Sheet

1 of 20

Gilbert Cell Mixer Design Tutorial

J P Silver

E-mail: [email protected]
1 ABSTRACT
3 MIXER DEFINITIONS
Frequency translation in a system, is performed by a
non-linear device known as a mixer. There are vari-
(1) Conversion Gain: This is the ratio (in dB) between
ous topographies from simple single ended, single
the IF signal (usually the difference frequency between
balanced mixers to more complicated double & triple
the RF and LO signals) and the RF signal.
balanced mixers that provide better isolation from
the Local Oscillator (LO) and spurious.
(2) Noise Figure: Noise figure is defined as the ratio of
The most popular double-balanced mixer used in SNR at the IF port to the SNR of the RF port.
RFIC designs is the Gilbert Cell mixer.
(i) Single sideband (SSB): This assumes the only
The design of this mixer is the subject of this paper.
noise from the signal ω1 and not the image frequency
ω1-1, this would be the case if a band-pass filter was
2 INTRODUCTION added in front of the mixer eg.

Mixers are non-linear devices used in systems to trans- RF = 1694 MHz, LO = 1557MHz to give an IF of
late (multiply) one frequency to another. All mixer types 137MHz.
work on the principle that a large Local Oscillator (LO)
RF drive will cause switching/modulating the incoming Also an image IF will add to 137MHz from an RF of
Radio Frequency (RF) to the Intermediate Frequency 1420MHz ie 1557MHz-1420MHz = 137MHz
(IF).
(ii) Double sideband (DSB): In DSB both side-
The multiplication process begins by inputting two sig- bands are available thus it has twice as much power
nals: available at the IF port compared to the SSB signal. As a
result, it’s conversion loss is 3dB less than that of an
SSB signal, as shown:
a = Asin(ω 1t + φ1 ) and signal b = Bsin(ω 2 t + φ 2 )
P(IF) DSB = 2 P(IF) SSB and conversion loss is given by
The resulting multiplied signal will be:

a.b = ABsin (ω 1.t + φ1 ).sin(ω 2 .t + φ 2 ) (LC)DSB = (LC)SSB − 3(dB)

This can be multiplied out thus: or in terms of loss ratios (LC)DSB =

(LC)SSB
2
1
Using this trig identity .. sinAsinB = - [cos (A + B ) − cos (A - B )]
2

Where A = (ω 1.t + φ 1 ) and B = (ω 2 .t + φ 2 )

AB
=- [cos((ω1.t + φ1 ) + (ω 2 .t + φ2 )) − cos((ω1.t + φ1 ) − (ω 2 .t + φ2 ))]
2

AB
=- [cos((ω1 + ω 2 )t + (φ1 + φ2 )) − cos((ω1 − φ1 )t − (φ1 − φ2 ))]
2

(iii) DSB to SSB Noise Figure conversion

Sum frequency (re- Difference frequency
moved by filtering) ie I.F
 Sheet
2 of 20

1dB Compression
T
Fm = 1 + (LC - 1) point (dBm) (Output)
To
Where T = temperatu re of mixer,
To = room temperatu re (273 °K)

For DSB , (LC )DSB =

(LC )SSB IF Output
2 Power
(dBm)

⎛ (LC )SSB ⎞ T
Therefore, Fm (DSB) = 1 + ⎜ − 1⎟
⎝ 2 ⎠ To
1dB
At room temperatu re, ie T = To

Fm (DSB) =
(LC )SSB in other word s
RF Input Power 1dB Compression point
(dBm) (dBm) (Input)
2

Fm (DSB) is half or 3dB less than Fm (SSB)

Figure 1 Typical gain compression characteristic
for a non-linear Amplifier/Mixer, showing the
Fm (DSB) = Fm (SSB) − 3(dB) measurement of the 1dB compression point.

(3) Isolation: These parameters define how much signal For typical double balanced mixers this figure is ~ 6dB
leakage will occur between pairs of ports. ie RF to below the LO power level. (So performance can be im-
LO,LO to IF and RF to IF. So if for example RF to IF proved by overdriving the LO port).
isolation was specified at 35dB this means that the RF at
the IF port will be 35dB lower than the RF applied to RF The 1dB compression point gives rise to the dynamic
port. range of the mixer, which is the difference between the
1dB compression point and the minimum discernable
(4) Linearity signal (MDS – this is dependant on the noise floor of the
device).
(i) 1dB Compression point: Like other non-resistive
networks, a mixer is amplitude-nonlinear above a certain (ii) Intermodulation (IM3) performance: This pa-
input level resulting in a gain compression characteristic rameter is the same as specified for amplifiers and meas-
as shown in Figure 1. ured in a similar way. It is measured by applying two
closely spaced input tones at frequencies F1 and F2.
Above this point the If fails to track the RF input power
level – normally a 1dB rise in RF power will result in a Third order products from the mixing of these tones with
1dB rise in the IF power level. The 1dB compression the LO (at frequency FLO) occur at frequencies given by:
point is measured by plotting incident RF power against (2F1±F2)±FLO and (2F2±F1)±FLO. In the case of the
IF power as shown in the figure above. mixer, the third order products of most interest are (2F1-
F2)- FLO and (2F2-F1)-FLO as they fall in, or close to the
Most mixers have the 1dB compression point specified IF band.
at the input ie the single-tone input signal level at which
the output of the mixer has fallen 1dB below the ex- The IM3 performance is often summarised by giving the
pected output level. 3rd Order Intercept point (IM3 Intercept) as shown in
the compression characteristic of Figure 2, where the
IM3, IM5 plots intersect with the extrapolated gain plot
(blue dotted line). As a rule of thumb the IM3 intercept
point is approximately 10dB above the 1dB compression
point.
 Sheet
3 of 20

This figure of merit gives an indication of the mixer’s This results in the following equations for 1dB gain
signal handling capability. In particular it provides an compression point and IM3 [1].
indication of the levels of third order products a mixer is
likely to produce under multi-tone excitation. V 2
I =W .C . od
DSAT vsat ox V + E .L
od sat

IM3 Intercept point (dBm) Output

With

V = Voltage overdrive = Vgs - Vt and

IM3 Gradient od
=1:3
E = Velocity saturation field strength given by
sat

IF Output 2V µ
E = sat Where µ = o
Power sat eff
µ 1 + ϑ.V
(dBm) eff od
Fundamental RF
Input IM5 Gradient
2
=1:5 ⎛ µ .Vod ⎞⎛ µ .V ⎞
V .L
P1dB ~ 0.29 sat Vod ⎜ 1 + 1 ⎟⎜ 1 + 1 od ⎟ - (1)
µ .Rs ⎜ 4 V .L ⎟ ⎜ 2 Vsat .L ⎟⎠
1 ⎝ sat ⎠⎝

to convert to dBm = 10 log (1000*P1dB)

