Professional Documents
Culture Documents
V
o
V
i
V
i
P
o,1 dB
P
i,1 dB
Gain compression occurs because eventually the
output signal (voltage, current, power) limits, due to the
supply voltage or bias current.
If we plot the gain (log scale) as a function of the input
power, we identify the point where the gain has dropped
by 1 dB. This is the 1 dB compression point. Its a very
important number to keep in mind.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 3/29 p. 3/29
Apparent Gain
Recall that around a small deviation, the large signal
curve is described by a polynomial
s
o
= a
1
s
i
+ a
2
s
2
i
+ a
3
s
3
i
+
For an input s
i
= S
1
cos(
1
t), the cubic term generates
S
3
1
cos
3
(
1
t) = S
3
1
cos(
1
t)
1
2
(1 + cos(2
1
t))
= S
3
1
_
1
2
cos(
1
t) +
2
4
cos(
1
t) cos(2
1
t)
_
Recall that 2 cos a cos b = cos(a + b) + cos(a b)
= S
3
1
_
1
2
cos(
1
t) +
1
4
(cos(
1
t) + cos(3
1
t))
_
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 4/29 p. 4/29
Apparent Gain (cont)
Collecting terms
= S
3
1
_
3
4
cos(
1
t) +
1
4
cos(3
1
t)
_
The apparent gain of the system is therefor
G =
S
o,
1
S
i,
1
=
a
1
S
1
+
3
4
a
3
S
3
1
S
1
= a
1
+
3
4
a
3
S
2
1
= a
1
_
1 +
3
4
a
3
a
1
S
2
1
_
= G(S
1
)
If a
3
/a
1
< 0, the gain compresses with increasing
amplitude.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 5/29 p. 5/29
1-dB Compression Point
Lets nd the input level where the gain has dropped by
1 dB
20 log
_
1 +
3
4
a
3
a
1
S
2
1
_
= 1 dB
3
4
a
3
a
1
S
2
1
= 0.11
S
1
=
4
3
a
1
a
3
4
3
a
1
a
3
1
t + s
3
2
cos
3
2
t + 3S
2
1
s
2
cos
2
1
t cos
2
t+
3s
2
1
S
2
cos
2
2
t cos
1
t
The rst two terms generate cubic and third harmonic.
The last two terms generate fundamental signals at
1
and
2
. The last term is much smaller, though, since
s
2
S
1
.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 14/29 p. 14/29
Blocker (III)
The blocker term is therefore given by
a
3
3S
2
1
s
2
1
2
cos
2
t
This term adds or subtracts from the desired signal.
Since a
3
< 0 for most systems (compressive
non-linearity), the effect of the blocker is to reduce the
gain
App Gain =
a
1
s
2
+ a
3
3
2
S
2
1
s
2
s
2
= a
1
+ a
3
3
2
S
2
1
= a
1
_
1 +
3
2
a
3
a
1
S
2
1
_
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 15/29 p. 15/29
Out of Band 3 dB Desensitization
Lets nd the blocker power necessary to desensitize
the amplier by 3 dB. Solving the above equation
20 log
_
1 +
3
2
a
3
a
1
S
2
1
_
= 3 dB
We nd that the blocker power is given by
P
OB
= P
1 dB
+ 1.2 dB
Its now clear that we should avoid operating our
amplier with any signals in the vicinity of P
1 dB
, since
gain reduction occurs if the signals are larger. At this
signal level there is also considerable intermodulation
distortion.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 16/29 p. 16/29
Series Inversion
Often its easier to nd a power series relation for the
input in terms of the output. In other words
S
i
= a
1
S
o
+ a
2
S
2
o
+ a
3
S
3
o
+
But we desire the inverse relation
S
o
= b
1
S
i
+ b
2
S
2
i
+ b
3
S
3
i
+
To nd the inverse relation, we can substitute the above
equation into the original equation and equate
coefcient of like powers.
S
i
= a
1
(b
1
S
i
+ b
2
S
2
i
+ b
3
S
3
i
+ ) + a
2
( )
2
+ a
3
( )
3
+
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 17/29 p. 17/29
Inversion (cont)
Equating linear terms, we nd, as expected, that
a
1
b
1
= 1, or b
1
= 1/a
1
.
Equating the square terms, we have
0 = a
1
b
2
+ a
2
b
2
1
b
2
=
a
2
b
2
1
a
1
=
a
2
a
3
1
Finally, equating the cubic terms we have
0 = a
1
b
3
+ a
2
2b
1
b
2
+ a
3
b
3
1
b
3
=
2a
2
2
a
5
1
a
3
a
4
1
Its interesting to note that if one power series does not
have cubic, a
3
0, the inverse series has cubic due to
the rst term above.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 18/29 p. 18/29
Cascade
IIP2
A
IIP3
A
IIP2
B
IIP3
B
IIP2
IIP3
G
A
V
G
A
P
Another common situation is that we cascade two
non-linear systems, as shown above. we have
y = f(x) = a
1
x + a
2
x
2
+ a
3
x
3
+
z = g(y) = b
1
y + b
2
y
2
+ b
3
y
3
+
Wed like to nd the overall relation
z = c
1
x + c
2
x
2
+ c
3
x
3
+
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 19/29 p. 19/29
Cascade Power Series
To nd c
1
, c
2
, , we simply substitute one power series
into the other and collect like powers.
