EECS 142/242M

Lecture 14: Effect of Feedback on Distortion

Prof. Ali M. Niknejad

University of California, Berkeley

Copyright c 2011 by Ali M. Niknejad

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 1/28 – p.

 Effect of Feedback on Disto

sǫ so
si + a

We usually implement the feedback with a passive network

Assume that the only distortion is in the forward path a

so = a1 sǫ + a2 s2ǫ + a3 s3ǫ + · · ·

sǫ = si − f so

so = a1 (si − f so ) + a2 (si − f so )2 + a3 (si − f so )3 + · · ·

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 2/28 – p.

 Feedback and Disto (cont)

We’d like to ultimately derive an equation as follows

so = b1 si + b2 s2i + b3 s3i + · · ·

Substitute this solution into the equation to obtain

b1 si + b2 s2i + b3 s3i + · · · = a1 (si − f b1 si − f b2 s2i − f b3 s3i + · · · )

+ a2 (si − f b1 si − f b2 s2i − f b3 s3i + · · · )2
+ a3 (si − f b1 si − f b2 s2i − f b3 s3i + · · · )3 + · · ·

Solve for the first order terms

b1 si = a1 (si − f b1 si )

a1 a1
b1 = =
1 + a1 f 1+T

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 3/28 – p.

 Feedback and Disto (square)

The above equation is the same as linear analysis (loop gain

T = a1 f )
Now let’s equate second order terms

b2 s2i = −a1 f b2 s2i + a2 (si − f b1 si )2

 2
f a1
b2 (a + a1 f ) = a2 1−
1+T

b2 (1 + T )3 = a2 (1 + T − T )2 = a2
a2
b2 =
(1 + T )3
Same equation holds at high frequency if we replace with T (jω)

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 4/28 – p.

 Feedback and Disto (cube)

Equating third-order terms

b3 s3i = a1 (−f b3 s3i ) + a2 (−f b2 2s3i ) + a3 (si − f b1 si )3 + · · ·

1 a3
b3 (1 + a1 f ) = −2a2 b2 f +
1+T (1 + T )3
−2a2 f a2 a3
b3 (1 + T ) = +
1 + T (1 + T )3 (1 + T )3

a3 (1 + a1 f ) − 2a22 f
b3 =
(1 + a1 f )5

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 5/28 – p.

 Second Order Interaction

The term 2a22 f is the second order interaction

Second order disto in fwd path is fed back and combined with
the input linear terms to generate third order disto
Can get a third order null if

a3 (1 + a1 f ) = 2a22 f

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 6/28 – p.

 HD2 in Feedback Amp

1 b2
HD2 = 2 som
2 b1

1 a2 (1 + T )2
= 3 2 som
2 (1 + T ) a1
1 a2 som
=
2 a21 1 + T
1 a2
Without feedback HD2 = 2 a21 som

For a given output signal, the negative feedback reduces the

1
second order distortion by 1+T

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 7/28 – p.

 HD3 in Feedback Amp

1 b3 2
HD3 = 3 som
4 b1

1 a3 (1 + T ) − 2a22 f (1 + T )3 2
= 5 3 som
4 (1 + T ) a1
 2

1 a3 2 1 2a2 f
= s 1 −
4 a31 om (1 + T )4 a3 (1 + T )
| {z }
disto with no fb

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 8/28 – p.

 Feedback versus Input Attenuation

si so so
f a

Notice that the distortion is improved for a given output signal

level. Otherwise we can see that simply decreasing the input
signal level improves the distortion.
Say so1 = f si with f < 1. Then

so = a1 so1 + a2 s2o1 + a3 s3o1 + · · · = a1 f si + a2 f 2 s2i + a3 f 3 s3i + · · ·

|{z} | {z } | {z }
b1 b2 b3

But the distortion is unchanged for a given output signal

1 b2 1 a2
HD2 = 2 som = 2 som
2 b1 2 a1

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 9/28 – p.