2
RF Input Power 8 Vsat .L ⎛ µ .Vod ⎞⎛ µ .V ⎞
(dBm) 1dB Compression point PIIP3 ~ Vod ⎜1 + 1 ⎟⎜1 + 1 od ⎟ - (2)
3 µ .Rs ⎜ 4Vsat .L ⎟⎠⎜⎝ 2Vsat .L ⎟⎠
(dBm) (Input) 1 ⎝

Figure 2 IM3 gain compression characteristic, as a µ ≅ µ + 2ϑ Vsat.L

1 o
rule of thumb the IM3 intercept point is approxi-
mately 10dB above the 1dB compression point.
4 GILBERT CELL MIXER[1]
The IM3 and IM5 graphs will intercept the fundamental There are two types of mixer, passive and active. Gener-
graph at the intercept point. (Note the IM2 intercept ally passive types (although have better IM3 perform-
point will be different and usually a lot higher). ance) have higher conversion losses and hence higher
noise figures than active mixers.
Again, for mixers the measurement is referred to the
input (IP3,in) and is given by:- Additionally, there are single balanced mixers and dou-
ble balanced mixers. Single balanced mixers are much
IMR less complex, but have inferior performance in terms of
IPn,in = + Input power(dBm) RF to IF and LO to IF rejection, compared to double
( n − 1)
balanced mixers.
Where IMR = Intermodulation ratio (The difference in
Given below are the advantages and dis-advantages of
dB between the desired output and spurious signal) and
double balanced mixers.
n = the IM order.
Advantages:
Typically, for double balanced mixers IM3,in is ~ 14dB
(1) Both LO and RF are balanced, providing both LO
greater than the single tone 1dB compression point and
and RF Rejection at the IF output.
~ 8dB greater than the LO power.
(2) All ports of the mixer are inherently isolated from
each other.
We can get a rough estimate of the gain compression of
(3)Increased linearity compared to singly balanced.
the LNA a non-linear expression of the input and output
(4)Improved suppression of spurious products (all even
parameters can be expanded using Taylor’s theorem.
order products of the LO and/or the RF are suppressed).
(8)High intercept points.
 Sheet
4 of 20

(9) Less susceptable to supply voltage noise due to dif-

Vcc
ferential topography.
Rload
Disadvantages:

(1)Require a higher LO drive level.

(2)Require two baluns (although mixer will usually be Differential IF
connected to differential amplifiers). Output
(3)Ports highly sensitive to reactive terminations. M4

M5
The most popular active, double balanced mixer topog-
raphy in RFIC design is the Gilbert Cell mixer, the cir-
cuit of which is shown in Figure 3. This type of mixer Differential LO Input M6
M7

exploits symmetry to remove the unwanted RF & LO

output signals from the IF by cancellation.
M2 M3
4.1 MIXER OPERATION
Differential RF Input
The RF signal is applied to the transistors M2 & M3
which perform a voltage to current conversion. For cor-
rect operation these devices should not be driven into
M1
saturation and therefore, signals considerably less than
Tail current bias
the 1dB compression point should be used. Performance adjust
can be improved by adding degeneration resistors, on Vee
the source terminals of M2 & M3.

MOSFets M4 to M7 form a multiplication function, Figure 3 Basic circuit of the Gilbert Cell Double
multiplying the linear RF signal current from M2 and balanced mixer (DBM)
M3 with the LO signal applied across M4 to M7 which
provide the switching function.
The voltage gain of the mixer with source degeneration
is given by:
M2 and M3 provide +/- RF current and M4 & M7
switch between them to provide the RF signal or the
inverted RF signal to the left hand load. M5 & M7 ⎛ ⎞
V ⎜ ⎟
switch between them for the right hand load. RF ≈ ⎜
2 R L ⎟
V π ⎜ Rs + 1 ⎟
The two load resistors form a current to voltage trans- IF ⎜ ⎟
⎝ gm ⎠
formation giving differential output IF signals.
The gain of the first stage is determined by gm given by:

4.2 DESIGN GUIDELINES 2.K N .W.I D

Depending on the application the mixer may be designed gm =
with a low SSB noise figure, a particular gain or a high L
linearity. A good starting point is to use the differential
LNA and add the switching FET’s with the same W/L The voltage overdrive level ie (Vgs-VT) should be set at
ratio’s. around 0.2 to 0.4V. Depending on the current flowing
through the ‘LNA’ RF section we can determine the
As we found with the LNA, to increase the linearity of optimum W/L ratio for the LO switching section.
the mixer source, degeneration resistors (or inductors)
can be added to M2 & M3 sources as shown in Figure
4.
 Sheet
5 of 20

Other useful design equations are:

The design criteria for the mixer, is shown in Table 2-1.
2.ID
gm =
(Vgs − VT ) Parameter Specification Units
Frequency 2.45 to 2.85 GHz
Noise Figure (DSB) < 10 dB
gm IIM3 Intercept Point (Input) >20 dBm
gm' = Where σ = R S .gm
1+ σ Voltage Gain >8 dB
Power consumption <100 mW
Source impedance 50 ohms
K N .W
.(Vgs − VT )
2
ID = Load impedance 500 ohms
2.L Voltage Supply ± 2.5 V

The design stages for the basic mixer design with high Table 2-1 Required specification for the Bluetooth
linearity are: front end Gilbert Cell mixer

(1) Select a value for Rs. If we start with the LNA We designed the differential LNA to have 50-ohm input
design , Rs will be realized by an inductor Ls. and output impedances, hence the input match of the
(2) Using the design equations of section 4(ii) de- mixer will be 50 ohms also. The 500 ohm output imped-
cide on the IM3 value required to calculate the ance of the mixer is designed to match to an ‘off-chip’
voltage overdrive (Vgs-VT). – This is part of filter, and is most easily achieved by having 500-ohm
the LNA design process. mixer load resistors.
(3) Calculate gm and check for compliance of
mixer gain with a suitable load resistance. We choose an initial Rs value of 10 ohms and using the
(4) Determine LO switching W by selecting mini- CMOS14 process results in the following analysis:
mum L allowed and assuming that the over-
drive voltage (Vgs-VT) will be between 0.2 and Using the spice model data for the Agilent CMOS14 0.5um
0.4V. we have:

L = 0.6um,
µ0 = 433 cm^2/(V*s),
To M4 to M7 θ = 0.5,
Rs = 10 ohms,
VT = 0.67V
Vsat = 1.73E5 m/s
M2 M3
First convert numbers to metric format:
Differential RF Input
Rs Rs
µ0 = 433 cm^2/(V*s), = 0.433 m^2/(V*s
L = 0.6um, = 0.6E-6m.

µ ≅ 0.433 + 2x0.1x1.73E 5 .0.6E −6 = 453

M1 1
Tail current bias
adjust The equations of section 4(ii) were entered into a
Vee
spreadsheet, along with a range of Vod from 0.01 to 5V.

From Table 2-2 we can see that to achieve a minimum

Figure 4 Modified RF input circuit of mixer with
IM3 value of 20dBm we require a VOT of 1V. As Vt for
source degeneration added for improving the line-
arity. Note in the LNA design these resistances are this process is 0.67V, the value of Vgs will be (1-0.67) =
realised as loss-less inductors Ls. 0.33V.

