Download as pdf or txt
Download as pdf or txt
You are on page 1of 107

Design of CMOS Mixer

2021. 01. 08

Chang-Wan Kim
cwkim@dau.ac.kr

Department of Electronics Engineering


Dong-A University
Contents
• Introduction
-. Basic operation RF IF
• Mixer Topologies
-. Single-and Double-balanced Mixer
-. Voltage passive mixer
LO
• Noise Analysis in SBM
-. Thermal noise and flicker noise analysis
• Passive Mixer
-. Current-driven mixer, BPF
-. Mixer examples
• Summary

2
Zero-IF RF Transceiver
Down-Mixer VGA LPF

LNA A/D

A/D

VCO
Freq
/2 PLL
Divider

D/A

DA D/A

Up-Mixer VGA LPF

 Frequency up/down-conversion
 Three ports (RF, IF, LO)
 LNA, BB amp, /2-circuit
3
Basic Function
Down-conversion mixer

AS cos(RF  LO )t AS cos(RF  LO )t


VIF  
2 2
VRF  AS cos RF t

VLO  cos  LO t

Up-conversion mixer

VIF  AS cos IF t

AS cos(IF  LO )t AS cos(IF  LO )t


VRF  
2 2

VLO  cos LO t 4


Basic Operation in Time Domain
• The mixer operation : Multiplication of the RF signal by a rectangular
waveform (LO signal). implementation
VRF  AS cos RF t
S1

VLO
1

1 2 1 1 1 1 
t  
VLOx(t)   cos t  cos 3t  cos 5t  cos 7t  cos 9t  ... 
2  3 5 7 9 
LO harmonics 5
Basic Operation in Frequency Domain
• Since multiplication in time domain is convolution in the frequency
domain, we have Y ( f )  X ( f ) * V LO ( f )

 
VLO (t )   cn e jno t
V LO ( f )   cn ( f  nf 0 )
 

x(t )  As cos  s t X(f )

• After convolution of two signal, the conversion gain is “C1”


    
Y ( f )      cn ( f  nf 0 )  X ( f   ) d 
     
 
  cn  [ ( f  nf 0 ) X ( f   )]d 
 

  cn X ( f  nf 0 ) ; Frequency shift of X(f)


6
Frequency Domain
“C1”

|fIF| |fIF|

X(f)
f
-fRF dc fRF = fLO+|fIF| f

VLO ( f ) *
f
-fLO dc fLO

||

Y( f )
f
-fIF fIF
dc

7
LTV System
Modulated input voltage signal : vIN (t )  A(t ) cos( IN t   (t ))
LO voltage signal (single tone) : vLO (t )  ALO cos  LO t

VIN VOUT
here, vIN (t )  A(t ) cos( IN t   (t ))
 A(t ){cos  IN t cos  (t )  sin  IN t sin  (t )}
VLO

A(t ) ALO A(t ) ALO


vout (t )  cos((IN t  LOt )   (t ))  cos((IN t  LOt )   (t ))
2 2

• After frequency conversion, amplitude and phase information are still


preserved.
• We see that the modulated signal is translated to two new frequencies
(INLO) and (IN+ LO) not present at the input
; a time-variant (TV) system.
• But, mixer is a linear circuit with respect to the input RF and IF signal.
• Therefore, mixers are a LTV system.
8
Mixer Conversion Gain
• Voltage Conversion Gain is defined by

Vrms ,IF
VCG 
Vrms ,RF

• Power Conversion Gain is defined by

The IF power delivered to the load


PCG 
The available RF power from the source

• Only if the input impedance and the load impedance of the mixer are
equal each other, then the VCG and PCG are equal when expressed
in decibels.

9
Mixer Image Problem
• Even with ideal mixer, receiver suffers from image problem

• Example)
fLO = 1 GHz
fRF = 0.9 GHz and fIM = 1.1 GHz (image frequency)
Image degrades the noise figure of the receiver
 See next slide.

• Solutions)
-. Image reject filter ahead of mixer
-. Higher IF
-. Image rejection mixer

-. IQ demodulation

10
Mixer Image Problem
• Image problem (like IIP2, IIP3) considerably degrades the Rx sensitivity.

|fIF||fIF| |fIF||fIF|

f
-fRF -fimage dc fimage fRF f

*
f
-fLO dc fLO

||

f
-fIF fIF
dc

11
Mixer Noise Figure: SSB
• Consider a noiseless mixer with unit gain.
• The single-sideband noise figure (SSB NF) of a noiseless mixer is
equal to 3-dB.

SNRIN | at RF freq.
F 
SSBNF
SSB
SNROUT | at IF freq.

Sin Sin
N N in
SSB F  in  2
Sout 1  Sin
N out 1 Nin  N in

• The “SSB” means that the desired signal spectrum resides on only
one side of the LO, a common case in heterodyne systems

12
Mixer Noise Figure: DSB
• For direct-conversion systems, the double-sideband noise figure
(DSB NF) of a noiseless mixer is equal to 0-dB.

SNRIN | at RF freq.
DSB F
DSBNF
SNROUT | at IF freq.

Sin Sin
N N in
DSB F  in  1
Sout 1  Sin
N out 1 N in

• The “DSB” means that the desired signal spectrum resides on both
sides of the LO.
• The SSB NF = The DSB NF + 3-dB.

