 shows the SD mode in the frequency domain. the conversion products of negative frequency are folded to positive frequencies. Rubiola Tutorial on mixers February 2. (15). 2008 RF −ω0 −ω0 ω0 ω0 1/ℓ2 1/ℓ2 power loss ℓ2 /2 power loss ℓ2 ω ω power loss ℓ2 IF −2ω0 2ω0 ω Figure 9: Frequency-domain sketch of the linear synchronous detection. The model of Fig. Of course. Here. x and y are the same thing of A′ and A′′ of Eq. dependig on the sign of the LO. Yet. making use of twosided spectra. . the multi-harmonic frequency conversion mechanism. the frequency conversion mechanism is slightly different. Spectra LO ωl RF ante unw d ante unw d 3ωl 5ωl ω ω IF ω Figure 10: Signals are converted to IF by the harmonics at frequency multiple than the LO frequency. Using one-sided spectra.

2U (43) 1 plus a trivial factor 2U VL that accounts for losses. In this conditions. 4 . The signal at the output of the low-pass filter is4 (Fig. we emphasize the properties connected with the Cartesian-coordinate representation. (42) Eq. X(t) = 1 VL x(t) cos φL + y(t) sin φL 2U . 2008 E. 2U thus. Rubiola Tutorial on mixers VL [x cos φL + y sin φL ] 2U 19 vi = x cos ω0 t − y sin ω0 t RF IF internal LO signal vl = VL cos(ω0 t + φL ) equivalent to vl = VL [cos φL cos ω0 t − sin φL sin ω0 t] X= Figure 11: Linear synchronous detection. 2U Once again. (42) can be interpreted as the scalar product X= 1 VL (x. Let us now replace the LO signal vl′ (t) with vl′′ (t) = −VL sin(ω0 t + φL ) = −VL [sin φL cos ω0 t − cos φL sin ω0 t] . y) · (cos φL . the output signal is Y (t) = (44) 1 (45) vi (t) vl′′ (t) ∗ hlp U 1 = x cos ω0 t − y sin ω0 t VL − sin φL cos ω0 t − cos φL sin ω0 t ∗ hlp U (46) 1 = (47) VL − x sin φL + y cos φL + (2ω terms) ∗ hlp . 11) X(t) = 1 (39) vi (t) vl′ (t) ∗ hlp U 1 x cos ω0 t − y sin ω0 t VL cos φL cos ω0 t − sin φL sin ω0 t ∗ hlp = U (40) 1 = (41) VL x cos φL + y sin φL + (2ω terms) ∗ hlp .February 2. X(t) is the same thing of vo (t) of other sections. sin φL ) .

The theory of coherent communication is analyzed in [Vit66]. are commercially available from numerous manufacturers. Devices like that of Fig. by joining Equations (42) and (48). For example. The coefficient vi = x cos ω0 t − y sin ω0 t X= RF IF 1 2U VL is implied. with x and y quantized in 8 level (3 bits) each. we find X(t) Y (t) = 1 VL 2U cos φL sin φL − sin φL cos φL x(t) y(t) . the well known wireless standard 811g (WiFi) is a 64 QAM. thus. Section 7 provide more details on these devices. 13. Y) X φL x Figure 12: Cartesian-frame rotation.20 E. known as I-Q detectors. y) (X . Rubiola Tutorial on mixers February 2. (49) Equation (49) is the common form of a frame rotation by the angle φL in Cartesian coordinates (Fig. 2008 y Y (x. 12). The simultaneous detection of the input signal with two mixers pumped in quadrature is common in telecommunications. (48) Finally. The transmitted signal is of the form (37). Y (t) = 1 VL − x(t) sin φL + y(t) cos φL 2U . 5 . where QAM modulations are widely used5 . VL [x cos φL + y sin φL ] 2U (in-phase) VL [−x sin φL + y cos φL ] 2U (quadrature) LO vl′ = VL cos(ω0 t + φL ) Y = RF IF LO vl′′ = −VL sin(ω0 t + φL ) pump 90◦ Figure 13: Basic I-Q detector.

When a sinusoidal signal of appropriate power and frequency is sent to the LO port. the double balanced mixer can be used in a wide range of frequency (up to 104 ). 2008 E. Another reason to keep the third-order term is the adjacent channel interference. that is. 14. In fact. and a minor concern with ferrite cores. In this condition. In large-signal conditions. the coefficient a2 is nonzero because of the residual asymmetry in the diodes and in the baluns. each of which treated as above. In practice. and neglecting the harmonic terms higher than the first.3 Linearity A function f (·) is said linear [Rud76] if it has the following two properties f (ax) = af (x) f (x + y) = f (x) + f (y) . The coefficients a2 and a3 are never given explicitely. Instead. (52) The symmetric topology cancels the even powers of vi . that is. (21)] vo (t) = 1 VL Ai (t) cos (ωl − ωi )t − φi (t) + cos (ωl + ωi )t + φi (t) 2U The linearity of vo (t) vs. for the above polynomial can not be truncated at the second order. (50) (51) The same definition applies to operators. this problem is absent in microwave mixers. The coefficient a1 is the invese loss ℓ. Yet. well below saturation. the power at which the nonlinear term (a2 vi and a3 vi ) is equal to the linear term. (33)]. the output signal vo (t) is a linear function of the input vi (t). This can be easily proved for the case of simple conversion [Eq. This is the case of a superheterodyne receiver in which the incoming signal is an unmodulated sinusoid vi (t) = Vi cos ωi t. where it is linear in a wide range of power. Rubiola Tutorial on mixers 21 3. the mixer output signal can be expanded as the polynomial 2 3 vo (vi ) = a0 + a1 vi + a2 vi + a3 vi + . 4 Mixer loss The conversion efficiency of the mixer is operationally defined via the twotone measurement shown in Fig. or by using the internal LO signal of real mixers [Eq. the output signal consists of a pair of sinusoids of . . the mixer is linear. the intercept power (IP2 and 2 3 IP3) is given. . In practice. transformer nonlinearity should also be analyzed. which may exceed 1016 (160 dB). The LO sinusoid is set to the nominal saturation power. (31)]. vi (t) can also be demonstrated in the case of the multi-harmonic conversion. either by taking a square wave as the LO internal signal [Eq. . In principle. the Fourier series is a linear superposition of sinusoids.February 2.

according to the sign of vl (t). and Po is the power of the IF output at the selected freqency. The LO power is entirely wasted in switching the diodes. the instantaneous output power is conserved 1 2 1 2 vi (t) = v (t) . 6 . vo (t) = ±vi (t). free from resistive dissipation. In our previous articles we took ℓ = Pi /Po instead of ℓ2 = Pi /Po . One of these sinusoids. the ring of Figure 2 works as a loss-free switch that inverts or not the polarity of the RF. The loss is about constant in a wide range of power and frequency. for only a fraction of the input power is converted into the desired frequency. The lowest loss refers to the ideal case of the zerothreshold diode. found with the definition (53). Under this assumptions. usually |ωl − ωi | is selected. The upper limit of the RF power range is the saturation power. specified as the compression power P1 dB at which the loss increases by 1 dB.22 E. R0 R0 o (54) Nonetheless. frequency ωo = |ωl ± ωi |. There result a loss inherent in the frequency conversion process. The SSB power loss ℓ2 of the mixer is defined6 as Po 1 = ℓ2 Pi SSB loss ℓ (53) where Pi is the power of the RF input. Intrinsic SSB loss. 2008 Po = Pi Po = Pl saturated IF |ωi − ωl | Pi Figure 14: Definition of the SSB loss ℓ. The practical use is unchanged because ℓ is always given in dB. Rubiola Tutorial on mixers ωl Pl ≃ PS saturated LO RF IF ωi Pi ≪ PS not saturated |ωi − ωl | Po LO Spectra Pl ωl RF Po loss ℓ2 Pi ωi (Po ) ω ω |ωi + ωl | linear ω loss ℓ2 Po log-log scale February 2. the mixer splits the input power into the conversion products at frequency |ωi ± ωl | and higher harmonics. Of course. The specifications of virtually all mixes resort to this definition.