5 DESIGN EXAMPLE [1,2,3,4]

This section will describe the design of a Gilbert Cell
mixer that will connect to the C-S Differential LNA de-
signed for ‘Bluetooth’ described in another tutorial
 Sheet
6 of 20

Estimation of Input 1dB Compression point and IIIP3 vs Vod

Enter L 0.6 um 6E-07
Enter Vsat 2.40E+07 2.40E+07
Enter uo 4.33E+02 4.33E-02
Enter theta 0.5
Enter Rs 20 ohms u1 = 1.44E+01

Vod (V) 1dB Comp IIMP3

(dBm) (dBm)
0.01 -8.35 1.29 1.33E-03 1.45E-04 1.00250752 1.010055
0.1 2.13 11.77 1.33E-02 1.45E-03 1.02507517 1.102816
0.2 5.65 15.29 2.66E-02 2.89E-03 1.05015035 1.210662
0.3 7.90 17.54 3.99E-02 4.34E-03 1.07522552 1.323538
0.4 9.62 19.26 5.32E-02 5.78E-03 1.10030069 1.441444
0.5 11.05 20.68 6.65E-02 7.23E-03 1.12537587 1.56438
0.6 12.28 21.91 7.98E-02 8.67E-03 1.15045104 1.692346
0.7 13.37 23.00 9.31E-02 1.01E-02 1.17552622 1.825343
0.75 13.87 23.51 9.97E-02 1.08E-02 1.1880638 1.893727

Table 2-2 Calculated 1dB compression point & 3rd order intercept point.

Input referred 3rd Order Intercept Point vs Vod

40

35

30
Input 1dB comp (dBm)

25

20

15

10

0
0 1 2 3 4 5
Vod = Vgs-Vt (Volts)

Figure 5 Plot of Input referred 3rd order intercept point (IIM3 dBm) with voltage overdrive (Vgs-VT)
 Sheet
7 of 20

With voltage gain goal of 7.5dB we can calculate gm [2] Now with gm calculated we can W (Assuming we take
the minimum gate length to be 0.6um and assume a cur-
⎛ ⎞ rent of 3mA) ie.
⎜ ⎟ --- (1)
2 ⎜ RL ⎟
Conversion Gain C.G ≈
π ⎜ 1 ⎟ W
⎜ Rs + ⎟ g m 2 = 2.Kp. .IDS re - arrange to get W
⎝ gm ⎠ L

We can re-arrange to get gm for our voltage gain of

W =
gm 2 .L
=
(0.0199 ) x 0.6
2

7.5dB ie 2.Kp.I DS 2 x 171E − 6 x 3E − 3

Convert dB to conversion gain =

2.I D
W ~ 231 gm =
7.5
(Vgs − VT )
10 10 = 5 .62
∴ For a first run simulation we can assume all gate widths
⎛ ⎞ to be 231, however we need to ensure that the switching
⎜ ⎟ FETs are driven by a gate overdrive value of between
2 ⎜ 500 ⎟
5.62 ≈ 200mV to 400mV.
π ⎜ 10 + 1 ⎟
⎜ ⎟
⎝ gm ⎠
2.ID
gm =
(Vgs − VT )
Alternatively we could derive gm assuming we wish to
design for minimum noise as in the case of the LNA. To
find the optimum gate width (W) for low noise we can 2x3E- 3
gm = = 0.02 now find W from
use the following expression [3]: (0.3 )
1 W
W ≈ gm2 = 2.Kp. .IDS
opt 3ω.L.C .R
ox gen L

Where Rgen is the resistance of the source connected to

W=
gm2 .L
=
(0.02 ) x 0.6 ⇒ W ~ 233
2

the mixer input – typically we design this for 50 ohms. 2.Kp.IDS 2 x 171E − 6 x 3E − 3

Eox 3.46 * 8.85E −12 We can now run our first simulation by assuming all
Cox = = = 3.37E - 3
TOX 9.06E − 9 devices have W = 233um and L = 0.6um, tail current of
6mA, source resistors of 10 ohms and load resistors of
In this case W would be: 500 ohms.
1
W ≈
opt 3 x 2.5E 9 x2π x0.6E -6 x 3.37E −3 x50
5.1 BALUN DESIGN [5]
For the purposes of the simulation we need to convert
W ≈ 315um the differential inputs and outputs of the mixer to single
opt
For our example however, we will be designing the ended source and load impedances. The device that
mixer to have a specific gain and output load impedance, achieves this balanced to un-balanced transformation is
therefore rearranging the earlier equation (1) we can known as a ‘Balun’.
obtain gm ie
A typical balun circuit is shown in Figure 6, where the
−1 ‘Zin’ block allows the input impedance to be measured,
⎡ 2 RL ⎤
gm = ⎢ . − Rs⎥ resulting in the correct value of 50 ohms.
⎣ π Vgain ⎦
In this case the impedance transformation is 1:4 ie 50
−1 ohms to a differential impedance of 200 ohms.
⎡ 2 500 ⎤
gm = ⎢ . − 10 ⎥ = 0.0199
⎣π 5.62 ⎦ If we wish to add a load to each port then they will be
half this at 100 ohms – see example
 Sheet
8 of 20

S-PARAMETERS
S-PARAMETERS
S_Param
SP1 S_Param
Center=1.0 GHz SP1
Center=1.0 GHz
2
Span=1 GHz 2

Span=1 GHz R
1
1
-
T1
R1
Term R 1
1
-
T1 R=100 Ohm
Term1 R1 Term
Num=1 1
R=100 Ohm Term1
Z=50 Ohm Num=1 1

1
Z=50 Ohm
-
T2
3 1
-
3 T2

N TF3 3

N TF3 R
Zin TF1 R2
Zin TF1
Zin T1=1.414 R=100 Ohm
Zin T1=1
Zin1 T2=1.414
Zin1 T2=1
Zin1=zin(S11,PortZ1)
Zin1=zin(S11,PortZ1)

Figure 6 Schematic diagram of a balun transforma-

tion. In this example the grounded center tap does Figure 7 ADS example simulation of the 1:4 balun.
Note we will still get a 50 ohms input impedance is
not effect the transformation. The ‘Zin’ block allows
we disconnect the 100 ohm loads and connect a 200
the input impedance to be measured and will result ohm impedance across the output of the balun.
in the correct value of 50 ohms.

The general equation for balun transformer is:

6 SIMULATIONS [6]
The first simulation shown in Figure 10 shows a har-
2 monic balance simulation of the mixer to determine con-
Z ⎛N ⎞
OUT = ⎜ sec ⎟ version gain, DBB & SSB noise figure.
Z ⎜N ⎟
IN ⎜ prim ⎟
⎝ ⎠

Where Nsec = Number of turns on secondary winding &

Nprim = number of turns on the primary winding

Example: We have designed our differential mixer

inputs to be 100ohms. Determine the turns ratio for
the balun to match to a source impedance of 50ohms.