13
Noise Folding

-5f0 -3f0 -f0 f0 3f0 5f0

> 80%

• RF noises around nfLO are down converted to the IF band, further


increasing the output noise.
• Nonetheless, this effect is not serious because of
-. The limited RF input bandwidth.
-. The smaller magnitude of higher LO harmonics.
14
Mixer Port-Port Isolation
• The LO RF feed-through (dc-offset, LO leakage)
• The LO  IF feed-through (saturating IF stage)
• The RF  IF feed-through (serious in passive mixer)

RF IF

LO

15
Typical Mixer Specifications (Active mixer)
• Noise figure: 7 ~ 12 dB typical
• Linearity(IIP3/IIP2): -5 ~ 5 dBm/ > +45 dBm typical
• Conversion gain: 5 ~ 15 dB typical
• LO leakage
• LO input power: -3 ~ 0 dBm typical
• Port matching : RF port typical
• Port isolation
• Power consumption

16
Mixer Topologies
In terms of Conversion Gain, VDD

R R
 Active Mixer
-. Transconductor(driver) + switching stage VIF+ VIF-

-. dc-consumption (flicker noise) LO+ LO-


-. conversion gain, moderate NF/ linearity.
-. moderate LO power (3 ~ +0dBm)
VRF
iRF

 Passive mixer
-. only switching stage
-. no dc-current (less flicker noise) ILO+

-. conversion loss, high linearity


iRF
-. Higher LO power (+5 ~ +10dBm)
iIF

ILO-

17
Mixer Topologies
In terms of Balancing structure

 Double-balanced
-. Differential RF and LO
-. Gilbert-cell mixer
-. LO & RF isolation at the output

 Single-balanced
-. One of RF or LO is differential
-. LO or RF isolation at the output

 Unbalanced
-. RF and LO driven single-ended
-. Dual-gate mixer, exponential mixer, Square-law mixer

18
Active Mixer : SBM & DBM

iO1 iO2
iO1 iO2

M2 M3 M3
iD2 iD3 iD4 iD5 iD6
+ + iD3
vLOcosLOt M4 M5 M6
VLO cosLOt
- -
M1 M1 iD1 iD2 M2
iD1
VRF cosRFt +
vRFcosRFt
-

19
Single-Balanced Mixer (1/5)
• The single-balanced mixer accommodates a differential LO
but a single-ended RF signal.
iD2 SD2
1
iO1 iO2
t
M2
iD3 SD3
M3
iD2 iD3 1
+
VLO cosLOt
iD1 t
-
IC1
M1
iD1
VRF cosRFt
t
iD1  I D1  g m vRF cos RF t
- Assuming ideal 1 2 1 1 
switching SD2   sin LO t   sin(3LO t )  sin(5LO t )..
2  3 5 
1 2 1 1 
SD3    cos LO t   cos(3LO t )  cos(5LO t )..
2  3 5 
20
Single-Balanced Mixer (2/5)
I D1 g m vRF 2I 2g v
iO1  iD1  S D 2 
 cos  RF t  D1 cos  LO t  m RF cos  RF t cos  LO t  ...
2 2  
I g v 2I g v g v
 D1  m RF cos  RF t  D1 cos  LO t  m RF cos LO   RF   m RF cos LO   RF   ...
2 2   

I D1 g m vRF 2I 2g v
iO 2  iD1  S D 3 
 cos  RF t  D1 cos  LO t  m RF cos  RF t cos  LO t  ...
2 2  
I g v 2I g v g v
 D1  m RF cos  RF t  D1 cos  LO t  m RF cos LO   RF   m RF cos LO   RF   ...
2 2   

4 I D1 2 2
iO  iO1  iO 2  cos  LO t  g m vRF cos LO   RF   g m vRF cos LO   RF   ...
  

21
Single-Balanced Mixer (3/5)
• Current conversion loss
-. 2/π ( -3.9 dB) by the frequency conversion
-. RF power converted to IF, RF+IF, and higher order,
leading to conversion loss.

• No LO isolation at IF
-. Differential LO leakage
-. LO can saturate IF output  can degrade P1dB of the next stage
-. Parallel capacitors at the IF helps to suppress LO.

22
Single-Balanced Mixer (4/5)
• When M1 is turn-on, the mixer operates as the cascade amplifier.
-. M1 should work in the saturation region, not the triode region.
-. LO power normally -3 ~ +0 dBm. V DD
VDD

R R R R

VIF+ VIF- VIF+ VIF-


LO+ LO-
M1 M2
LO+ LO-
M1 M2
VRF M3
vRF cos RF t VRF id  g mvRF cos RF t
vRF cos RF t
M3

23
Single-Balanced Mixer (5/5)
1 1

iBB  gm vRF cos IF t

iBB  g m vRF cos IF t
 
VDD

R R

1 1
 g m RD vRF cos IF t  VIF VIF   g m RD vRF cos IF t
 
LO+ LO-
M1 M2

VRF
iRF  g m vRF cos RF t
vRF cos RF t

2
g v cos  IF t
• Current conversion gain; iBB  m RF
 
2
iRF g m vRF cos RF t 

2
g R v cos IF t
vBB  m D RF 2
• Voltage conversion gain;   g m RD
vRF vRF cos RF t  24
Double-Balanced Mixer (1/3)
• The double-balanced mixer accommodates differential signals for
both LO and RF.

iO1 iO2 SD3=SD6


iD3 = iD6
1

t
M3 iD4 = iD5 SD4=SD5
iD4 iD5 iD6 1
+ iD3
vLOcosLOt M4 M5 M6
t
iD1
- ID1
M1 iD1 iD2 M2
t
+ iD2
ID2
vRFcosRFt
- t

g m vRF g m vRF
iD1  I D  cos  RF t , iD 2  I D  cos  RF t  ...
2 2
1 2 1 2
S D3  S D6   cos  LO t  ..., SD4   cos  LO t  ...
2  2 
25
Double-Balanced Mixer (2/3)
1
iO1  iD1 S D 3  iD 2 S D 5  I D  2 g m v RF cos  RF t cos  LO t  ...

1 1
 ID  g m v RF cos  LO   RF t  g m v RF cos  LO   RF t  ...
 
1
iO 2  iD1 S D 4  iD 2 S D 5  I D  2 g m v RF cos  s t cos  o t  ...

1 1
 ID  g m v RF cos  LO   RF t  g m v RF cos  LO   RF t  ...
 
1 1
iO1  I D  g m v RF cos  LO   RF t  g m v RF cos  LO   RF t  ...
 
1 1
iO 2  I D  g m v RF cos  LO   RF t  g m v RF cos  LO   RF t  ...
 
