Upload
rakesh-rt
View
29
Download
9
Embed Size (px)
Citation preview
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 1/18
Electrical Engineering Department (ISY), Linköping University [email protected]
Tutorial-1 Low Noise Amplifier (LNA) Design
By Rashad.M.Ramzan
[email protected] Objective: Low noise amplifiers are one of the basic building blocks of any communication system. The purpose of the LNA is to amplify the received signal to acceptable levels with minimum self-generated additional noise. Gain, NF, non-linearity and impedance matching are four most important parameters in LNA design. The objective of this tutorial is to outline the basic tradeoffs between different amplifying topologies w.r.t gain, NF and impedance matching. After this comparison it is concluded that inductor degenerated common source topology gives the best performance to meet the gain, NF, and impedance matching goals with minimum power consumption in case of narrowband designs.
Goals: After this tutorial, students should be able to
• Calculate the gain, input impedance and NF of common gate, common source, and shunt feedback amplifiers.
• Understand the basic equations and tradeoff between different LNA topologies.
• Perform the calculation for inductor degenerated common source topology and understand the tradeoff between the gain, NF, and impedance matching.
A supplement tutorial LNA lab is also part of this course which takes the circuit from Problem-1.8 and guides through different analyses to design a practical LNA.
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 2/18
Electrical Engineering Department (ISY), Linköping University [email protected]
VDD
RL
VGS x(t) NMOS
DC-bias
Problem-1.1(Tutorial) NMOS transistor is racing horse in LNA design arena due to its higher mobility compared to PMOS transistors. Calculate the IP3 of NMOS CS amplifier shown below. Assume that NMOS transistor is in saturation.
a) Consider simplified square law model. (HW)
2( )2
nD GS T
KI V V= −
b) Consider the short channel effects as (Tutorial) 2
1
( )2 1 ( )
,0.2 0.1
n GS TD
GS T
GS T
K V VIV V
Velocity Saturation Mobility DegradationV V V and V
θθ
θ −
⎡ ⎤−= ⎢ ⎥+ −⎣ ⎦=
− = =
Observe that this transistor is not a very “short channel” device as θ << 1. c) What conclusion can be drawn from part b) about the bias current and transconductance
of the transistor for higher IP3?
Solution: a). Homework answer: ∞=3IP b).
)()()()( 33
2210 txtxtxxy αααα +++= --------------(1)
3
1321 3
4coscos)(αα
ωω =⇒+= IPAtAtAtx
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 3/18
Electrical Engineering Department (ISY), Linköping University [email protected]
( ))(12
2
TGS
TGSnD VV
VVKI−+
−=
θ
Here we assume a small signal x(t) around the bias (VGS – VT), so
( )[ ]))((1
)(2
2
txVVtxVVKI
TGS
TGSnD +−+
+−=
θ
we define VVV TGS Δ=− --------------- Bias voltage
[ ]1))((
)(2
2
+Δ+Δ+
=VtxVtxKI n
D θ --------------------------------------------------------------(2)
( )( )Vtx
VtxRKVRIV LnoLDo Δ++
Δ+=⇒=
)(1)(
2
2
θ and we put KRK Ln =
2
1<<θ so ( )Vtx Δ+)( is also small 21
11 ρρ
−=+
⇒
( )( )
2)(1
)(11 Vtx
VtxΔ+
−≈Δ++
θθ
( ) ( )⎟⎠⎞
⎜⎝⎛ Δ+−Δ+=
2)(1)( 2 VtxVtxKVo
θ
( ) ( )2
)()( 32 θKVtxVtxKVo Δ+−Δ+=
)(2
)(2
3
)(2
322
32
232
txKtxVKK
txVKVKVKVKVo
θθ
θθ
−⎟⎠⎞
⎜⎝⎛ Δ−+
⎟⎠⎞
⎜⎝⎛ Δ−Δ+Δ−Δ=
-----------------------------(3)
Comparing (1) & (3)