As we said earlier the ‘differential’ impedance will be

twice that required for each of the output ports, so we
need to transform from 50ohms to 200 ohms ie 1:4.
2
⎛N ⎞ N
200 ⎜ ⎟
= 4 = ⎜ sec ⎟ ∴ sec = 2
50 ⎜ Nprim ⎟ N
⎝ ⎠ prim

The circuit of the balun for ADS simulations is shown in

Figure 7.
 Sheet
9 of 20

IF_spectrum Eqn IF_spectrum=dBm(HB.Vout)

Spectrum at IF port Figure 8 Results from the simulation of Figure 10,
showing the output carrier spectrum, conversion
0
gain & noise figure, together with the mix table of
output carriers (explained more in the text).
-100 Note that for our example we have combined the
differential outputs using a ‘Balun’ and thus we
-200 have 3dB more conversion gain ~ 10dB where we
have designed for a single differential stage RF gain
of 7.5dB. Also note that the double sideband noise
-300 figure is ~3dB worse than the single sideband value,
as both the IF and ‘Image’ carriers are taken into
-400 account on the DSB noise figure.
0 2 4 6 8 10 12
freq, GHz In the simulation of Figure 10, balun (balanced to un-
balanced) transformers have been used to provide single
Eqn ConversionGain=dBm(mix(HB.Vout,{-1,1}))-RF_pwr ended inputs and outputs and will thus combine the dif-
freq ConversionGain ferential IF signals to give a 3dB increase in the differ-
250.0MHz 10.29 ential conversion gain ie 10dB (With the mixer designed
to give a differential gain of 7.5dB).
noisefreq NFdsb NFssb
250.0MHz 5.965 9.281
The next ADS harmonic balance simulation (Figure 12)
is set up to sweep the local oscillator input power and
Mix
freq plot the resulting DSB noise figure and conversion gain.
Mix(1) Mix(2)
0.0000 Hz 0 0 This simulation is useful when optimizing the LO
250.0MHz -1 1 switching circuit (ie W and V overdrive).
2.000GHz 2 -1
2.250GHz 1 0
2.500GHz 0 1
4.250GHz 3 -1
4.500GHz 2 0
4.750GHz 1 1
6.500GHz 4 -1
6.750GHz 3 0
7.000GHz 2 1
9.000GHz 4 0
9.250GHz 3 1
11.25GHz 5 0
11.50GHz 4 1
 Sheet
10 of 20

vcc

V_DC
I_Probe
SRC1
ID
Vdc=2.5 V
Var
Eqn VAR
VAR1
L=0.6 R R
Rs=10 R3 R4 Port
Rload=500 R=Rload Ohm R=Rload Ohm IF_Plus
W=231 Num=5
LO_W=231
RF_W=231

MOSFET_NMOS MOSFET_NMOS
MOSFET3 MOSFET4
Port Model=cmosn Model=cmosn
MOSFET_NMOS Length=L um
LO_Plus Length=L um MOSFET_NMOS Port
Num=1 Width=LO_W um MOSFET8 MOSFET7 Width=LO_W um IF_Minus
Model=cmosn vee vee Model=cmosn Num=6
vee Length=L um Length=L um vee
Width=LO_W um Width=LO_W um
Var
Port v2 Eqn VAR
LO_Minus VAR3
Num=2 Ibias=6
MOSFET_NMOS vcc
MOSFET_NMOS MOSFET6 I_DC
MOSFET1 Model=cmosn SRC7
Model=cmosn Length=L um Idc=(Ibias/10) mA
vg vee vee
Length=L um Width=RF_W um
R Width=RF_W um R
R1 R2
R=10k R=10k
R R
R5 R6 MOSFET_NMOS
C
R=Rs Ohm R=Rs Ohm C MOSFET10
C2
C5 Model=cmosn
C=10 pF
C=10 pF Length=L um
MOSFET_NMOS Width=(W/10) um
Port
Port MOSFET9 vee RF_Minus
RF_Plus Model=cmosn
Num=4
Num=3 Length=L um vee
Width=W um
V_DC vee
SRC4
Vdc=-2.5 V

BSIM3_Model
cmosn
NMOS=yes Wln=1 Cgso=2.79e-10 K3=68.279056 Vsat=1.174604e5 Eta0=0.1178659 Uc1=-5.6e-11
Idsmod=8 Ww=0 Cgdo=2.79e-10 K3b=1.252205 A0=0.9059229 Etab=2.603903e-3 Kt1=-0.11
Version=3.1 Wwn=1 Cgbo=2e-9 W0=1e-5 Keta=3.997018e-3 Dsub=0.751089 Kt2=0.022
Mobmod=1 Wwl=0 Xpart=0.5 Nlx=5.28517e-8 Lketa=-0.0143698 Drout=0.0428851 Em=4.1e7
Capmod=2 Tnom=27 Dwg=-7.483283e-9 Dvt0=6.5803089 Wketa=-5.792854e-3 Pclm=0.7319137 Xw=0
Rsh=2.8 Tox=1.01e-8 Dwb=1.238214e-8 Dvt1=0.9107896 Ags=0.1450882 Pdiblc1=2.091364e-3 Xl=-1e-7
Js=0 Cj=5.067009e-4 Nch=1.7e17 Dvt2=-0.1427458 Pags=0.0968 Pdiblc2=9.723614e-4
Lint=1.097132e-7 Mj=0.7549569 Vbm=-3.0 Ua=1e-12 B0=1.648829e-6 Pdiblcb=-0.5
Ll=0 Cjsw=4.437149e-10 Xj=1.5e-7 Ub=1.582544e-18 B1=5e-6 Pscbe1=2.541131e10
Lln=1 Mjsw=0.1 U0=433.6065339 Uc=1.831708e-11 Voff=-0.0850186 Pscbe2=5e-10
Lw=0 Pb=0.99 Vth0=0.6701079 Delta=0.01 Nfactor=1.2410485 Pvag=0.1945781
Lwn=1 Pbsw=0.99 Pvth0=8.691731e-3 Rdsw=1.28604e3 Cdsc=2.4e-4 Ute=-1.5
Lwl=0 Cjswg=2.2346e-10 K1=0.825917 Prdsw=-33.9337286 Cdscb=0 At=3.3e4
Wint=2.277646e-7 Mjswg=0.1 K2=-0.0316751 Prwg=0.0182608 Cdscd=0 Ua1=4.31e-9
Wl=0 Pbswg=0.99 Pk2=9.631217e-3 Prwb=-0.0586598 Cit=0 Ub1=-7.61e-18

