University of California, Berkeley Fall 2010

EE142, Problem Set #2 Solutions Amin Arbabian

Prof. Jan Rabaey

Problem 1
FM
Given X = xc (t) = Ac cos[ωc t + Φ(t)]
Substituting in Y = aX + bX 2 + cX 3 we get,
Y = axc (t) + bxc (t)2 + cxc (t)3

⇒ Y = aAc cos[ωc t + Φ(t)] + bA2c cos2 [ωc t + Φ(t)] + cA3c cos3 [ωc t + Φ(t)]

bA2c cA3c
⇒ Y = aAc cos[ωc t + Φ(t)] + {1 + cos[2ωc t + 2Φ(t)]} + {3cos[ωc t + Φ(t)] + cos[3ωc t + 3Φ(t)]}
2 4
After ltering, only the fundamental frequency remains. Hence, the ltered signal Yf il is given as :
3cA3c
Yf il = (aAc + )cos[ωc t + Φ(t)]
4
We can easily demodulate this signal, by nding its derivative followed by an envelope detector as shown in
Figure 1.

Figure 1: FM demodulator
For FM, the non-linear channel only changes the amplitude but keeps the phase intact. Hence, the transmitted
message is not aected.

AM
For AM, xc (t) = A(1 + m(t))cos(ωc t)
Therefore, proceeding in the same manner as above,
bA2 (1 + m(t))2 cA3 (1 + m(t))3
Y = A(1 + m(t))cos(ωc t) + {1+cos(2ωc t)} + {3cos(ωc t) + cos(3ωc t)}
2 4
After ltering,
3cA3 (1 + m(t))3
Yf il = {A(1 + m(t)) + }cos(ωc t)
4
The message signal gets corrupted by the channel.

Problem 2
We want to generate the signal xmod (t) = AC (1+mx(t))cos(ωc t). Figure 2 shows the block diagram of the modulator
and demodulator.

1
 xc(t)
u(t) v(t) z(t)
Combiner aX+bX2
x(t)

fc
MODULATOR

x(t) u(t) v(t) z(t)

aX+bX2

fc Wide BPF
DEMODULATOR f0=100 kHz

Figure 2: Block diagram for Modulation and Demodulation

Modulator
Referring to the block diagram, we have
u(t) = xc (t) + x(t)

Assume xc (t) = Acos(ωc t)

v(t) = au(t) + bu(t)2

⇒ v(t) = a{Acos(ωc t) + x(t)} + b{Acos(ωc t) + x(t)}2

⇒ v(t) = a{Acos(ωc t) + x(t)} + b{A2 cos2 (ωc t) + x(t)2 + 2Ax(t)cos(ωc t)}

A2
⇒ v(t) = a{Acos(ωc t) + x(t)} + b{ [1 + cos(2ωc t)] + x(t)2 + 2Ax(t)cos(ωc t)}
2
and
z(t) = aAcos(ωc t) + 2bAx(t)cos(ωc t)

2b
⇒ z(t) = aA{1 + x(t)}cos(ωc t)
a
Here, the modulation index m = 2b
a and we have control over the ratio ab .

Demodulator
The rst bandpass lter is to select the desired carrier frequency.
Therefore,
2b
u(t) = aA{1 + x(t)}cos(ωc t)
a

v(t) = au(t) + bu(t)2

2b 2b
⇒ v(t) = Aa2 {1 + x(t)}cos(ωc t) + Aab[{1 + x(t)}cos(ωc t)]2
a a

2b 2b
⇒ v(t) = Aa2 {1 + x(t)}cos(ωc t) + Aab{1 + x(t)}2 cos2 (ωc t)
a a

2
 2b Aab 2b
⇒ v(t) = Aa2 {1 + x(t)}cos(ωc t) + {1 + x(t)}2 [1 + cos(2ωc t)]
a 2 a

2b Aab 4b 4b2
⇒ v(t) = Aa2 {1 + x(t)}cos(ωc t) + {1 + x(t) + 2 x(t)2 }[1 + cos(2ωc t)]
a 2 a a
Since x(t) is a tone at 100 kHz, by adding a bandpass lter at this frequency, we get
Aab 4b
z(t) = x(t)
2 a

⇒ z(t) = 2Ab2 x(t)

Problem 3
The transition frequency ωt for a MOSFET is dened as the frequency at which the short circuit current gain of the
amplier goes to unity. In order to simplify the problem, we consider the eect of each of the parasitic resistances
separately.

