Professional Documents
Culture Documents
September 2004
Introduction
A convenient starting point is the common-emitter amplier. Our understanding of this amplier (up to EC1) is that it can provide a fair amount of voltage gain. We apply a small input at the base and we expect to get a pretty wide swing at the collector side. Thats amplication. But, so far, we have assumed that the signals are not of very high frequency such that the parasitic capacitances in the transistor do not have any signicant effects. Remember that the impedance of a capacitor is inversely proportional to wC. So, if C is very small, it is practically an opencircuit at moderately low frequency. But as soon as we increase the signal frequency, the capacitor becomes less and less like an open-circuit, and in fact its impedance begins to drop. Therefore, at high frequencies, we must take into account the presence of parasitic capacitances in order to get a fuller picture of the voltage gain.
C.K. Tse: Frequency Response of Transistor Ampliers
VCC
RL input + VBE IC + output VCE
RL
IC + vCE + + ~
+ RB2 + vBE
+ ~
vo
vin
coupling capacitors (large enough so that they become shortcircuit at signal frequencies)
C.K. Tse: Frequency Response of Transistor Ampliers 3
+ vin
RB1 ||RB2
+ ~ vbe
rp
gmvbe
~
ro
RL
+ vo
Small-signal parameters of BJT: transconductance ic/vbe current gain ic/ib input resistance of BJT vbe/ib output resistance of BJT
gm = q I Ic = c kT VT Ic at room temperature 25mV
gm V ro = A Ic
r =
This is not an ohmic resistor!
+ vin
RB1 ||RB2
+ ~ vBE
rp
gmvBE
~
ro
RL
+ vo
Parameters of CE amplier:
if Rs 0 and ro
o=gmr
0
i arg c (deg) ib
0 45o 90o 135o 180o C.K. Tse: Frequency Response of Transistor Ampliers
f (in logscale)
But this zero gives further negative phase shift instead of positive phase shift. In control theory, this kind of zero is called right-half-plane (RHP) zero.
Why?
Lets look at the circuit. We omit the biasing circuit, and focus on the short-circuit current gain.
+VCC
Ic ic ib Ib
VEE
Small-signal model:
ib + vbe
B
r C
ic
ro gmvbe
We insert a parasitic capacitor across each pair of terminals of the BJT. But since the output is shorted, we can omit the one across C and E in this case.
E
The current ic, from KCL, is Hence, we get
ic = gm v be v be jC = v be ( gm jC )
C 1 j ic gm = gm r ib 1+ j (C + C ) r
7
This is consistent with the frequency response of |ic/ib| we saw earlier. The pole frequency is
fp = 1 (in Hz) 2 (C + C ) r
gm (in Hz) 2C
gmr
fz =
fz
8
=
2 T
1 C 2 2 2 (C + C ) r g m
(gm r )
(C + C ) r2
2
( gm r )
gmr
gm (C + C )
or f T =
gm 2 ( C + C )
fp transition frequency
C.K. Tse: Frequency Response of Transistor Ampliers
fT
fz
9
bo b1
A: B:
gm gm = o (C + C ) (C + C ) gm gm = 1 (C + C ) (C + C ) gm gm = 1(C + C ) (C + C )
B C
C:
gm o (C + C )
gm 1 (C + C )
gm 1(C + C )
||
2 fT The transition frequency fT is a very important parameter for studying the frequency response of a BJT amplier. It tells us how (and r) changes as frequency increases.
2 fT
10
G
Cgs
Cgd
id
ro gmvgs
The main difference between MOSFET and BJT is that there is no r in the model of MOSFET or r = .
S
|(j)|
We can easily nd the transfer function of the current gain as: Cgd 1 j id gm ( j ) = = gm ig j (Cgs + Cgd )
2 fT
11
VCC
RB1 R
s
RL
Ic +
vo
RB2
vin
Recall that the voltage gain ay low frequencies is v o RB1 || RB 2 = [ gm ( RL || ro )] gm RL if Rs 0 and ro v in ( RB1 || RB 2 ) + Rs
C.K. Tse: Frequency Response of Transistor Ampliers 12
B
+ vbe r C
C
Cce ro RL
vin
RB
gmvbe
E
p1 p2 z
Questions
How many poles? How many zeros? Where are the poles and zeros?
p1
Most important is the rst dominant pole p1 which limits the gain!
C.K. Tse: Frequency Response of Transistor Ampliers
p2
?
13
o
z2
x
p2
x
p1
C.K. Tse: Frequency Response of Transistor Ampliers
o
z1
real
14
1+ F( j ) = Ao 1+
where Ao is the low-frequency gain.
j 1 z2 j 1+ p1
j z1 j p2
o
z2
x
p2
x
p1
C.K. Tse: Frequency Response of Transistor Ampliers
o
z1
real
15
40 dB/dec
We know all poles are LHS poles, but zeros can be LHS or RHS. z2
z1 p1 p2
20 dB/dec
What is the difference between a RHS zero and a LHS zero? To see the difference, we have to look at the phase.
