Low Noise Amplifiers

1
Jayanta Mukherjee
Transceiver Performance
Typical GSM Receiver

• Minimum detectable signal (Pi,mds) of -102 dBm

• A min BER of 10-3
• BER of 10-3 corresponds to a SNRout of about 9-12 dB (GMSK)

Pi,mds = -174 dBm + 10 log[Δf (Hz)] + F (dB) + SNRout,min (dB), Δf = 1 MHz

Pi,mds SNRout,min
Δf

Required range for the noise figure of the receiver < 3 dB

2
Jayanta Mukherjee
Gain
• Gain large enough to minimize noise contribution of subsequent
stages.
• This leads to compromise between NF and linearity . Higher gain will
degrade linearity but improve NF.
• For heterodyne output of LNA matched to i/p of mixer (50 ohms).
Here gain implies power gain.
• However where uniform matching across the chain cannot be done,
voltage gain is used.

3
Jayanta Mukherjee
Input and Output Matching

Maximum gain occurs for simultaneous conjugate match at the input and
output (if the device is stable).
 Input matching: 50 ohms
Z in  R0
- Reflection coefficient in 
Z in  R0
R
- For Zin  R 0  R we have  in 
2 R0  R
- For  in of around -17 dB (0.141) we need R  16.4 ohms.

 Output matching : 50 ohms for heterodyne transceiver.

4
Jayanta Mukherjee
Stability
Stern Stability factor :
2 2 2
1    S11  S22
K with   S11S22 - S12S21
2 S21 S12

If K  1 and   1 for all frequencie s the circuit is unconditionally stable for all passive sources and loads.
Unconditional stability is not required if the source and load impedances ZS and Z L are known e.g. heterodyne receiver.

Stability is achieve d if

Re Zin  ZS   0 and Re Z out Z L   0

In practice good grounding and power supply decoupling (using decoupling capacitors)
is required to reduce the wire inductance and establish the AC grounding.

5
Jayanta Mukherjee
Linearity Peak power may be 2 dB above average. Hence LNA should have high enough linearity
to avoid spreading from Tx to Rx till -18 dBm i/p power. So say P1dB of -15dBm can
provide good compromise between spreading and Rx signal detection.

• With i/p power levels being lower than P1dB,

usually linearity is not a problem.

• Wideband receivers like UWB, SDR and

-50 dB cognitive can pose a problem on linearity
since a strong interferer in the presence of
IM distortion can spread and affect the
desired band.

6
Jayanta Mukherjee
Bandwidth

• BW should be large enough to accommodate band.

• Less than 1 dB variation over band.
• BW may be switched using various techniques like N path filtering,
switching tank.

7
Jayanta Mukherjee
BJT equations

IC

VA

ro=VA/IC

8
Jayanta Mukherjee
More generally, Parameter Typical values
 complex rb 100 ohms
Z 
gm
gm 40 mA/V
Cμ 12 fF
Cπ 1fF
ro 100 Kohms
rc low
rπ 1 Kohms

9
Jayanta Mukherjee
CE BJT noise insignificant
Collector resistance
Zπ
Base resistance Thermal noise
Thermal noise (insignificant)

Collector
Shot noise

Base shot
Small signal equivalent circuit
noise Not a physical resistor (active region)

kT IC
2
ib,n  2qI B Δf 2
, ic,n  2qI C Δf  2kTg m Δf , VT  
q gm
q  charge of electron (absolute value)  1.6 x 10-19 Coulomb
10
Jayanta Mukherjee
Equivalent noise current

From this
From this circuit
circuit
When I/P is open,

i 
2
n  2
i
c,n  2 2
i
complex b , n  / 2
complex i2
b ,n 
ic2,n
 complex
2
 ib2,n  2qI b f 
2qI c f
 complex

2kTg m f
 complex
When I/P is shorted
2
2
Z v 2
 complex 2 vr2,b
io2  g m2 vn2  n
g m Z   ic2,n Ignoring rb
Z  rb Z  rb
2  Z  rb
2
complex