Eect of gate resistance rg

rg Cgd
io

iin Cgs gmvgs Cdb

With rg alone

Figure 3: Small signal circuit with gate resistance alone

Figure 3 shows the small signal circuit considering the gate resistance alone.
We have a i0 as the current through Cgd and the controlled source gm vgs .
iin
vgs =
jω(Cgs + Cgd )

Therefore,
gm jωCgd
i0 = iin − iin
jω(Cgs + Cgd ) jω(Cgs + Cgd )

i0 gm − jωCgd
⇒ =
iin jω(Cgs + Cgd )

Therefore, the unit gain frequency

gm
ωt =
Cgs + Cgd
Since rg does not appear in the expression, it has no eect on the transition frequency.

3
 vg Cgd
io

iin Cgs gmvgs Cdb

vs

rs
With rs alone

Figure 4: Small signal circuit with source resistance alone

Eect of gate resistance rs

Figure 4 shows the small signal circuit considering the source resistance alone.
An approximate way to look at this circuit is to see that the eective transconductance of the circuit is Gm =
1+gm rs and since the circuit is a source follower, node vs = 1+gm rs . Due to bootstrapping, the eective gate source
gm g m rs

capacitance is given as Cgs,ef f = 1+gCgs

m rs
.
Hence, using the previous expression and replacing the gm and Cgs by their eective values we get
Gm
ωt =
Cgs,ef f + Cgd

gm
1+gm rs
⇒ ωt = Cgs
1+gm rs + Cgd

gm
⇒ ωt =
Cgs + Cgd (1 + gm rs )

Hence, if rs is signicant it reduces the transition frequency.

An accurate expression for the short circuit gain by solving the nodal equation is given as
gm
i0 1+(gm +jωCgs )rs − jωCgd
= Cgs
iin jω( 1+(gm +jωC + Cgd )
gs )rs

which gives the same transition frequency as before.

Eect of gate resistance rd

vg Cgd vx rd io

iin Cgs gmvgs Cdb

With rd alone

Figure 5: Small signal circuit with drain resistance alone

Figure 5 shows the small signal circuit considering the drain resistance alone.

4
 iin = jωCgs vg + jωCgd (vg − vx )

and
vx
jωCgd (vx − vg ) + gm vg + jωCdb vx + =0
rd
Solving the above two equations, we get
1
jω(Cgd + Cdb ) + rd
vg = vx
jωCgd − gm

and since i0 = − vrdx , we have

i0 gm − jωCgd
= jω{rd Cdb (Cgs +Cgd )+rd Cgs Cgd }
iin jω(Cgs + Cgd (1 + gm rd ))[1 + ]
Cgs +Cgd (1+gm rd )

We see that the gm

ωt ≈
Cgs + Cgd (1 + gm rd )

but there is also another pole at Cgs +Cgd (1+gm rd )

rd [Cdb (Cgs +Cgd )+Cgs Cgd ] which moves in as rd increases. Hence, ωt reduces
further as rd increases.

Problem 4
To nd the low frequency gain of the amplier we must rst nd the loop gain. We can ignore Cs and treat C1 as
a short. To nd the closed loop gain we rst nd the loop gain, T = Af . Figure 6 shows the circuit with loading.

Figure 6: Amplier circuit with loading

If we assume that the gain from the second stage is unity as it is a source follower, we have
vout
A= = −gm1 R1 R2
is

if b 1
f= =
vout R2

5
 T = Af = gm1 R1

Now, we can nd the closed loop gain of the amplier and assuming that T is very large we have,
gm1 R1 R2
ACL = − ≈ −R2 (1)
1 + gm1 R1
We nd the open loop input and output impedances taking into account loading from R2 .
1
Zin,OL = R2 ||
jωCgs1

1
Zout,OL = R2 ||
gm2
Since the feedback is shunt-shunt, it reduces both the input and output impedances approximately by the loop
gain.