21 2
32 VKVK Δ−Δ=θα , VKK Δ−=
23
2θα ,
23θα K
−=
⎟⎠⎞
⎜⎝⎛ Δ−
Δ=
Δ−Δ×== 2
2
3
13 32
38
2
232
34
34 VV
K
VKvKAIP θθ
θ
αα
θθVVAIP
Δ=
Δ=
3162
38
3 As 1<<θ 23 VΔ can be ignored.
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 4/18
Electrical Engineering Department (ISY), Linköping University [email protected]
( )θθ
TGSIP
VVVA−
=Δ
=3
163
163 -------------------------------(4)
Please, note that this formula only holds for small value of θ. For larger θ values compare the lecture notes. In any case a large gate bias voltage (VGS – VT) improves IP3.
Put -1V 0.1 V,2.0 ==Δ θV
VoltsAIP 27.31.02.0
316
3 ==
dBmmWdBmIIP 201501.
227.3log10)(3
2
≅⎥⎥
⎦
⎤
⎢⎢
⎣
⎡
⎥⎥⎦
⎤
⎢⎢⎣
⎡⎟⎟⎠
⎞⎜⎜⎝
⎛=
c). From ( ))(12
2
TGS
TGSnD VV
VVKI−+
−=
θ the NMOS transconductance can be found as
( )( )2)(1
)(2)(2 TGS
TGSTGSnm VV
VVVVKg−+
−+−=
θθ
.
By comparison of those two formulas we find
2)(2)(1)( TGS
TGS
TGSTGS
m
D VVVVVVVV
gI −
≅−+−+
×−=θθ
and hence, (4) can be rewritten as
m
DIP g
IAθ3
323 ≅
As shown, IIP3 is decided by the ratio ID/gm which is constant for a given gate bias voltage. Using e.g. a wider transistor does not change this ratio and only the power consumption is increased.
Problem-1.2 (Tutorial) It is preferred in current RF designs that the input of LNA be matched to 50 Ω. The easiest way is to shunt the gate with a resistor of 50 Ω.
a) Calculate the gain, input impedance and NF in absence of gate noise. Assume that Rsh=RL for NF derivation.
b) What are the disadvantages of shunt resistor with reference to gain and NF?
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 5/18
Electrical Engineering Department (ISY), Linköping University [email protected]
gmVgs
S
D Vout
RLdi2
RL is noiseless
G
Rsh
RshnV ,2
RsnV ,2 Rs
VDD
RL
Rsh
Vin
Vout
Zin
Rs
(Biasing not shown)
Solution: a). (Please read assumption in the problem statement carefully)
sourceinputtoduenoiseOutputpowernoiseoutputTotalF =
fkTRV sRsm Δ= 4,2 LmGate
RgGain −=
fkTRV shRshm Δ= 4,2 ⎟⎟
⎠
⎞⎜⎜⎝
⎛+
=shs
shLm RR
RRgA for Rsh = Rs
fgkTi md Δ= γ42 2
Lm
RgA −=
Using superposition, considering one at a time and shorting / opening other sources. 2
22,
2,
2⎟⎟⎠
⎞⎜⎜⎝
⎛+
××=shs
shLmRsnRson
RRRRgVV
222
,2
,2
⎟⎟⎠
⎞⎜⎜⎝
⎛+
××=shs
sLmRshnRshon
RRRRgVV
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 6/18
Electrical Engineering Department (ISY), Linköping University [email protected]
Lddno RiV 22,
2 ×=
Rson
doRshon
Rson
dnoRshonRson
VVV
VVVVF
,2
,2
,2
,2
,2
,2
,2
1 ++=
++=
( )
( ) ( )2
222
2
2
222
2
222
4
4
4
41
shs
sLmS
Lm
shs
sLms
shs
shLmsh
RRRRgfkTR
RfgkT
RRRRgfkTR
RRRRgfkTR
F
+×
×Δ
×Δ+
+×
×Δ
+×
×Δ+=
γ
In case of impedance match Rs = Rsh
smLms
mL
S
LSms
mL
RgRgR
gR
RRRg
R
gRF
γγγ 42
4
2
4
11 22
2
2
222
2
+=×
×+=
××
++=
b). - Poor Noise Figure
- Input signal attenuated by voltage divider
- Rsh adds extra noise.