2 2

 vn2  vr2,b  ic2,n   Z  rb

 complex 2
 
 vr2,b  ic2, n
Z
complex
2
 4kTrb f 
2kT f
gm
11
Jayanta Mukherjee
BJT Noise Figure Input referred o/p noise

SNRin S i / Ni 1 GA N i  N a N a / GA v 2n  Rs2 in2

F    1  1
SNRout Sout / N out G A Ni Ni vs2
At lower frequencies,  complex  
2kT
4kTrb 
v 2n  Rs2 in2 gm Rs2 2qI c
F  1  1   complex approximated to 
v 2
s
4kTRs  4kTRs
at lower frequencies
rb 1 g R I Iq
F  1   m s , Taking 2kTg m  2qI c ,  g m  c  c
Rs 2 g m Rs 2 VT kT
At higher frequencies,
g m Rs  1   r C 
2 2 2
rb 1 g m Rs rb 1 g m Rs  rb 1
F  1    1   = 1   
Rs 2 g m Rs 2  complex Rs 2 g m Rs 2 complex
2
Rs 2 g m Rs 2 r2C2

2

rb 1 g R
 1   m s 1   2 r2C2 
Rs 2 g m Rs 2

12
Jayanta Mukherjee
Optimal Source Impedance
We had earlier seen that, F(Ys ) vs2
Rs
2
Rn 2 4 Rn Γ s - Γ opt
 Fmin  Yopt  Ys  Fmin 
Gs  2

1  Γ s 1  Γ opt
2

Proof :
Let, vs2  4 kTRs , vn2  4kTRn , in2  4kTGn
N a / GA v2  i2 R2 R  Gn Rs2 R
F  1  1 n n s  1 n  1  n  Gn Rs
Ni vs2 Rs Rs
dF
For optimum value of Rs ,  0, from which the optimum value of R s is given by, 
dRs

Rn vn2
Rs,opt   2
Gn in
Rn
 Fmin  1  Gn Rs,opt   1  2 RnGn
Rs,opt

Rn   Rs Rs  Rn Rs
2 2
Rn  2
 F - Fmin   Gn Rs  2 Rn Gn  1     2   1  Rn Rs Gopt  Gs
Rs Rs   Ropt  R  R R
 opt
 s opt

2
 F  Fmin  Rn Rs Gopt  Gs
Rn
More generally F Ys   Fmin 
2
Yopt  Ys
Gs 13
Jayanta Mukherjee
Optimal Rs for a BJT
 1  1 1 
v  4kTRn f  2kT  2rb 
2
n  f  Rn   2rb  
 g m  2  g m

2kTg m gm
in2  4kTGn f  f  Gn 
 complex 2  complex

Rn  1  2 g m rb 
Rs ,opt  
complex
, NFmin  1  2 RnGn  1  1  2 g m rb  /  complex
Gn gm

To decrease NFmin we need to:

1. Decrease rb (increase transistor size)

2. Decrease gm (decrease IC )
3. Increase β (very little scope)

14
Jayanta Mukherjee
Conjugate Matching and Noise Matching
• Noise matching does not yield the maximum gain (conjugate match)
• Ideal Target:
o RS,opt = Zin* = 50 ohms for simultaneous conjugate, noise and 50 ohm
impedance match.
• Methods:
o Adjust transistor size and bias to obtain RS,opt = 50 ohms (noise
match)as much as possible.

o If no further improvement can be done, then simply match i/p so that

Zin=50 ohm

15
Jayanta Mukherjee
CE BJT linearity

This voltage corresponds to -23 dBV or -10 dBm across 50 ohms. Additional linearization is required for larger IIP3.