Rubiola Tutorial on mixers 23 In the described conditions. Hence. (55) Only the first term of the above contributes to the down-converted signal at 4 the frequency ωb = |ωi − ωl |. .February 2. 15. The peak amplitude of this term is VL = π V. π 3 5 . (61) ≃ 1. at the cost of larger loss (2–3 dB more). A scheme is proposed in Fig. the internal LO signal is a unit square wave (VS = 1 V). 2008 E. the SSB loss can be of 4. it may be necessary to measure the conversion loss.5 dB [OMK80]. Some mixers. The difference is due to the microstrip baluns that match the nonlinear impedance of the diodes to the 50 Ω input over the device bandwidth. the LO saturation power is between 5 and 10 mW (7–10 dBm). 26). and 9 dB for 3-octave units. whose Fourier series expansion is vl (t) = 4 1 1 cos ωl t − cos 3ωl t + cos 5ωl t − . or of two diode rings (see Fig. . The loss of most HF/UHF mixers is of about 5–6 dB in a band up to three decades. When the frequencies multiple of the LO frequency are exploited to convert the input signal. Generally. optimized for best linearity make use of two or three diodes in series.57 (3. . The advantage of these mixers is high intercept power. This is due to the low loss and to the large bandwidth of the tranmission-line transformers. vo (t) = 1 vl (t) vi (t) U 4 = cos(ωl t) Vi cos(ωi t) π 4 1 = Vi cos[(ωi − ωl )t] + cos[(ωi + ωl )t] π 2 2 = Vi cos[ωb t] rubbing out the USB π (56) (57) (58) (59) The RF and IF power are Pi = Vi2 2R0 and Po = 1 4Vi2 2R0 π 2 (60) from which the minimum loss ℓ = ℓ= π 2 Pi /Po is minimum SSB loss. + . and need larger LO power (up to 1 W). . . The loss of microwave mixer is usually between 6 dB for the 1-octave devices. In the case of a narrow-band mixer optimized for conversion efficiency.92 dB) SSB loss of actual mixers.

.24 E. the output signal is vo (t) = 1 Vi VL cos ωi t + cos ωl t ∗ hbp (t) U 1 1 = Vi VL cos(ωi − ωl )t U 2 Po = when the input power is Pi = 7 (64) (65) The output power is 1 1 V 2V 2 2R0 4U 2 i L 1 V2 . and select the lower7 output frequency ωb = |ωi − ωl |. For the purpose of analytical calculus. (63) Measuring the output power. VL can be derived 2 to the power of the output product. Derivation of the internal LO voltage from the loss. Rubiola Tutorial on mixers synthesizer common reference February 2. Let vi (t) = Vi cos [ωi (t) + φi ] (62) the RF input. The internal LO signal is vl (t) = VL cos(ωl t + φl ) . we can drop the phases φ and φl . it holds that VL < π V. Hence. The by equating the output power Pi /ℓ usefulness of this approach is in that ℓ is always specified. the amplitude VL of the internal LO signal is often 4 needed. 2R0 i (66) (67) Some experimental advantages arise from taking ωb = |ωi −ωl | instead of ωb = |ωi +ωl |. With real (lossy) mixers. 2008 RF ωi P ≪ PS not saturated LO ωl P = PS saturated IF ωb = ωi − kωl odd k power meter synthesizer spectrum analyzer Spectra LO ωl RF filter bandwidth IF ωb power loss ℓ2 3 ωb ωi ω 3ωl 5ωl ω ω Figure 15: Measurement of the mixer loss in harmonic coversion.

we obtain 1 2 2 1 1 1 2 Ai = A V . (68) 2 2R ℓ 2R0 4U 2 i L 0 hence 2U Internal LO peak amplitude. while the intrinsic loss ℓ = π/2 yields VL = 4/π ≃ 1. In a narrow power range. dBm 0 Mixer MCL TFM−10514 −10 Parameter: LO input power −20 +10 dBm +7 dBm −30 +4 dBm +1 dBm −40 −2 dBm −50 −5 dBm −60 −8 dBm −11 dBm −14 dBm −17 dBm −20 dBm −70 −80 −90 −50 −40 −30 −20 −10 0 +10 RF input power. This fact is often referred as power desensitization (also LO desensitization. (69) VL = ℓ Interestingly. 12). the LO power has little or no effect on the output signal. the loss of most mixers is close to 6 dB. for VL ≃ 1 V. Combining the two above Equations with the definition of ℓ [Eq. dBm Figure 16: Conversion loss measured at various LO power levels (1990 p.27 V. or pump desensitization). Rubiola Tutorial on mixers 25 IF output power. What if the LO power differs from the nominal power? When the LO input is saturated. say ±2 dB . (66)]. 2008 E.February 2.

According to the model of Fig. the input power affects the duration of the wavefronts. which is more reproducible than the exponential law of the forward current. As a side effect of loss. The mixer can no longer be described as a simple switch that invert or not the RF signal. and in turn the cancellation of even hamonics. Rubiola Tutorial on mixers February 2. the LO clipper limits the internal voltage to ±VS . 5 Saturated Modes When both RF and LO inputs are saturated. a circuit may be sensitive to the LO power if stray input signals are not filtered out properly. which turns the input sinusoid into a trapezoidal waveform. The internal Schottky diodes exhibit exponential i = i(v) characteristics. and drops abruptly some 10 dB below the nominal LO power. Strong oddorder harmonics of the two input frequencies are present. Finally. the conversion loss changes slightly. Special care is recommanded with high-level mixers. white noise increases. Hence. Insufficient LO power may also impair symmetry. at each instant the largest signal controls the switch.26 E. 2008 from the nominal power. and that up conversion of the near-dc flickering of the junction takes place during this transition time . the mixer behavior changes radically. Figure 16 shows an example of output power as a function of the RF power. where amplitudes are constant. the roles are interchanged continuously. and noise also varies. flicker noise increases. depending on the LO sign. phase noise . Increasing the fraction of time in which the exponential law dominates emphasizes the asymmetry of the diodes. where the small size limits the heat evacuation. changing the LO power affects the dc voltage at the IF output. The conversion efficiency 1/ℓ is reduced. Saturation means that amplitude has little effect on the output. As a result. This can be a serious problem when the mixer is used as a synchronous converter or as a phase detector. Of course. in which the nominal LO power of of 50 mW or more. Too high LO power may increase noise. Whereas this phenomenon is still unclear. In the case of saturated modes. and in turns the harmonic contents. which is inherent in harmonic generation. and in the miniaturized mixers. Below the nominal LO power. 5. Instead. and the the ring is unable to switch. for saturated modes are useful in phase detectors or in frequency synthesis. The physical explanation is that saturated current is limited by the diode bulk resistance. and damage the mixer. while even-order harmonics are attenuated or cancelled by symmetry. A further consequence of saturation is phase noise multiplication. for various LO power levels. hence lower LO power is not sufficient to saturate the diodes. we guess that this is due to the increased fraction of period in which the diodes are neither open circuit or saturated. and sets the polarity of the other one.

