Professional Documents
Culture Documents
Nonlinearity
&
Noise
Yet
another
lecture
from
the
road
Recap
&
Perspec=ve
• An
overview
–
where
have
we
been
recently:
– General
linear
system
theory
• S-‐parameters,
but
also
equivalent
small-‐signal
models
like
the
hybrid-‐π
model,
etc.
• Give
linear
response
of
circuit/system;
can
use
superposi=on
&
Fourier
analysis
to
determine
output
for
arbitrary
input
signal
• “Gain,
phase”
at
each
frequency;
no
new
frequencies,
no
signal
components
that
were
not
present
in
the
input
signal
• Strictly
applies
only
for
systems
governed
by
linear
differen=al
equa=ons
(any
order,
but
constant
coefficients)
• Approximately
applies
to
most
systems
if
signals
are
small
(equivalent
to
approxima=ng
func=on
with
first
two
terms
(constant
plus
linear)
in
Taylor
series)
Recap
&
Perspec=ve
(cont.)
• Overview
(con=nued):
– Nonlinear
effects
• Many
important
systems
are
not
well
approximated
with
a
linear
descrip=on
• Examples:
almost
any
component
under
high
power
condi=ons
(e.g.
hea=ng);
amplifiers
driven
with
large
inputs
(or
designed
for
switching-‐
mode
opera=on
for
high
efficiency);
devices
for
harmonic
genera=on,
mixing,
or
detec=on
(nonlinear
response
is
desired)
• Our
approach:
Taylor
series
expansion
of
transfer
characteris=c,
resul=ng
in
polynomial
representa=on
of
response
– We
simplified
to
neglect
memory,
history-‐dependent
effects
(e.g.
hea=ng);
assumed
output
depends
only
on
instantaneous
input
value.
More
advanced
approaches
exist
to
handle
this
• Conclusions:
saw
harmonic
genera=on
(plus
DC
shi^),
intermodula=on
products
(e.g.
sum
&
difference
frequencies)
• Figures
of
merit:
P1dB
(gain
compression),
PIP3
(third
order
intermodula=on)
Nonlinear
Figures
of
Merit
-‐
Review
• Gain
compression
–
P1dB:
– Single
input
tone
– Map
output
power
(at
input
frequency)
vs.
power
of
input
signal
– Linear
theory:
output
power
propor=onal
to
input
power;
nonlinear
effects
tend
to
cause
satura=on:
# 3 k 2&
vo! (t ) = k1 A %1− 3
A ( cos (ω t )
$ 4 k1 '
– Note:
input
is
power
in
single
tone;
output
power
“counted”
is
only
the
power
at
this
frequency
(i.e.,
the
harmonic
power
is
not
included)
Nonlinear
Figures
of
Merit
-‐
Review
• Gain
compression
–
P1dB:
# 3 k 2&
vo! (t ) = k1 A %1− 3
A ( cos (ω t )
– Defini=on:
Input
1
dB
compression
point,
$ 4 k1 '
P1dB,
is
input
power
at
which
output
is
1
dB
below
the
linear
case.
– Note:
some
data
sheets
will
report
the
output
1
dB
compression
point
(e.g.
the
y-‐axis,
rather
than
x-‐axis).
Depends
on
intended
applica=on
Nonlinear
Figures
of
Merit
-‐
Review
• Intermodula=on
–
PIP3:
– Saw
complicated
rela=onship
with
two
input
tones
– Figure
of
merit:
for
comparison
of
components–
PIP3
– Two
input
tones,
equal
amplitude
(for
figure
of
merit);
small
enough
that
gain
compression
can
be
neglected:
v
in (t ) = A !"cos (ω1t ) + cos (ω 2 t )#$
– Output
power
at
close-‐in
intermodula=on
product
frequencies
vs.
power
of
input
signal.
Near
ω1
and
ω2,
have:
3
vo! (t ) = k1 A "#cos (ω1t ) + cos (ω 2 t )$% + k3 A 3 "#cos ( 2ω 2 − ω1 ) t + cos ( 2ω1 − ω 2 ) t $%
4
Nonlinear
Figures
of
Merit
-‐
Review
• Intermodula=on
–
PIP3
(con=nued):
– Purpose
of
PIP3
figure
of
merit:
quan=fy
rela=onship
between
Pin,
Pout
(at
signal
frequency),
and
intermodula=on
products
– Map
power
in
“desired”
frequencies
(first
term)
to
power
in
intermodula=on
products
(second
term)
vin (t ) = A !"cos (ω1t ) + cos (ω 2 t )#$
3
vo! (t ) = k1 A "#cos (ω1t ) + cos (ω 2 t )$% + k3 A 3 "#cos ( 2ω 2 − ω1 ) t + cos ( 2ω1 − ω 2 ) t $%
4
1 2 Pin
=
power
at
each
input
tone
Pin ∝ A
2
1 Pd
=
power
at
each
desired
output
Pd ∝ k12 A 2 ∝ Pin tone
2
1 2 9 6 2 9
Pim ∝ k3 A = k3 APin3
2 16 4 Pim
=
power
at
each
intermod
output
tone
Nonlinear
Figures
of
Merit
-‐
Review
• Intermodula=on
–
graphically:
– Graph
Pd,
Pim
vs.
