ECE 255, Frequency Response

19 April 2018

1 Introduction
In this lecture, we address the frequency response of amplifiers. This was
touched upon briefly in our previous lecture in Section 7.5 of the textbook.
Here, we delve into more details on this topic. It will be from Sections 10.1 and
10.2 of the textbook.

Figure 1: The frequency response of a discrete circuit is affected by the cou-

pling capacitors and bypass capacitors at the low frequency end. At the high-
frequency end, it is affected by the internal capacitors (or parasitic capacitances)
of the circuit (Courtesy of Sedra and Smith).
Printed on April 19, 2018 at 15 : 33: W.C. Chew and S.K. Gupta.

1
Figure 2: Internal or parasitic capacitances become important at high frequency
for both the MOSFET transistor (top) and the BJT transistor (bottom) (Cour-
tesy of Sedra and Smith).

2
Figure 3: For integrated circuits (IC) where there are no coupling and bypass
capacitor needed, e.g., in differential amplifiers, there is no low-frequency cut-
off, and the circuits work down to very low frequency or DC. Then only internal
capacitors affect the high-frequency response (Courtesy of Sedra and Smith).

As shown in Figure 1, the gain of the amplifier falls off at low frequency
because the coupling capacitors and the bypass capacitors become open circuit
or they have high impedances. Hence, they have non-negligible effect at lower
frequencies as treating them as short-circuits is invalid.
At high frequencies, as shown in Figure 2, the simple hybrid-π model or T
model is insufficient to describe the internal physics of the devices. Any two
pieces of metal close to each other become a capacitance at higher frequencies.
Internal capacitances need to be added, to account for capacitive coupling be-
tween these metal pieces. For instance, for the MOSFET, they give rise to a
capacitance Cgs between the gate and the source, and a capacitance Cgd be-
tween the gate and the drain. Similar parasitic capacitances need to be added
for the BJT, causing these devices to depart from their ideal model. Again, this
causes the amplifier gain to fall off at higher frequency.
The operational frequency range or bandwidth of the amplifier is delineated
by fL , at the lower end of the operating frequency range, and fH , at the upper
end. These frequencies are defined as the frequencies when the gain drops below
3 dB1 of the midband gain of the amplifier. For IC’s as shown in Figure 3 when
no coupling capacitors are used, e.g., for differential amplifier design, then fL is
zero. Hence, bandwidth BW is defined as

BW = fH − fL , (discrete-circuit amplifiers), (1.1)

BW = fH , (integrated-circuit amplifiers) (1.2)

A figure of merit for amplifier design is the gain-bandwidth product, defined

as
GB = |AM |BW (1.3)
1 This is also the half-power point.

3
where |AM | is the midband gain. The gain-bandwidth product is often a con-
stant for many amplifiers. It can be shown to be a constant when the amplifier
has only one pole for example. In other words, |AM | increases when BW de-
creases, and vice versa causing GB to remain constant.

2 Low-Frequency Response of Discrete-Circuit

Common-Source and Common-Emitter Am-
plifiers
We will study the effects of the coupling and bypass capacitors on the low-
frequency performance of the common-source (CS) and the common-emitter
(CE) discrete-circuit amplifiers in this lecture.

4
Figure 4: (a) A discrete-circuit common source (CS) amplifier. (b) The small-
signal model with the coupling and bypass capacitor in place (Courtesy of Sedra
and Smith).

2.1 The CS Amplifier

Figure 4(a) shows a CS (common source) amplifier with coupling capacitors
and bypass capacitors. First, we replace the discrete-circuit amplifier with its
small-signal model, but keeping the capacitors in the circuit as shown in Figure
4(b). To determine the amplifier gain Vo /Vsig , we write it as
   
Vo Vg Id Vo
= (2.1)
Vsig Vsig Vg Id

5
1. Fist we invoke the voltage-divider formula which also works for complex
impedances. To this end, writing RG = RG1 k RG2 , one gets
!
Vg RG RG s
= = (2.2)
Vsig RG + sC1C1 + Rsig RG + Rsig s + CC1 (RG1 +Rsig )

In the above, s is the Laplace transform variable, and is related to the

frequency ω by s = jω. Thus, one can see that a pole exists at in the
complex ω plane at

ωP 1 = j/[CC1 (Rsig + RG )] (2.3)

and there is a zero at ω = s = 0. This is due to the DC blocking effect of

the coupling capacitor.
2. In addition, the drain current Id and the source current Is are the same,
namely,
Vg gm YS
Id = Is = 1 = Vg (2.4)
gm + Z S gm + YS
where ZS is the source impedance, which is the analog of the source re-
sistance in previous discrete model in Chapter 7, and YS is the source
admittance. The above yields Id /Vg , an important factor in (2.1), viz.,