16
Jayanta Mukherjee
Inductive Degeneration

 1 
vb  iin  rb  
ve  sC   g 
  iin  g m vb 'e  iin 1  m 
Le s Le s  sC  Gm = ic/vin can be made dependent only on Le
vb 1 L
Z in   rb   sLe  g m e (Prove it)
iin sC C
with proper choice of Le , g m and Cπ we can select :
Hence linearity is improved
1
sLe  0
sCπ
Le
Z in  rb  g m  50 Ω
Cπ 17
Jayanta Mukherjee
Neutralization and Cascoding
• K indicates that stability improves as |S12| or (|z12| or |y12|) decreases.
• This can be accomplished by neutralizing the input-output capacitance path:
L1 is selected to resonate with Cμ at the frequency of interest Problem: In RFIC the floating inductor introduces parasitic capacitances
loading the input and output nodes.
• Reduced feedback can be achieved with the cascode configuration.

18
Jayanta Mukherjee
Cascode BJT LNA

• Current fed to both transistor by the same bias line.

• Low distortion due to the inductance Le.
• Stability improves (since back propagation of of signal is minimized).
• Slight degradation in Noise Figure.
19
Jayanta Mukherjee
MOSFET characteristics
Model as given in the book by Johns and Martin

Saturation region

20
Jayanta Mukherjee
Model as given in the book by Johns and Martin

 = 0.02 V -1 , Lov  25 nm
gm  0.42 mS for W/L=10, ID = 100 μA, μnCox = 90 μA/V2
1 1
rds  ro    500 Kohms
λI D 0.02  100  10 6
C gs  23 W L Cox  W Lov Cox  0.49 f F
21
Jayanta Mukherjee
MOSFET noise (CS configuration) Saturation region small signal model

s
insignificant
Iout
id2  4kTg m

Gate induced thermal noise

1
v r2, g  4 kTR g , Rg  (Typical value 20 ohms), (where g d0 is the
5gd 0
drain - source conductanc e at zero V DS ) is not a real resistor,

 g
i  v
2 2 2 It only responds to noise currents and voltages but not to

 
d r,g m 2 2
v 
2
n , g i Z in  i  i  i / g Zin
2 2
m n
2
d
2
n
2
d
2
m
determinis tic signals.
g m2
vn2  in2  Rs  Rg  R  Rg  in2 4kT  g m  Rs  Rg  4kT  g m
2 2 2
vr2, g id2 s Rg
 F  1  1    1   2
v2
s v 2
s g m2 vs2 vs2 Rs 4kTRs g m2 4kTRs g m2 Z in
Rg  ( Rs  Rg )2 
 1   2
Rs Rs g m Rs g m Z in
Rg  
At dc |Zin |    F  1   , If gate induced thermal noise is ignored, F  1 
Rs Rs g m Rs g m
1 Rg   R  Rg  2 2C 2
Rg  
  2 
s gs 2
At higher frequencies we consider |Zin |  , F  1    1  1   2
R  R C
Rs Rs g m 
gs 
ω Cgs Rs Rs g m Rs g m
s g

  
Can be made low by tuning R s  Typically 50 ohm) 
22
Jayanta Mukherjee
Alternate Derivation
+
vgs gmvgs
-

I out  g m v gs  i d - - - - (1)
V s  R s  R g   v gs  sC gs  v gs  v gs sC gs ( R s  R g )  1 
Now substituti ng v gs from eqn (1) above we get,

Vs 
I out  id 
sC ( R s  R g )  1
gs
gm

When i d  0 (noiseless case)

I out gm gm
   G m (say! )
Vs sC gs ( R s  R g )  1 1  j  C gs ( R s  R g )

23
Jayanta Mukherjee
Contd ….
G m represents the overall transcond uctance between i/p voltage
and output current.

2
Hence total o/p referred noise current due to v r,g and v s2 is given by,

2
4 kT(R s  R g )  G m

2
4 kT(R s  R g )  G m  i d2
Hence, NF  2
4 kTR s  G m

 1
Rg
Rs


Rs g m

1   2 C gs2 R s  R g 
2


24
Jayanta Mukherjee
CS MOSFET
• Higher Linearity using inductive degeneration.
• Wideband matching difficult.

25
Jayanta Mukherjee
CG Noise

Input impedance is low:

1
Zin  (prove it)
g m  g mb  sCgs

  
equal to 0

vn2
 in2 Rs2 in2 Rs2 id2 Rs2
F  1  1  1
vs2 4kTRs 4kTRs
4 kTg m  gm
1  1   1 

  2.2 dB (for   2 / 3)
4kTGs Gs
 assuming that Gs  g m  I / P matching ]
For R s =50 ohm
G s is only 0.02
which will cause
NF to rise.

  is (2/3) only for long channel devices. Zin

Rs
 Short channel MOSFETs have larger value of  .
 Parasitics and other noise sources will add additional noise. Vin
 Easier 50 ohms matching over wideband. However resistive matching will add 3 dB noise.
i
 G m = d =-  g m  g mb  . 26
vin Jayanta Mukherjee
Scaling Rules
In classical noise theory the noise figure is a function of the source admittance Ys
and can be written as,
R 2
F(Ys )  Fmin  n Ys  Yopt
Gs

 
f 1 gm
2 fT 
Fmin  1   1 c 2 C gs
fT
  2 / 3,   4/3, c  j0.395 (Typical values) (see book by Thomas Lee)
These expressions indicate that Fmin is minimized for large fT  large g m . This indicates that we should :
 Select the shortest gate length available.
 Use the largest current allowable by the power budget.
 Use the largest width allowable within power budget.

27
Jayanta Mukherjee
Conjugate Matching and Noise Matching
• Noise matching does not yield the maximum gain (conjugate match)
• Ideal Target:
o ZS,opt = Zin* = 50 ohms for simultaneous conjugate, noise and 50 ohm
impedance match.
• Methods:
o Adjust transistor size and bias to obtain ZS,opt = 50 ohms (noise and
conjugate match) as much as possible.

o If no further improvement can be done, then simply match i/p so that

Zin=50 ohm (impedance match)

28
Jayanta Mukherjee
Common Source Input Matching with feedback capacitor

 1 
Starting from : vdg sCF  vds   sCL   g m v gs  0, vdg  vds - vgs
 RL 
1
and rearranging : vgs(sC F -g m )  vds(sC F   sCL )
RL
 v 
iin  v gs sCgs  vgd sCF  v gs  sCgs  sCF  ds sCF   Yin v gs
 vgs 
 
results in the impedance Yin -sC gs being :
1  sCL RL  g m RL
Yin'  Yin  sCgs  sCF
1  s(CL  CF ) RL 29
Jayanta Mukherjee
…Contd
 
Re Yin'  RLC F  2
CF  g m RL (C L  C F )
1  RL2 (CL  C F ) 2  2
RL2C L (CL  C F ) 2  1  g m RL
 
Im Y  C F 
'
in
1  RL2 (C L  C F ) 2  2
If g m R L  1, C L  C F and   1/(R LC L ), the expression reduces to :

 
Re Yin' 
g m CF
2 CL
 Re [Yin ]  Need to make this equal to 1/ 50 ohms-1

Alternatively a resistance in series/par allel can be added but that will increase NF.

   g R 
 
  g R  
Im Yin'  C F  1  m L   Im Yin   C F 1  m L   CGS   Need to design L g such
2  2 
   
that this value is canceled.

 Matching is narrowband
 C L can change o/p loading.
30
Jayanta Mukherjee
Inductive Degeneration
+
id

vin

vs vg  iin (1/ C gs s )  g  -
  iin  g m vgs  iin 1  m 
Ls s Ls s  Cgs s 
 
vg 1 L
Z in    Ls s  g m s iD gm 1 1
iin Cgs s C gs Gm    for  L 
vin 1   2CgsL s   j gm L s j r Ls  r Cgs
r s

With proper choice of Ls , g m and Cgs we can select:

1
Ls s  0  Improves linearity (due to negative feedback).
Cgs s
 Matching does not lead to increase in NF since components are reactive.
Ls or add a resistance in series to   Try to derive the NF of the inductively degenerated CS LNA.
Z in  g m  50   
Cgs  match i/p to 50 ohms  31
Jayanta Mukherjee
Linearity of degenerated CS stage
I D  1Vin   2Vin2   3Vin3 Here Vin refers to the ac component
 ID 1 2 ID 1 3 ID
1  , 2  , 3 
 Vin V
2  Vin2 Vin0
6  Vin3 Vin0
in0