still remaining in a safe operating range until the “absolute maximum ratings” are approached. and the ratio ωl /ωi is not too close to the the ratio of two small integers (say. and the cancellation of even harmonics. the actual power is hardly higher than 1. The model of Fig. Yet. and only partially in′ ′′ fluenced by VP and VP . A bandpass filter selects the upper or the lower frequency of (72). Rubiola Tutorial on mixers 27 multiplication takes place in both LO and RF. the model fails in predicting amplitude because the ring is no longer a multiplier. In saturated modes the specified maximum power at the RF port is always exceeded. the mixer leaves the “normal” linear operation. From Eq. 2 ′ ′′ Example 1 Replacing VP and VP with VL yields VO = 1 U VL . Let the input signals ′ vi (t) = VP cos ωi t (70) . The output amplitude is lower than expected. 5 describes some characteristics. Yet. The main output signal consists of the pair of sinusoids vo (t) = VO cos(ωl − ωi )t + VO cos(ωl + ωi )t (72) that derives from the product vi (t) vl (t). the saturated amplitudes VP and VP should be equal. When this maximum power is exceeded. • the input frequencies are not equal. (71) vp (t) = ′′ VP cos ωl t ′ ′′ If possible. 5. as it emphasizes the internally clipped waveforms. The unsuitability of the model of Fig.February 2. Let us 2 consider typical mixer that has a loss of 6 dB when the LO has the nominal power of 5 mW (7 dBm).1 Saturated Frequency Converter (SC) Mode The conditions for the mixer to operate in SC mode are • the LO and the RF ports are saturated by sinusoidal signals. (69) we get VL ≃ 1 V.25 mW (+1 dBm).5 mW (+4 dBm) with R0 = 50 Ω. Read page 7. 5 to predict amplitude can be seen in the following example. 5–7). • the output is band-passed. . thus we expect 2 VO = 250 mV. and an output power VO /2R0 = 2. 2008 E. Yet. the output amplitude VO is chiefly due to the internal structure of the mixer.

Another reason is that input phase fluctuations are multiplied by h and k.e. Rubiola Tutorial on mixers February 2. and drops abruptly outside the bandwidth. thus ωl = pω0 and ωi = qω0 integer p>0. One reason is that the positive and negative subscripts of Vhk make the spectrum measurements unambiguously identifiable. The contition on the ratio ωl /ωi two output frequencies ωh′ k′ and ωh′′ k′′ do not degenerate in a single spectral line. 5. This problem is explained in Section 5. and wrong results may be obtained discarding the sign. We recommend to keep the sign of h and k. at least for small h and k. i. the output signal is vo (t) = odd h. with positive h and k. 5–7 max.2 Degenerated Frequency Converter (DC) Mode The conditions for the mixer to operate in DC mode are the following • the LO and the RF ports are saturated by sinusoidal signals. q>0.. hωl +kωi > 0. • the output is band-passed.2. 2008 ωl RF ωi IF 3ωl 5ωl ω 3ωi ωo = hωl + kωi 3ωi ω ω −1 3 1 −3 5 3 −3 −1 1 1 −1 −3 3 5 5 −1 3 1 1 3 −1 5 5 1 3 3 1 5 h k Figure 17: Frequency conversion with a saturated mixer.k Vhk cos(hωl + kωi )t positive frequencies ωhk = hωl + kωi > 0 . p=q . • the input frequencies are not equal. Other authors write the output frequencies as |±hωl ±kωi |. Figure 17 shows an example of spectra involving harmonics. (73) where the sum is extended to the positive output frequencies. Vhk decreases more rapidely than the product |hk|. When ωl and ωi are multiple of a common frequency ω0 . (74) . Accounting for the harmoncs. and the ratio ωl /ωi is equal or close to the the ratio of two small integers (say.).28 Spectra LO E.

thus even-order harmonics. Fortunately. and the few cases of interest can be anlyzed individually. 2008 E. The generic output term of frequency nω0 derives from the vector sum of all the terms for which hp + kq = n . frequency (hp + kq)ω0 . For this reason. q = 2. • for a given ωl ωi pair. Thus f0 = 5 MHz. there is no point in devlopping a sophisticated theory. (75) thus vn (t) = h. integer n. the vn (t) interact with one another. several output frequencies nω0 exist. and we select the output frequency fo = 5 MHz with an appropriate bnd-pas filter. The combined effect of saturation and symmetry produces strong odd-order harmonics hωl and kωi ωl : 3ωl : vl1 = V1 cos(pω0 t + φl ) vl3 = V3 cos(3pω0 t + 3φl ) ωi : 3ωi : vi1 = V1 cos(qω0 t + φi ) vi3 = V3 cos(3qω0 t + 3φi ) ··· ··· ··· hωl : vlh = Vh cos(hpω0 t + hφl ) ··· ··· ··· ··· ··· ··· kωi : vik = Vk cos(kqω0 t + kφi ) ··· ··· ··· inside the mixer.k pair : hp+kq=n Vhk cos(nω0 t + hφl + kφi ) (76) Reality is even more complex than (76) because • some asymmetry is always present. and groups of terms collapse into fewer terms of frequency nω0 . all the cross products appear. p = 1. with amplitude Vhk . The output signal (76) results from the following terms hfl + kfi = nf0 −1×5 + 1×10 = 5 +3×5 − 1×10 = 5 −5×5 + 3×10 = 5 +7×5 − 3×10 = 5 −9×5 + 5×10 = 5 ··· hp + kq = n −1×1+1×2 = 1 +3×1−1×2 = 1 −5×1+3×2 = 1 +7×1−1×2 = 1 −9×1+5×2 = 1 ··· vn (t) V−1 1 cos(ω0 t−φl +φi ) V3 −1 cos(ω0 t+3φl −φi ) V−5 3 cos(ω0 t−5φl +3φi ) V7 −3 cos(ω0 t+7φl −3φi ) V−9 5 cos(ω0 t+7φl −3φi ) ··· . each one described by (76). Due to nonlinearity. • each term of (76) may contain an additional constant phase φhk . The following example is representative of the reality. and phase hφl + kφi . Example 2 The input frequencies are fl = 5 MHz and fi = 10 MHz. Rubiola Tutorial on mixers 29 the sum (73) degenerates.February 2. therefore the sum (76) can be accurately estimated from a small number of terms. while almost all the difficulty resides in parameter measurement. After time-domain multiplication. the amplitudes Vhk decrease rapidly with |hk|. and n = 1.

V3 −1 = 0. a.k pair : hp+kq=n Both Vn and φn are function of φl and φi . as in V = V ejφ .u.u. 2008 ℜ V−1 1 Vn −ϕl 2 output voltage. Let ϕ the fluctuation of the static phase In this section we use uppercase boldface for phase vectors. and phase gain as a function of the static phase φl . V−1 1 = 1 a. Bottom: output voltage.. Only V−1 1 and V3 −1 are taken into account.. (76) becomes Vn = Vhk . fi = 10 a. (−14 dB) φi = 0 1 phase gain 0 –1 –2 –3 –2 –1 0 Phi 1 2 phase φl 3 Figure 18: Simplified picture of degenerated frequency conversion.30 ℑ E.u.u. fl = 5 a. 8 . thus function of the phase relationship between the two inputs. Rubiola Tutorial on mixers ℑ V−1 1 −ϕl V3 −1 3ϕl Vn ℜ V3 −1 3ϕl February 2. with φi = 0. thus 1 √ Vn ejφn = 2 1 √ Vhk ejφhk 2 with φhk = hφl + kφi . 5.u. Top: phasor representation. V is the rms voltage. fn = 5 a.2 a. (77) h.3 Phase Amplification Mechanism Introducing the phasor (Fresnel vector) representation8 Eq.u.