2 2
iO1  iO 2  g m v RF cos  LO   RF t  g m v RF cos  LO   RF t  ...
 

26
Double-Balanced Mixer (3/3)
• No RF-to-IF and LO-to-IF feedthrough.
• For the same conversion gain compared to SBM, but twice dc power
consumption/chip area and a higher NF.
• Linearity
-. High even-order linearity
-. Moderate/poor IIP3 non-linearity due to a tail current source.

27
Voltage Switching Passive Mixer (1/3)

Ron

• Instead of switching currents, we can also switch the voltage signals.


-. TX mixer or BPF
• RF voltage signal is divided by Ron and RIF at LO rate.
So this kind of mixer has “the conversion loss”, not the gain.
• This mixer requires good switches that turn on hard (low on-resistance)
and turn off well (good isolation).
• Switches work in the deep-triode region when they turn-on.
• Large LO power
28
T.H. Lee, Cambridge, “The design of CMOS Radio-Frequency Integrated Circuits”,
Voltage Switching Passive Mixer (2/3)

• A practical implementation uses MOS transistors as switches.


• The larger devices for their lower on-resistance Ron.
Trade-off
• The smaller devices for their low feed-through capacitance.

29
T.H. Lee, Cambridge, “The design of CMOS Radio-Frequency Integrated Circuits”,
Voltage Switching Passive Mixer (3/3)

“Turn-on” “Turn-off”

• To reduce port-coupling, smaller device size is preferred.


30
T.H. Lee, Cambridge, “The design of CMOS Radio-Frequency Integrated Circuits”,
Noise Analysis in SBM
VDD

RL RL

VIF+ VIF-

LO+ LO-
M2 M3

M1
VRF

31
Noise Sources in the SBM
(1) Load
VDD
-. Thermal noise
-. 1/f noise (only for active loads)
RL RL
(2) Switching pairs
VIF+ VIF- -. Thermal noise
-. 1/f noise
LO+ LO-
M2 M3 (3) Transconductance stage
-. Thermal noise
M1
VRF -. 1/f noise

(4) LO power level

32
Thermal Noise Analysis
• For the noise analysis, the differential LO waveform is assumed.
• In order words, the output current is differential.
VLO(t)
1
T
2
T 0 T 3T T
t

4 4 2
1

   4  cos t  1 cos 3t  1 cos 5t  1 cos 7t  1 cos 9t  ...
VLOx(t)
t
 3 5 7 9 

33
Thermal Noise of the load stage
• Thermal noise (4kTR) for resistive load
• 1/f noise
-. Polysilicon resistor : 1/f-noise free.
-. PMOS load : less 1/f noise than that of NMOS

VDD

RL RL

VIF+ VIF-

LO+ LO-
M2 M3

M1
VRF

34
Thermal Noise of the Transconductance stage
• Thermal noise only at RF and odd harmonics of the LO are down
converted. V DD

RL RL

4 1 1  VIF+ VIF-
in ,out |ind  ind   cos  LO t  cos 3 LO t  cos 5 LO t    
 3 5  LO+ LO-
M2 M3
here, ind  4 kT g m   f ; white noise M1
VRF

4 1 1 
 in ,out |ind  4 kT g m   cos  IF t  cos  IF t  ... 
 2 3 
2 2
in2,out |ind  4 1 1 1   4  1  
2
  4 kT g m     1  2  2  ...   4 kT g m       4 kT g m
f   2  3 5    2  8 
“Noise folding”

Here, ind is the drain thermal noise


in,out is the differential output current
35
Thermal Noise of the Transconductance stage
• White noise in the input transistor is indistinguishable from the RF signal.
• The LO freq. and its odd harmonics will down-convert the respective
components of white noise to the IF.

“Noise folding”

1 Odd LO harmonics

1/3
i2 1/5
n ,out
 4kT  g m …
f
LO 3LO 5LO … Freq.
Downconverted
Thermal noise

~80% ~10% ~10%


Thermal noise
0 IF RF 3LO 5LO … Freq.
36
Thermal Noise of the Switching Stage
• When M2 is on and M3 off, noise from M2 is only considered.
• The transconductance operates as the current source Zs for M2.
-. Now, M2 is the cascode device.
• In the cascode TR, the output noise current is given by
VDD

RL RL

VIF+ VIF-

I nd 2 LO+ LO-
I n ,out  M2 M3

1 g m 2 Z S VRF
M1

Here, Ind2 : the drain thermal noise of M2

For high value of ZS for a given gm2,


the output noise contribution by
the drain thermal noise of M2 can be ignored.

37
Thermal Noise of the Switching Stage
• When M2 or M3 is off, it does not contribute to the output noise.
Thus, When the LO amplitude is high, the noise contribution of the
switching pair is usually lower than that of the transconductance stage.
• But, in the real SBM, there exits a time-interval  when M1 and M2 are
both “turn-on”. Then, M2 and M3 contribute to the output noise current.
io ,n |M 2  G (t )vn 2 (t )
The PSD of the output current
  
i2
|   G (t ) dt  vn 2 (t ) dt   G (t ) 2 dt  vn22
2 2
LO o,n M 2
  


2
2 I / V 2 Ts vn22  2 2 I / V 2 Ts  4kT
T T Gm
2I/V
I
2LO  4kT
Even Function A
I
io2,n |M 2 &3  8kT vLO(t)=2AsinLOt
 A

38
Thermal Noise of the Switching Stage
• Frequency translations of white noise originating in switches

Noise folding

2LO- IF

4LO- IF
2LO+IF

4LO+IF
IF

0 2LO 4LO 6LO

Frequency translations of white noise from M2 and M3

39
Total Output Noise
• The total output noise PSD of the SB-Mixer
8kT  I 2
S ( f )  4kTRS g R  4kT  g m 3 R 
o 2 2 2
 RL  8kTRL
n m3 L L
 A

Source Resistor driver Switches Load

• The Noise Factor

1   2 I 2 
F  1    2 
RS  g m 3  A g m 3 g m 3 RL 
2

40
High LO power improves NF & Gain
• LO Power
Conversion gain

NF

41
Flicker Noise Analysis

42
Flicker Noise in Zero-IF receivers
• In the direct conversion receivers, immediately after down-conversion,
IF signal is still weak!