Pin,
usually
on
log-‐log
scale
(all
powers
in
dBm)
• Reminder:
dBm
=
10*log10(P/1
mW)
1
Pin ∝ A 2
2 Pd ( dBm ) = Gain ( dB) + Pin ( dBm )
1
Pd ∝ k12 A 2 ∝ Pin
2 Pim ( dBm ) = offset + 3Pin ( dBm )
1 9 9
Pim ∝ k32 A 6 = k32 APin3
2 16 4
– PIP3:
intercept
between
linear
(Pd)
and
intermod
(Pim)
terms
– Intercept
is
“fic==ous”;
in
prac=ce,
based
on
low
power
data
(avoid
gain
compression);
real
lab
data
includes
everything…
Intermodula=on
Analysis
• Measurement:
– Measure
Pd,
Pim
at
several
(low)
levels
of
Pin
• Can
easily
separate
Pd,
Pim
(on
spectrum
analyzer)
because
at
different
frequencies
• Use
slope
of
1
for
Pd
vs.
Pin
(in
dBm)
• Use
slope
of
3
for
Pim
vs.
Pin
(in
dBm)
• Find
intercept
point;
input
IP3
(PIP3,
IIP3)
or
output
IP3
(OIP3)
can
be
projected
• Analysis:
2
Pim ! Pin $
– “Intermodula=on
Ra=o”:
IMR = =# &
Pd " PIP3 %
– Convenient
rela=on:
• Relate
expected
intermodula=on
products
from
(known)
IIP3
and
Pin
• Find
IIP3
given
measured
Pim,
Pd
Noise
• Linear
and
nonlinear
analysis
relates
output
to
input
s=mulus
– Linear:
small
signals;
non-‐linear:
large
signals
• Circuits,
systems
also
produce
outputs
independent
of
input:
noise
– Ul=mately,
noise
limits
our
ability
to
resolve/recover/process
very
small
signals
• Noise
is
fundamental
–
cannot
be
eliminated;
but
can
be
managed
• Sources
of
noise:
– Thermal
noise:
random
mo=on
of
carriers
(electrons,
holes)
in
resis=ve
material
– Shot
noise:
cause
by
random
=ming
of
events
• Current
is
made
of
up
of
flow
of
electrons,
but
they
have
some
“jiler”
in
when
they
arrive;
this
generates
shot
noise
– Flicker
or
1/f
noise:
trapping/detrapping,
o^en
defect
or
surface
related;
has
~1/f
noise
power
spectral
density
Thermal
Noise
• Let’s
look
at
thermal
noise
in
a
resistor:
en = 0
en2 = 4kTBR
en2
– Power
available:
kTB
=
Independent
of
R
4R
– Power
transfer?
If
both
resistors
at
same
temperature,
net
flow
=
0
(equal/opposite
flows).
If
at
different
temperatures,
power
from
hot
to
cold
(alempts
to
equilibrate
the
system).
You
knew
that.
Modeling
Noise
(cont.)
• How
about
complex
impedances?
– One
can
show
that:
Z ( f ) = R ( f ) + jX ( f )
en2 = 4kT ∫ R ( f ) dF
B
– No
noise
from
reactances
(no
loss
or
dissipa=on,
no
noise)
• Numerical
example:
– Noise
voltage
across
at
1
MΩ
resistor
in
bandwidth
of
100
MHz
(e.g.
typical
oscilloscope
input)
4kT
=
1.6x10-‐20
J
(T=290
K)
en2 = 4kTBR = 1.6 ×10 −20 ⋅10 6 ⋅108 = 1.6 ×10 −6 V 2
Noise
Figure
• Cau=on:
Input SNR
– From
defini=on
#1:
Noise Factor ≡
Output SNR Ts =To =290 K
– Looks
like
noise
figure
should
be
how
much
(in
dB)
the
SNR
degrades
because
of
the
noise
of
the
two
port
– This
is
not
strictly
true:
note
that
this
is
true
only
if
Ts=To
– As
we
saw,
Ts
can
be
an
effec=ve
temperature
with
no
obvious
connec=on
to
“thermometer”
temperatures
(e.g.
if
signal
came
from
an
antenna,
etc)
• So
noise
figure
should
be
thought
of
as
a
“test-‐based
metric”
– In
the
lab,
can
test
the
SNR
with
Ts=To,
and
find
NF
– In
real
systems,
Ts
is
almost
never
To,
so
the
actual
SNR
change
can
be
quite
different
Noise
Figure
and
LNA
Design
• How
is
this
related
to
our
LNA
design
approach?