Id gm YS
= (2.5)
Vg gm + YS

Moreover,
1 1
YS = = + sCS (2.6)
ZS RS
Hence, by letting s = jω again, the bypass capacitor introduces a pole in
the complex ω-plane at

gm + 1/RS
ωP 2 = j (2.7)
CS
A zero occurs at
1
sZ = − (2.8)
CS RS
or at
j
ωZ = (2.9)
C S RS
Usually, |ωP 2 |  |ωZ |.
3. To complete the analysis, using the current-divider formula,
!
RD RD s
Io = −Id 1 = −Id 1 (2.10)
RD + sCC2 + RL RD + RL s+ CC2 (RD +RL )

6
 Since Vo = Io RL , one gets, from (2.10), that

Vo Io RD RL s
= RL = − 1 (2.11)
Id Id RD + RL s + CC2 (RD +RL )

A pole is introduced at
j
ωP 3 = (2.12)
CC2 (RD + RL )

while a zero occurs at s = 0.

In aggregate,
   
Vo s s + ωZ s
= AM (2.13)
Vsig s + ωP 1 s + ωP 2 s + ωP 3

where
RG
AM = − gm (RD k RL ) (2.14)
RG + Rsig
Here, AM is the overall midband gain.

3 Bode Plots
To simplify the plotting of the frequency response, it is best to do it with Bode
plots. They are log versus log or log-log plots or dB versus log-of-the-frequency
plots. Then amplitudes are converted to dB with the formula that

GAIN in dB = 20 log10 (GAIN) (3.1)

where GAIN is the voltage gain. To this end, one takes the log of (2.13) to
arrive at that
     
Vo s s + ωZ s
log = log |AM | + log + log + log
Vsig s + ωP 1 s + ωP 2 s + ωP 3
(3.2)
One sees that product of three functions now becomes sum of three functions in
the logarithm world. Therefore, if one can figure out the salient feature of the
plot of each of the terms, then one can add them up to see the aggregate plot.

7
 s
Figure 5: Bode plot of the function f1 (s) = s+ωP 1 (Courtesy of Sedra and
Smith).

For simplicity, one looks at the first function in (2.13), namely, the function
f1 (s) = s+ωs P 1 . One sees that at the high-frequency end, f1 (s) → 1 as s → ∞.
Thus, on a log scale, it tends to zero as s → ∞ or ω → ∞ since s = jω.
At the low-frequency end, however, f1 (s) ∼ s/ωP 1 as s → 0. On a dB-log
scale, we have in dB,

|f1 (s)|dB ∼ 20 log10 |ω/ωP 1 | = 20 log10 |ω| + 20 log10 |ωP 1 |,

ω → 0, after letting s = jω. The above means that it is a straight line which
has a slope of 20 on a dB versus log-frequency plot as shown in Figure 5.
One can see that when ω = ωP 1 , or s = jωP 1 , then

jωP 1 j 1 1
|f1 (s = jωP 1 )| = = = =√ (3.3)
jωP 1 + ωP 1 j+1 |j + 1| 2
The above implies that
1
|f1 (s = jωP 1 )|2 = (3.4)
2
or that ω = ωP 1 is the half-power point or on the dB scale, the 3 dB point.

8
 s+ωZ
Figure 6: Bode plot of the function f2 (s) = s+ωP 2 (Courtesy of Sedra and
Smith).

Now one looks at the second function in (2.13), namely, the function
s + ωZ
f2 (s) = .
s + ωP 2
By the same token, it can be shown that the Bode plot of the function f2 (s) is
as shown in Figure 6. If the two points ωZ and ωP 2 are far apart, then when
|s|  ωZ ,
s
f2 (s) ≈
s + ωP 2
and the Bode plot of f2 (s) resembles that of f1 (s) as shown in the right-hand
side of Figure 6. When |s|  ωP 2 , then
s + ωZ ωZ s + ωZ
f2 (s) ≈ =
ωP 2 ωP 2 ωZ
The above implies that f2 (s) grows with increasing ω as shown on the left-hand
side of Figure 6. The 3 dB point, however, is at ω = ωZ , and the slope of the
straight line connecting the pole and the zero points is 20 dB/decade, as shown
in Figure 6.
Given the characteristic of the Bode plot of a simple rational function, the
aggregate Bode plot can be obtained by adding the individual plots together as
shown in Figure 7. The Bode plots can be approximated by staggered piecewise
constant and linear functions when we add them. This is possible when the
poles and zeros are far apart. In this figure, fL is an important number since it
decides when the mid-band gain of the amplifier starts to deteriorate when the
frequency drops.

9
 It is to be noted that if we add two straight line plots, y1 = a1 x + c1 and
y2 = a2 x + c2 , then the resultant plot of y = y1 + y2 = (a1 + a2 )x + c1 + c2 .
Hence, one just adds the slopes of the two straight line plots in Bode plots. So
if two lines have a slope of 20 dB/decade each, the resultant aggregate slope is
40 dB/decade.

