Mixers

∗
Liam Devlin

1. Introduction
Mixers are frequency translation devices. They allow the conversion of signals between a high frequency (the
RF frequency) and a lower Intermediate Frequency (IF) or baseband. In communications systems the RF is the
transmission frequency, which is converted to an IF to allow improved selectivity (filtering) and an easier
implementation of low noise and high gain amplification. This paper details the design of mixer circuits,
concentrating on low cost Printed Circuit Board (PCB) based designs using discrete Surface Mount Technology
(SMT) components.

2. The Fundamentals
The non-linear behaviour of a mixing device is used to realise the mixing function. Diodes, Field Effect
Transistors (FETs) and bipolar transistors can all be used as mixers and are all covered in this paper. Figure 1
shows the typical I-V characteristics of a Schottky diode, which can be described by equation (1).

I = a1V + a 2V 2 + a3V 3 + a 4V 4 + ......... (1)

I (mA)
150
125
100
75
50
25
0
0 0.25 0.5 0.75 1 1.25 1.5
Vf (volts)

Figure 1: Typical forward I-V Characteristics of a diode

If the diode is excited by two sinusoids, Cos(ω1t) and Cos(ω2t) the current through the diode is given by
equation (2).

I = a1 (Cos(ω 1t ) + Cos (ω 2 t )) + a 2 (Cos(ω 1t ) + Cos (ω 2 t )) 2 + .... (2)

When expanded this contains the term 2a2Cos(ω1t)Cos(ω2t) which has the trigonometrical relationship shown
in (3). It is either the sum or difference term that is the desired output of a mixer.

2Cos (ω1t )Cos(ω 2 t ) = Cos ((ω 1 − ω 2 )t ) + Cos ((ω 1 + ω 2 )t ) (3)

∗
Liam Devlin is with Plextek Communications Technology Consultants, London Road, Great Chesterford, Essex,
CB10 1NY Tel: +44 (0)1799 533200 Fax: +44 (0)1799 533201 Email: [email protected]
Diodes are “square-law” devices, which means the function describing their non-linear behaviour has a strong a2
component. This means that if excited correctly they should be able to produce a strong mixing product. Thus
the basic mixer design entails injecting the signals to be mixed and extracting the desired mixing product whilst
maximising the efficiency of the conversion. One significant problem with mixers is that in addition to the
wanted product, there are also numerous unwanted spurious products, often referred to as “spurs”. Figure 2
depicts the spectral output of a downconverting mixer. The Local Oscillator (LO) is mixed with the wanted RF
signal to produce a copy of the
RF signal at the difference
frequency (the IF). In general
the mixer will generate outputs IF
at a range of frequencies given (RF-LO)
by mRF ± nLO. The spectrum LO
shown in Figure 2 has an LO Image
frequency below the IF, this is (LO-IF) RF
known as low-side injection. 2IF
2RF-LO 2LO 2RF
One frequency of particular 3IF
importance is the image
frequency. This is 2IF away
Figure 2: Mixer spectral output
from the RF and will be
converted directly to the same
IF frequency as the RF. Noise and unwanted signals present at this frequency can severely degrade the system
performance. Filtering and/or image reject mixers (covered later in this paper) are normally incorporated to
address this problem. More detailed information on the system design can be found in [1].

In the case of upconverting mixers the input signal is the IF and the desired output signal is either the product or
difference of the LO and IF frequencies, depending whether high-side or low-side injection is being used. If the
wanted output is LO+IF, the difference product (LO-IF) is termed the unwanted side-band, or image and must
be rejected by filtering or the use of an image-reject mixer.

Most mixers incorporate some form of filtering which helps to reduce the levels of the unwanted spurious
outputs. Another commonly used technique, which helps reduce spurious outputs, is the use of balanced mixer
designs. More detail on balanced mixer design is included in Section 4.

3. Mixer Terminologies
Listed below are some of the terms used in referring to mixers or mixing performance:

Conversion loss: The ratio of the wanted output signal level to the input, normally expressed in dB.

Noise Figure: The ratio of the Signal to Noise Ratio (SNR) at the input compared to the SNR at the output,
measured at 290K. To avoid ambiguity this paper will use the term noise figure to refer to the value of this ratio
in dB and the term noise factor to refer to the value as an absolute ratio.

Double Sideband (DSB) Noise Figure: Includes noise and signal contributions at both the RF and the
image frequencies.

Single Sideband (SSB) Noise Figure: No image signal is included although image noise is included.
Provided the mixer performance is the same at the image and the wanted frequencies, the SSB noise
factor = twice the DSB noise factor.

Compression: For small input signal levels, each dB increase in signal level results in a dB increase in the
output signal level. As the input signal level continues to increase, the conversion loss of the mixer will
eventually start to increase. The 1dB compression point is the input signal level at which the conversion loss has
increased by 1dB. Mixers should be used “backed-off” from the 1dB compression point as in addition to
distortion of the wanted signal, operation at or close to it would give rise to significant increases in the levels of
the spurious outputs.