Values of 10–50 kHz are useful in the HF/UHF bands. The output phase fluctuation is ϕn = ∂φn ∂φn ϕl + ϕi . (77)] as a function of the input phase.33 and 2. it is better to use two vector . the mixer must be measured in the same conditions (RF and LO power) of the final application. Parameter Measurement. (76) by means of a spectrum analyzer. One input signal is set at a frequency δ off the nominal frequency ωl (or ωi ). Figure 20 provides an example. and up to 1 MHz at higher frequencies. by means of a vector voltmeter. the three signals must be converted to the same frequency ω0 with approprate dividers. With V3 −1 /V−1 1 = 0. Of course. The second method consists of the direct measurement of Vn [Eq. The first method is the separate measurement of the coefficients Vhk of Eq. it is not very accurate because it hides the phase errors φhk that may be present in each term. from which the phase gain is derived. 2008 E. 1 mode was present. the vectors are opposite. This gives amplitude and phase. The experimentalist not aware of degeneracy may obtain disappointing results when low-order harmonics are present. Figure 18 shows a simplified example in which a 5 MHz signal is obtained by mixing a 5 MHz and a 10 MHz. while it would be −1 (constant) if only the −1. and all the terms of Eq. (76) are visible as separate frequencies. For φl = 0. Rubiola Tutorial on mixers 31 φ. and the amplitude is at its minimum. accounting only for two modes (10 − 5 and 3×5 − 10). On the other hand. For the measurement to be possible. The combined effect of V−1 1 and V3 −1 yields a large negative φn . A phase fluctuation is therefore attenuated. the phase gain ∂φn /∂φl spans from −0. The offset δ must be large enough to enable the accurate measurement of all the spectral lines with a spectrum analyzer. A small negative φn results from V−1 1 and V3 −1 pulling in opposite directions. and the amplitude is at its maximum. In this condition degeneracy is broken. For φl = π/4 ≃ 0. The phase gain/attenuation mechanism is a consequence of degeneracy.2 (−14 dB). While one vector voltmter is sufficient. 19). φl or φi .785.February 2. ∂φl ∂φi (78) where the derivatives are evaluated in the static working point. The deliberate exploitation of degeneracy to manage phase noise is one of the most exhotic uses of the mixer. The effect on phase noise was discovered studying the regenerative frequency dividers [ROG92]. There are two simple ways to measure the parameters of a degenerated frequency converter (Fig. but small enough not to affect the mixer operation. There follows that the input phase fluctuations φl and φi are amplified or attenuated (gain lower than one) by the mixer. as in the above example. the vectors are in phase. This method is simple and provides insight.

to be determined a posteriori. most frequency synthesizers can be adjusted in phase even if this feature is not explicitely provided. two frequency synthesizers and three frequency dividers. the dividers can not be replaced with commercial synthesizers because a synthesizer generally accepts only a small set of round input frequencies (5 MHz or 10 MHz). Rubiola Tutorial on mixers divider ÷p synthesizer common reference synthesizer ωl = qω0 P = PS adjustable saturated phase ωi = pω0 P = PS saturated divider ÷n divider ÷q February 2.32 E. The trick consists of misaligning the internal quartz oscillator when the instrument is locked to an external frequency reference. Figure 21 shows an example of direct measurement. The drawback of the direct measurement method is that it requires up to two vector voltmeters. compared to the calculated values. In some cases good results are obtained with resistive power splitters located close to the mixer because these splitters are not directional. based on the first method. 2008 vector voltmeter ω0 vector voltmeter nω0 synthesizer common reference synthesizer ωi = pω0 RF P = PS saturated LO ωl = qω0 + δ P = PS saturated IF nω0 spectrum analyzer Figure 19: Parameter masurement of a degenerated frequency converter. If the internal phase locked loop does not contain an integrator. about in quadrature.4 Phase Detector (PD) Mode The mixer works as a phase detector in the following conditions • the LO and the RF ports are saturated by sinusoidal signals of the same frequency ω0 . Interestingly. 5. In the general case. the misalignamet turns into a phase shift. voltmters because the measurement accounts for the reflected waves in the specific circuit. .

2 2 (79) from which one obtains a sinusoid of frequency 2ω0 . Rubiola Tutorial on mixers center 239. The output signal of an actual mixer is a 2 distorted sinusoid of frequency 2ω0 plus a dc term. (ii) the IF circuit is usually designed for frequencies lower than the RF and LO. dBm +10 0 −10 −20 −30 −40 −50 −60 −35 −45 −32 −40 −49 −38 −48 frequency kHz 0 250 11νL − 5νR 7νL − 3νR 9νL − 4νR −70 −250 5νR − 9νL 4νR − 7νL 3νR − 5νL 2νR − 3νL νR − νL νL 3νL − νR 5νL − 2νR Figure 20: Amplitude. the . and phase gain in a degenerated frequency converter. the maximum of |vo (t)| is about independent of ϕ. After low-pass filtering.4 MHz span 500 kHz res bw 1 kHz video filter 3 kHz +1 −8 33 IF ootput power. and a dc term − 1 sin φ 2 that is equal to − 1 φ for small φ.February 2. A better choice is to use a reversed mode. (80) V2 and V0 are experimental parameters that depend on the specific mixer and on power. hence V2 decreases as the absolute value of the dc term increases. When the PD mode is used close to the quadrature conditions. Using the 2ω0 output signal to double the input frequency is a poor choice because (i) the quadrature condition can only be obtained in a limited bandwidth. The product of such input signals is cos ω0 t + φ cos ω0 t − π 2 = 1 1 sin 2ω0 t + φ − sin φ . which can be approximated by vo (t) = V2 sin 2ω0 t + φ − V0 sin φ . • the output is low-passed. Due to saturation. 2008 E. the deviation of dc response from sin φ can be ignored. phase.

degrees Figure 21: Amplitude. and phase gain in a degenerated frequency converter. This condition occurs at some random—yet constant— phase a few degrees off the quadrature conditions. 2008 × calculated amplitude + measured amplitude φo . This involves setting the phase between the inputs to an appropriate value. It is often convenient to set the input phase for zero dc output. Due to diode asymmetry. to be determined experimentally. output signal is9 vo = −kφ φ + Vos . Rubiola Tutorial on mixers ⋄ calculated phase measured phase February 2. mV E. Exploiting the asymmetry of the entire v(i) law of the diodes.34 IF amplitude. therefore to make the mixer insensitive to amplitude modulation. which compensate for Vos . (80)]. the input power affects Vos . (81) where kφ is the phase-to-voltage gain [the same as V0 in Eq. phase. in a range where the mixer characteristics are virtually unaffected. . degrees phase gain ∇ calculated phase gain ∆ measured phase gain φi − φl . the major problem is that there are distinct 9 The phase-to-voltage gain is also written as kϕ (with the alternate shape of ϕ) because it is used with the small fluctuations ϕ. it is often possible to null the output response to the fluctuation of the input power. Figure 22 shows an example of phase detector charactaristics. and Vos is the dc offset that derives from asymmetry. In our experience. The IF output can be loaded to a high resistance in order to increase the gain kφ .

dBm Figure 22: Example of phase detector characteristics: output voltage as a function of ϕ (data are from a handbook Macom) and phase-to-voltage gain as a function of power (measured).February 2. degrees 500 400 Phase-to-voltage gain. dPl dvo .5 −1.0 0 45 90 135 360 Phase difference. V 50 Ω 0 −0. mV/rad 300 DB0218LW2 term. to 50 Ω 0 −15 −10 −5 0 +5 +10 Input power. Rubiola Tutorial on mixers 35 100 Ω 0.5 Output voltage. to 10 kΩ 200 100 M14A term. where the side effects of the offset are detrimental. 2008 1. dPi dvo . In some cases the nulls occurr within some 5◦ from the quadrarure. . in other cases farther. AM sensitivities dvo .0 1 kΩ E. d(Pl + Pi ) (82) and that nulling one of them is not beneficial to the other two.