- At high frequency, shunt L is needed to tune out Cgs
- Reduced gain.
Problem-1.3 (Tutorial) Another approach to get 50 Ω input impedance match is shunt feedback amplifier shown below.
a) Calculate the gain, input impedance and NF neglecting the gate noise. The gate-drain,
gate-bulk, and gate-source capacitance can be neglected as well. b) What are the disadvantage of shunt feedback amplifier with reference to gain and NF?
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 7/18
Electrical Engineering Department (ISY), Linköping University [email protected]
gmVgs
Vout
RLnDI 2
(Equivalent noise model ignoring gate noise), RL is noiseless
Rs
RsnV ,2
RFnV ,2 RF
VDD
RL Vout
Zin
Rs
(Biasing not shown)
RF
Vin
gmVgs
Vout
RL
Rs RF
Vgs
iin
Vin
RL
Vout
Iin Rs
RFIin
Solution:
fkTRVfgkTI SRSmnD Δ=Δ= 4,4 22 γ
sourceinputtoduepowernoiseOutputpowernoiseinputTotal
VAVF
RStotv
outn ==2
,2
,2
Here Av,tot = Gain from Vin to Vout
Again using superposition theorem
RStotv
outDnoutRFnoutRSn
RStotv
outn
VA
VVV
VAVF
2,
2
,2
,2
,2
2,
2
,2 ++
==
Gain Calculation
( ) outFSinin VRRiV ++=
( ) Lgsminout RVgiV −=
oFings VRiV +=
( )LSmLFS
LmL
in
outtotv RRgRRR
RgRVVA
+++−
==1
,
If RF>>RS & gmRF>>1
Lm
F
SmL
F
S
Lmtotv Rg
RRgR
RR
RgA −≅+
+++
−= 11
,
Lmtotv RgA −≅,
Also Lm
LFin Rg
RRZ++
=1
By ignoring Cgs, we have considered real part only.
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 8/18
Electrical Engineering Department (ISY), Linköping University [email protected]
gmVgs
outRFV ,2
RL
Rs RF
Vgs
i RFV 2
gmVgs
outnDV ,2
RL
Rs RF
Vgs
i
DnI ,2
For source resistance
nRStotvoutnRS VAV 2,
2,
2 = ---------------(1) For feedback resistance
outRFRFFSgs VViRiRV ,+−=−=
( )gsmLoutRF VgiRV −=,
( )
( )SmF
LRF
smL
FSRFoutRF Rg
RRV
RgRRRVV +=
++
+= 1
11
1,
( )2
,2
,,2 1 ⎥
⎦
⎤⎢⎣
⎡+= Sm
F
LRFnoutRFn Rg
RRVV ---------------------(2)
Similarly
0,, =+
+++FS
outnDgsmnD
L
outnD
RRV
VgIR
V
FS
outnDSgs RR
VRV
+= ,
LnD
FS
Sm
FSL
DnoutnD RI
RRRg
RRR
IV ≈
++
++
=11
,,
So,
22,
2LnDoutnD RIV = ------------------------------------------(3)
Combaining (1) (2) & (3)
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 9/18
Electrical Engineering Department (ISY), Linköping University [email protected]
( )
RSntotv
LnD
RSntotv
SmF
LRFn
VARI
VA
RgRR
VF
,2
,2
22
,2
,2
2
,2 1
1 +⎥⎦
⎤⎢⎣
⎡+
+=
Lmtotv RgA −=, , fkTRV SRSn Δ= 4,2 , FRFM kTRV 4,
2 = & mnD gkTI γ42 =
SmSmF
S
RgRgRRF γ
+⎟⎟⎠
⎞⎜⎜⎝
⎛++=
2111
b).