16
Phase shift
40 dB/dec
A LHS pole comes with negative phase shift. One LHS pole gives 90. p1 A LHS zero comes with positive phase shift. One LHS zero gives +90. A RHS zero comes with negative phase shift. One RHS zero gives 90.
0
90 180 270
z1 p2
z2
20 dB/dec
o
z2
x
p2
x
p1
C.K. Tse: Frequency Response of Transistor Ampliers
o
z1
real
17
B
+ vbe r C
C
Cce ro RL
vin
RB
gmvbe
E
For this circuit, there are two independent capacitors, hence the circuit is second order. Hence, there must be TWO LHS poles. Moreover, there is at least ONE zero due to C because gmvbe can pull all its current through C resulting in zero current owing to RL. Lets nd the frequency where this happens!
j 1 v o RB1 || RB 2 z = [ gm ( RL || ro )] v in ( RB1 || RB 2 ) + Rs j j 1+ 1+ p1 p2
C.K. Tse: Frequency Response of Transistor Ampliers 18
slope = 20 dB/dec
Basically, if we see a node along the signal path which has (i) a substantial resistance toward the input side; and (ii) a substantial capacitance to ground, then there will be a pole!
Req
Ceq
19
B
+ vbe r C
C
Cce ro RL
vin
RB
gmvbe
E
To nd the poles, we examine the two nodes B and C. Our aim is to nd the equivalent RC lters at B and C. First consider node B. We can assume that (before roll-off) the current source gmvbe ows in the output load RL // ro such that vCE = gm ( RL || ro )v be So, the voltage across the capacitor C is
v BC = v be [1+ gm ( RL || ro )]
C.K. Tse: Frequency Response of Transistor Ampliers 20
Rs
+
B
+ vbe r C C[1+gm(RL||ro)]
vin
RB
vin
RB
E
Clearly, at node B, we can see that (i) (ii) capacitor C is pumping current equal to capacitor C is pumping current equal to
iC = jC v be [1+ gm ( RL || ro )] iC = jC v be
iC + iC = jC v be [1+ gm ( RL || ro )] + jC v be = j{C [1+ gm ( RL || ro )] + C }v be
Ceq = C [1+ gm ( RL || ro )] + C
21
B
+ vbe r C
CM
C[1+gm(RL||ro)]
vin
RB
E
Observe that C has been EXPANDED by a factor of [1+gm(RL||ro)] which is just the dc gain.
22
B
+ vbe r C
CM
C[1+gm(RL||ro)]
vin
RB
E
A pole can be found from the equivalent low-pass lter at node B. This is the Miller effect pole. Clearly, the equivalent R and C are Ceq = C + CM = C + C[1+gm(RL||ro)] Req = Rs || RB || r This pole is
p1 = 1 1 = Ceq Req {C + C [1+ gm ( RL || ro )]}( Rs || RB || r )
23
B
+ vbe r C
C
Cce ro RL
vin
RB
gmvbe
Next consider node C. We can assume that the signal at B looking from C is comparatively small and that node B is essentially grounded. Therefore, we can approximate that C goes to ground at C. Clearly, the combined capacitance to ground is C + Cce. Also, since the equivalent resistance is RL || ro, the pole is
p2 =
(C
+ Cce )( RL || ro )
B
+ vbe r C
C
Cce ro RL
vin
RB
gmvbe
E
To nd the zero, we can simply try to nd the frequency (on complex plane) where the response is 0. This is quite easy! Almost by inspection, we have, when output is 0,
v bc = v be
gm v be = v be sC
s=
gm C
25
B
+ vbe r C
CM
C[1+gm(RL||ro)] gmvbe C + Cce ro RL
vin
RB
E
The complete transfer function for the gain is note the -ve sign (its RHP)
j 1 v o RB1 || RB 2 z = [ gm ( RL || ro )] v in ( RB1 || RB 2 ) + Rs j j 1+ 1+ p1 p2
p1 = 1 1 = Ceq Req {C + C [1+ gm ( RL || ro )]}( Rs || RB || r )
p2 =
(C
+ Cce )( RL || ro )
z=
gm C
26
z p1 p2
90 180 270 C.K. Tse: Frequency Response of Transistor Ampliers 27
p1
p2 z
pole splitting
1M
10M
100M
Cgd (pF)
100 10 1 0.1 1M 10M 100M
p1
These are not Bode plots! They actually show the positions of the poles and the RHP zero for different values of gm and Cgd. Note the pole splitting when gm is greater than a certain value. This actually improves the stability, as we will see when we discuss feedback later.
p2
28
vi
vo = Avi
29
Discussion
Is there any way to beat Miller? Direction 1: kill C A capacitor ying over a signal input and a swinging amplied output. We can kill Miller if we can either stop the input or the output from moving! The question is how to do it, with the signal amplication still maintained. Direction 2: kill R How about Rs, which is clearly an evil that causes the roll off? Can we make it as small as possible? What kind of amplier should be used to buffer the input so that Rs can be smaller? The process for making Miller disappear is called broadbanding an amplier. I will tell you more about beating Miller in the nal-year elective High Frequency Circuit Design.
30