The internal sat10 The mixer saturation current. When impedance-matching is not needed. 6 Reversed Modes The mixer can be reversed taking the IF port as the input and the RF port as the output (Fig. due to symmetry. should not be mistaken for the diode reverse saturation current. 24. Therefore. The major difference versus the normal modes is the coupling bandwidth: the output is now ac-coupled via the RF balun. of amplitude proportional to ii . 5) but for the input-output filters. the LO pump forces the diodes to switch.36 E. in phase with vp (t). The average resistance of D1 and D3 increases. . The LO signal makes the diodes switch. of phase opposite to vp (t). exactly as in the normal modes. • the IF input current is lower than the saturation current10 IS . The mixer can be represented as the system of Fig. 2008 RF out vo (t) RL RF load IF input RG IF source ii (t) Figure 23: Reversed-mode modulator. which can be of some mA. 23). the resistance of D2 and D4 averaged over the period decreases.1 Linear Modulator (LM) The mixer works as a LM in the following conditions • the LO port is saturated by a sinusoidal signal. a negative ii produces an output voltage proportional to ii . which is similar to the LC model (Fig. and the conduction angle of D2 and D4 increases. When a positive current ii is present. no signal is present at the RF output. The latter can be in the range from 10−15 A to 10−15 A. a small voltage vo (t) appears at the RF output. As usual. Rubiola Tutorial on mixers mixer LO source RG LO input vp (t) D3 D4 D2 D1 February 2. and their conduction angle decreases. while the input is in most cases dc-coupled. the IF input can be driven with a current source. 6. • a near-dc signal is present at the IF input. Similarly. At zero input current.

By virtue of Eq. 25. Similar results were obtained testing other mixers. VL can not be derived from the reverse loss. signal 37 IF input ii (t) not saturated ii ≪ IS LO input vp (t) saturated P ≥ PS multiplier RF out vo (t) 50 Ω idle clipper active clipper 50 Ω internal LO signal vl (t) = ±VS 50 Ω Figure 24: Reverse-mode model of a mixer. In such conditions the nominal SSB loss is ℓ = 2 (6 dB). [Eq. (17)]. After Eq. VL = 1 V. This is close to the measured value of 75 mV. when ii ≈ 12 mA dc. Beyond ii = 3 mA. 2008 E. U (83) (84) Example 3 The LO signal of a mixer (Mini-Circuits ZFM-2) is a sinusoid of frequency fl = 100 MHz and power P = 5 mW (7 dBm). the mixer lives gradually the linear behavior. (69). When the input current is ii = 2 mA dc. and saturates at some 230 mV rms of output signal. . or expanded as Eq. the mixer operates as a linear modulator described by vo (t) = 1 vi (t) vl (t) U 1 = vi (t) VL cos ωl t . (69) provide useful approximation of reverse behavior. which is not documented. measuring some mixers we found that the ‘conventional’ (forward) SSB loss ℓ and Eq. urated LO signal can be approximated with a sinusoid vl (t) = VL cos ωl t. we expect an output signal of 100 mV peak. Rubiola Tutorial on mixers dc or sinus. thus 71 mV rms.2 Reverse Linear Converter (RLC) The mixer works as a RLC in the following conditions • the LO port is saturated by a sinusoidal signal. (32).February 2. The latter is obtained fitting the the low-current experimental data of Fig. Strictly. (84). Thus. the input voltage is vi = R0 ii = 100 mV with R0 = 50 Ω. Nonetheless. Reciprocity should not given for granted. 6.

• an optional filter selects one of the beat products. I input RF output term. and the internal LO amplitude VL can be estimated using Eq. (69) and the ‘conventional’ SSB loss ℓ. mVrms 160 140 120 100 80 60 40 20 0 0 2 4 6 8 10 IF input current. not shown in Fig. USB . the output signal is vo (t) = 1 vi (t) vl (t) (85) U 1 = Ai (t) cos ωi t + φi (t) VL cos(ωl t) (86) U 1 . This mode is similar to the LM mode. the output signal is 1 VL Ai (t) cos (ωl − ωi )t − φi (t) 2U 1 vo (t) = VL Ai (t) cos (ωl + ωi )t + φi (t) 2U vo (t) = LSB. mA February 2. Input is driven with a current source. • LO and the IF separated in the frequency domain. which is not saturated. VL Ai (t) cos (ωl − ωi )t − φi (t) + cos (ωl + ωi )t + φi (t) = 2U (87) The model of Fig. Letting vi (t) = Ai (t) cos[ωi (t) + φi (t)] the input. to 50 Ω (vol. 2008 Mini Circuits ZFM-2. Rubiola Tutorial on mixers 200 180 RF voltage. • a small narrowband signal is present at the IF input.38 E. Output is terminated to 50 Ω. or (88) (89) . If an external bandpass filter. is present. 5 p. 24. 24 still holds. 52) 12 Figure 25: Gain of a mixer used ad a modulator.

Rubiola Tutorial on mixers 39 under the obvious condition that the signal bandwidth fits into the filter passband. no more than 5–7). 6. 2008 E. • the output is band-passed. Let vp = VP cos ωl t the LO input signal.). ii = ±Ii the IF input current. 6. • the input frequencies are equal.3 Digital Modulator (DM) Mode The mixer works as a DM in the following conditions • the LO port is saturated by a sinusoidal signal. The RSC mode is similar to the SC mode. • a large near-dc current is present at the IF input.February 2. • the input frequencies are not equal. 5-7 max. The only difference between SC and RSC is the input and output bandwidth. because IF and RF are interchanged. and VO the saturated output amplitude.4 Reverse Saturated Converter (RSC) Mode The mixer works in the RSC mode under the following conditions • the LO and the IF ports are saturated by sinusoidal signals. Equation (90) represents a BPSK (binary phase shift keying) modulation driven by the input current ii . for the explanations given in Section 5.5 Reverse Degenerated Converter (RDC) Mode The mixer works in the RDC mode when • the LO and the IF ports are saturated by sinusoidal signals. and the ratio ωl /ωi is not too close to the the ratio of two small integers (say. • the output is band-passed. 6. .1 also apply to the RSC mode. (90) where sgn(·) is the signum function. • the RF output is bandpassed. or the ratio ωl /ωi is equal or close to the the ratio of two small integers (say. which is saturated. The output signal is vo (t) = sgn(ii ) VO cos ωl t .

It is to be made clear that when two equal input frequencies (ωi = ωl = ω0 ) are sent to the input. 2008 The RDC mode is similar to the DC mode (Section 5. and n = 2. . Of course. p = 1. Some mixers are explicitely designed to operate in the phase detector mode.1) but for the trivial difference in the input and output bandwidth. the output preamplifier can hardly be noise-matched to an input resistance lower than a few hundreds Ohms. typically 500 Ω. Example 4 The input frequencies are fl = fi = 5 MHz. The fancy name “variable attenuator” is sometimes used. The output signal results from the vector addition of several beat signals. a large 2ω0 signal is always present at the RF output.1). the 11 I never come across a phase detector whose residual flicker is documented. The IF bandwidth reduction that results from the increased output impedance is not relevant in practice. q = 1. Thus f0 = 5 MHz. and we select the output fo = 10 MHz. In some cases such devices are actually generalpurpose mixers documented for phase detector operation. the reverse mode differs significantly from the normal mode. But in the reversed modes no dc output is permitted because the RF port is ac coupled. which is the most relevant parameter for a number of measurements. The residual flicker. The output signal [Eq. Yet. Rubiola Tutorial on mixers February 2. In fact. A mixer can be designed and documented to be used in a reverse mode as an analog modulator (See Sec. The main advantage of this higher impedance is a lower residual white noise of the system. Often the IF output impedance is larger than 50 Ω. resulting from the vector addition of several signals. this condition would turn the degenerated converter mode into the phase-detector mode. In the DC mode. Analog Modulator / Variable Attenuator. each one with its own phase and amplitude. is usually not documented11 . (76)] results from the follwoing terms hfl + kfi = nf0 +1×5 + 1×5 = 10 +3×5 − 1×5 = 10 −1×5 + 3×5 = 10 +5×5 − 3×5 = 10 −3×5 + 5×5 = 10 ··· hp + kq = n +1×1+1×1 = 2 +3×1−1×1 = 2 −1×1+3×1 = 2 +5×1−3×1 = 2 −3×1+5×1 = 2 ··· vn (t) V1 1 cos(ω0 t+φl +φi ) V3 −1 cos(ω0 t+3φl −φi ) V−1 3 cos(ω0 t−φl +3φi ) V5 −3 cos(ω0 t+5φl −3φi ) V−3 5 cos(ω0 t−3φl +5φi ) ··· 7 Special Mixers and I-Q Mixers Phase Detector. as the roles of IF and RF are interchanged. 6.40 E. which makes the RDC mode an efficient frequency doubler.