The linear terms, as expected, are given by
c
1
= b
1
a
1
= a
1
b
1
The square terms are given by
c
2
= b
1
a
2
+ b
2
a
2
1
The rst term is simply the second order distortion
produced by the rst amplier and amplied by the
second amplier linear term. The second term is the
generation of second order by the second amplier.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 20/29 p. 20/29
Cascade Cubic
Finally, the cubic terms are given by
c
3
= b
1
a
3
+ b
2
2a
1
a
2
+ b
3
a
3
1
The rst and last term have a very clear origin. The
middle terms, though, are more interesting. They arise
due to second harmonic interaction. The second order
distortion of the rst amplier can interact with the linear
term through the second order non-linearity to produce
cubic distortion.
Even if both ampliers have negligible cubic,
a
3
= b
3
0, we see the overall amplier can generate
cubic through this mechanism.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 21/29 p. 21/29
Cascade Example
In the above amplier, we can decompose the
non-linearity as a cascade of two non-linearities, the G
m
non-linearity
i
d
= G
m1
v
in
+ G
m2
v
2
in
+ G
m3
v
3
in
+
And the output impedance non-linearity
v
o
= R
1
i
d
+ R
2
i
2
d
+ R
3
i
3
d
+
The output impedance can be a non-linear resistor load
(such as a current mirror) or simply the load of the
device itself, which has a non-linear component.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 22/29 p. 22/29
IIP2 Cascade
Commonly wed like to know the performance of a
cascade in terms of the overall IIP2. To do this, note
that IIP2 = c
1
/c
2
c
2
c
1
=
b
1
a
2
+ b
2
a
2
1
b
1
a
1
=
a
2
a
1
+
b
2
b
1
a
1
This leads to
1
IIP2
=
1
IIP2
A
+
a
1
IIP2
B
This is a very intuitive result, since it simply says that
we can input refer the IIP2 of the second amplier to
the input by the voltage gain of the rst amplier.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 23/29 p. 23/29
IIP2 Cascade Example
Example 1: Suppose the input ampliers of a cascade
has IIP2
A
= +0 dBm and a voltage gain of 20 dB. The
second amplier has IIP2
B
= +10 dBm.
The input referred IIP2
B
i
= 10 dBm20 dB = 10 dBm
This is a much smaller signal than the IIP2
A
, so clearly
the second amplier dominates the distortion. The
overall distortion is given by IIP2 12 dB.
Example 2: Now suppose IIP2
B
= +20 dBm. Since
IIP2
B
i
= 20 dBm20 dB = 0 dBm, we cannot assume
that either amplier dominates.
Using the formula, we see the actual IIP2 of the
cascade is a factor of 2 down, IIP2 = 3 dBm.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 24/29 p. 24/29
IIP3 Cascade
Using the same approach, lets start with
c
3
c
1
=
b
1
a
3
+ b
2
a
1
a
2
2 +b
3
a
3
1
b
a
a
1
=
_
a
3
a
1
+
b
3
b
1
a
2
1
+
b
2
b
1
2a
2
_
The last term, the second harmonic interaction term,
will be neglected for simplicity. Then we have
1
IIP3
2
=
1
IIP3
2
A
+
a
2
1
IIP3
2
B
Which shows that the IIP3 of the second amplier is
input referred by the voltage gain squared, or the power
gain.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 25/29 p. 25/29
LNA/Mixer Example
A common situation is an LNA and mixer cascade. The
mixer can be characterized as a non-linear block with a
given IIP2 and IIP3.
In the above example, the LNA has an
IIP3
A
= 10 dBm and a power gain of 20 dB. The mixer
has an IIP3
B
= 20 dBm.
If we input refer the mixer, we have
IIP3
B
i
= 20 dBm20 dB = 40 dBm.
The mixer will dominate the overall IIP3 of the system.
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 26/29 p. 26/29
Example: Disto in Long-Ch. MOS Amp
v
i
V
Q
I
D
= I
Q
+i
o
I
D
=
1
2
C
ox
W
L
(V
GS
V
T
)
2
i
o
+I
Q
=
1
2
C
ox
W
L
(V
Q
+v
i
V
T
)
2
Ignoring the output impedance we have
=
1
2
C
ox
W
L
_
(V
Q
V
T
)
2
+ v
2
i
+ 2v
i
(V
Q
V
T
)
_
= I
Q
..
dc
+C
ox
W
L
v
i
(V
Q
V
T
)
. .
linear
+
1
2
C
ox
W
L
v
2
i
. .
quadratic
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 27/29 p. 27/29
Ideal Square Law Device
An ideal square law device only generates 2nd order
distortion
i
o
= g
m
v
i
+
1
2
C
ox
W
L
v
2
i
a
1
= g
m
a
2
=
1
2
C
ox
W
L
=
1
2
g
m
V
Q
V
T
a
3
0
The harmonic distortion is given by
HD
2
=
1
2
a
2
a
1
v
i
=
1
4
g
m
V
Q
V
T
1
g
m
v
i
=
1
4
v
i
V
Q
V
T
HD
3
= 0
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 28/29 p. 28/29
Real MOSFET Device
0
200
400
600
Effective Field
M
o
b
i
l
i
t
y
Triode CLM DIBL SCBE
R
o
u
t
k
V
ds
(V)
2
4
6
8
10
12
14
0 1 2 3 4
The real MOSFET device generates higher order
distortion
The output impedance is non-linear. The mobility is
not a constant but a function of the vertical and
horizontal electric eld
We may also bias the device at moderate or weak
inversion, where the device behavior is more
exponential
There is also internal feedback
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 9 p. 29/29 p. 29/29