Figure 9 ADS schematic of the Gilbert Cell Mixer circuit. For this simulation the gates of the RF MOSFETS 1 & 6 are
grounded (via 10K resistors that block any RF signal) and to ensure correct bias the tail MOSFET 9 is connected to a
negative supply. The tail current is set by the ‘bias’ variable – in this case the controlling MOSFET 10 has a width =
W/10, hence the bias = bias/10. The RF, LO and IF ports are terminated with a port so that the circuit will appear as a 6
port device when used in simulations as a sub-circuit..
 Sheet
11 of 20

HARMONIC BALANCE

HarmonicBalanc
HB1
MaxOrder=5
Freq[1]=LO_freq Basic_Gilbert_Cel Balun3Port
Freq[2]=RF_fre lX1 CMP2
Balun4Port
Order[1]=5 CMP3
Order[2]=1 IF_Plus

InputFreq=RF_fre LO_Plus

IF_Minus
Vout
NLNoiseMode=ye Term
FreqForNoise=IF_fre
RF_Minus

LO_Minus
Term3
NoiseNode[1]="Vout RF_Plus
Num=3
"UseKrylov=yes Z=500 Ohm
EquationName[1]
V_DC
SRC6
Var VAR Vdc=vg V
Eqn VAR6 vin
P_1Tone P_1Tone
LO_pwr=5 PLO
Balun3Port PRF
vg=1.0 Num=1 CMP1 Num=2
LO_freq=2250 Z=50 Ohm Z=50 Ohm
MH
RF_freq=2500 P=dbmtow(LO_pwr) DC
P=dbmtow(RF_pwr)
IF_freq=250
MH Freq=LO_freq
DC Freq=RF_freq
RF_pwr=-30
MH
dB DC1

Figure 10 Harmonic balance simulation of the Gilbert Cell Mixer. As the inputs and outputs are differential balun trans-
formers have been added to convert to single ended inputs and outputs. The 500-ohm load Term3 correctly terminates
the mixer 500-ohm output impedance. The RF frequency is set to 2500MHz (RF_freq), Local oscillator frequency to
2250MHz (LO_freq) , resulting in an IF frequency of 250MHz (IF_Freq).. For correct switching of the LO transistors
the variable vg needs to be set to 1V – running the simulation this gives Vgs across the switching transistor of ~ 1V.
 Sheet
12 of 20

freq Mix
The resulting plots of DSB noise figure and conversion LO_pwr=-20.000, Mix(1)
0.0000 Hz 0
gain vs local oscillator input power are shown in 250.0MHz 0
(Figure 11). 2.000GHz 1
2.250GHz 1
DSB Noise Figure 2.500GHz 1
nf(3)[0]
4.250GHz 2
30 4.500GHz 2
4.750GHz 2
6.500GHz 3
25 m2 6.750GHz 3
LO_pwr= 5.000 7.000GHz 3
20 nf(3)[0]=9.277 8.750GHz 4
9.000GHz 4
9.250GHz 4
15 11.00GHz 5
m2 11.25GHz 5
10 11.50GHz 5

LO_pwr=-20.000, Mix(2)
5 0.0000 Hz 0
-20 -15 -10 -5 0 5 10 250.0MHz 1
2.000GHz -1
LO_pwr 2.250GHz 0
2.500GHz 1
Conversion Gain 4.250GHz -1
4.500GHz 0
15 4.750GHz 1
m1 6.500GHz -1
10 6.750GHz 0
7.000GHz 1
5

Table 2-3 From the simulation of Figure 12 a table was gen-

0 m1
erated giving values of mix(1) and mix(2) and their associ-
LO_pwr= 5.000
ated frequencies. Thus if we select vout with and index of
-5 real(Vout_tone)=10.281
{0,1} we will be selecting the IF frequency at 250MHz.
-10 The plots of Figure 11 show us that to maintain opti-
-20 -15 -10 -5 0 5 10 mum conversion gain (and minimum DSB noise figure)
real(Vout_tone) LO_pwr we need an LO drive power in excess of 0dBm.

For the 3rd ADS simulation we want to simulate the

Figure 11 Results from the simulation shown in Figure gain compression characteristics of the mixer. The simu-
12, showing DSB noise figure and conversion gain vs lation for this is shown in Figure 15 with the resulting
LO drive level. plots shown in Figure 13. The top plot shows the
graphical gain compression characteristic. Here we have
a plot of linear gain shown in blue (Line) and a plot of
the compressing IF gain response. As the input RF is
increased the IF gain begins to decrease as the mixer
saturates. Where the IF gain deviates by 1dB from the
linear gain is the gain compression point. Markers 1 & 2
differ by 1dB for the same input power of –4.5dBm. The
output power at gain compression given by marker 3 is
4.382dBm.

The variable ‘inpwr’ is generated from the ‘Gain com-

pression simulation box giving a slightly higher input
power level of –3.189dBm. To calculate the output
power at gain compression we add the linear gain (Gain
variable) and subtract 1dB (as the IF plot has dropped
1dB from the linear gain value).
 Sheet
13 of 20

Meas
MeasEqn
HARMONIC BALANCE Eqn
meas1
HarmonicBalance Vout_tone=dBm(mix(Vout,{0,1}))-RF_pwr[0]
HB1 Var
VAR
Eqn
Freq[1]=LO_freq VAR6
Order[1]=5 LO_pwr=5
SS_MixerMode=yes vg=1.0
SS_Freq=IF_freq LO_freq=2250 MHz
UseAllSS_Freqs=yes IF_freq=250 MHz
MergeSS_Freqs=yes RF_pwr=-30 _dBm Basic_Gilbert_Cell Balun3Port
InputFreq=LO_freq+IF_freq X1 CMP2
Balun4Port
FreqForNoise=IF_freq CMP3
NoiseNode[1]="Vout" IF_Plus

SortNoise=Sort by value LO_Plus

IF_Minus
Vout
SweepVar="LO_pwr"
Start=-20
RF_Minus Term
LO_Minus
Stop=10 RF_Plus Term3
Step=1 Num=3
Other=OutVar="RF_pwr" Z=500 Ohm
V_DC
SRC6
Vdc=vg V
P_1Tone
PLO
Balun3Port
Num=1
CMP1
Z=50 Ohm vin
P=dbmtow(LO_pwr) P_1Tone
Freq=LO_freq PRF
Num=2
Z=50 Ohm
Freq=
P_USB=dbmtow(RF_pwr)

Figure 12 This ADS Harmonic simulation is setup to sweep the Local oscillator power (LO_pwr) from –20dBm to
+10dBm in 1dBm steps. The RF input frequency = LO_freq (2250MHz) + IF_freq (250MHz) = 2500MHz (Set to a power
(RF_pwr) of –30dBm. This simulation is slightly different to the one shown in Figure 10 in that upper sideband is se-
lected (by setting the power variable P_USB in ‘PRF one tone source). In the Harmonic simulator box, setting
SS_Mixer_Mode to ‘yes’ causes the simulator to solve for all small-signal mixer sidebands. This default option requires
more memory but delivers more accurate results. The measurement equation Vout_tone selects the IF frequency (in
dBm) using the mix function then substracts the RF_pwr (at fundamental frequency) to give conversion gain. In order to
select the IF frequency at 250MHz we need to know what the mix terms are, the first value 0 is mix(1) and the second
value 1 is mix(2) if you plot a table of mix(1) and mix(2) you will find this equates to an output carrier at 250MHz – this
is shown in Table 2-3.
 Sheet
14 of 20