. used as a power splitter. Improved Impedance-Matching Mixers. as compared to the single-diode ones. need large LO power. In some cases low intermodulation performance must be achieved at any cost. 2008 E. The 90◦ hybrid junction. and show higher loss. High linearity is achieved by forcing the diodes to switch abruptly in the presence of a large pump signal. has the useful property that the input (output) is always impedance matched when the isolation port is correctly terminated and the two outputs (inputs) are loaded with equal impedances. The BPSK modulator differs from the analog modulator in that the IF input is saturated (See Sec.3). and the output signal changes sign when the input current is negative. Special mixers are used. 26 (classes I-III). High Linearity Mixers.February 2. the device may differ from a general-purpose mixer mostly in the documentation. This property is exploited joining two equal double-balanced mixers to form the improved mixer of Fig. Rubiola Tutorial on mixers Diode circuit LO power 7 to 13 dBm 41 Class I Class II Type 1 Class II Type 1 Class III Type 1 13 to 24 dBm 13 to 24 dBm 20 to 30 dBm Class III Type 2 20 to 30 dBm Class III Type 3 20 to 30 dBm Figure 26: Diode assemblies of high linearity mixers. These mixers. mixer operation is more general than that of a simple attenuator because the mixer input current can be either positive or negative. based on a ring in which the diodes are replaced with the more complex elements shown in Fig. up to 1 W. based on the same idea. BPSK Modulator. Other schemes are possible. 6. Once again. 27 (Class IV mixer).

at the IF output of the mixers are 1 v1 = √ VI VL sin a cos b 2U 1 v2 = √ VI VL cos a cos b . 29 divides the IF components. thus high dynamic range and low distortion. in which the LSB and the USB are converted into the same IF frequency ωb . Image-Rejection Mixer. The double-double-balanced mixer (Figure 28) shows high 1 dB compression point. Rubiola Tutorial on mixers Class IV −90° 1 √ VI 2 February 2. enabling the selection of the LSB or the USB input (RF) signal. Other schemes are possible. and high isolation.42 E. The scheme of Fig. RF input LO input IF out Figure 28: Double-double-balanced mixer. Double-Double-Balanced Mixers. The converted signals. Let us go back to the frequency conversion system of Fig. 6. 2008 RF input VI cos a sin a RF IF 1 √ V V sin(a + b) 2 2U I L 1 + 2√U VI VL sin(a − b) 0° IF out 1 2U VI VL sin(a + b) 1 + 2U VI VL sin(a − b) 0° 0° 1 √ VI 2 cos a RF IF 1 √ V V 2 2U I L −90° R0 VL sin b LO −90° −90° sin(a + b) 0° 180°0° 1 + 2√2U VI VL sin(a − b) VL cos b LO R0 0° 0° LO input R0 VS cos√ b VS > 2VL Figure 27: Improved impedance-matching mixer. This device is sometimes called triple balanced mixer because it is balanced at the three ports. Let us for short a = ωi t and b = ωl t the instantaneous phase of the RF and LO signal. 2U .

thus 1 v1 = √ VI VL sin(a − b) + sin(a + b) 2 2U 1 v2 = √ VI VL cos(a − b) + cos(a + b) 2 2U . Rubiola Tutorial on mixers IF out 43 −90° 1 √ VI 2 sin a RF IF v1 = √1 VI VL 2U sin a cos b −90° USB ′′ v1 ′ + v2 0° 0° 1 √ VI 2 cos a RF IF v2 = √1 VI VL 2U cos a cos b 0° 0° VL cos b LO LSB ′ v1 ′′ + v2 −90° −90° R0 VL cos b LO 180 0° ° 0° 0° R0 LO input VS cos√ b VS > 2VL ω a < ωb LO ωa RF ωb USB out rejected ωa − ωb LSB out ωb − ωa rejected ωa + ωb ω USB out LSB out rejected ωa − ωb ωa − ωb ω RF ω LO ω a > ωb ω ωa ω ωb ω ωa + ωb rejected ω ωa + ωb ω ωa + ωb Figure 29: Image-rejection mixer.February 2. The non- . 2008 RF input VI cos a E. and the phase of the negative-frequencies signal by +90◦ . The rotated signals are ′′ v1 = 1 4U VI VL 1 4U VI VL 1 4U VI VL 1 4U VI VL − cos(a − b) − cos(a + b) a>b a<b a>b a<b + cos(a − b) + cos(a + b) + sin(a − b) + sin(a + b) − sin(a − b) − sin(a + b) ′′ v2 = √ which also account for a factor 1/ 2 due to energy conservation. The path of the hybrid junction labeled ‘−90◦ ’ turns the phase of the positive-frequency signals by −90◦ .

Let us assume that the LO frequency ωl and the IF center frequency ωIF are given.4 MHz (101.7 = 122. which is taken in by the mixer.7 = 91). The output signals are ′′ ′ vUSB = v1 + v2 = 1 4U VI VL 0 sin(a − b) + sin(a + b) a>b (USB taken in) a<b (LSB rejected) (91) a>b (USB rejected) ′ ′′ vLSB = v1 + v2 = 0 1 4U VI VL cos(a − b) + cos(a + b) a<b (LSB taken in) (92) The unwanted sideband is never cancelled completely. the conversion loss is increased by the loss of the input power splitter. Example 5 The IF filter of a FM receiver has a bandwidth of 300 kHz centered at 10.44 E.7 + 10. In the best case. while the image-rejection mixer converts only one of these channels. 30) is a different arrangement of the same blocks used in the image-rejection mixer. The main purpose of this device is to modulate a carrier by adding only one sideband. the required LO power is increased by 3–4 dB. Of course. we tune the local oscillator to 101. All explanations are given on the scheme. I-Q Detectors and Modulators. the desired one (91 MHz) and the image frequency at 122. A mixer down-convert to IF two channels. 2008 rotated signals are 1 VI VL sin(a − b) + sin(a + b) 4U 1 ′ v2 = VI VL cos(a − b) + cos(a + b) 4U ′ v1 = . operating as a modulator. shown in Fig.7 − 10. 30.4).7 MHz (101. The I-Q mixer can be reversed. The SSB modulator (Fig.7 MHz. SSB Modulator. in Fig. Rubiola Tutorial on mixers February 2. Yet. The main reason to prefer the image-rejection mixer to a (simple) mixer is noise. which is of 3–4 dB.2 is commercially available in (at least) two practical implementations.4 MHz). as the simple mixer did . The mixer converts both |ωl −ωIF | and |ωl + ωIF | to ωIF . The two-axis synchronous detector introduced in Section 3. only noise is present at the image frequency (122. the noise of the electronic circuits is present at both frequencies. For the same reason. either LSB or USB. 32. A rejection of 20 dB is common in practice. yet not by the image-rejection mixer. In order to receive a channel at 91 MHz.