Power spectral density


IF signal

1/f
1/f
Thermal noise
f
1/f corner freq.

43
Flicker Noise of the Transconductance stage
• 1/f noise is up-converted to the LO and its odd harmonics,
for the perfectly matched mixer.
4 1 1 
in ,out |1/ f  ( g m  vn ,1/ f )   cos  t  cos 3 t  cos 5 LO t     ; differential LO
 LO
3
LO
5 

Odd LO harmonics

1/f noise
Thermal noise
0 LO 3LO 5LO … Freq.
mismatch

(1) Feed-through
(2) Up-converted 1/f noise

IF

0 LO 3LO 5LO … Freq.


44
Noise Contribution of the transconductance stage
Odd LO harmonics


LO 3LO 5LO … Freq.

1/f noise
~80% ~10% ~10%
Thermal noise
0 RF 3LO 5LO … Freq.

Down-converted
Thermal noise (noise folding)
+1/f noise (feedthrough)
IF

0 LO 3LO 5LO … Freq.


45
Flicker Noise of the Switching stage
• The transconductance stage is replaced by the ideal current source I.

Vn
-. Input referred 1/f noise voltage
for both M2 and M3

-. Freq. of Vn << LO.

vLO(t)=2AsinLOt

46
Darabi, JSSCC2000, “Noise in RF-CMOS Mixers: A simple physical model”
Direct Mechanism of the flicker noise
• When not including flicker noise of the switching pair vn,
the differential output current Iout is a square wave at LO with zero dc
value. VDD

RL RL
[ the differential output current ]
+ OUT -
at LO Vn
M2 ON
I1 M2 M3
+
t LO
-
--1I
M3 ON
I CP

4 1 1 
I out  I   cos  LO t  cos 3 LO t  cos 5 LO t      Only LO feed-through
 3 5 

47
Darabi, JSSCC2000, “Noise in RF-CMOS Mixers: A simple physical model”
Direct Mechanism of the flicker noise
M2 Ⓐ
ON M2 ON
I
• The slowly varying flicker noise vn
(a) modulates the zero-crossing time
t

-I
of the switching pair M2 and M3.
M3 ON M3 ON


I
• Iout w/o vn
(b)
t
-I

• Noise pulse at 2Lo with 2I has


2I Ⓒ
the average dc component!
Δt 2Lo
1 I  Small DC current
(c) io ,n (t )    vn (t )  Large LO swing
t
 A  Large device size

-2I (a) Mixer output current w/ Vn


(b) Mixer output current w/o Vn
(c) Noise pulse at the mixer output by Vn

48
Darabi, JSSCC2000, “Noise in RF-CMOS Mixers: A simple physical model”
Direct Mechanism of the flicker noise
• The flicker noise Vn from switching TRs appears at the mixer output
(IF) without frequency translation and is up-converted to 2LO, 4LO,


I
io ,n ( f )   vn ( f  2nf LO )
  A
2I
io ,n ( f ) 
I I
Vn ( f ) Vn ( f )
A A
No translated Up-converted
1/f noise 1/f noise

0 LO 2LO (rad/sec)

49
Darabi, JSSCC2000, “Noise in RF-CMOS Mixers: A simple physical model”
Indirect Mechanism of the flicker noise
• Indirect mechanism: f (fLO, Cp)

M2 M3

Vn
t at LO

TLO

Vp
VH
I CP

50
Darabi, JSSCC2000, “Noise in RF-CMOS Mixers: A simple physical model”
Indirect Mechanism of the flicker noise
• Indirect mechanism: f (fLO, Cp)
Vn M2 ON M3 OFF M2 ON

LO
LO

T/2
t at LO T

M2
LO
Vp
VH
I CP
LO

• The average output noise current

2  Small Cp
io ,n  CP vn  Large device size 2LO
T  Low LO freq.
Feedthrough+2LO, 4LO, 6LO…..
51
Darabi, JSSCC2000, “Noise in RF-CMOS Mixers: A simple physical model”
Summary of the SBM noise
• Flicker noise
-. Flicker noise in the transconductor partially leaks to the output due
to some offset.
-. Flicker noise in switching stage appears at IF output and
even-order LO harmonics by two mechanisms.

1 I 2
io ,n (t )    vn (t ) io ,n  CP vn
 A T
• Thermal noise
8kT  I 2
S ( f )  4kTRS g R  4kT  g m 3 R 
o 2 2
 RL  8kTRL
2
n m3 L
 A
L

VDD

R R

VIF+ VIF-

LO+ LO-

VRF

52
The Current Bleeding Techniques(1/16)
TSMC 0.13um CMOS
VDD=1.5 V

100 

VT=375mV 2.3mA

VGS = 0.55V 1.27V


1.25V
40/0.35um

gm,SW=9.7mS
2.3mA
0.7V

0.9V 40/0.13um
0.35V
RF= 5-GHz 0.55V
40/0.13um
Voltage gain = 0-dB
gm=20mS
VT=418mV
53
The Current Bleeding Techniques(2/16)

NF(SSB) = 13-dB@10-MHz
14-dB@1-MHz

fc = 253-KHz

54
The Current Bleeding Techniques(3/16)
VDD=1.5 V

100 
0.93V

200/0.35um
0.7mA
1.6mA

1.44V
1.25V

VGS = 0.45V 0.8V


1/gm,SW=4.5mS 2.3mA

Voltage gain = -4.3-dB

55
The Current Bleeding Techniques(4/16)

NF(SSB) = 15-dB@10-MHz
15.6-dB@1-MHz

fc = 220-KHz

56
The Current Bleeding Techniques(5/16)
The DB-Mixer

The DB-Mixer with bleeding

 1/f noise improved from fc=253 kHz to 220 kHz


 Conversion gain degrades 0-dB  4.3-dB
 Thermal noise degrades 14-dB 15.6-dB @ 1MHz

-. 1/f noise a little improved due to small dc current in the


switching devices (direct mechanism).
-. Reduction of the conversion gain because more RF current
flows into Cp by the increase of the impedance of the SW TRs
seem from the RF stage.  Thermal noise degraded.