 BJT With Emitter Degeneration

IC
The total input signal applied to
the base of the amplifier is
vi
RE vi + VQ = VBE + IE RE
VQ

The VBE and IE terms can be split into DC and AC currents

(assume α ≈ 1)

vi + VQ = VBE,Q + vbe + (IQ + ic )RE

Subtracting bias terms we have a separate AC and DC equation

VQ = VBE,Q + IQ RE

vi = vbe + iC RE

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 10/28 –p

 Feedback Interpretation

The AC equation can be put into the following form

vbe = vi − ic RE

Compare this to our feedback equation

sǫ = si − f so

The equations have the same form with the following

substitutions
sǫ = vbe
so = ic
si = vi
f = RE

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 11/28 –p

 BJT with Emitter Degen (II)

Now we know that

2 3
ic = a1 vbe + a2 vbe + a3 vbe + ···

where the coefficients a1,2,3,··· come from expanding the

exponential into a Taylor series

1 IQ
a1 = gm a2 = ···
2 Vt2

With feedback we have

ic = b1 vi + b2 vi2 + b3 vi3 + · · ·

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 12/28 –p

 Emitter Degeneration (cont)

The loop gain T = a1 f = gm RE

gm
b1 =
1 + gm RE

1


q 2
2 kT IQ
b2 =
(1 + gm RE )3
 
2 

1 q 2
q 3 2 IQ RE 
1 kT IQ  2 kT
b3 = 4 1 − 1 q
3 
4 · 6 (1 + gm RE ) IQ (1 + gm RE )
6 kT

For large loop gain gm RE → ∞

q 3
−1 kT IQ
b3 =
12 (1 + gm RE )4

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 13/28 –p

 Harmonic Distortion with Feedback

Using our previously derived formulas we have

1 b2
HD2 = 2 som
2 b1

1 iˆc 1
=
4 IQ 1 + gm RE
1 b3 2
HD3 = 3 som
4 b1
!2 3gm RE
1 iˆc 1− 1+gm RE
=
24 IQ 1 + gm RE

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 14/28 –p

 Harmonic Distortion Null

We can adjust the feedback to obtain a null in HD3

HD3 = 0 can be achieved with

3gm RE
=1
1 + gm RE

or
1
RE =
2gm

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 15/28 –p

 HD3 Null Example

HD3
-40

-45

-50

-55

-60

-65

-70
0 20 40 60 80 100

1 RE
RE =
2gm

Example: For IQ = 1mA, RE = 13Ω

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 16/28 –p

 BJT with Finite Source Resistance

IC
RS

vi
RE
VQ

vi + VQ − IB RB = VBE + IE RE
Assuming that α ≈ 1, β = β0 (constant). Let RB = RS + rb
represent the total resistance at the base.
 
RB
vi + VQ = VBE + IC RE +
β0
The formula is the same as the case of a BJT with emitter
′
degeneration with RE = RE + RB /β0
A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 17/28 –p
 Emitter Follower

IC

vo
vi
RL
VQ

The same equations as before

with RE = RL

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 18/28 –p

 Common Base

RE IC

vi
RL
−VQ
VCC

Same equation as CE with RE feedback

vi − VQ + IC RE = −VBE

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 19/28 –p

 N Tones in One Shot

Consider the effect of an m’th order non-linearity on an input of

N tones
N
!m
X
ym = An cos ωn t
n=1

N
!m
X An ωn t −ωn t


ym = e +e
n=1
2

N
!m
X An ωn t
ym = e
2
n=−N

where we assumed that A0 ≡ 0 and ω−k = −ωk .

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 20/28 –p

 Product of sums...

The product of sums can be written as lots of sums...

X X X X
= () × () × () · · · × ()
| {z }
m−times

N N N
X X X A k 1 A k 2 · · · Ak m j(ωk1 +ωk2 +···+ωkm )t
= ··· × e
2m
k1 =−N k2 =−N km =−N

Notice that we generate frequency component

ωk1 + ωk2 + · · · + ωkm , sums and differences between m
non-distinct frequencies.
There are a total of (2N )m terms.