Zin,OL R2 || jωC1gs1
Zin,CL = =
1+T 1 + gm1 R1

1
Zout,OL R2 || gm2
Zout,CL = =
1+T 1 + gm1 R1
There are mainly two poles in this system. The output pole is very far o, hence it is neglected. One pole is at
the input of the rst stage, while the other is at the output of the rst stage. The feedback action decreases the
loading from R2 but increases the eective value of Cgs1 . As long as Cgs1 is small relative to Cs , we can ignore it
to simplify.
1+T 1+T
ω1 = ≈ (2)
R2 (Cs + Cgs1 (1 + T )) R2 C s
The second pole is dened by R1 , Cdb1 , Cgd1 and Cgd2 , since Cgs2 is bootstrapped.
1
ω2 = 1
R1 (Cdb1 + Cgd2 + Cgd1 (1 − Av1 ))

If we assume that the loop gain is high, we must set R2 = 500Ω. Also, we require a bandwidth of atleast 3GHz.
Hence, ω1 being the dominant pole (as Cs is high), we set ω1 = 2π(3 GHz). Hence from (2), we get T = 0.885. The
loop gain is denitely very low and hence our approximation is not valid. We need to use the exact expression of
gain and bandwidth and calculate R2 and T . We also design it for 3.5 GHz to give us sucient room as we have
not accounted for the second pole. Using these values and using (1) and (2), we get R2 = 727Ω and T = 2.199.
From here on, the amplier must be designed to provide the required loop gain. As there are no other speci-
cations (such as noise) given, and the number of design variables are much more, there are many possible design
solutions to this problem. We must note that the overall gain provided by the amplier (common source stage and
source follower) should be 2.199. Also, R1 must be selected so that the second pole is further away from ω1 . It
must also be kept in mind that the value of R1 aects the headroom for M1 .
Due to velocity saturation, the transistors do not follow a square-law characteristic. In fact, for minimum
channel-length devices, gm begins to saturate to approximately 1.25mS/µm for Vov > 0.2 (ie.: the drain current
grows linearly with Vov ). One possible design that meets the given specications is shown in Figure 7.
The ac response of the amplier is shown in Figure 8.
The eye diagram with various data rates is shown in Figures 9, 10 and 11.
The eyes are wide open and the bits easily discernible for the 6 Gbps data stream.
As the data rate is increased beyond the bandwidth of the amplier, the eyes begin to close. This is shown
in the eye diagram for a 10 GHz frequency random data stream. The bits are still discernible, however the noise
margins are signicantly reduced.

6
 Vdd

1.4k

20u/90n vout
2u/90n
3.6 mA
Cs is 300
770

Figure 7: Circuit diagram of amplier

Finally, as the frequency is increased further to 20 GHz, the eyes close completely and the individual bits are
no longer discernible.

The group delay of the amplier represents the delay of dierent frequency components through the amplier.
A at group delay means that all frequency components experience the same delay through the amplier. If the
group delay varies throughout the bandwidth of the signal of interest, then dierent frequency components of the
signal will experience dierent delays through the amplier, this is called dispersion.

7
User: sivavth Date: Sep 21, 2010 12:34:27 AM PDT ee142_hw2 ckt1 schematic : Sep 21 00:25:34 2010 71

Sep 21, 2010 AC Response

dB20(VF("/net040"))
55.0
M0(4.027MHz,
M0(4.027MHz,
M0(4.027MHz, 54.05dB)
M0(4.027MHz, 54.05dB)
54.05dB)

52.5

M1(3.451GHz,
M1(3.451GHz,
M1(3.451GHz, 51.04dB)
M1(3.451GHz, 51.04dB)
51.04dB)

50.0
Y0 (dB)

47.5

45.0

42.5

40.0
1 05 1 06 1 07 1 08 1 09 1 01 0
freq (Hz)

Figure 8: Amplier AC response

Figure 9: Data at 3 GHz (6Gbps)

8
Figure 10: Data at 10 GHz

Figure 11: Data at 20 GHz