57
The Current Bleeding Techniques(6/16)

iRF 1/ g m1 Cp

g m1  g m 2

iRF 1/ g m 2 Cp ro Output
Impedance of
PMOS

58
The Current Bleeding Techniques(7/16)
VDD=1.5 V

400
0.93V

200/0.35um
1.6mA

1.26V
1.25V

VGS = 0.45V 0.8V


1/gm,SW=4.5mS

Voltage gain = 7.3-dB

59
The Current Bleeding Techniques(8/16)

NF(SSB) = 12-dB@10-MHz
12-dB@1-MHz

fc = 113-KHz

60
The Current Bleeding Techniques(9/16)
The DB-Mixer with bleeding

the Opt. DB-Mixer with bleeding

 1/f noise improved 220-kHz113-kHz


 Conversion gain improved 4.3-dB  7.3-dB
 Thermal noise improved 15.6-dB 12-dB

(1) Conversion gain improved due to higher value of resistive


loads.  Thermal noise & 1/f noise improved

61
The Current Bleeding Techniques(10/16)

400

at 5-GHz

Voltage gain = 11.4-dB

62
The Current Bleeding Techniques(11/16)

NF(SSB) = 10-dB@10-MHz
10-dB@1-MHz

fc = 33-KHz

63
The Current Bleeding Techniques(12/16)
The Opt. DB-Mixer with bleeding
The Opt. DB-Mixer with bleeding+L
 1/f noise improved 113-kHz33-kHz
 Conversion gain improved 7.3-dB  11.4-dB
 Thermal noise improved 12-dB 10-dB

(1) Conversion gain improved due to tuning out the Cp by L


 Thermal & flicker noise improved
(2) 1/f noise also improved due to no Cp (indirect mechanism)

64
The Current Bleeding Techniques(13/16)

Voltage gain = 11.7-dB

65
Jinsung Park, MTT-J, 2006, “Design and analysis of low flicker noise CMOS mixers for DCR”
The Current Bleeding Techniques(14/16)

NF(SSB) = 9-dB@10-MHz
9-dB@1-MHz

fc = 30-KHz

66
The Current Bleeding Techniques(15/16)
The Opt. DB-Mixer with bleeding+ 1L

The Opt. DB-Mixer with bleeding+2L


 1/f noise a little improved 33-kHz30-kHz
 Conversion gain almost same 11.4-dB  11.7-dB
 Thermal noise improved 10-dB  9-dB

(1) Conversion gain not much improved.


-. The leakage of RF current into the PMOS is not dominant.
(2) 1/f noise almost same.
-. PMOS devices not dominant
(3) Thermal noise improved
-. due to no contribution of the bleeding devices

67
The Current Bleeding Techniques(16/16)

at resonant freq.

iRF 1/ g m 2 Cp L ro
thermal noise by ro

at resonant freq.

iRF 1/ g m 2 Cp L
no thermal noise by ro

68
The BLIXER: Noise-Cancellation
The BLIXER = Balun+LNA+MIXER
-. The mixer has been stacked on top of the LNA.
-. The noise of CG is cancelled when gmPRCG=gmNRCS.

Stephan C. Blaakmeer, JSSCC2008, “The BLIXER, a wideband Blaun-LNA-I/Q-Mixer Topology”


69
The BLIXER: Noise-Cancellation

DSB NF

Stephan C. Blaakmeer, JSSCC2008, “The BLIXER, a wideband Blaun-LNA-I/Q-Mixer Topology”


70
Current-Reused Low-Noise Balun Mixer
The Proposed Noise-Canceling Topology
 Large dc-power consumption  Current-Reused
 Noise and Distortion canceling in RF range  After Downconversion

ac ground

CBY

CP LP LN CN
+ IF

i o+ io
in MP + LO
gmP
MN
+ gmN
RS
vin LS CIN
I BIAS
vS CBY

VDD

71
Chang-Wan Kim,“A 2.4GHz Current-Reused CMOS Balun-Mixer,” IEEE MWCL, July. 2009.
Current-Reused Low-Noise Balun Mixer
(1) A current-reused transconductance stage using noise canceling
to reduce dc power consumption.
For ZIN =1/gmP (= RS)

 the output signal voltages


vo+ = +gmP(vs/2)ZP
vo- =  gmN(vs/2)ZN,

 the output noise voltages


due to the CG input transistor MP
vn,o+ = +(in/2)ZP
vn,o- = +(in/2)RSgmNZN.

 If gmPZP = gmNZN, the output signal voltages are fully differential and the
output Noise voltages due to MP are cancelled by taking the differential output.
 If gmN is scaled up as NgmP, ZP should be NZN for both output-balancing
and noise-canceling.
72
Chang-Wan Kim,“A 2.4GHz Current-Reused CMOS Balun-Mixer,” IEEE MWCL, July. 2009.
A 2.4 GHz Current-Reused CMOS Balun-Mixer
• The noise-canceling occurs after frequency down-conversion

ZP = Z1+Z2
Z N = Z1

73
Chang-Wan Kim,“A 2.4GHz Current-Reused CMOS Balun-Mixer,” IEEE MWCL, July. 2009.
A 2.4 GHz Current-Reused CMOS Balun-Mixer
Features of the proposed topology
(1) the switching stage is ac-coupled for small dc currents
-. a higher conversion gain
-. reduction of the direct-mechanism of 1/f noise from
the switching stage
(2) L1 and L2 tune out all parasitic capacitances at RF freq.
-. all signal and noise currents are moved into the S/W stage
-. the indirect-mechanism of 1/f noise from the S/W transistors
is also mitigated.