Figure 7: Additive sum of the Bode plot of a rational function with three poles
and one zero away from the origin (Courtesy of Sedra and Smith).

4 Rational Functions, Poles and Zeros

In general, the gain function is a rational function of s. A rational function is a
function which is the ratio of two polynomial functions, namely,

a0 + a1 s + . . . + aM sM
A(s) = (4.1)
b0 + b1 s + . . . + bN sN
By simple expansion,

aN (s − s1 )(s − s2 ) . . . (s − sN ) = a0 + a1 s + . . . + aN sN (4.2)

which is a polynomial. The above also implies that an N -th order polynomial
(right-hand side) can be written in a factorized form (left-hand side) in terms of
factors of the roots. It also says that an N -th order polynomial has N zeros. In
other words, a rational function of two polynomials, an M -th order polynomial
in the numerator, and an N -th order polynomial in the denominator, will have

10
M zeros and N poles. More important, it turns out that many functions can
be well approximated by rational functions. Therefore, it is possible to write a
voltage gain function as
(s − s1 )(s − s2 ) . . . (s − sM )
A(s) = A0 F (s) = A0 (4.3)
(s − s1 )(s − s2 ) . . . (s − sN )

4.1 The Meanings of the Zeros and Poles

The deeper meanings of the zeros and poles can be contemplated by looking at
(2.13) more carefully.

4.1.1 Meanings of Zeros

At the zeros of the right-hand side, it is seen that Vo = 0 when Vsig 6= 0. It
implies that no signal can be sent from the input to the output. Something is
blocking this signal! This happens when the coupling capacitors have infinite
impedances, which happens at s = 0 for both CC1 and CC2 . In other words at
ω = 0, they are open circuited, and the gain of the amplifier is zero.
The other zero is obtained when ZS = ∞ or when the source impedance in
Figure 4 is infinite. This is precisely what happens when ω = ωZ in (2.13). At
this frequency, IS = 0, because of the infinite source impedance. When this is
the case, the gain of the amplifier is again zero.

4.1.2 Meanings of Poles

To understand the poles, note that at these frequencies, the right-hand side of
(2.13) becomes infinite. That means that Vo 6= 0 even if Vsig → 0. Or non-zero
signal exists in the circuit even if Vsig = 0. This is possible if the capacitors are
charged, and their charges are relaxing or discharging via the resistors connected
to them even when Vsig = 0. These are called the natural solutions or the
resonant solutions of a system. In mathematics, these are the homogeneous
solutions of a system of equations.
An example of a natural resonant system is the tank circuit as shown in
Figure 8. One can connect a voltage source Vsig to the loop of this tank circuit.
The voltage source will charge the capacitor in the loop. The output can be
taken to be the voltage drop across either the inductor or the capacitor. When
one switches off the voltage source replacing it with a short circuit, the capacitor
discharges, giving rise to non-zero voltage and current in the loop. Thus, the
output will be nonzero even when the input is zero because of the resonant so-
lution! The resonant solutions form the poles of the system and on the complex
s plane they are at
j
s = ±√ (4.4)
LC
or they are along the imaginary axis. A Laplace component of a signal has time
dependence of the form exp(st), and hence, the natural solution will have time
dependence of the form exp(st) = exp(±jt/(LC)). They are sinusoidal signals.

11
Figure 8: An LC tank circuit resonating at its natural frequency via its natural
solution.

One can repeat this gedanken experiment (thought experiment) with an RC

circuit as shown in Figure 9, and the same meaning will prevail but instead, the
pole is at
1
s=− (4.5)
RC
namely, on the negative real axis. A Laplace component of a signal has time
dependence of the form exp(st), and hence, the natural solution will have time
dependence of the form exp(st) = exp(−t/(RC)) which are exponentially de-
caying.

Figure 9: A capacitor charging and discharging. When it is discharging, it does

so via its natural solution.

The charging and discharging of the capacitor is shown in Figure 9. When

the capacitor is discharging, the input signal is zero, and it is discharging via
the natural solution of the circuit.

12
5 Rigorous Method for Finding Time Constants

Figure 10: A capacitor charging and discharging via a general circuit.

When a capacitor is connected to a complicated circuit, it can discharge into

this complicated circuit, or relax its charge into this complicated circuit. Hence,
finding its relaxation time can be rather daunting. Figure 10 shows a capacitor
connected to a somewhat general circuit. The general circuit can be a lot more
complicated that what has been shown!
Imaging now that the capacitor and the rest of the circuit form a loop from
which one can write a KVL. Assuming that a voltage Vsig is inserted into this
loop, and a current I is flowing. Then the KVL equation is

I [1/(sC) + Z(s)] = Vsig (5.1)

One can take I as the output of this system, and that Vsig as the input. The
question is that when Vsig = 0, or the input turned off, can I, the output be
nonzero. Yes, it can be if
1/(sC) + Z(s) = 0 (5.2)
From this equation, one can solve for the frequency s at which a natural solution
can exist. In general, Z(s) can be a very complicated function, and solving (5.2)
is non-trivial, unless the circuit is very simple. Here, Z(s), for example, can be
found by the test current method, making it a very complicated function of s.

