Upload
vuongkhue
View
220
Download
2
Embed Size (px)
Citation preview
EECS 142
Lecture 3: High-Speed Amplifiers and Tuned Amplifiers
Prof. Ali M. Niknejad
University of California, Berkeley
Copyright c© 2005 by Ali M. Niknejad
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 1/44 – p. 1/44
CS/CE Amplifier Bandwidth
RL
Cdb + CL
Cµ
+vs
−
+vo
−RS
Due to Miller multiplication, the input cap is usually thedominant pole
ω−1
0≈ Rs(Cin + |Av|Cµ)
ω−1
0= RsCin(1 + µ|Av|) ≈ RsCinµ|Av|
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 2/44 – p. 2/44
CS/CE Amplifier Gain-Bandwidth
Assuming the voltage gain is given by thelow-frequency value of gmRL, we have
ω−1
0= RsCinµgmRL = (gmRs)(gmRL)
Cin
gmµ
ω−1
0= |Av|2
Rs
RLω−1
T µ
The amplifier has a bandwidth reduction factor of A2v
ω0 × |Av|2 = ωT ×(
RL
Rs
)
× 1
µ
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 3/44 – p. 3/44
Bandwidth Example
Say we need a gain of 60 dB (Av = 1000) and RL
Rs= 2.
The technology has a capacitance ratio of µ = 0.2:
ω0|Av|2 = 106ω0 = ωT × 2 × 5
ω0 =ωT
105
Compare this to a current mirror amplifier. When wefollow the “normal” gain-bandwidth tradeoff, we have
ω0 =ωT
Ai=
ωT
1000
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 4/44 – p. 4/44
Common Base Amplifier
RL
+vs
−
RS
roCin
Cµ
gmvin
+vin
−
+vs
−
RS
RL
Write KCL at base node of circuit
vs + vin
Rs+gmvin+sCinvin = 0 vs = vin(1+gmRs+sCinRs)
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 5/44 – p. 5/44
Common Base Amp (cont)
And write KCL at the output node
(sCo +1
RL)vo + gmvin + sCµvo = 0
vo
(1
RL+ s(Co + Cµ)
)
= gmvin
The voltage gain is a product of two terms
Av =vo
vs=
gmRL
1 + s(Co + Cµ)RL
vx
vs
Av =GmRL
(1 + s(Co + Cµ)RL)(1 + sRsCin
1+gmRs)
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 6/44 – p. 6/44
Common Base Bandwidth
Note the transconductance is degenerated,Gm = gm/(1 + gmRs). Note that the input capacitance isalso degenerated by the action of series feedback.
Unlike a CE/CS ampilifier, the poles do not interact (dueto absence of feedback capacitor)
First let’s take the limit of high loop gain, gmRs ≫ 1
Av =RL
Rs
(1 + s/ωT )(1 + s/ωL)
where ωL = ((Co + Cµ)RL)−1 is the pole at the output.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 7/44 – p. 7/44
Matched Common Base Amp
The common-base amplifier has the nice property thatthe input impedance is low (roughly 1/gm) andbroadband, thus easily providing a termination to thedriver (a filter, the antenna, or a previous stage). If weassume that Rs = 1/gm, we have
Av =1
2gmRL
(1 + s/2ωT )(1 + s/ωL)
The 3dB bandwidth is thus most likely set by the timeconstant at the load.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 8/44 – p. 8/44
Shunt Feedback Amp
+vs
−
RS
RL
RF
vo
VCC
C∞
The shunt-feedback amplifier is a nice high-frequencybroadband amplifier building block. The action of theshunt feedback is used to lower the input impedanceand to set the gain.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 9/44 – p. 9/44
Shunt Feedback Gain / Input Resis
The in-band voltage gain and input impedance is givenby (see PS 1)
Av =RF
Rs
Rin = (1 +RF
RL)
1
gm
For an input match, Rs = (1 + RF
RL) 1
gm, or gmRs = (1 + RF
RL)
Since the voltage gain sets RF , the input impedancematch determines the required transconductance gm
(and hence the power dissipation)
A bipolar version will dissipate much less power due tothe higher intrinsic gm
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 10/44 – p. 10/44
Shunt Feedback Amp BW
The amplifier is broadband and approximately obeysthe classic gain-bandwidth tradeoff Avω0 ≈ ωT
A zero-value time constant analysis identifies thedominant pole
τ1 = Cin
(
Rs||rπ||RF (1 + RL
RF)
1 + gmRL
)
τ2 = Cµ
(
RF ||RL(Rs||rπ)
Rs||rπ|| 1
gm||RL
)
ω−3dB ≈ 1
τ1 + τ2
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 11/44 – p. 11/44
Shunt Feedback/CC Cascade
If the shunt-FB amplifier needs to drive a lowimpedance load, a broadband voltage buffer is needed
As shown below, an emitter follower (or source follower)provides the solution (note this is a fast pnp). Note thebuffer is broadband (gain ≈ 1) and only loads the coreamplifier by the degenerated input capacitanceCin2/(1 + gm2RE)
+vs
−
RS
RL
RF
vo
VCC
C∞
RE
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 12/44 – p. 12/44
Series RLC CircuisZ
L
R
C+vs
−
+vo
−
The RLC circuit shown is deceptively simple. Theimpedance seen by the source is simply given by
Z = jωL +1
jωC+ R = R + jωL
(
11
ω2LC
)
The impedance is purely real at at the resonantfrequency when ℑ(Z) = 0, or ω = ± 1√
LC. At resonance
the impedance takes on a minimal value.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 13/44 – p. 13/44
Series Resonance
vR
vC
vL
vs
ω < ω0 ω = ω0
vR
vC
vL
vs
ω > ω0
vR
vC
vL
vs
It’s worthwhile to investigate the cause of resonance, orthe cancellation of the reactive components due to theinductor and capacitor. Since the inductor and capacitorvoltages are always 180◦ out of phase, and onereactance is dropping while the other is increasing,there is clearly always a frequency when themagnitudes are equal.