Consequently, the noise contribution by switching transistors


M3-M6 can be ignored when the switching pair is hard
switched.

74
Chang-Wan Kim,“A 2.4GHz Current-Reused CMOS Balun-Mixer,” IEEE MWCL, July. 2009.
A 2.4 GHz Current-Reused CMOS Balun-Mixer
• The IF output noise voltage due to the drain noise of MP
[( Z1  Z 2 )  g mN RS Z1 ] [( Z1  Z 2 )  g mN RS Z1 ]
v IF ,n    in v IF , n    in
2 2
Note that the output noise voltages (vIF,n+ and vIF,n) are not
common-mode, but are rather differential for each other. However,
when gmP(Z1+Z2) = gmNZ1, the output noise voltages can be still
removed at each output node respectively (more better !).
• The IF output signal voltage
[g mP (Z1  Z2 )  g mN Z1 ] [g mP ( Z1  Z 2 )  g mN Z1 ]
v IF   vs v IF    vs
2 2
The IF output signals are inherently balanced by switching operation
in the mixer, regardless of the noise-canceling condition (more better !).

• For gmP(Z1+Z2) = gmNZ1, the conversion gain of the proposed mixer


is simply given as 2/(Z1+Z2)/RS, directly in proportion to Z1+Z2.
75
Chang-Wan Kim,“A 2.4GHz Current-Reused CMOS Balun-Mixer,” IEEE MWCL, July. 2009.
Implementation
• To more reduce noise factor, gmN = NgmP and Z1+Z2 = N Z1.
• The simulated SSB NF and input-referred P1dB for N = 1, 1.5, and 2,
when RF = 2.4 GHz and Z1+Z2 = 800 .

In this work, N = 2
SSB NF (dB)

gmN = 2gmP = 40mS


Z1 = 400//2pF
Z2 = 400//2pF

NF is improved in proportion to N, whereas P1dB decreases with N,


for the constant conversion gain.
76
Chang-Wan Kim,“A 2.4GHz Current-Reused CMOS Balun-Mixer,” IEEE MWCL, July. 2009.
Implementation
Differential Conversion gain

TSMC 0.13m

SSB NF

77
Chang-Wan Kim,“A 2.4GHz Current-Reused CMOS Balun-Mixer,” IEEE MWCL, July. 2009.
Performance Summary

This work [5]1) [7]


RF Freq.(GHz) 2.0–2.7 0.5–7 0.1–3.85
S11(dB) < –17 < –10 < –10
Conversion Gain (dB) 13.5 18 20 2)
SSB NF (dB) 8 7.5–8.5 3) 8.4–11.5
IIP3(dBm) 6 3 3.2 4)
DC Power (mW) 5.3 16 9.78
Supply Voltage 1.5 V 1.2 V 1.2 V
# of on-chip inductor 3 15) 4
Chip area (mm2) 0.9 0.01 0.88
CMOS Tech. 0.13 m 65 nm 90 nm
1)I/Q–Blixer, 2) including voltage gain of an IF amp., 3) DSB NF+3dB, 4) P1dB+9.6 dB,
5) off-chip inductor.

78
Chang-Wan Kim,“A 2.4GHz Current-Reused CMOS Balun-Mixer,” IEEE MWCL, July. 2009.
Current-Driven Passive Mixer

IN OUT

High Z Low Z

LO

79
50% duty-cycle, Current-driven Passive Mixer
• The current conversion gain is ~ 2/ (-3.9dB).
• No dc current. No 1/f noise?

ILO+
 1 2 1
iIF  cos  RF t  [  {cos  LO t  cos 3LO t  ...}]
2  3
1 1 1
iRF  cos  RF t  cos(RF   LO )t  cos( RF  LO )t  ...]
2  
iIF

C 1 2 1

iIF  cos RF t  [  {cos LO t  cos 3 LOt  ...}]
2  3
1 1 1
ILO-  cos  RF t  cos( RF   LO )t  cos(RF  LO )t  ...]
2  

ILO+ 2
cos( RF   LO )t
iIF 2
  ( 3.9dB )
ILO- iRF cos RF t 

50% duty-cycle LO

80
D% duty-cycle, Current-driven Passive Mixer
• The current conversion gain for d% duty-cycle passive mixer
 2
ILO+

50% LO
2 1 d  0.5  G
 ILO-

G  sin  d  
 2d d  0.25  2 2 ILO+
25% LO
G
  ILO-

• +3 dB gain improvement by using 25% duty-cycle, theoretically


• PSS Simulation result;
-. CG= -9.24dB for 50% LO
CG= +1.4dB
-. CG= -7.84dB for 25% LO

RF=2.05GHz, LO=2GHz 81
1/f noise in the passive mixer
• The time-varying drain channel current causes 1/f noise to appear
around dc as well as LO harmonics in the passive mixer.

• In MOSFET, larger LO power and longer turn-on time


increase the 1/f noise.

82
50% duty-cycle, IQ Passive Mixer
• Because of IQ overlap time, interaction between I and Q path degrades
the receiver performances; high-and low-side gain imbalance (image
current), noise, I/Q mismatch problems…

ILO+

TLO/4

I_iIF
Image current C TLO/4
LNA
iRF ILO-

QLO+

iRF ZL

Q_iIF

QLO-

83
50% duty-cycle, IQ Passive Mixer
• Image current (due to IQ cross-over) exacerbates the gain imbalance
between the high- and low-side conversion gain.