HB.P_IF the resistor Rs has an effect on the gain compression and

10 IM3 (due to de-generation) in which gain is exchanged
m3 for linearity [3].
RFPwr=-4.500 m1
HB.P_IF=4.382 m3 The simulated results agrees with the predicted IIM3
5 1dB shown in Table 2-2, however the resulting IIM3 and
Line gain compression point not only depends on the voltage
overdrive of the input transistors but also (to a lesser
extent) the switching transistors. To achieve the required
0
linearity the bias current was increased from 6mA to
9mA. This of course will degrade the DSB noise figure
slightly and was now found to be 10.3dB (from 9.2dB).
-5 m1
RFPwr=-4.500 m3
Line=5.303 freq=250.0MHz
Spectrum=-21.268 Eqn Spectrum=dBm(Vout)
-10 Broadband Spectrum
m3
0
-30 -25 -20 -15 -10 -5 0 5 -50
-100 Spectrum
RFPwr
-150
Eqn Line=HB1.HB1.HB.RFPwr+Gain[0] Gain[0] -200
9.803 -250
-300
-350
Eqn Gain=10*log10(HB.P_IF-HB1.HB1.HB.RFPwr) -400
-450
Output_Gain_Comp inpwr -500
5.613 -3.189 dBm 0 2 4 6 8 10 12
freq, GHz
Eqn Output_Gain_Comp=Gain-1+inpwr[0] Spectrum Near IF Frequency
0 m1
-10 m1
-30
freq=250.0MHz
dBm(Vout) dBm(Vout)=-21.268
Figure 13 The gain compression from the simulation
-50
shown in Figure 15. The Line equation generates a linear
-70
gain line shown in blue. The red plot shows the non-linear m2
-90 freq=250.2MHz
IF gain compression characteristic. Markers1 & 3 are
-110 m2 dBm(Vout)=-123.700
placed at the same input power where the two lines di-
verge by 1dB. The input power at this point is defined by -130
marker3 at –4.5dBm. Using the ‘Gain compression’ result -150
– variable ‘inpwr’ yields a slightly higher value of – 249.5 250.5
3.19dBm. The output power where gain compression oc- freq, MHz
curs is simply inpwr plus the linear gain –1dB (as the ‘lin-
ear’ gain at this point has dropped by 1dB). TOIoutput
Eqn TOIoutput=1.5*m1-0.5*m2 29.948
th rd
The 4 and final ADS simulation calculates the 3 Or-
der intercept point of the mixer. From the gain compres- ConvGain Input_IM3 Output_IM3
8.732 21.216 29.948
sion simulation we can estimate the IM3 intercept point
to be ~ 10dB above the Output Gain compression point
– in this case 10 + 4.382 = 14.38dBm (or 15.61dBm if
you take the other calculation). Figure 14 Plot of the resulting IM3 simulation shown in
Figure 16. The first plots show all the mixing products and
Again there is another useful simulation tool that will harmonics attenuated by the IF band-pass filter. The sec-
ond plot shows the two input carriers together with their
directly calculate the IM3 of the mixer ie the IP3out
IM3 products, resulting in an intercept point of ~ 30dBm
function. using a graphical method. The table shows the direct cal-
culation of the Intercept point using the IP3out simulation
So far we have designed the mixer to achieve our speci- box in conjunction with the Harmonic balance simulator.
fication in terms of DSB noise figure, gain, output im-
pedance. But there appears to be some difference in the
calculated IM3 and the simulated IM3. We know that
 Sheet
15 of 20

HARMONIC BALANCE GAIN COMPRESSION

HarmonicBalance XDB
HB1 HB2
MaxOrder=4 Var VAR MaxOrder=4
Eqn
Freq[1]=Flo MHz VAR1 Freq[1]=Flo MHz
Freq[2]=Frf MHz LOPwr=0 Freq[2]=Frf MHz
Order[1]=7 RFPwr=0 Order[1]=7
Order[2]=3 Frf=2500 Order[2]=3
UseKrylov=yes Flo=2250 UseKrylov=yes
SweepVar="RFPwr" Fif=Frf-Flo GC_XdB=1
Start=-30 GC_InputPort=2
Stop=6 Meas MeasEqn GC_OutputPort=3
Eqn
Step=0.25 meas2 GC_InputFreq=Frf MHz
P_IF=dBm(mix(Vout,{-1,1})) GC_OutputFreq=Fif MHz
GC_InputPowerTol=1e-3
GC_OutputPowerTol=1e-3
GC_MaxInputPower=100
Balun4Port Basic_Gilbert_Cell
X1 Balun3Port
CMP3
IF_Plus
CMP2
LO_Plus
IF_Minus Vout
Term
RF_Minus
LO_Minus Term3
RF_Plus
Num=3
Z=500 Ohm

V_DC
SRC6
P_1Tone Vdc=vg V
PORT1 P_1Tone
Num=1 Var VAR
Eqn PORT2
VAR6 Balun3Port
Z=50 Ohm Num=2
P=dbmtow(LOPwr) vg=0.4 CMP1
Z=50 Ohm
Freq=Flo MHz P=dbmtow(RFPwr)
Freq=Frf MHz

Figure 15 ADS simulation to derive the gain compression characteristic of the mixer. The P_IF function produces and
ideal straight line of linear gain. Thus when the compressed gain of the mixer is added to the graph, the point at which
the two lines diverge by 1dB is the gain compression point. To simply the process of finding the gain compression point
the ‘Gain Compression’ simulation block is added to the schematic. The two Input frequencies are declared ie RF & LO,
together with the output frequency Fif. The ports have also to be defined, in this case the input port (RF) is port2, and
the output port (IF) is port 3.
 Sheet
16 of 20

VAR Var
Eqn
HARMONIC BALANCE VAR1
HarmonicBalance RF_freq=2500 MHz
HB2 Rs=20
MaxOrder=4 vg=0.9
Freq[1]=LO_freq LO_freq=2250 MHz BPF_Chebyshev
Freq[2]=RF_freq+fspacing/2 IF_freq=RF_freq-LO_freq BPF3
Freq[3]=RF_freq-fspacing/2 fspacing=100kHz Fcenter=IF_freq
Order[1]=3 RF_pwr = -30 dBm BWpass=5 MHz
Order[2]=3 Basic_Gilbert_Cell Balun3Port Apass=0.5 dB
Order[3]=3 X1 CMP2 Ripple=0.1 dB
UseKrylov=yes N=3
IF_Plus
LO_Plus
IF_Minus