 ID
I D  K VGS  VTH  , g m   2 K VGS 0  VTH   2 K   RS I D 0  VTH 
2

 VGS Vin0

where Vin 0 and I D 0 are the dc bias values

VGS  Vin  RS I D  I D  K Vin  RS I D  VTH 
2

I D  I D  I D  I 
  2 K Vin  RS I D  VTH  1  RS   2 K   RS I D 0  VTH   1  RS D 
Vin  Vin  Vin     Vin 
Vin0 gm  Vin0 

I D gm
 1   ,
Vin Vin 0
1  g m RS
Similarly, we can find,
1 2ID K 1 3 I D 2 K 2 RS
2   , 3  
2 Vin2 1  g m RS  6 Vin3 1  g m RS 
3 5
Vin 0 Vin 0

From these equations, IIP3 and IIP2 can be obtained as

4 α1 ( 1  g m RS )2 2 gm  g
AIIP 3   , AIIP 2  1  m 1  g m RS 
3 α3 K 3 RS 2 K
MOSFET Cascode LNA
CG stage

For  high, Cgs acts as a short. Hence noise current will be shorted.
For  medium, Cgs acts as low impedance

ro 1 id
vgs   (id  g m vgs )  vgs 
jC gs g m  jCgs
id id
g m vgs    id
j Cgs j
1 1
gm T
Hence the drain noise current does not pass through ro .
For  low, Cgs acts as high impedance,
• Cascode noise does not appear at o/p so noise id ro
vgs   (id  g m vgs ) ro  vgs 
analysis same as inductor degenerated CS stage. 1  g m ro
• Cascode device is unilateral so stability improves id g m ro
g m vgs   id
• Linearity improved by Ls . 1  g m ro
Hence again the drain noise current does not pass through ro .
33
Jayanta Mukherjee
Common Source with Resistive Feedback
f assumed to be 1 for all cases

I X  g mVX
VX 1 1
   Rin , For input matching, RS  Rin  RS 
I X gm gm
X ac circuit Also, Vout  VX - I X RF  VX - g mVX RF
V out R
  1 F (1)
VX RS
Vin R in V R in
Further, VX   X  (2)
R in  R S Vin R in  R S
From (1) and (2),
V out  RF   R in  1  RF 
Av   1     1   (i/p matched)
Vin  RS   R in  R S  2  RS 

small signal model

Common Source with Resistive Feedback (Noise
analysis) 1
R   R  R  (Prove it !)
2
out F S

 RF  RS 
2

V 2
I R 2 2
 4kT  g m ,

n,out,M1 d out
4
+ i/p referred
voltage noise due
to M1
2 2
Vn2,out ,M 1 4kT 4kT 2 1  RF  kTRS2  RF 
V 2
 RS2 , Vn,out,R
2
 V 2
A2
 R  1    1  
 n,in,RF
RF F n,in,RF v
RF
S
4  RS  RF  RS 
contribution of R F 
Rin - to onput referred
noise voltage
current noise
of R F

Hence,
 RF  RS 
2 2
kTRS2  RF 
4kTγg m  1  
2
Vn,out,M  Vn,out,R
2
4 RF  RS 
NF  1  1 F
 1
Av2  4kTRS  1  RF 
2

1    4kTRS 
4  RS 
+ 2
 R  RF   RS 
 1   g m RS  S   
 RS  RF   R F 
+ Vn2,out ,RF
   For R S  R F  50  
Vn2,in ,RF
-  1   For R F high 
- For,   2 / 3, and R F high, the NF is equal to 2.2 dB
Some other topologies

36
Jayanta Mukherjee
Noise cancelling LNA

X
A1 is the gain of noise voltage of M1 at the o/p.
It can be shown that the noise voltage of auxillary
amplifier can be used to cancel the noise of M1.