The Type-2 detector seems to work better than the Type-1 because the 180◦ junction exhibit higher symmetry and lower loss than the 90◦ junction. with similar characteristics.February 2. . Type 1 modulator I input x RF IF VL x cos a −90° 0° 0° VL cos a LO y R0 √1 VL 2U Q input RF IF VL y cos a RF out [x cos a + y sin a] −90° VL cos a LO 0° 0° 180 0° ° LO input Type 2 modulator I input x VS cos√ a VS > 2VL R0 RF IF VL x cos a 0° 0°180° VL cos a LO y R0 √1 VL 2U Q input RF IF VL y sin a RF out [x cos a + y sin a] 0° VL sin a LO −90° −90° 0° 0° LO input VS cos√ a VS > 2VL R0 Figure 31: I-Q modulators. Figure 33 gives an idea of actual loss asymmetry. Rubiola Tutorial on mixers 1 − 2√2U VI VL sin(b − a) 45 RF out −90° 1 √ VI 2 sin a IF RF 1 + 2√2U VI VL sin(b + a) 0° USB 1 2U VI VL 0° 0° 1 √ VI 2 cos a 0° 0° VL cos b LO IF RF 1 √ V V sin(b − a) 2 2U I L 1 + 2√2U VI VL sin(b + a) sin(b + a) LSB −90° 180° R0 VL sin b LO −90° −90° 1 − 2U VI VL sin(b − a) 0° 0° R0 LO input VS cos√ b VS > 2VL Figure 30: SSB modulator. 31. 2008 IF input VI cos a E. Other configurations of I-Q detector/modulator are possible.1). Some power loss and asymmetry is more tolerated at the LO port. A number of I-Q modulators are available off the shelf. which is saturated. 6. (Sec. shown in Fig. of a few degrees. In addition. that is a deviation from quadrature. there can be a phase error.

based on Mini Circuits mixers and power splitters. Rubiola Tutorial on mixers Type 1 detector −90° 1 √ [x cos a 2 February 2.8 8. it is worth pointing out that the phase relationships shown in Figures 32–31 result from a technical choice. Mar 2002 Figure 33: DSB loss of a home-made VHF I-Q detector. dB 6.6 7.2 7. 6.8 7.0 8.2 3 4 5 6 7 8 9 10 11 12 13 14 15 LO power. 2008 − y sin a] RF IF I out + √1 V x 2 2U L + 2a terms RF input x cos a − y sin a 0° 0° R0 VL cos a LO 1 √ [x sin a 2 + y cos a] RF IF Q out + √1 V y 2 2U L + 2a terms −90° VL cos a LO 0° 0° 180 0° ° LO input Type 2 detector 0° 1 √ [x cos a 2 VS cos√ a VS > 2VL R0 − y sin a] RF IF I out + √1 V x 2 2U L + 2a terms RF input x cos a − y sin a 180°0° R0 VL cos a LO 1 √ [x cos a 2 − y sin a] RF IF 0° Q out − √1 VL y 2 2U + 2a terms VL sin a LO −90° −90° 0° 0° LO input VS cos√ a VS > 2VL R0 Figure 32: I-Q detectors.6 6.4 7.6 p.4 6. for they should not be given .46 E.0 7.2 DSB loss. Finally. dBm Vol.0 ER−67 I−Q detector 6.9.

Inputs and output of the mixer only approximate the nominal impedance. dBm third-order intercept 1 dB compression point primary response two-tone intermodulation noise level RF input power. Impedance matching. there are two possible choices.February 2. the impedance mismatching depends on frequency and power. and for a beat note 2π ωb = 2π |ωs − ωb | of some 1–5 kHz. . IF output power. hence inspection is recommended. or a lock-in amplifier can be used instead of the dual-channel FFT analyzer. for reflection are present in the circuit. dBm Figure 35: . Q leads I or Q lags I. I have some preference for 1 1 ωs > ωl . The FFT analyzer is used to measure the phase of the signal Q versus the reference signal I. Rubiola Tutorial on mixers dual channel FFT analyzer I 47 (ref) common reference synthesizer ωs P ≪ PS (linear) ωl P = PS (saturated) RF I−Q detector synthesizer Q ωb = |ωs − ωl | (signal) LO Figure 34: Understanding the phase relationships inside an I-Q detector. a vector voltmeter. A phasemeter. The book [Raz98] is a good reference. The experimentalist may come across unclear or ambiguous documentation. Figure 34 shows a possible method. for granted. 2008 E. for they are listed quickly only for the sake of completeness. Letting the phase of the LO arbitrary. 8 Non-ideal behavior Most of the issues discussed here resort to the general background on radiofrequency and microwave background. In practice.

The mixer picks up noise from a number of frequency slots sometimes difficult to predict. Offset. [Eq. is often the main concern. This is due to the fact that. of about 10 dB below the LO power. when two strong adjacent-channel signals are present at ∆ω and 2∆ω off the received frequency ωi . and afterwards with the double balanced mixer. In ‘synchronous detector’ mode. The mixer behavior deviates from the ideal linear model of Section 3. The same problem is present in the in ‘phase detector’ mode. Significantly lower noise was later obtained with the Schottky diode [Bar67. In radio 3 engineering the cubic term. and even in lownoise electronics. Nonetheless in the design electronics. Of course. Ker79a.48 E. in the case of I-Q devices the quadrature accuracy is relevant. which causes interference.3. the output differs from the expected value by a dc offset. of the order of 1 dB. Ker79c. At high input power. where also the RF power affects the offset. . This occurs because of saturation. which depends on the LO power and of frequency. Noise pick-ups from various frequency slots is probably the major practical issue. More recent and complete analysis of the mixer noise is available in [HK78a. 2008 Isolation and crosstalk. The presence of the USB/LSB pair makes the image-rejection mixer . HK78b. . for the input-output relationship is of the form vo (vi ) = 2 3 a0 + a1 vi + a2 vi + a3 vi + . (52). Internal phase shift. hence the SSB loss increases. Rubiola Tutorial on mixers February 2. The 1 dB compression power is defined as the compression power at which the loss increases by 1 dB (Figure 14). Ker79b]. The mixer is almost always preceded by an amplifier. Gew71]. Nowadays mixers exhibit low noise figure. Ber58]. and to the other input as well. a preselector filter can not fix the problem. the mixer starts saturating. The presence of a small phase lag at each port inside the mixer has no effect in most application. Non-linearity. 2. 9 Mixer Noise The mixer noise were studied since the early time of radars [TW48. isolating the LO port is relevant because of power. here repeated]. a conversion product falls exactly at ωi . the mixer noise is often a second-order issue because: 1. a3 vi . 3. A fraction of the input power leaks to the output. 1 dB compression point. Being ∆ω ≪ ωi . Often.