NF ↓ gmRS ↑ & RF ↑ usually Ω= 50SR
- Better performance than CS amplifier
- RF induces noise
- At higher f ↑ a shunt inductor needed to tune out Cgs
- Broadband Amp @ Lower frequency
- To make NF ↓ RF > RS and gmRS >> 1
Problem-1.4 (HW) Common gate amplifier also offers 50 Ω input impedance match and solves the input matching problem.
c) Calculate the gain, input impedance and NF in absence on gate noise. Neglect gate drain
and gate to bulk and gate to source capacitance. a) What are the disadvantage of common gate amplifier with reference to gain and NF?
Problem-1.5 (Tutorial) The disadvantages of the amplifiers discussed in Problem-2, 3 & 4 can be circumvented by using the source degenerated LNA shown below.
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 10/18
Electrical Engineering Department (ISY), Linköping University [email protected]
VDD
RL
VS VoutRs
(Biasing not shown)
Ls
Lg
VS Rs
Ls
Lg iin Voutio
gmVgs
Vin Vgs
Zin
a) Calculate the input impedance. This inductor source degenerated amplifier presents a
noiseless resistance for 50Ω for input power match. How we can cancel the imaginary part of complex input impedance so that the LNA presents 50Ω real input resistance at input port.
b) Calculate the NF in absence on gate noise. Neglect gate drain and gate to bulk and gate to source capacitance.
c) Cgd bridges the input and output ports. The reverse isolation of this LNA is very poor. Why reverse isolation is important? Suggest the modification to improve reverse isolation.
Solution: a). From model above we can write
( ) soinsginin Ljicj
iLjLjiV ωω
ωω +⎟⎟⎠
⎞⎜⎜⎝
⎛++=
1 ---------------(1)
gsinmgsmo Cj
igVgiω
1×== --------------------------------------(2)
Substituting (2) in (1)
( )⎥⎥⎦
⎤
⎢⎢⎣
⎡+++=
gs
sm
gssginin C
LgCj
LLjiVω
ω 1
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 11/18
Electrical Engineering Department (ISY), Linköping University [email protected]
Vin
Rs Lg + Ls
gs
sm C
Lg Zin VgsCgs
Reference: For series RLC Circuit
R L
C VC Vin
RCRL
CL
RQ
o
os ω
ω 11===
and inSC VQV =
frequency of current gain equal 1
( )gs
sm
gssg
in
inin C
LgCj
LLjiVZ +++==
ωω 1
( )gs
sm
gssgin C
LgCj
LLjZ +++=ω
ω 1
For matching Lg + Ls are canceled out by Cgs. So at frequency of interest
( ) ( ) gssgo
gsosgo CLLC
LL+
=⇒=+11 2ω
ωω
And sgs
mS L
CgR =Ω= 50
Notes: a). Ls is typically small and may be realized by the bond wire for source. b). Lg can be implemented by spiral/external inductor.
b). From part a)
( )gs
sm
gssgin C
LgCj
LLjZ +++=ω
ω 1
We can draw this circuit as Here
( ) ( )STS
sgo
gs
SmS
sgoin LR
LL
CLg
R
LLQ
ωωω
+
+=
+
+=
gs
mT C
g≅ωQ
gsgs
SmSo
in
CC
LgRQ
⎟⎟⎠
⎞⎜⎜⎝
⎛+
=
ω
1 for match load
gs
SmS C
LgR =
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 12/18
Electrical Engineering Department (ISY), Linköping University [email protected]
RL Vin
Vout Rs
Ls
Lg
Zin Vgs
gssoin CR
Qω2
1=
Gain
inings VQV =
gs
outm V
Ig =
minin
mgs
in
outm gQ
VgV
VI
G ===
minm gQG =
so, Lmin
out RGVV
−= where minm gQG =
Noise Figure:
sourceinputtodueoutputatpowernoiseoutputatpowernoiseTotal
F =
For this calculation we ignore channel noise.