Resonance occurs when ωL = 1
ωC .A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 14/44 – p. 14/44
Quality Factor
So what’s the magic about this circuit? The firstobservation is that at resonance, the voltage across thereactances can be larger, in fact much larger, than thevoltage across the resistors R. In other words, thiscircuit has voltage gain. Of course it does not havepower gain, for it is a passive circuit. The voltage acrossthe inductor is given by
vL = jω0Li = jω0Lvs
Z(jω0)= jω0L
vs
R= jQ × vs
where we have defined a circuit Q factor at resonanceas
Q =ω0L
R
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 15/44 – p. 15/44
Voltage Multiplication
It’s easy to show that the same voltage multiplicationoccurs across the capacitor
vC =1
jω0Ci =
1
jω0C
vs
Z(jω0)=
1
jω0RC
vs
R= jQ × vs
This voltage multiplication property is the key feature ofthe circuit that allows it to be used as an impedancetransformer.
It’s important to distinguish this Q factor from theintrinsic Q of the inductor and capacitor. For now, weassume the inductor and capacitor are ideal.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 16/44 – p. 16/44
More of Q
We can re-write the Q factor in several equivalent formsowing to the equality of the reactances at resonance
Q =ω0L
R=
1
ω0C
1
R=
√LC
C
1
R=
√
L
C
1
R=
Z0
R
where we have defined the Z0 =√
LC as the
characteristic impedance of the circuit.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 17/44 – p. 17/44
Circuit Transfer Function
Let’s now examine the transfer function of the circuit
H(jω) =vo
vs=
R
jωL + 1
jωC + R
H(jω) =jωRC
1 ω2LC + jωRC
Obviously, the circuit cannot conduct DC current, sothere is a zero in the transfer function. The denominatoris a quadratic polynomial. It’s worthwhile to put it into astandard form that quickly reveals important circuitparameters
H(jω) =jωR
L1
LC + (jω)2 + jωRL
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 18/44 – p. 18/44
Canonical FormUsing the definition of Q and ω0 for the circuit
H(jω) =jωω0
Q
ω20
+ (jω)2 + j ωω0
Q
Factoring the denominator with the assumption thatQ > 1
2gives us the complex poles of the circuit
s± =ω0
2Q± jω0
√
11
4Q2
The poles have a constant magnitude equal to theresonant frequency
|s| =
√√√√
ω20
4Q2
/
+ ω20
(
11
4Q2
/)
= ω0
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 19/44 – p. 19/44
Root Locus
Q → ∞
Q <1
2
Q >1
2
"""""
"""""
""""""""""
""" !!!!!
A root-locus plot of the poles as a function of Q. AsQ → ∞, the poles move to the imaginary axis. In fact,the real part of the poles is inversely related to the Qfactor.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 20/44 – p. 20/44
Circuit Bandwidth
0.5 1 1.5 2 2.5 3 3.5 4
0.2
0.4
0.6
0.8
1
Q = 1
Q = 10
Q = 100
H(jω0) = 1
∆ω
ω0
As we plot the magnitude of the transfer function, wesee that the selectivity of the circuit is also relatedinversely to the Q factor.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 21/44 – p. 21/44
Selectivity
In the limit that Q → ∞, the circuit is infinitely selectiveand only allows signals at resonance ω0 to travel to theload.
Note that the peak gain in the circuit is always unity,regardless of Q, since at resonance the L and Ctogether disappear and effectively all the source voltageappears across the load.