Third Order Intercept Point. This is a figure of merit to give an indication of the mixer’s signal handling
capability. In particular it provides an indication of the levels of third order products a mixer is likely to
produce under multi-tone excitation. It is measured by applying two closely spaced input tones at frequencies F 1
and F2. Third order products from the
mixing of these tones with the LO (at
frequency FLO) occur at frequencies
given by: (2F1±F2)±FLO and
F 2 -F LO F 1 -F LO
P IF (2F2±F1)±FLO. In the case of a
downconvert mixer, the third order
products of most interest are (2F1-F2)-
∆L FLO and (2F2-F1)-FLO as they fall in, or
close to the IF band. Figure 3 depicts
(2F 2 -F 1 )-F LO (2F 1 -F 2 )-F LO the IF output spectrum of a
downconvert mixer under two-tone
excitation.
Figure 3: IF spectrum for mixer third order intercept
The third order intercept point itself is
point measurement an entirely imaginary point, at which
the third order product becomes as
large as the direct downconverted product. The level of the third order products rises at three times the rate of
increase of the input signal level and fundamental output level. The mixer’s output referred third order intercept
point (TOIout) is given by equation (4), all values are in dB and it is the dB value of ∆L which is divided by 2.

∆L
TOI out = PIF + (4)
2

With mixers, the third order intercept point is often referred to the input, which just requires adding the
conversion loss to TOIout.

Linearity. The linearity of a mixer refers to its signal level handling ability. Thus a mixer with high linearity
will have a high TOI.

Spur’s. An abbreviation of spurious product. The term is used to describe any unwanted mixing product.

Sub-harmonic mixer. This is a mixer circuit designed to accept an LO input at a fraction (often a half) of the
desired LO mixing frequency.

Harmonic mixer. This is just another term for sub-harmonic mixer but is more commonly used for circuits
employing higher multiples of the input LO to produce the mixing LO.

Pump. A term sometimes used to describe the LO drive. The LO input is said to be “pumping” the mixer.

Image frequency. For high side injection (FLO > FRF) this is FLO + FIF, for low side injection (FLO < FRF) it is
FLO - FIF. In downconvert mixers, it is a frequency that is converted directly to IF along with the IF itself. In
upconvert mixers it is an unwanted sideband which, without additional filtering, is usually at a similar level to
the wanted signal.
Image-reject mixers. A more complex mixer configuration, which has the advantage of providing inherent
cancellation of the image signal.

Image enhancement. A method for reducing the conversion loss of a mixer by terminating the image frequency
in an appropriate reactive impedance. Should be used with caution as the resultant mixer can have severely
degraded intermodulation performance [5]. Also, the exact image impedance is normally found empirically.

4. Diode Mixers
Most modern diode mixer designs use Schottky diodes. The main reason for this is that the Schottky diode is a
majority carrier device which means it has a higher switching speed than p-n junction diodes [2]. In-expensive
plastic packaged diodes are now available, which are suitable for designing mixers up to around 13GHz.
Manufactures normally specify the intended application of a particular diode and the selection of a suitable
diode is a vital step in diode mixer design. It is also common for manufacturers to refer to diodes as low,
medium or high barrier. The higher the barrier height, the higher the forward voltage required to turn the diode
on. The exact definition of what constitutes a low, medium or high barrier is open to the manufacturer’s
interpretation. However, broadly speaking, for a forward current of 1mA, low barrier diodes require a forward
voltage of around 0.2 - 0.3V, medium 0.4 - 0.5V and high 0.6 - 0.7V. The higher the barrier, the higher the LO
drive which will be required to obtain low loss mixing but the resultant mixer should have greater linearity.

The electrical equivalent circuit for a packaged Schottky diode is shown in Figure 4. Also shown in Figure 4 is
a typical RF Schottky diode in a SOT23 package; with a pencil tip for size comparison, Lp and Cp are the
packaging parasitics. Rs is the parasitic series resistance of the diode and Cj and Rj are the non-linear
components of the Schottky diode junction. The non-linearity of Rj is responsible for the square law behaviour
of the diodes DC characteristics (Figure 1).

C j(V) R j (V)

Rs

Cp

Lp

Figure 4: Equivalent circuit of a packaged Schottky diode and a photograph of a SOT23

packaged diode, with pencil tip for size reference
Most diode mixer designs utilise unbiased diodes, however forward biasing of the diodes, so a small DC current
flows, can offer reduced conversion losses. This is particularly the case when limited LO drive is available. The
diode is biased to have a quiescent operating point close to the region of maximum non-linearity in its operating
characteristics which allows the diodes square law characteristic to be traversed with lower levels of LO drive.
 4.1. Single-ended Diode Mixers
Mixers, which utilise a single diode as the mixing element, have no inherent isolation between the mixer ports
and are known as single-ended designs. Figure 5 shows a basic block diagram of a single-ended mixer.

One of the main difficulties with single-ended RF

designs is that the LO and RF inputs must be
separated with a diplexer filter. They are IF
normally relatively closely spaced and separating LO
the two frequency bands can be problematic.
Coupled with this, the fact that no inherent
spurious suppression is afforded by this
topology, it is not surprising that few modern
diode mixer designs are single-ended. The Figure 5: Basic block diagram of a single-ended
exception to this is high mm-wave frequency mixer
designs, which are still often realised with a
single diode.