Due to image current


1
Z L (LO  m )[ Z C* ( LO  m )  RSW ]I RF e jRF
I BB , I   jI BB ,Q  
2
 2  * 2   2 
 2 Z (   )  Z (   )  R  Z ( )   Z (   )  R  Z ( ) 
   Z ( ) 
L LO m C LO m SW
 2 BB m C LO m SW
 2 BB m 2 BB m

A. Mirzaei, JSSCC2009, “Analysis and optimization of current-driven passive mixers …” 84


50% duty-cycle, IQ Passive Mixer
• Simulation Results; LO=2GHz, RF= 1.8GHz to 2.2GHz

low-side gain < High-side gain

85
50% duty-cycle, IQ Passive Mixer
• Isolation I and Q path by using two separate Gm cells reduce this
gain imbalance.
ILO+

I_iIF

C
Image current
iRF ILO-

QLO+
X No image current
iRF ZL

Q_iIF

QLO-

86
50% duty-cycle, IQ Passive Mixer
• Switches connect I and Q path when the in-phase and quadrature
LOs overlap.
• Since the mixer output impedance seen from the BB amp is RSC //Rsw,
which is low (~RSW), the input-referred low-frequency noise from
the passive mixer and BB amp are boosted by the BB amp.
R2
ILO+
- VBB+
+
ILO-
Cpar
R2
IRF QLO+
+
- VBB-
QLO-
Z(RF) R2
R2
RSC=1/fLOCpar 2RSW
- vn,out
+
vn,in * 87
25% duty-cycle, IQ Passive Mixer
• Because only one switch turns on at TLO/4(NO IQ overlap-time), many
problems in the 50% IQ passive mixer are considerably mitigated;
more gain, lower NF, better IQ crosstalk.
ILO+

ILO+
I_iIF

ILO-
iRF
ILO-
QLO+
C QLO+

iRF ZL
QLO-
Q_iIF

QLO-

88
25% duty-cycle, IQ Passive Mixer
• No image current. The gain imbalance is considerably mitigated
(almost equal).

High-side band conversion gain


2
Z L ( LO  m ) I RF e j (RF  / 4)
I BB , I   jI BB ,Q  
 2 
 Z (   )  Z (   )  R  Z BB (m ) 
L LO m C LO m SW
 2

Low-side band conversion gain


2
Z L* (LO  m ) I RF e  j (RF  /4)
I BB , I   jI BB ,Q  
 * 2 
 Z L (LO  m )  Z C (LO  m )  RSW   2 Z BB (m ) 
*

A. Mirzaei, JSSCC2009, “Analysis and optimization of current-driven passive mixers …” 89


25% duty-cycle, IQ Passive Mixer
• PSS+PAC simulation results
• The gain imbalance is considerably reduced by using the 25% LOs.

50% LO 25% LO

90
50%/25% LO, IQ Mixer Comparision
• PSS Simulation results (current-gain and NF)
-. 50% LO: -18.37dB (NF=10 dB)  CG= +7dB
-. 25% LO: -11.43dB (NF=8 dB)  NF=2dB
• The connection between I and Q path in 50% LO mixer considerably
reduces the conversion gain.

50% LO: NF= 10dB

25% LO: NF= 8dB

RF=2.05GHz, LO=2GHz

91
Input Impedance of the Current-driven Mixer
ILO+
LPF  TLO
1, 0t
ZBB(S) ILO+
S I  (t )   4
I_iIF
0, TLO
 t  TLO
ZBB(S)
 4
ILO- T
vRF ILO-
S I  (t )  S I  (t  2 LO )
Zin 4
iRF QLO+
QLO+ T
BPF ZBB(S)
SQ  (t )  S I  (t  LO )
4
Q_iIF
T
ZBB(S) QLO- SQ  (t )  S I  (t  3 LO )
4
QLO-

In I+ path, the baseband current iBB,I+(t) is


iBB, I  (t )  S I  (t )iRF (t )

In I+ path, the baseband voltage VBB,I+(t) across the baseband impedance


ZBB(S) is given by
vBB,I  (t )  [ S I  (t )iRF (t )] * z BB (t )
A. Mirzaei, IEEE Trans C&S2010, “Analysis and optimization of direct-conversion receivers with 25% duty…” 92
Input Impedance of the Current-driven Mixer
• Since Rsw appears only at TLO/4, it can be presented by a constant resistor.
As a result of no reverse isolation in the passive mixer, the baseband
voltage is frequency-translated to around LO and its odd harmonics.
• The resulting RF voltage at the RF side of the mixer is following.
ILO+

vRF (t )  RSW iRF (t )  S I  (t )  [( S I  (t )iRF (t )) * z BB (t ) ZBB(S)

I_iIF

 S I  (t )  [( S I  (t )iRF (t )) * z BB (t ) ZBB(S)


 SQ  (t )  [ ( SQ  (t )iRF (t )) * zBB (t ) vRF

iRF
Rsw ILO-

QLO+

 SQ  (t )  [( SQ  (t )iRF (t )) * zBB (t ) ZBB(S)

Q_iIF

ZBB(S)

• The Fourier transform of vRF(t) is obtained.


QLO-

  an amiRF (  ( n  m)LO )
vRF ( )  RSW iRF ( )  4  
m  n  Z BB (  nLO )
n  m  4k ( 25% LO )

A. Mirzaei, IEEE Trans C&S2010, “Analysis and optimization of direct-conversion receivers with 25% duty…” 93
Input Impedance of the Current-driven Mixer
• After ignoring higher order harmonics, and only considering
the iRF at LO+ m (n+m=0)

vRF ( )
 RSW  4  an Z BB (  n LO )
2
Zin ( ) 
iRF ( ) n 

• The input impedance seen from the mixer system is actually


the baseband impedance ZBB(s) shifted to the integer harmonics of
the LO along with a scaling factor |an|2.

ZBB()
0

2 1
2 2
 2 9 2
Zin()
-3fLO -2fLO -fLO 0 fLO 2fLO 3fLO

A. Mirzaei, IEEE Trans C&S2010, “Analysis and optimization of direct-conversion receivers with 25% duty…” 94
Input Impedance of the Current-driven Mixer
• Simulation Results; fLO = 2 GHz, ZBB=100  ||10 pF

- 3dB BW= 470M @ 2 GHz

95
N-path BPF
• Band-pass Filter = Down-converter + LPF + up-converter

1)Down-conversion 2) Low-pass filtering 3) Up-conversion

interferer

RF signal

fs dc dc fs

Vs Vout

fclk fclk

Finally, |Vout/Vs| is a band-pass filter response @ fs.