OUTnRS
OUTnD
OUTnRS
OUTnDOUTnRS
VV
VVVF
,2
,2
,2
,2
,2
1+=+
=
LDnOUTnD RiV 2,
2,
2 = fgkTi mDn Δ= γ4,2
LmRSnOUTnRS RGVV 22,
2,
2 = fkTRV SRSn Δ= 4,2 & minm gQG =
2222,
22,1
LminRSn
LDn
RgQV
RiF += mDn gkTi γ4,
2 = , SRSn kTRV 4,2 =
21
inSm QRgF
γ+=
Notes: - Very good NF value - Narrow band matching
- NF ↓with 2Q
- The Q is dependent upon Lg + Ls, Ls usually small so Q depends mainly upon Lg
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 13/18
Electrical Engineering Department (ISY), Linköping University [email protected]
VDD
RL
VS Vout Rs
Ls
Lg
VDD
LD
VS Rs
Ls
Lg CL RL generates noise so replace
RL with LD so that’s
LDo CL
1=ω
Reverse Isolation
Vout
Lo
VDD
LD
Rs
Ls
Lg
CL
Vb
(Final Design)
Cgd
C). Drawbacks i). The CL can be considered the input capacitance of the following mixer or filter. ii).
Reverse isolation depends upon capacitance between output and input. To make it less the cascode architecture can be used.
Problem-2.6 (HW) Fill-in the Table below, use the data from Problem-2.4, 2.5, 2.6 and 2.7
Type of LNA Zin Noise Factor Gain NF (dB)
Shunt Resistor Rsh 42m Sg Rγ
+ 2m Lg R−
Common Gate
Shunt Feedback
Source Degenerated
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 14/18
Electrical Engineering Department (ISY), Linköping University [email protected]
a) Calculate the NF for all above amplifiers. Assume γ=2, gm = 20mS, Rs = 50Ω, RF =
500Ω, and Qin = 2. b) Which is the best topology for Narrow Band LNA design at high frequency?
Problem-1.7 (Tutorial) Real Design: We will design the inductor-source-degenerated LNA shown in Fig below to meet the specification outlined for IEEE802.11b standard. The first cut approximate values are calculated as a starting point for simulation. In LAB3: Design of LNA you will take the same design and modify these component values to meet the specification.
LNA Specification:
NF < 2.5 dB, Gain > 15dB, IIP3 > -5dBm, Centre Frequency = 2.4 GHz
S11 < -20dB, S22 < -10dB, Load Capacitance = 1pF
Technology Parameters for 0.35um CMOS: 2 2 20.35 , 170 , 4.6 , 58 , 2eff n ox ox p oxL m C A V C mF m C A Vμ μ μ μ μ γ= = = = =
Solution:
Technology 0.35μm CMOS: ⎪⎭
⎪⎬⎫
⎪⎩
⎪⎨⎧
===
==
mLmmFC
VACVAC
effox
oxpoxo
μγ
μμμμ
35.0,2,6.4
,58,1702
22
Design Parameters
NF < 2.5 dB, Gain > 15dB, IIP3 > -5dBm, f0 = 2.4 GHz
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 15/18
Electrical Engineering Department (ISY), Linköping University [email protected]
Component Description
Ls – Matches input impedance
Lg – Sets the Resonant Frequency fO = 2.4 GHz
M3 – Biasing transistor which forms current mirror with M1
Ld – Tuned output increases the gain and also work as band pass filter with CL
M2 – Isolates tuned input from output to increase reverse isolation, also reduces the effect of Miller capacitance Cgd
CB – BC blocking capacitor chosen to have negligible reactance at fO = 2.4 GHz
RBIAS – Large enough so that its equivalent current noise is small enough to be ignored. (Don’t consider it as voltage noise source. Why??)
Design Procedure
Size of M1:
From the noisy two-port theory (see the course book or lecture notes):
( )Ω
=−=50
115
2CCG gsopt γδαω ----------------------(1)
mmLCCWpFC effOXgsMgs 42/34 1 ≈≈⇒≅ ( not feasible – huge size, huge power ! )
We will not go for global minimum noise figure. Instead, we will look into the constraint power design approach.
Solution:
LNA NF will be optimized for given power that is less than the global minimum.