IF5. by which noise is converted to the IF band from the sidebands of frequencies multiple of the LO frequency.... Flicker (1/f ) noise is generally not documented. in the presence of white noise at the RF port (Fig.. that is.. All the references found about the mixer noise are limited to classical white noise. 10 Where to learn more Our approach. p. The output slots IF1. 7). 29.. which consists of identifying and analyzing the modes of Table 2. RF ω N 4ω0 ω RF1 RF2 RF3 RF4 RF5 RF6 RF7 RF8 RF9 ω0 2ω0 3ω0 IF IF1 IF2 IF3 2N/ℓ IF4 IF5 IF6 IF7 IF8 . and IF3 are down-converted from the input slots RF3+RF4. 15 and Fig. The flicker behavior of mixer may depend on the operating mode. RF2+RF5.. is original. . . The first one is the multi-harmonic frequency conversion (Fig. 3.February 2. . 10 p. 43) appealing. . the general rule is that flicker noise is a near-dc phenomenon. in the ‘synchronous detector’ mode (Sec. the near-dc flicker is up-converted by non-linearity and brougt to the output. . N/ℓ 4ω0 ω ω0 power loss ℓ2 /2 2ω0 power loss ℓ2 3ω0 Figure 36: A step appears in the conversion of white noise. 5. . respectively.. while the flicker noise is not considered. . At higher frequencies. Two phenomena deserve attention. Then. Thus. after increasing to infinity the number of frequency slots so that their width is dω. . as listed in Table 2 (p. thermal and shot noise. 18). the conversion power loss is ℓ2 /2. the output slots IF4. IF2. powered by the LO pump. for the loss is ℓ2 . there are no specific references. come from RF7. 36). The analytical proof follows exactly the graphical proof. A lot can be learned from the data sheets of commercial mixers and from . or available at the output. and RF1+RF6. Thus. . 2008 E.. where the dc signal is taken at the output. 7 p. The second phenomenon is a step in the output noise spectrum at the LO frequency. . RF8. Rubiola Tutorial on mixers 49 (Fig.2) and in the ‘phase detector’ mode (Sec. .4). Only a graphical proof is given here.. Spectra LO ω0 . Yet.

Barber. IRE Trans. Rubiola Tutorial on mixers February 2. Newkirk and R. Burlington. Macom [M/A] and Mini-Circuits [Minb. Anzac. 188–220]. References [Anza] Anzac. (1958). A few books about radio engineering contains a chapter on mixers. Some radio amateur handbooks provide experiment-oriented information of great value. 1990. A. [Anzb] [Bar67] [Ber58] 12 http://www. and Chapter 7 (Microwave Mixer Design). about basic mixer theory. written by D. 15 (1967). Anza]. chapter 6 (Mixers) of Rohde & al.50 E. Noise figure and conversion loss of the schottky barrier mixer diode. Watkins Johnson12 application notes [Hena. The classical book written by S.. Biphase and quadriphase digital modulators. IEEE Trans. 629–635. Recent editions of the the ARRL Handbook [Str99] contain a chapter on mixers (chapter 15 in the 1999 edition). pp. [MV00. Adams-Roussell Co. Inc. 1990. Inc. MA. learning in this way requires patience because manufacturer tend to use their own notation. Transmission-line transformers and baluns are described in [Sev01]. Mark R. and because of the commercial-oriented approach. Saul M Bergmann. In RF and Microwave Signal Processing Components Handbook. hard to find elsewere. We also found useful the Anzac [Anzb. of Vendelin & al. Karlquist. 277–318]. 324–326. In RF and Microwave Signal Processing Components Handbook. We found useful chapter 3 (mixers) of McClaning & al. Henb] provide useful general description and invaluable understanding of intermodulation [Che]. One aspect of minimum noise figure microwave mixer design. 2008 the accompaining application notes. The non-specialist may be interested at least in the first part. one may surprised by the difference between standpoints. pp. MA. most of which published in the IEEE Transactions on Microwave Theory and Technology and other IEEE Journals.[VPR90].. full of practical information and common sense. Mina] application notes. Double balanced mixers. 261–344]. [KBR80. chapter 7 (Mixers) of Krauss & al. Kollberg [Kol84] collects a series of articles. [RWB96. L. 11. Adams-Roussell Co. pp. Burlington. Theory Tech.wj. Microw. no. Another problem is that the analysis is often too simplified. A book edited by E. Maas [Maa93] is a must on the subject. which makes difficult to fit technical information into theory. This collection covers virtually all relevant topics. Microw. Unfortunately.com/technotes/ . Theory Tech. Reading books and book chapters on mixers.

2. M/A-Com. [HK78b] . 55–61. no. Noise figure for a mixer diode. . W. . CA. 1980. Mixers. 1996. NY. Maas. Microw. In RF Microwave and Millimeter Wave Handbook. 27 (1979). and Frederick H. Brooklyn. Noise and loss in balanced and subharmonically pumped mixers: Part i—theory. [Ker79b] . Palo Alto. 135–140. 2. 1997. IEEE Trans. Bostian. 27 (1979). IEEE Trans. [Ker79c] [Kol84] [M/A] [Maa93] [Mina] . Conversion loss and noise of microwave and millimeter-wave mixers: Part 1—theory. Modern mixer terms defined. 1997–98. Palo Alto. Gewartowski. Shot-noise in resistive-diode mixer and the attenuator noise model. no. 26 (1978). 27 (1979). New York. Watkins Johnson Company. Theory Tech. Bert C. Kollberg (ed. In RF/IF Designer’s Handbook. 12. Theory Tech. 938–943. Theory Tech. UK. 944–950. [Gew71] [Hena] [Henb] [HK78a] Daniel N. Microwave and millimeter-wave mixers. CA. Microw. 481. A. part 1 and 2. 49–55. Mini-Circuits. Palo Alto. 1993. Held and Anthony R. In RF and Microwave Design Handbook. Dunstable. no. IEEE Trans. Microw. Microwave mixers. Noise and loss in balanced and subharmonically pumped mixers: Part ii—application. CA. no. Erik L. Theory Tech. Solid state radio engineering. S. Watkins Johnson Company. Conversion loss and noise of microwave and millimeterwave mixers: Part 2—experiment. Mixers: Part 1. Raab. characteristics and performance. Krauss. Mixers: Part 2. 2008 [Che] E. 5. IEEE Trans. Theory Tech. Kerr. Charles W. IEEE. 1984. no. [KBR80] Herbert L. 12. theory and technology. Ltd. Microw. J. application note m562. Microw. Theory Tech. Kerr. Microw. no. 1997–98. IEEE Trans. [Ker79a] Anthony R.February 2. In RF and Microwave Design Handbook. 29 (1971). 1997–98. In RF and Microwave Design Handbook. Rubiola Tutorial on mixers 51 Daniel Cheadle. Watkins Johnson Company. Henderson. 2.). New York. Artech House. 26 (1978). Selecting mixers for best intermod performance. John Wiley & Sons. IEEE Trans.

McGraw Hill. RF microelectronics. Masayoshi. vol. Henry C. N.52 [Minb] [MV00] E. [Sev01] [Str99] [TW48] [Vit66] Jerry Sevick. 2000. . W2FMI. McGraw Hill. 1999. [ROG92] Enrico Rubiola. Brooklyn. The ARRL handbook. 41 (1992). A. New York. Anthony M. Atlanta. T. New York. 3. Theory Tech. 2008 Mini-Circuits. 2001. and M. [Raz98] Behzad Razavi. Principles of mathematical analysis. IEEE Trans. McGraw Hill. 1998.). Pavio. 1948. IEEE Trans. Noble. 28 (1980). 1976. Crystal rectifiers. McGraw Hill. [OMK80] H. NJ. A. Radio receiver design. Dean R Straw. Prentice Hall. 1966. Andrew J. and Jacques Groslambert. [RWB96] Ulrich L. 3. [VPR90] George D. 1990. [Rud76] Walter Rudin. 1997. Newington CT. 353–360. Viterbi. Principles of coherent communication. John Wiley & Sons. Microw. 180–185. 1996. NY. K-band integrated double-balanced mixer. N6BV (ed. Bucher. Rubiola Tutorial on mixers February 2. Published yearly. Meas. and Ulrich L Rohde. Atlanta. GA. New York. no. Communications receivers: Principles and design. Jerry Whitaker. In RF/IF Designer’s Handbook. Upper Saddle River. Understanding mixers. Noble. Instrum. GA. Phase noise in the regenerative frequency dividers. Kevin McClaning and Tom Vito. 15. Torrey and Charles A. Marcel Olivier. Radiation Laboratory Series. and T. no. Microwave circuit design. Transmission line transformers. American Radio Relay League. Kozo. Rohde. Ogawa. Whitmer. Vendelin.