A step-by-step procedure for designing a single ended mixer is given below:

1. Choose a suitable diode for the application. Factors effecting this choice include operating frequency,
available LO drive, cost versus performance trade-offs and package style.
2. Design the IF filter, the techniques described in [3] can be used. In addition to having low insertion loss
it is important that it presents a high input impedance at the LO and RF frequency's. See also the
comments on matching, below.
3. Design the RF and LO filters [3], in addition to having low loss and providing a diplexing action which
gives isolation between the two inputs, the common output of these filters must provide a high
impedance across the IF frequency band. See also the comments on matching, below.
4. Large signal simulators are now commonly available and most manufacturers supply large signal
models for their diodes. It is strongly recommended that, when possible, a large signal analysis of the
mixer be carried out prior to fabrication.

Matching: If a diode is considered as a switch, being either open or short-circuit, then impedance matching
between the mixer ports and the diode is not possible and indeed not necessary. However, it is more appropriate
to think of a mixer diode as a square-law device. The impedance that the diode presents is a time varying
impedance, dependent on the LO level and frequency. It is the time-averaged value of the diode’s impedance,
which must be used if matching is attempted. If an accurate large signal model and the packaging parasitics are
available, simulation of the LO-dependant diode impedance is possible. For those without access to a large-
signal simulator an estimate of the time-averaged value of Rj(V) and Cj(V) can be made. Matching to the diode
can improve the performance of a mixer but it must be addressed with care. It is important to note that the filter
requirements detailed in steps 2 and 3, above must still be satisfied with any matching networks present.

Linearity: The best way to improve the linearity of a diode mixer is to increase the LO drive level. Higher
barrier-height diodes should be used for best performance, provided that adequate LO drive is available.
Techniques such as image-enhancement should be avoided as this can degrade linearity [5].

4.2. Single-balanced Diode Mixers

A single-balanced diode mixer uses two diodes. Either the LO drive or the RF signal is balanced (applied in
anti-phase), adding destructively at the IF port of the mixer and providing inherent rejection. The level of
rejection is dependent on the amplitude and phase balance of the balun, providing the balanced drive, and the
matching between the two diodes. A rejection of 20 to 30dB is normally possible for good discrete designs.
Other advantages of a singly-balanced design are rejection certain mixer spurious products, depending on the
exact configuration, and suppression of Amplitude Modulated (AM) LO noise. AM noise could be a significant
problem in early microwave and mm-wave receivers where the available LO sources were very noisy. Modern
wireless transceivers tend to make use of synthesised LO drives and the LO phase noise gives more of a
problem than the AM noise.

One disadvantage of
balanced designs is that RF
0°°
they require a higher LO
°
drive level. Figure 6 shows 0

IF
a block diagram of a single-
balanced mixer. It utilises LO 0°°
an anti-podal diode pair. -180°°
Matched diode pairs, in
various configurations, are R F sho rt-circu it
readily available in low-
cost plastic packages.
Other configurations of Figure 6: Block diagram of a single-balanced mixer
single-balanced mixers are
possible, more details can
be found in [2].

For the topology shown in Figure 6, the LO drive to the two diodes is in anti-phase (balanced) and the RF signal
is in-phase. If the mixing products are at mRF ± nLO, this mixer will reject all products where m is even. If the
RF drive were in anti-phase and the LO in-phase, all spurious products with n even would be rejected. The anti-
phase signal is also cancelled at the IF port. Because the LO drive should be at a significantly higher level than
the RF signal, it is often chosen as the anti-phase signal to increase the LO to IF isolation. However, it is also
important to consider the spurious rejection properties.

The RF short-circuit, shown at the IF port in Figure 6, is

required for the mixer to function. Although shown
explicitly here, it is normally incorporated in the IF low
pass filter design. If the RF impedance at the IF port were
high, the RF signal voltage across the diodes would be
small and the mixer's conversion loss very high. The LO
signal, however, does not require a low impedance at the
IF port. Because the LO is a balanced signal across the
diode pair, the common port of the diodes is a virtual
earth to the LO. The LO drive across the two diodes adds
destructively to a null at the common port, as if it were
grounded. In most cases, however the LO and the RF are
comparatively close in frequency and the RF short circuit
will also be a good short circuit at the LO frequency.

The design procedure for a balanced diode mixer is

similar to that for a single-ended. The only difference
being that the balun structure providing the RF and LO
isolation and its design must be considered as part of the
Figure 7: Photograph of a single-
RF/LO filter design. One option for the balun realisation
balanced diode mixer using a Rat-Race is a Rat-Race coupler, as described in [4]. This is a very
balun popular option at microwave frequencies where it is a
comparatively small structure that can be
produced inexpensively on a printed substrate.
Figure 7 shows a photograph of a prototype
realisation of such a mixer. It covers the
HIPERLAN RF band of 5.15 to 5.35GHz with a
700MHz IF, measured conversion loss is 7dB and
the input referred 1dB compression point is
+5dBm for an LO drive level of +8dBm. A SOT-
23 packaged anti-podal diode pair has been used
and a 1pF 0603 capacitor is used to realise the RF
short at the IF port, whilst forming the first
element of the IF low-pass filter.