In this case the optimum transistor width is
Soxeffoopt RCL
Wω3
1=
while the minimum power-constraint NF :
LD RREF
Ls
Lg
M2
Vout M1
CL = 10pF
M3 RBIAS
RS CB Vin
VDD
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 16/18
Electrical Engineering Department (ISY), Linköping University [email protected]
Tp
Tp FF
ωω
ωω
αγ 6.514.21 min,min, +=⇒+= ---------------(A)
As compared to (A) from (1) we can derive
( )TT
CCFωωγ
ωω 3.211
521 2
min +=−+=
T
Fωω3.21min += -------------------------------------------------(B)
(A) is minimum NF for a given power consumption.
(B) is global minimum noise figure.
In practice the difference is usually 0.5dB to 1dB (no big deal for Lower Power)
Step - 1:
mAII 521 == (Limited Power consumption)
Step - 2:
SoxeffM RCL
W0
1 31
ω=
oM m
Wωμ ××××
=506.435.03
11
⎪⎭
⎪⎬
⎫
⎪⎩
⎪⎨
⎧
==
===Ω=
GHzff
mLVACmmFCR
ooo
effoxn
oxS
4.2,2
,35.0,170,6.4,50 2
πω
μμμ
41 109.3 −×=MW
mWM μ390109.3 41 =×= −
Step - 3:
oxeffMgs CLWC 11 32
=
pFmCgs 41.06.435.039032
1 =×××= μμ
11
1 2 DMM
oxnm IL
WCg ⎟⎠⎞
⎜⎝⎛= μ or
TGS
DMm VV
Ig
−= 1
1
2 (for short channel model)
VmAmgm 43535.0
39017021 =×⎟⎠⎞
⎜⎝⎛××= μ
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 17/18
Electrical Engineering Department (ISY), Linköping University [email protected]
SecradGpF
VmACg
gs
mT 104
41.043
1
1 ==≈ω
Assuming 2=γ
Now T
oFωω6.51min +=
dBG
GF 55.2104
4.226.51min ≈+=π
dBNF 55.2≈
This NF is very close to the specified value. If we increase ID then Tω should increase slightly as well and hence, a lower NF value can be achieved at expense of more power.
Step - 4:
Source and gate inductance such that they cancel Cgs and set Ω50 input impedance
SecradGfoo 154.222 === ππω
From previous problem
STgs
SmdTransformeS L
CLg
RR ω≅==
nHG
RLT
SS 5.0
10050
≅==ω
nHLS 5.0= can be implemented using the bond wire.
Now 1
20
1
gssg C
LLω
=+
( )nH
pFGLL sg 81.10
41.0151
2 =×
=+
nHLg 10≈
Step - 5:
pFCC
L LLo
d 112 == Q
ω
( )nH
pFGLd 4.4
11512 ≅×
=
nHLd 4.4=
Spring 2009: Radio Frequency Integrated Circuits (TSEK03) 18/18
Electrical Engineering Department (ISY), Linköping University [email protected]
Step - 6:
Size of M3 is chosen to minimize power consumption
mAIkRmW REFM 6.02,70 33 =⇒Ω== μ
Ω= kRBIAS 2 (Large enough so that it’s equivalent current noise can be neglected)
pFCB 10= ( Ω≈ 6.6CX so good value @ 2.4G Ω== 6.62
1
BoB Cf
Xπ
)
Step - 7:
Size M2 = M3
So that they can have shared Drain Area..
(Note: We will simulate same LNA circuit in LAB # 2)
Problem-1.8 (Point to Ponder): Connecting two inductor-source-degenerated LNAs as shown beneath makes a differential LNA. The differential LNA has the advantage of a higher common mode rejection ratio and less sensitivity to the ground inductance variation Ls compared to its single-ended counterpart.
a) Compare intuitively the NF of single ended and differential amplifier if both have same power consumption.
b) If low power is not a parameter on interest, which LNA has lower NF?
Instructions: For hand calculation of NF you can ignore the gate noise of the device and noise generated by the load resistance RL.