RF_Minus
Vout
P_1Tone LO_Minus
RF_Plus

PORT3
Num=3 Balun4Port Term
Z=50 Ohm CMP3 Term3
P=dbmtow(5) Num=3
Freq=LO_freq Z=500 Ohm
Balun3Port
V_DC CMP1
SRC6
Vdc=vg V P_nTone
PORT1
Pni

IP3out

Num=1
IP3out Z=50 Ohm
P0

ipo1 BPF_Chebyshev Freq[1]=RF_freq+fspacing/2

Output_IM3=ip3_out(Vout,{-1,1,0},{-1,2,-1}) BPF2 Freq[2]=RF_freq-fspacing/2
Pout=dBm(mix(Vout,{-1,1,0})) Fcenter=RF_freq P[1]=dbmtow(RF_pwr)
ConvGain=Pout-RF_pwr BWpass=RF_freq/20 P[2]=dbmtow(RF_pwr)
Input_IM3=Output_IM3-ConvGain Apass=0.5 dB
Ripple=0.5 dB
N=3

Figure 16 ADS simulation to determine the 3rd order intercept point of the mixer. In this simulation two RF carriers are
set to one 50kHz below the RF_freq of 2500MHz and the other one 50kHz above the RF_freq. The spacing of the carri-
ers set by fspacing is therefore 100kHz. The clever part of the simulation is the IP3out where the IM3 output intercept
point (dBm) is calculated from the two IF frequencies at 250MHz + 50KHz and –50KHz. Note the filter is added to the
IF to filter out the effects of RF breakthrough.
 Sheet
17 of 20

Luckily noise and IIM3 are fairly independent of each There are a number of issues when using an inductor Ls
other as IIM3 is not particularly sensitive to L and W, so one being the circuit will now be more frequency de-
important for noise optimization is not a variable of pendant. A second issue is that inductors used in CMOS
IIM3! tend to have low Quality (Q) factors – typically 2 to 3.
Thirdly the addition of the source inductor will have a
It is of course possible to improve the linearity of the significant effect on the input impedance (Re) and will
mixer by adding voltage follower buffer stages on each need to be checked to ensure the impedance is not nega-
output, this will have the added advantage in that the tive (which may cause instability!).
frequency response of the mixer will not be altered by
the following stage input capacitance (assumimg induc- We can set up a S-parameter simulation to alter the
tive loading!). value of Ls to determine the input resistive impedance.
To ensure the simulation will work the switching tran-
We can now compare our simulation results to that of sistors are biased on using two voltage sources.
the specification. The table below (Table 4) shows the
original key parameters. The S-parameter simulation is shown in Figure 18 with
the resulting impedance plot shown in Figure 17.
Parameter Specification Predicted Units
Frequency 2.45 to 2.85 2.45 to 2.85 GHz
Noise Figure < 10 10.3 dB m1
IM3 Intercept >20 21.2 dBm real(Zin1) freq=2.503GHz
Point (Intput) real(Zin1)=51.008
Voltage Gain >8 8.2 dB Ls=1.080000
120
Power consump- <100 90 mW
tion
100
Source impedance 50 50 ohms
Load impedance 500 500 ohms 80
Voltage Supply ± 2.5 ± 2.5 V
60
m1
Table 4 Predicted performance of the mixer against 40
the original specification.
20
7 DESIGN COMPLETION 0
The design so far, generally meets the specification
2.0 2.2 2.4 2.6 2.8 3.0
given and the design process has shown how gain, noise
figure and linearity are interdependent and thus trade freq, GHz
offs need to be made at the beginning of the design Figure 17 Input impedance result from the simulation
process. In particular the fact reducing the W/L ratio shown in Figure 18. The value of the source de-generating
increase the overdrive voltage and hence linearity but inductor (Ls) has been varied between 0.5nH and 5nH. We
reduces the gain and increases the noise figure! see that with an inductance of ~1.1nH the input impedance
is 50 ohms.
There are a number of things that can be done to im-
prove the design including replacing Rs with a lossless With a source de-generation inductor of value 1.1nH
inductor, using separate tail current sources. results in an input impedance of ~ 50 ohms (Rs
=18ohms). If 100 ohm input impedance is required (If
7.1 LOSSLESS SOURCE DEGENERATION connecting to a balun) then Ls = 3nH.
We saw how the introduction of the source resistor Rs
greatly improved the linearity of the mixer. However, The simulation shows that the mixer circuit is stable as
using a resistor will cause a voltage drop and hence an the value of Re is positive and NOT negative (the result-
needed increase in the voltage rails to ensure the devices ing input return loss S11 is negative as it should be)
remain in saturation. Another important problem with
using a resistor is the noise it will generate.

The solution is to use a lossless component – an induc-

tor.
 Sheet
18 of 20

PARAMETER SWEEP

ParamSweep
V_DC
Sweep1
SweepVar="Ls" SRC7
SimInstanceName[1]="SP1" Vdc=vg V
SimInstanceName[2]=
SimInstanceName[3]=
SimInstanceName[4]= Basic_Gilbert_Cell Balun3Port
SimInstanceName[5]= X1 CMP2
SimInstanceName[6]=
Start=0.1 IF_Plus

Stop=5 LO_Plus
Vout
IF_Minus
Step=
RF_Minus Term
LO_Minus
RF_Plus Term2
N Num=2
Zin Z=500 Ohm
Zin
Zin1 V_DC
Zin1=zin(S11,PortZ1) SRC6
Vdc=vg-0.5 V
Var VAR
Eqn
VAR6 Term
LO_pwr=5
S-PARAMETERS Term1
Ls=1 Term Num=1
Rs=18 S_Param Term3 Z=50 Ohm
vg=1.0 SP1 Num=3
LO_freq=2250 MHz Center=2.5 GHz Z=50 Ohm
IF_freq=250 MHz Span=1 GHz
RF_pwr=-30 _dBm

Figure 18 S-Parameter simulation to determine the effect of varying the source de-generation inductance (Ls) on the in-
put impedance (Re). The Parameter sweep box is set to vary the variable Rs from 0.1 to 5nH and applies to the simulator
‘SP1’. This is to allow an optimum value of Ls to be used for an input match ie Ls = 1nH for 50 ohms and Ls=3nH for
100 ohms
 Sheet
19 of 20

When we add an inductor we are in fact adding an in- In order to ensure a stable design (ie with no risk of os-
ductor with an added series resistance dependant on the cillation) it was highlighted that the input impedance
Q (Quality factor) of the inductor: should be checked to ensure the input impedance is posi-
tive and not negative.
ω. L Finally, an alternative design scheme was given to allow
Q = a source resistor to be used without degrading the volt-
2. R age swing capability of the mixer, by using two separate
current sources in lieu of the single ‘tail’ current source
For example at 2.5GHz the series R will be ~ 23ohms in the conventional design.
for a 5nH inductor.