Figure 8 shows a 1.5GHz single-balanced mixer

realised on an FR4 substrate with a lumped Figure 8: Photograph of a single-balanced
element balun. This had a conversion loss of 9dB mixer with lumped element balun
for an LO drive of +8dBm.

The ceramic resonator filter to the bottom left of Figure 8 is the RF image filter. At the output of this filter is
the lumped element RF filter of the mixer. It connects to the common port of the antipodal diode pair (in the
SOT23 package). The IF port of the mixer (at the top centre of Figure 8) is also connected to the common port
of the diode pair via a filter. The output of the RF filter needs to present an open circuit to the output of the IF
filter and vice-versa. The LO input is to the bottom right and a simple lumped element balun providing a
differential drive across the diode pair.

4.3. Double-balanced Diode Mixers

A double-balanced diode mixer normally make use of four diodes in a ring or star configuration with both the
LO and RF being balanced. All ports of the mixer are inherently isolated from each other. Matched diode rings
(fabricated in close proximity on the same substrate material) are readily available in SOT143 plastic packages.
The advantages of a double-balanced design over a
single balanced design are increased linearity, a
improved suppression of spurious products (all
even order products of the LO and/or the RF are
suppressed) ant the inherent isolation between all LO d b
ports. The disadvantages are that they require a
higher level LO drive and require two baluns.
Figure 9 shows a block diagram of a double-
balanced quad-ring diode mixer. Details of the star c
topology can be found in [2].
IF
The operation of a double balanced mixer is best
understood by considering the diodes as switches.
The LO alternately turns the right hand pair and RF
left hand pair of diodes on and off in anti-phase.
Points ‘a’ and ‘c’ are virtual earths to the RF Figure 9: Block diagram of a double-balanced
signal and can be considered as connected to diode mixer
ground. Thus points ‘b’ and ‘d’ (the balanced RF
signal) are alternately connected to ground (at
points ‘a’ and ‘c’). This means an in-phase RF signal and an anti-phase RF signal are alternately routed to the
IF port under control of the LO. Thus the signal at the IF port is effectively the RF signal multiplied by an LO
square wave of peak magnitude ±1.
This action is easily demonstrated using simple mathematical processing software. Figure 10 shows a sinusoidal
voltage waveform at a frequency of 1GHz, this is the RF waveform. Figure 11 shows a square wave at a
frequency of 870MHz, this is the LO switching waveform. Multiplication of the two will produce a waveform
wit a strong component at the difference frequency (IF) of 130MHz.

Vrf 0
n

0 5 10 15 20
t
n

Figure 10: RF voltage waveform versus time in ns

Vlo 0
n

1
0 5 10 15 20
t
n

Figure 11: LO voltage waveform versus time in ns

Figure 12 shows the result of multiplying the RF and LO waveforms. A low frequency sinusoid is clearly
visible. This is a replica of the RF signal (i.e. a sinusoid) translated to the IF frequency of 130MHz. Although
this method of mixer analysis
provides a qualitative
understanding of how the 1
mixer functions, it is not
adequate to predict the RF
functionality. Ideal square
wave multiplication, such as Vif 0
n
this, results in a conversion
loss of 3.9dB. In practice
diode-ring mixers have
additional losses (in the 1
baluns and diodes) and
0 5 10 15 20 25 30 35 40
imperfections which increase
t
the conversion loss actually n
achieved. A loss of between Figure 12: IF voltage waveform (Vrf*Vlo) versus time in ns
6 and 8dB is typical for a
well designed diode ring mixer. In order to predict accurately the mixer’s performance, large signal circuit
simulation must be performed.

The block diagram in Figure 9 shows the differential RF and LO signals provided using wire-wound ferrite
transformers. Wire-wound transformers can be used at frequencies up to over 2GHz but lower cost printed or
lumped element baluns are often implemented in practical mixers. At higher frequencies wire wound
transformers become impractical and printed and/or lumped baluns become the norm. Care should be taken to
consider how the performance of these baluns differs from wound transformers; additional filtering may be
necessary. An overview of practical balun configurations is given in Section 5.

4.4. Double doubly balanced diode

mixers

As the name implies, a double doubly balanced

mixer is an interconnection of two double balanced
mixers. Figure 13 shows a block diagram. Increased
linearity is the main advantage of the double doubly
RF LO
balanced topology (or treble balanced as it is also
referred to). The reason for this is easily understood;
the incident power is simply shared amongst a twice
as many diodes, thus increasing the signal handling
capability by 3dB. The main disadvantage is also
obvious, increased complexity: A total of 3 baluns
and 8 diodes are required. In addition to this, a
higher level of LO drive (3dB more) must be
provided. An alternative approach to realising high
linearity mixers is the FET resistive mixer, detailed
in Section 6. This can yield even higher linearity
than the double doubly balanced topology whilst IF
having a simpler circuit configuration. Figure 13: Block diagram of a double doubly
balanced mixer
The transformer configuration shown in Figure 13 is
from [5]. In practice such a complex arrangement of
wire wound transformers is unlikely to be used and a more practical approach is to combine a pair of double
balanced mixers using hybrid power combiners/splitters.