Auxiliary amplifier
Current reuse LNA
VDD
• Not cascode. Both M1 and M2 act as CS stages.

• Additional gain due to 2 stages but no additional bias current needed.

R2 LD
• Voltage head room reduced.

Vout • Less reverse isolation than cascode

RF o/p of M1
as input of
M2 M2 Same bias current for M1 and M2 hence the term “current reuse”

C1 R1
C2
X No RF coupling between drain of M1 and source of M2

Vin M1

38
Jayanta Mukherjee
N Path Filtering
1
Normalized LO magnitude (i.e. VDD=1)  a1  1  j  ( Prove!)
2
iBB

+
RSW
VBB 

- S1 (t )  ae n
jnLO t
 LO1 VDD
iBB n  LO1

0
+   jn
RSW
VBB
S 2 (t )  ae n
2
e jnLOt  LO 2
LO2
VDD
n  0
- 
VDD
iRF
iBB S3 (t )  ae
n 
n
 jn
e jnLOt  LO3 LO3
0
+
RSW
R
in   VDD
ae
jn
jnLO t
VBB
S 2 (t )  n
2
e  LO 4 LO4
- n 
0
iBB 0 T/4 T/2 3T/4 T
RSW + Fourier series coefficients
2
VBB LO 
- T

39
Jayanta Mukherjee
N Path Filter (..Contd) RSW
iBB
+
VBB
S1 (t ) 


ae n
jnLO t
 LO1
n 
-  
iBB  jn

+
S 2 (t )  ae n
2
e jnLOt  LO 2
RSW n 
VBB 
-
iBB
S3 (t )  ae
n 
n
 jn
e jnLOt  LO3
iRF 
When switch 1 is ON R in
RSW +
S 2 (t )  

an e
jn
2
e jnLOt  LO 4
VBB
n 
iBB (t )  S I (t )iRF (t ) -
iBB

Voltage across z BB (t ) is given by, RSW +

VBB

v BB (t)=iBB (t )  z BB (t )   S I (t )iRF (t )   z BB (t )  z BB acts like a LP filter -

4 VDD
vRF (t )  RSW iRF (t )   S k (t )   S k (t )iRF (t )  z BB (t ) LO1 0
k 1
 
LO Baseband VDD
LO2 0
Fourier Transform of Sk (t )   S k (t )iRF (t )  z BB (t ) is given by, V
LO3 DD
   0
 j ( n  m )( k 1)
 e
m  n 
2
an am I RF (  (n  m)LO )Z BB (  n LO ) (Prove it!) V
LO4 DD
0
0 T/4 T/2 3T/4 T
Here k can be any of 1,2,3,4
40
Jayanta Mukherjee
In order to make the summation terms in the expression of v RF (t) to be in phase,
take (n+m)=4p, where p=0,1,2 .... Then v RF (t) becomes,
 
VRF ( )=RSW I RF (t )  4   an am I RF (  (n  m) LO ) Z BB (  nLO )
m  n 

Considering only the terms which contain I RF ( ), i.e. n  m  0, we get,


VRF ( )=RSW I RF ( )  4  an I RF ( ) Z BB (  nLO )  am  a- n  an*
2
(Prove it!)
n 

VRF ( ) 
 RSW  4  an Z BB (  nLO )
2
 Z IN ( ) 
I RF ( ) n 

… … .. … … ..

0 -2LO -LO 0 LO 2LO

ZBB ZIN
41
Jayanta Mukherjee
 … … .. … … ..

0 -2LO -LO 0 LO 2LO

ZBB()

Considering only these 2 bands

2
Z IN ( )  RSW   Z BB (  LO )  Z BB (  LO )
2
• Base band narrow band filtering is shifted to rf frequencies. This way high Q filtering
at RF is obtained even though we are only using a physical low frequency, low Q, filter.
• Directly channel can be filtered.
• Mixer first topology can also be implemented, if filtering of interferers is good enough.
• This scheme took some to become popular because good switching was not available earlier
at RF frequencies. 42
Jayanta Mukherjee