Therefore, when simulating the circuit with a source

degenerating inductor we use the INDQ model if we 9 REFERENCES
know the unloaded Q of the inductor (Depends on the
topography & process) or if we wish to determine the Q [1] Thomas Lee, “The Design of CMOS Radio-
required we can use an ideal inductor in series with a Frequency Integrated circuits”, Cambridge University
resistor (SRL series resistor inductor model). Press, second edition 2004, ISBN 0-521-835389-9,
Chapter 13.
7.2 ALTERNATIVE DE-GENERATION SCHEME
[2] T Soorapanth, T.H Lee, “RF Linearity of Short-
Channel MOSFETs”, IEEE Journal of Solid State Cir-
The design to date, has used resistors followed by induc-
cuits, vol. 32, no. 5, May 1997.
tors to provide source de-generation in order to meet the
given linearity specification. The source resistor (Rs)
[3] Pham B, “A 1.9GHz Gilbert Mixer in 0.18u CMOS
was substituted with a source inductor (Ls) to improve
for a cable tuner”, Department of Electronics, Carleton
the voltage swing of the mixer. There is however, an-
University, Canada, 2002-2003
other topography that uses a de-generating source resis-
tor but doesn’t degrade the voltage swing [7]. And the
[4] S Long, Agilent EEsof EDA - DesignSeminar: RFIC
circuit is shown in Figure 19. In this configuration there
MOS Gilbert Cell Mixer Design, June 1999
is one de-generating resistor connected between the two
sources, together with a separate current source. (Note
[5] P Vizmuller, “RF Design Guide – Systems, Circuits
that for best performance theses current sources need to
and Equations”, Artech House, 1995, ISBN 0-89006-
be well matched).
754-6, p69 - p76.
Each of the two current sources set the current through
[6] ADS – Advanced Design System, RF/Microwave
each side of the differential stage and are set to Itail x 2.
CAD, Agilent EEsof EDA, http://eesof.tm.agilent.com/
As the potential at each source should be the same
(hence the need for well-matched current sources) there
[7 Thomas Lee, “The Design of CMOS Radio-
will be no voltage drop across the de-generating resistor
Frequency Integrated circuits”, Cambridge University
(Rs), thus no degrading the voltage swing of the mixer.
Press, second edition 2004, ISBN 0-521-835389-9,
p425.
8 CONCLUSION
This tutorial has described the design process of a MOS
Gilbert cell (differential) mixer with ADS simulations
given to predict the various circuit parameters of gain,
noise figure and linearity
The design process depends on the applications key
driving parameter, where the mixer could be designed
for optimized Gain/Output load, DSB Noise figure or
Linearity.
As with the differential LNA, the design equations for
predicting Intercept point & Compression point where
given and showed that this parameter could be opti-
mized, by using source de-generation.
 Sheet
20 of 20

vcc

V_DC
I_Probe
SRC1
ID
Vdc=2.5 V

Var
Eqn VAR
VAR1
L=0.6 R R
Rload=500 R3 R4 Port
W=220 R=Rload Ohm R=Rload Ohm IF_Plus
LO_W=270 Num=5
RF_W=150

MOSFET_NMOS MOSFET _NMOS

MOSFET3 MOSFET 4
Port Model=cmosn Model=cmosn
MOSFET _NMOS Length=L um
LO_Plus Length=L um MOSFET _NMOS Port
Num=1 Width=LO_W um MOSFET 8 MOSFET 7 Width=LO_W um IF_Minus
Model=cmosn vee vee Model=cmosn Num=6
vee Length=L um Length=L um vee
Width=LO_W um Width=LO_W um
Port v2 Var
Eqn VAR
LO_Minus VAR3
Num=2 Ibias=4.0
MOSFET _NMOS vcc
MOSFET_NMOS MOSFET 6 I_DC
MOSFET1 Model=cmosn SRC7
Model=cmosn Length=L um Idc=(Ibias/10) mA
vg vee vee
Length=L um Width=RF_W um
Width=RF_W um
DC_Feed DC_Feed
DC_Feed1 DC_Feed2

MOSFET _NMOS
C
R C MOSFET 10
C2
R5 C5 Model=cmosn
C=10 pF vee R=Rs Ohm C=10 pF Length=L um
vee Width=(W/10) um
Port
Port
RF_Minus
MOSFET _NMOS
RF_Plus
Num=4
MOSFET 11
Num=3 MOSFET_NMOS vee
Model=cmosn MOSFET9 V_DC
Length=L um Model=cmosn SRC4 vee
Width=W um Length=L um Vdc=-2.5 V
Width=W um

BSIM3_Model
cmosn
NMOS=yes Wln=1 Cgso=2.79e-10 K3=68.279056 Vsat=1.174604e5 Eta0=0.1178659 Uc1=-5.6e-11
Idsmod=8 Ww=0 Cgdo=2.79e-10 K3b=1.252205 A0=0.9059229 Etab=2.603903e-3 Kt1=-0.11
Version=3.1 Wwn=1 Cgbo=2e-9 W0=1e-5 Keta=3.997018e-3 Dsub=0.751089 Kt2=0.022
Mobmod=1 Wwl=0 Xpart=0.5 Nlx=5.28517e-8 Lketa=-0.0143698 Drout=0.0428851 Em=4.1e7
Capmod=2 Tnom=27 Dwg=-7.483283e-9 Dvt0=6.5803089 Wketa=-5.792854e-3 Pclm=0.7319137 Xw=0
Rsh=2.8 Tox=1.01e-8 Dwb=1.238214e-8 Dvt1=0.9107896 Ags=0.1450882 Pdiblc1=2.091364e-3 Xl=-1e-7
Js=0 Cj=5.067009e-4 Nch=1.7e17 Dvt2=-0.1427458 Pags=0.0968 Pdiblc2=9.723614e-4
Lint=1.097132e-7 Mj=0.7549569 Vbm=-3.0 Ua=1e-12 B0=1.648829e-6 Pdiblcb=-0.5
Ll=0 Cjsw=4.437149e-10 Xj=1.5e-7 Ub=1.582544e-18 B1=5e-6 Pscbe1=2.541131e10
Lln=1 Mjsw=0.1 U0=433.6065339 Uc=1.831708e-11 Voff=-0.0850186 Pscbe2=5e-10
Lw=0 Pb=0.99 Vth0=0.6701079 Delta=0.01 Nfactor=1.2410485 Pvag=0.1945781
Lwn=1 Pbsw=0.99 Pvth0=8.691731e-3 Rdsw=1.28604e3 Cdsc=2.4e-4 Ute=-1.5
Lwl=0 Cjswg=2.2346e-10 K1=0.825917 Prdsw=-33.9337286 Cdscb=0 At=3.3e4
Wint=2.277646e-7 Mjswg=0.1 K2=-0.0316751 Prwg=0.0182608 Cdscd=0 Ua1=4.31e-9
Wl=0 Pbswg=0.99 Pk2=9.631217e-3 Prwb=-0.0586598 Cit=0 Ub1=-7.61e-18

Figure 19 Gilbert Cell mixer with source de-degeneration (to improve linearity) but does not degrade the voltage swing
of the mixer. The two ‘tail’ current sources need to be well matched for best performance.