4.5. Sub-harmonic mixers Diode Mixers

A sub-harmonic, or sub-harmonically pumped, mixer has an LO input at FLO/n but is designed to maximise the
conversion efficiency of an FLO product. They are useful at higher frequencies when it can be difficult to
produce a suitable LO signal (low phase noise, tuning range and output power all become more difficult to
achieve with increasing frequency, whilst cost increases). It is also possible to design frequency multipliers
using transistors and/or diodes and to multiply an LO input before using it to drive a fundamental mixer. Such
architectures are in common use but are not true sub-harmonic mixers. Figure 14 is a circuit configuration,
which can be used to realise a sub-harmonic diode mixer. For an LO input at F LO/2, the output is maximised at
FLO ± FRF (or FLO ± IF when used as an upconverter).

The circuit makes use of an anti-parallel diode pair and provided the diodes are identical it has no fundamental
mixing response. It also benefits from the fact that the FRF and FLO are normally relatively close in frequency
(for a comparatively low IF). Thus the short circuit λLO/2 stub at the LO port is a quarter of a wavelength long
at the input frequency of FLO/2 and so is open circuit. However, at FRF this stub is approximately a half
wavelength long, so providing a short circuit to the RF signal. Conversely, at the RF input the open circuit
λLO/2 stub presents a good open circuit to the RF but is a quarter wavelength long at the frequency FLO/2 and so
is short circuit. The IF is normally far enough away from the RF frequency to allow easy realisation of an IF
filter presenting an open circuit output to the RF port.

RF

IF LO/2

λLO/2 λLO/2

Figure 14: Block diagram of a sub-harmonic diode mixer

A photograph showing a practical implementation of this type of sub-harmonic mixer is shown in Figure 15. It
uses an antiparallel diode pair in a low cost SOT 23 package. The IF port is to the left, going into a low pass
filter ending in an 0402 chip inductor which is open circuit resonant at the RF frequency of 11.7 to 12.7GHz.
The IF frequency is 1 to 2GHz and the F LO/2 input is fixed at 5.35GHz (FLO=10.7GHz).

Figure 15: Photograph of a 12GHz sub-harmonically pumped mixer

The conversion loss versus RF frequency has been measured for this mixer at three LO input power levels
(+5dBm, +8dBm and +10dBm), a graph of the results is shown in Figure 16. For LO drive levels of 8 or
10dBm the conversion loss is between 9.5 and 11dB, which is only slightly more than would have been achieved
with a fundamental diode mixer design, with the advantage of only having to generate an LO signal at half the
actual LO frequency. One disadvantage to the mixer compared to a fundamental design is increased spurious
products. This is not only due to the fact that the mixer is not balanced but also that there are additional
spurious products due to spurs with F LO/2 products.

20
Conversion Loss (dB)

15

LO=+5dBm
10 LO=+8dBm
LO=+10dBm

0
11.7 11.9 12.1 12.3 12.5 12.7
RF Frequency (GHz)
Figure 16: Measured conversion loss versus frequency for sub-harmonic mixer

5. Baluns
A balun is used to transform a signal between BALanced and UNbalanced modes. An unbalanced signal is
referenced to a ground plane, as in a coaxial cable or microstrip. A balanced signal is carried on two lines and is
not referenced to a ground plane. Each line can be considered as carrying identical signal but with 180° of phase
difference. A comprehensive presentation of balun design is beyond the scope of this paper but an overview of a
number of practical implementations is given below and references are provided.

5.1. Wire-wound transformers

A wire-wound transformer provides an C o m m o n -m o d e

in p u t D ifferen tia l o u tp u t
excellent balun and it has been used to
represent the balun in all of the mixer
topologies presented here. Miniature wire-
wound transformers are commercially available
covering frequencies from low kHz to beyond Figure 17: Centre-tapped transformer as a balun
2GHz [6]. They are often realised with a
centre-tapped secondary winding, if grounded this provides a short circuit to even-mode (common-mode) signals
whilst having no effect on the differential (odd-mode) signal.

Wire-wound transformers are more expensive than the printed or lumped element baluns described below,
which find greater adoption in practical mixer designs. It should be noted that most of these lumped element and
printed baluns do not provide the centre-tapped ground to even mode signals and this fact must be accounted for
in the mixer design.
 5.2. Printed baluns

There are a wide range of

printed balun topologies [7] they C o m m o n -m o d e
in p u t
have the advantage of being
inexpensive, realised as they are D ifferen tia l o u tp u t
on the Printed Circuit Board
(PCB) or Microwave Integrated
Circuit (MIC) substrate. On the Figure 18: Simple coupled line balun
downside they can be quite
large, particularly at lower RF frequencies. The rat-race coupler shown in Section 4.1 is commonly used at
microwave frequencies for bandwidths of up to around 10-20%.