5.1 The Relaxation Times—Simple Case

It turns out that the relaxation times of the capacitors in Figure 4 can be found
rather easily, because the relaxation of the charge of the capacitors happens
independently of each other. The discharge from one capacitor is not received
by another capacitor, due the the configuration of the circuit.
As aforementioned, when a capacitor C is connected in series to a resistor
R, the charge in the capacitor relaxes via the time constant τ = RC, or its
RC time constant. In other words, charges in the capacitor decay according to
exp(−t/τ ) = exp(−t/(RC)).

13
 The relaxation solutions are depicted in Figure 11. There are three such
solutions for this case, each associated with the relaxation of each capacitor.
Hence, in this circuit that is being studied, the number of poles is equal to the
number of capacitors in the circuit, because each capacitor has its own relaxation
time.
If only CC1 is charged and the other capacitors are uncharged, then the only
way for the capacitor CC1 to discharge is via the series connected resistors Rsig
and RG . Hence the relaxation time is as such, and that is the location of the
pole ωP 1 as expressed in (2.3). This discharging circuit is shown in Figure 11(a).
The capacitor can be viewed to be connected in series with the resistors.
If only CS is charged, then Vg = 0 always since no current can flow through
the node at Vg , and Vsig = 0. So the other end of the 1/gm resistor is connected
to a virtual ground.
Then CS is discharging via the circuit shown in Figure 11(b), giving rise to
the second pole ωP 2 as given (2.7).
If only CC2 is charged, then it can only discharge via the circuit shown in
Figure 11(c), giving rise to the third pole ωP 3 shown in (2.12). Again, the
capacitor is discharging via the resistors that are connected in series to it.

5.2 Selecting the Bypass and Coupling Capacitors

The relaxation time of the circuit is determined by the RC time constants. Since
usually, CS sees the smallest resistor which is (1/gm k RS ), and hence it has the
smallest time constant or the highest frequency pole, it usually determines fL
in the discrete circuit. Then the coupling capacitors CC1 and CC2 are picked to
make the other poles far enough so that the simple analysis previously discussed
can be used.

14
Figure 11: The three circuits that determine the relaxation time associated
with the three capacitors for the MOSFET CS amplifier of Figure 4 (Courtesy
of Sedra and Smith).

6 The Method of Short-Circuit Time Constants

For a general circuit, the relaxation times of the capacitors are difficult to find.
However, an approximate method can be used to unscramble the coupling be-
tween the capacitors. One is usually interested in determining fL , which corre-
sponds to the highest-frequency pole of the system. Then one can go through
the following procedure to find the highest-frequency pole approximately.
1. One makes the assumption that at this highest frequency fL , all the ca-
pacitors are short-circuited except for the capacitor of interest.
2. This approximation definitely removes the inter-capacitor coupling.
3. It is seen that at the highest frequency pole, the approximation is a good
one.
4. One then steps through the capacitors and find their respective relaxation
frequencies that are associated with them with this approximation.
5. Even though at the other lower frequency poles, the approximation is not
a good one, in practice, fL is calculated using
n
X 1
ωL ≈ (6.1)
i=1
Ci Ri

6. The above sum will be dominated by the smallest relaxation time or the
highest frequency pole, which is best approximated by the short-circuit
time-constant calculation.

15
 The above sum gives a rough estimate of the relaxation frequency of the
highest frequency pole.

6.1 The CE Amplifier

Figure 12: (a) A discrete-circuit common-emitter amplifier. (b) The correspond-

ing small-signal equivalent circuit (Courtesy of Sedra and Smith).

The way to analyze MOSFET discrete-circuit amplifier at the low frequency

end can be used to analyze a BJT discrete-circuit amplifier. This circuit is
shown in Figure 12. The analysis is very similar to before, except that there
is a difference when one finds the relaxation times of the capacitors. Here,

16
the relaxation circuits cannot be easily divided into three independent ones,
because of the inter-coupling effect between the capacitors CC1 and CE . The
base current is related to the emitter current, or the discharge of capacitor CC1
is received by capacitor CCE .
To remove the coupling, the short-circuit time-constant method is used to
find their time constants. These relaxation circuits are shown in Figure 13.
Their relaxation times can be easily found, and then used in (6.1) to estimate
the approximate ωL or fL .

Figure 13: Circuits for determining the short-circuit time constants for the CE
amplifier shown in Figure 12 in this case (Courtesy of Sedra and Smith).

17