The selectivity of the circuit lends itself well to filterapplications. To characterize the peakiness, let’scompute the frequency when the magnitude squared ofthe transfer function drops by half
|H(jω)|2 =
(
ωω0
Q
)2
(ω2
0ω2)2
+(
ωω0
Q
)2=
1
2
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 22/44 – p. 22/44
Selectivity BandwidthThis happens when
(ω2
0 ω2
ω0ω/Q
)2
= 1
Solving the above equation yields four solutions,corresponding to two positive and two negativefrequencies. The peakiness is characterized by thedifference between these frequencies, or thebandwidth, given by
∆ω = ω+ ω− =ω0
Q
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 23/44 – p. 23/44
Selectivity Bandwidth (cont)
The normalized bandwidth is inversely proportional tothe circuit Q.
∆ω
ω0
=1
Q
You can also show that the resonance frequency is thegeometric mean frequency of the 3 dB frequencies
ω0 =√
ω+ω−
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 24/44 – p. 24/44
Parallel RLCY
L RCi si o
The parallel RLC circuit is the dual of the series circuit.By “dual” we mean that the role of voltage and currentsare interchanged.
Hence the circuit is most naturally probed with a currentsource is. In other words, the circuit has current gain asopposed to voltage gain, and the admittance minimizesat resonance as opposed to the impedance.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 25/44 – p. 25/44
Duality
The role of capacitance and inductance are alsointerchanged. In principle, therefore, we don’t have torepeat all the detailed calculations we just performed forthe series case, but in practice it’s worthwhile exercise.
The admittance of the circuit is given by
Y = jωC +1
jωL+ G = G + jωC
(
11
ω2LC
)
which has the same form as before. The resonantfrequency also occurs when ℑ(Y ) = 0, or whenω = ω0 = ± 1√
LC.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 26/44 – p. 26/44
Duality (cont)
Likewise, at resonance the admittance takes on aminimal value. Equivalently, the impedance atresonance is maximum.
This property makes the parallel RLC circuit animportant element in tuned amplifier loads. It’s alsoeasy to show that at resonance the circuit has a currentgain of Q
iC = jω0Cvo = jω0Cis
Y (jω0)= jω0C
isG
= jQ × is
where we have defined the circuit Q factor at resonanceby
Q =ω0C
GA. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 27/44 – p. 27/44
Current Multiplication
The current gain through the inductor is also easilyderived
iL = jQ × is
The equivalent expressions for the circuit Q factor aregiven by the inverse of the previous relations
Q =ω0C
G=
R
ω0L=
R1√LC
L=
R√
LC
=R
Z0
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 28/44 – p. 28/44
Phase Response
The phase response of a resonant circuit is also relatedto the Q factor. For the parallel RLC circuit the phase ofthe admittance is given by
∠Y (jω) = tan−1
(
ωC(1 1
ω2LC
)
G
)
The rate of change of phase at resonance is given by
d∠Y (jω)
dω
∣∣∣∣ω0
=2Q
ω0
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 29/44 – p. 29/44
Phase Response
0.2 0.5 1 2 5 10
-75
-50
-25
0
25
50
75
100
ω/ω0
Q = 100
Q = 10
Q = 2
Q = 1
2
A plot of the admittance phase as a function offrequency and Q is shown. Higher Q circuits go througha more rapid transition.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 30/44 – p. 30/44
Circuit Transfer Function
Given the duality of the series and parallel RLC circuits,it’s easy to deduce the behavior of the circuit. Whereasthe series RLC circuit acted as a filter and was onlysensitive to voltages near resonance ω0, likewise theparallel RLC circuit is only sensitive to currents nearresonance
H(jω) =iois
=voG
voY (jω)=
G
jωC + 1
jωL + G
which can be put into the same canonical form asbefore
H(jω) =jωω0
Q
ω20
+ (jω)2 + j ωω0
Q
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 31/44 – p. 31/44
Circuit Transfer Function (cont)
We have appropriately re-defined the circuit Q tocorrespond the parallel RLC circuit. Notice that theimpedance of the circuit takes on the same form
Z(jω) =1
Y (jω)=
1
jωC + 1
jωL + G
which can be simplified to
Z(jω) =j ω
ω0
1
GQ
1 +(
jωω0
)2
+ j ωω0Q
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 32/44 – p. 32/44
Parallel Resonance
At resonance, the real terms in the denominator cancel
Z(jω0) =j R
Q
1 +
(jω0
ω0
)2
︸ ︷︷ ︸
=0
+j 1
Q
= R
It’s not hard to see that this circuit has the same halfpower bandwidth as the series RLC circuit, since thedenominator has the same functional form
∆ω
ω0
=1
Q
plot of this impedance versus frequency has the sameform as before multiplied by the resistance R.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 33/44 – p. 33/44
Tuned Amplifiers
Cdb + CL
Cµ
+vs
−
+vo
−
RLL The RLC loaded amplifier is atuned amplifier. A transconduc-tance device drives a shunt RLCload which results in a voltagegain.