Possibly the simplest printed balun is the

C o m m o n -m o d e
coupled line balun [8], also called a
in p u t parallel-line balun shown in Figure 18. The
D ifferen tia l o structure is a quarter of a wavelength long
at the centre frequency. It is capable of
bandwidths of over an octave, provided the
Figure 19: Coupled line balun using multiple coupled coupling between the lines is high enough.
lines In practice this is not normally the case for
the simple edge coupled balun shown in
Figure 18. A more practical approach is to
use multiple coupled lines as shown in Figure 19 or, where multi-layer substrate processing is available, to
adopt a broad-side coupler topology as in Figure 20. This broadside-coupled implementation is often referred to
as a parallel plate balun.

C o m m o n -m o d e
in p u t
D ifferen tia l o u tp u t

Figure 20: Coupled line balun, using broadside coupler structure

An improvement on the parallel-line balun is a printed version of the “Marchand Balun”. This is derived from
the co-axial balun, described by Nathan Marchand in 1944 [9]. The printed version of the Marchand balun is
shown in its simplest form in Figure 21. This is more tolerant to low even mode impedance (low coupling ratio)
than the parallel line balun and has a wider bandwidth.

O pen
C o m m o n-m o d e C ircu it
in pu t

D ifferen tia l o u tp u t
 Figure 21: Printed Marchand Balun

As with the parallel line balun, improved performance is obtained

D ifferen tia l o u tp u t
if multiple planar section are used [10] or if a broadside coupling
topology is adopted [11]. One draw back to using these printed
baluns at lower RF frequencies is their size. One technique to
reducing the size is to include lumped elements and printed
structures, as shown in Figure 22. This allows acceptable balun
performance to be achieved with significant area reduction [12].
As with the parallel line and Marchand baluns, the use of
broadside, rather than edge, coupling will yield tighter coupling
and improved performance.

5.3. Lumped Element Baluns

Lumped element baluns are based around the fact that the
insertion phase through a low pass filter lags the insertion phase
C o m m o n -m o d e
through a high pass filter [13]. It therefore possible to design low in p u t
pass and high pass filters that have a relatively constant 180°
difference in insertion phase. A wide range of topologies is Figure 22: Reduced size printed
possible and for narrow band designs, very simple structures can balun
be adequate [14]. Figure 23 shows a lumped element implementation of a rat-race splitter/combiner [7]. Signals
incident on the Σ port split equally in amplitude and phase, whilst signals incident on the ∆ port split equally in
phase but have a 180° phase difference. Design equations are shown in the figure, Zo is t he system impedance.

2C 2C

Σ In /O u t1

L L
C C
In /O u t2 ∆

C L C

1
ωo L = = 2Zo
ω oC

Figure 23: Lumped element realisation of a rat-race splitter

5.4. Active Baluns

Active baluns have a number of disadvantages:

• Degraded intermodulation performance for the resultant mixer

• Amplitude and phase balance is normally poorer than for passive designs
• Discrete realisations sensitive to package parasitics
 • DC bias is required
• The output impedances of each port can be significantly different

Figure 24 shows three possible active balun realisations.

3 -p or t p h ase splitter C -B /C -E a m p lifier L on g -ta il p air

Figure 24: Active balun topologies

The 3-port phase splitter only really works well at lower frequencies but can be quite compact. The C-B/C-E
balun can work well over quite wide bandwidths provided parasitics are well modelled. The long-tail pair is
commonly used in integrated realisations, closer inspection will reveal it is extremely similar in architecture to
the C-B/C-E topology.

6. FET Mixers
FETs can be used in mixers in both active and passive modes. Active FET mixers are transconductance mixers
using the LO signal to vary the transconductance of the transistor. They have the advantage of providing the
possibility of conversion gain
rather than loss and can also IF O u tp ut
have lower noise figures than
passive designs. Figure 25 RF and LO
In p u t
shows the simplest realisation of
a transconductance mixer,
biasing circuitry has been R F sh o rt-circu it
omitted for clarity. The RF (and
LO) short circuit at the drain is
important to ensure that the Figure 25: Simple transconductance mixer
value of Vds is not moved
significantly from its DC bias point by the applied LO. This ensures the magnitude of the time varying
transconductance is maximised so optimising the conversion gain. Unfortunately it also means that this mixer
topology is not well suited to realising upconverters.

The topology of Figure 25 has the disadvantage that some form of diplexing is required to separate the RF and
LO inputs which are incident on the same port. For this reason dual gate FET mixers are often used. This
toplogy is essentially a cascode arrangement of two transistors as shown in Figure 26, although in practice four
terminal dual gate FET devices are sometimes used.
The RF input is applied to the bottom device which is matched using the well-known techniques developed for
amplifier design, the LO signal is applied to the top device, which is often resistively matched. One advantage
this structure has is that the LO and
IF O u tp u t RF signals are inherently isolated. It
can be used to develop compact
L O In p u t mixers with conversion gain, as
described in [15]. Although the
potential of conversion gain rather
R F s h o rt-c irc u it than loss, which the
R F In p u t transconductance mixer offers, is
attractive the downside is that they
tend to have lower linearity than
passive designs.