• Recall:
4R Γ S − Γ opt
2
n
F = Fmin + 2
Z o 1+ Γ opt (1− Γ ) S
2
• This
shows
explicitly
how
F
depends
on
Zs
(Zs
<-‐>
Γs)
– This
is
important
if
you’re
doing
detailed
circuit
design
(e.g.
of
an
amplifier
to
meet
a
specific
noise
figure
target)
• But
for
system
design
or
analysis,
the
Zs
is
usually
already
defined
and
fixed
– Block
diagram-‐level
interconnec=ons;
not
re-‐designing
the
individual
components
– Example:
making
a
system
by
interconnec=ng
available
“50
Ω”
components
• In
this
case,
TE
is
sufficient
(if
TE
is
for
the
Zs
in
ques=on)
Noise
Figure
and
Loss
• A
special-‐case
two-‐port
is
the
matched
alenuator
• Passive
device—just
a
resistor
network.
Typically
designed
to
have
input
&
output
impedance
matched
to
Zo,
with
a
specified
alenua=on
(e.g.,
3
dB,
6
dB,
etc).
• Overall
noise?
TE 2
– Can
show:
TE = TE1 +
GA1
TE F −1
F = 1+ = F1 + 2
TO GA1
– Careful:
G’s
are
available
power
gains,
TE’s
must
be
for
actual
impedances
presented
System
Noise
Example
• Consider
two
amplifiers
in
cascade:
– Note:
usually
want
the
lowest
TE
amplifier
in
front.
But
not
always—
the
gains
also
play
a
role
Receiver
Sensi=vity
• For
radio
receivers,
sensi=vity
is
limited
by
noise
floor
• Define:
minimum
detectable
signal
(MDS)
for
given
SNR
– O^en
choose
0
dB
as
the
threshold
(though
other
choices
are
possible
depending
on
the
system)
• Example:
• Results:
– NF
=
8
dB
à
F=6.31
à
TE
=
1540
K
– Top
=
290
K
+
1540
K
=
1830
K
S S
– Nout
=
k
Top
B
GA;
Sout
=
Sin
GA;
SNRout = out = in
N out kTop B
– Se•ng
SNRout=1
à
Sin
=
5.3x10-‐17
W
=
-‐132.8
dBm
– Effec=ve
noise
floor
of
the
receiver;
for
reference,
thermal
noise
at
290
K
=
-‐174
dBm/Hz
Noise
Figure
Measurement
• Basic
idea:
measure
output
noise
power
for
two
different
source
temperatures
– From
this,
can
separate
contribu=on
from
source
and
from
two-‐port
2
Pim ! Pmax $
IMR = =# &
Pd " PIP3 %
Pim = Pmin ⋅ G; Pd = Pmax ⋅ G
2
PminG Pmin ! Pmax $
IMR = = =# &
PmaxG Pmax " PIP3 %
Spur
Free
Dynamic
Range
2
! Pmax $
• Since
IMR = PminG = Pmin =# &
PmaxG Pmax " PIP3 %
it
follows
(just
re-‐arranging)
that:
3
Pmax 1/3 2/3
Pmin = 2 ; Pmax = Pmin ⋅ PIP3
PIP3
• Remember:
– Pmin
comes
from
noise
analysis,
so
is
known
– SFDR
=
Pmax/Pmin
• Final
result:
2/3
! PIP3 $
SFDR = # &
" Pmin %
Spur
Free
Dynamic
Range
• Careful:
previous
page
was
all
in
MKS
(or
similar)
units
• Usually
specify
these
things
in
dB:
2/3
! PIP3 $
SFDR = # &
" Pmin %
Becomes:
2"
SFDR(dB) = # PIP3 ( dB) − Pmin ( dB)$%
3
• Not
complicated—just
be
careful
• One
final
note:
the
book
gets
the
same
results,
but
from
another
path;
Pozar
does
the
analysis
from
the
point
of
view
of
the
output
power
(vs.
input
like
done
here).
Congratula=ons
• You
made
it
to
the
end.