Av = gmZ(jω) =gm
Y (jω)
The peak gain occurs at resonance
Av,max = gmReff
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 34/44 – p. 34/44
Tuned Amp (cont)
Reff is the loaded resistance of the tank
Reff = RL||ro||Rx,L||Rx,C
In order to maximize the gain, we employ high-Qinductors and capacitors in the load and omit theexplicit load resistance RL. The peak gain is thus
Av,max ≈ gm(Rx,L||Rx,C)
Assuming the Q factor is dominated by the inductor, agood assumption for monolithic IC inductors, we have
Av,max ≈ gmRx,L = gmQLωL
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 35/44 – p. 35/44
Shunt-Series Transformation
Rs
Xs
RpXp
The key calculation aid is the series to paralleltransformation. Consider the impedance shown above,which we wish to represent as a parallel impedance.
We can do this at a single frequency as long as theimpedance of the series network equals the impedanceof the shunt network
Rs + jXs =1
1
Rp+ 1
jXp
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 36/44 – p. 36/44
Transformation (cont)
Equating the real and imaginary parts
Rs =RpX
2p
R2p + X2
p
Xs =R2
pXp
R2p + X2
p
which can be simplified by using the definition of Q
Qs =Xs
Rs=
R2pXp
RpX2p
=Rp
Xp= Qp = Q
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 37/44 – p. 37/44
Transformation (cont)
Which shows that
Rp = Rs(1 + Q2)
andXp = Xs(1 + Q−2) ≈ Xs
where the approximation applies under high Qconditions.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 38/44 – p. 38/44
Q Calculation
We used the series to parallel transformation tocalculate the effective shunt resistance due to theinductor
Reff = (1 + Q2L)Rx,L ≈ Q2
LRx,L = Q2L
ωL
QL= QL × ωL
The gain is maximized at a fixed bias current andfrequency by maximizing the product QL × L. So intheory, there is no limit to the voltage gain of theamplifier as long as we can maximize the quality factorQL.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 39/44 – p. 39/44
Selection of L
Note that the capacitance of the circuit is notdetrimental since it is resonated away with the shuntinductance. In other words L is chosen such that
L =1
ω20Ceff
where Ceff = Cdb + (1 |A−1v |)Cµ + CL.
The ability to tune out the parasitic capacitances in thecircuit is a major advantage of the tuned amplifier. Thisis especially important as it allows low-power operation.Another important advantage of the circuit is that thereis practically no DC voltage drop across the inductor,allowing very low supply voltage operation.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 40/44 – p. 40/44
Tuned Amplifier Advantages
Another less obvious advantage is the improved voltageswing at the output of the amplifier. Usually the voltageswing is limited by the supply voltage and the Vds,sat ofthe amplifier.
In this case, though, the voltage can swing above thesupply. Since the average DC voltage across aninductor is zero, the output voltage can swing aroundthe DC operating point of Vdd.
This is a major efficiency boost for the amplifier and isan indispensable tool in designing power amplifiers andbuffers.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 41/44 – p. 41/44
Cascode Tuned Amplifier
+vs
−
+vo
−
M1
M2
Cdb + Cµ2 + CLCµ1
RLL Besides the obvious ad-vantage of boosting theoutput impedance, thusmaximizing the Q of theload, the cascode devicesolves a major stabilityproblem of the amplifier.
We’ll show that a feedback Cµ path can easily lead tounwanted oscillations in the amplifier. The cascodetuned amplifier effectively has zero feedback and thus ismuch more stable. The loss in voltage headroom is asmall price to pay considering the improved headroomafforded by the inductor.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 42/44 – p. 42/44
Bandwidth
It’s interesting to note that the bandwidth of the circuit isstill determined by the RC time constant at the load.
BW =ω0
Q=
ω0
ω0RC=
1
RC
The ultimate sacrifice for high frequency operation in atuned amplifier is that the amplifier is narrow-band withzero gain at DC. In fact, the larger the Q of the tank, thehigher the gain and the lower the bandwidth.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 43/44 – p. 43/44
How high can we go?
To win some of the bandwidth back requires othertechniques, such as shunt peaking and distributedamplifiers. Other techniques (some invented here atBerkeley) can also help out.
But can we tune out parasitics and design amplifiersoperating at arbitrarily high frequency?
Based on the simple analysis thus far, it seems that forany given frequency, no matter how high, we can simplyabsorb the parasitic capacitance of the amplifier with anappropriately small inductor (say a short section oftransmission line) and thus realize an amplifier at anarbitrary frequency. This is of course ludicrous and we’llre-examine this question in future lecture.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 3 p. 44/44 – p. 44/44