When used in passive mode, the

Figure 26: Dual gate FET mixer topology FET is used as a switch. Its
suitability for switch realisation
stems from the fact that its drain-source resistance behaves as a voltage variable resistor, the resistance being
set by the gate-source voltage [16]. When used as a switch, a FET is operated with the drain and source at zero
volts DC. The RF signal path is drain to source and the gate is the control terminal. It can be represented by the
simplified equivalent circuit shown in Figure 27.

Rg

Vg

Figure 27: Simplified equivalent circuit of a passive switching FET

A simple FET switching mixer, which can provide high linearity for moderate LO drive levels, is shown in
Figure 28. The gates of the FETs are biased part way between 0V and pinch off, this allows the LO signal to
move the FETs between their “on” and “off” states. At lower RF frequencies FET gates have a high input
impedance and the load for the differential LO signal is thus approximately 2Rg (Figure 28). By setting Rg to a
moderately high value say (200 or 300Ω), increased gate voltage swing can be obtained for the same LO level
as compared to driving a 100Ω differential load. At higher frequencies, the input capacitance of the FET gate
presents a lower reactance and the LO voltage swing will be reduced for the same LO power level.

FET switching mixers will not function well if the gates are left unbiased. If the LO signal is large enough to
turn the FETs “off” on the negative cycle, it will drive the gate-source junction in to forward bias on the
positive cycle. It is vital that the gate bias voltage is set appropriately if optimum mixer performance is to be
obtained. For discrete implementations this gives a problem as the specified range of pinch-off voltages for the
FETs can be very wide (-0.5V to –2.5V is a typical range). Whilst integrated designs can overcome this
problem with on-chip bias circuitry, for discrete designs there are two solutions: Select on test resistors can be
used to set the bias or a supply of FETs with a reduced range of pinch-off voltages can be agreed with the
manufacturer. Both solutions have cost penalties.
 RF

IF

Rg Rg

G ate B ias

LO

Figure 28: Circuit diagram of a FET based switching mixer

A practical implementation of this switching mixer is shown in Figure 29. Composite printed/lumped baluns,
with broadside coupling, are used for the RF and LO. The IF is extracted from the centre point of the RF balun
through an inductor (bottom left of the photograph)
which is open-circuit resonant at the RF frequency.
This mixer was part of an early GSM handset
design. It had a conversion loss of 8dB and an input
1dB compression point of +8dBm for an LO signal
level of +5dBm.

Double-balanced FET quad ring mixers, analogous

to the diode-ring mixer (Section 7) can also be
used. An additional IF balun is required, as shown
in Figure 30. The LO signal switches Q1 and Q3
Figure 29: Photograph of a FET based on and off in anti-phase with Q2 and Q4. The effect
switching mixer of this is that the RF signal and a 180° phase
shifted version of the RF signal are alternately
routed through to the IF port. As with the diode ring, this means the IF output is effectively the RF signal
multiplied by an LO square wave of peak magnitude ±1. The additional cost and complexity of this topology
means it is not a popular choice for discrete realisations, although it has been used successfully on integrated
designs [17].
 Q1 Q2

LO IF

Q4 Q3

RF
Figure 30: FET quad ring mixer

7. BJT Mixers
Discrete bipolar mixers tend to find applications in low cost, low power receivers such as discrete
implementations of pager front ends. Designs can be compact, inexpensive and have conversion gain, however
they tend to have poor linearity. Figure 31 shows two typical implementations of low cost, discrete bipolar
mixers.

IF O u tp ut
IF O u tp ut
L O In pu t L O In pu t

R F Inp u t
R F Inp u t R F sho rt-circu it
R F sho rt-circu it

Figure 31: Low cost discrete bipolar mixer topologies

There is a wide range of commercially available Si bipolar integrated RF receivers and transceivers. The mixers
they contain differ significantly from the discrete implementations described above. The transistors fabricated
close to each other on an IC behave very similarly (they are well matched) and the die area they occupy is
smaller than that occupied by passive components [18]. This leads to different circuit topologies being exploited
with the almost universal choice for mixer realisation being the double-balanced “Gilbert Cell” [19] shown in
Figure 32. A long-tail differential pair amplifies the RF input to the mixer. This determines the gain of the
mixer and limits its linearity. The differential outputs of this amplifier are switched, by the LO signal,
alternately to each of the differential IF outputs. Once again it is essentially a multiplication of the RF by ±1 at
the LO frequency. This circuit relies on the different transistors being well matched and a discrete realisation is
not practical.
 IF output

LO input

RF input

Bias

Figure 32: Double-balanced ("Gilbert Cell") mixer

8. Image Reject Mixers

Image reject mixers comprise two balanced mixers, of any topology, driven in quadrature by the RF signal. The
LO drive to each mixer is in-phase and the IF output is combined in quadrature. Figure 33 shows a block
diagram of an image reject mixer, the arrows representing the relative phases of the respective signals. M1 and
M2 are the conversion gain/loss of the two mixers as a factor.
M 1 ⋅ ( RF − LO )
2
M 1 ⋅ ( LO − RF )
M 2 ⋅ ( LO − RF ) M 2 ⋅ ( RF − LO )
2
2 2
RF LO M 1 ⋅ ( LO − RF ) M 1 ⋅ ( RF − LO)
2 2 2 2
RF
IF o u tp u t is R F -R L O (im a g e )