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 21/28 –p

 Example
Let’s take a simple example of m = 3, N = 2. We already know
that this cubic non-linearity will generate harmonic distortion
and IM products.
We have (2N )m = 43 = 64 combinations of complex
frequencies. ω ∈ {−ω2 , −ω1 , ω1 , ω2 }. There are 64 terms that
looks like this (HD3 )

ω1 + ω1 + ω1 = 3ω1

ω1 + ω1 + ω2 = 2ω1 + ω2
(IM 3)
ω1 + ω1 − ω2 = 2ω1 − ω2
(Gain compression or expansion)

ω1 + ω1 − ω1 = ω1

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 22/28 –p

 Frequency Mix Vector

Let the vector ~k = (k−N , · · · , k−1 , k1 , · · · , kN ) be a 2N -vector

where element kj denotes the number of times a particular
frequency appears in a given term.
As an example, consider the frequency terms

ω2 + ω1 + ω2 

ω1 + ω2 + ω2 ~k = (0, 0, 1, 2)

ω2 + ω2 + ω1 

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 23/28 –p

 Properties of ~k

First it’s clear that the sum of the kj must equal m

N
X
kj = k−N + · · · + k−1 + k1 + · · · + kN = m
j=−N

For a fixed vector k~0 , how many different sum vectors are there?
We can sum m frequencies m! ways. But the order of the sum is
irrelevant. Since each kj coefficient can be ordered kj ! ways,
the number of ways to form a given frequency product is given
by
m!
(m; ~k) =
(k−N )! · · · (k−1 )!(k1 )! · · · (kN )!

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 24/28 –p

 Extraction of Real Signal
Since our signal is real, each term has a complex conjugate
present. Hence there is another vector k~0′ given by

k~0′ = (kN , · · · , k1 , k−1 , · · · , k−N )

Notice that the components are in reverse order since

ω−j = −ωj . If we take the sum of these two terms we have
n o
2ℜ ej(ωk1 +ωk2 +···+ωkm )t = 2 cos(ωk1 + ωk2 + · · · + ωkm )t

The amplitude of a frequency product is thus given by

2 × (m; ~k) (m; ~k)

m
= m−1
2 2

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 25/28 –p

 Example: IM3 Again

Using this new tool, let’s derive an equation for the IM3 product
more directly.
Since we have two tones, N = 2. IM3 is generated by a m = 3
non-linear term.
A particular IM3 product, such as (2ω1 − ω2 ), is generated by
the frequency mix vector ~k = (1, 0, 2, 0).
3!
(m; ~k) = =3 2m−1 = 22 = 4
1! · 2!
So the amplitude of the IM3 product is 3/4a3 s3i . Relative to the
fundamental
3 a3 s3i 3 a3 2
IM3 = = si
4 a1 si 4 a1

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 26/28 –p

 Harder Example: Pentic Non-Linearity

Let’s calculate the gain expansion/compression due to the 5th

order non-linearity. For a one tone, we have N = 1 and m = 5.
A pentic term generates fundamental as follows

ω1 + ω1 + ω1 − ω1 − ω1 = ω1

In terms of the ~k vector, this is captured by ~k = (2, 3). The

amplitude of this term is given by

5! 5·4 2m−1 = 24 = 16
(m; ~k) = = = 10
2! · 3! 2
10 5
So the fundamental amplitude generated is a5 16 Si .

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 27/28 –p

 Apparent Gain Due to Pentic

The apparent gain of the system, including the 3rd and 5th, is
thus given by

3 10
AppGain = a1 + a3 Si2 + a5 Si4
4 16

At what signal level is the 5th order term as large as the 3rd
order term?
r
3 2 10 4 a3
a3 Si = a5 Si Si = 1.2
4 16 a5
For a bipolar amplifier, we found that a3 = 1/(3!Vt3 ) and
a5 = 1/(5!Vt5 ). Solving for Si , we have
√
Si = Vt 1.2 × 5 × 4 ≈ 127 mV

A. M. Niknejad University of California, Berkeley EECS 142 Lecture 13 p. 28/28 –p