A. Ghaffari, RFIC2010, “A differential 4-path Highly linear Widely Tunable On-chip Band-pass filter” 96
N-path BPF
• CMOS Implementation; Switched RC N-path filter
• Switches are on/off by non-overlapping multi-phasing clocking.
• R appears at Ts/N.

DMIX RC-LPF UMIX Non-overlapping


Multi-phasing clocking
R
TS
C1
R
Vs Vout
C2
TS/N

CN

A. Ghaffari, RFIC2010, “A differential 4-path Highly linear Widely Tunable On-chip Band-pass filter” 97
N-path BPF
• R appears at TS/N
• Passive mixer has no reverse isolation.
• In the passive mixer, down and up-conversion process occur
simultaneously. One passive mixer has been used.
• Vout is the output voltage across one capacitor.
Vout

+ +
C1 C1
R - R -
Vs Vout Vs
C2 C2

CN CN

[ Final version ]
A. Ghaffari, RFIC2010, “A differential 4-path Highly linear Widely Tunable On-chip Band-pass filter” 98
N-path BPF
• RC >> Ton
• IQ passive mixer can be used as a superior tunable RF BPF.
Rs VI+
VV FS=FCLK
out
BPF VQ+
Rsw Vin
+
Vin
VQ-
VI-
+
C VIF
- - VBPF

N-path

1 [ Time domain ]
RC >> Ton
𝑅
𝑅 +𝑅

fLO 2fLO [ Frequency domain ] 99


A. Ghaffari, RFIC2010, “A differential 4-path Highly linear Widely Tunable On-chip Band-pass filter”
N-path BPF Simulation

470 MHz 700 MHz

LC BPF
Q: 5~15

Q ~ 100

100
N-path BPF using IQ passive mixer
• Equivalent RLC model of the N-path BPF

RS

+ +
Vin RS Vout
- -
fclk

fclk

A. Ghaffari, RFIC2010, “A differential 4-path Highly linear Widely Tunable On-chip Band-pass filter” 101
LNA with a N-path filter
• The N-path filter has been imbedded in the feedback path of the LNA.
-. The low value of CF can be used in the feedback path.
• Open-loop for the RF signal, but closed-loop for blockers
f0

CF SW1

- -
 A0 Vout A0 Vout
+ + 0 + +
Iin RS Iin RS
- -

102
J. Park, JSSCC2014, “Channel Selection at RF using Miller Bandpass filters”
RF receiver with 25% LO current-driven Passive Mixer
To suppress ZRF1
ZRF2
Impedance transform ZBB2

3MHz offset
Blocker RF
RF dc
in RF domain LNA GM Down-Conversion Mixer

VDD (1.5 V)

M9 TIA
RD LD LD RD
M10

Resonant frequency
Bonding wire Bonding wire
VDD (1.5 V)
IBB 4th-order LPF
control <3:0>
CD CD M11
M7 M8
CAC M12
RF RF CAC
Impedance
transform M2
VB M4 CAC CAC
CAC M13 TIA
RF
Input LG Bonding wire
CAC M14
M3 M5 M6
M1
IBB 4th-order LPF
CIN
LNA Gain RIN M15
Control LS
M16
<2:0>

Input Matching Control <3:0>

LO_QN2
LO_QP2
LO_IN2
LO_IP2
M25 M26 M27 M28 M29 M30 M31 M32 25% duty-cycle quadrature LO signal

LO_IP
ZBB1
LO_QP
`
LO_IN

CBB CBB CBB CBB LO_QN

dc
RF BPF
103
Chang-wan Kim, ETRIJ2018, “470-MHz-698-MHz IEEE 802.15.4m Compliant RF CMOS Transceiver”
RF receiver with 25% LO current-driven Passive Mixer
ZRF1 ZRF2
2 2
Normalized Transfer Function (dB)

Normalized Transfer Function (dB)


0 0
-2 4.2 dB
5.8 dB -2
@3MHz @3MHz
-4
-4
-6
BW1dB = 2 MHz -6 BW1dB = 2.4 MHz
-8

-10 -8

-12 -10
Nomarlized Attenuation (dB)

Nomarlized Attenuation (dB)


-4 -2 0 2 4 -4 -2 0 2 4
Offset Frequency (MHz) Offset Frequency (MHz)

zero
TIA with 4th order LPF Overall RF front-end Receiver (LNA to TIA)
10
Normalized Transfer Function (dB)

Normalized Transfer Function (dB)


0
0

-10 38.6 dB
28.5 dB -20 @3MHz
@3MHz
-20

-40
-30
BW1dB = 2 MHz
BW1dB = 2.2 MHz
-40
-60
-50
Nomarlized Attenuation

Nomarlized Attenuation

-4 -2 0 2 4 -4 -2 0 2 4
Offset Frequency (MHz) Offset Frequency (MHz)

104
Chang-wan Kim, ETRIJ2018, “470-MHz-698-MHz IEEE 802.15.4m Compliant RF CMOS Transceiver”
RF receiver with 25% LO current-driven Passive Mixer
• -97 dBm RF signal and –20 dBm blocker at 3-MHz offset
• The final baseband spectrum at the VGA output (700mVpp)

-20dBm Blocker
20 dBm (maximum)

TV signal
(blocker 1)
-97dBm RF Signal
IF signal 3

3 MHz

blocker

105
Chang-wan Kim, ETRIJ2018, “470-MHz-698-MHz IEEE 802.15.4m Compliant RF CMOS Transceiver”
Beyond Mixer
• LO path circuits are more important!
- I/Q amplitude and phase compensation
- 25% duty-cycle LO generation (small size and low-power)

• Careful Layout
- LO leakage and IIP2
- Mixer IIP2 determines the overall Rx IIP2.

• IIP2 Calibration circuit

106
Thank you !

107

You might also like