In-phas e LO
–3dB, 90° Hybrid splitter –3dB, 90° Hybrid

RF LO
2 2 IF o u tp u t is L O -R F (w a n te d )

M 2 ⋅ ( LO − RF ) M 2 ⋅ ( RF − LO )
2 2
M 1 ⋅ ( LO − RF ) M 1 ⋅ ( RF − LO )
M 2 ⋅ ( LO − RF )
2 2
2
M 2 ⋅ ( RF − LO )
2

Figure 33: Block diagram of an image reject mixer

To achieve perfect image cancellation, the mixers must be identical and the amplitude balance and phase shift of
all quadrature and in-phase splitters perfect. An integrated solution will yield higher image rejection than a
discrete and image reject mixer ICs are commercially available. With care, a discrete implementation should be
able to achieve over 20dB of image rejection. The rejection of a discrete implementation can be improved if
tuning of the circuit is carried out but this is not normally a viable option for high volume commercial products.
9. Summary
This paper has attempted to explain the operation of RF mixers and to provide guidelines for their design.
Diode, FET and BJT implementations have been considered and a number of practical examples presented.

10. References
[1] Hunter, M.T.J. “The Basics of System Design”, Proceedings of the IEE Tutorial Colloquium on “How to
Design RF Circuits”, Wednesday 5th April 1999, Savoy Place, London

[2] Maas, S.A. “Microwave Mixers”, Artech House, ISBN 0-89006-605-1

[3] Walker, J.L.B. “Filters”, Proceedings of the IEE Tutorial Colloquium on “How to Design RF Circuits”,
Wednesday 5th April 1999, Savoy Place, London

[4] Walker, J.L.B. ”Improvements to the Design of the 180° Rat Race Coupler and its Application to the
design of Balanced Mixers with High LO to RF Isolation”, IEEE MMT-S Digest, 1997, Vol. II, pp 747-
750

[5] Maas, S.A. “Two-Tone Intermodulation in Diode Mixers”, IEEE MTT-Transactions, Vol. MTT-35, No.
3, March 1987, pp 307-314

[6] Nishizuka, N et al, “Analysis of Frequency Characteristics of Small-Sized Wide-Band Compound

Transformers”, IEEE Transactions on Magnetics, Vol. 34, No. 4, July 1998

[7] Sturdivant, R, “Balun Designs for Wireless, …Mixers, Amplifiers and Antennas”, Applied Microwave,
Summer 1993, pp 34-44

[8] Cho, C and Gupta, K.C., “A New Design Procedure for Single-Layer and Two-Layer Three-Line
Baluns”, IEEE Transactions on Microwave Theory and Techniques, Vol. 46, No. 12, December 1998,
pp 2514-2519

[9] Marchand, N. “Transmission-Line Conversion”, Electronics December 1944, pp 142-145

[10] Schellenberg, J. and Hien, D, “Low-Loss Monolithic Baluns for K/Ka-Band Applications”, IEEE, MTT-
S Digest, 1999

[11] Tutt, M.N., et al, “A Low Loss, 5.5GHz – 20GHz Monolithic Balun”, IEEE MMT-S Digest, June 1997,
pp 933 - 936

[12] Ojha, S.P., Branner, G.R. and Kumar, B.P., “A Miniaturized Lumped-Distributed Balun for Modern
Wireless Communication System”, IEEE MTT-S digest, 1997, pp 1347-1350

[13] Devlin, L.M. “Digitally Controlled, 6 Bit, MMIC Phase Shifter for SAR Applications”, Proceedings of
the 22nd European Microwave Conference, 1992 pp 225-230

[14] Miron, D.B., “The LC Immittance Inverter”, RF Design, January 2000, pp 20-26

[15] Tsironis, C et al, “Dual-Gate MESFET Mixers”, IEEE MTT Transactions, Vol. MTT-32, No. 3, March
1984, pp248-255

[16] Devlin, Liam, “The Design of Integrated Switches and Phase Shifters”, Proceedings of the IEE Tutorial
Colloquium on “Design of RFICs and MMICs”, Wednesday 24 th November 1999, pp 2/1-14
[17] Devlin, L.M. Buck, B.J., Dearn, A.W., Clifton, J.C., Frier A.A.G., Geen, M.W., “A High Volume, Low
Cost, Plastic Packaged, 2.4GHz Transceiver MMIC”, Proceedings of the third annual Wireless
Symposium, Santa Clara, CA, 1995, pp 121-125

[18] Devlin, L.M., “RF ICs for Commercial Wireless Applications”, IEE tutorial colloquium on the Design of
RFICs and MMICs, 26th Nov. 1997, pp 2/1-2/11

[19] Gilbert, Barrie, “A Precise Four-Quadrant Multiplier with Subnanosecond Response”, IEEE Journal of
Solid-State Circuits, Dec. 1968, pp365-373