Académique Documents
Professionnel Documents
Culture Documents
SSB DSB AM PM
(a) (b) (c) (d)
Phase Noise
Leakage
PA
CH 1 CH 2 CH 3 CH 4 CH 5 CH 6
VCO
Channel Spacing 200 kHz
Desired
IF LO RF1 RF2
Y (f ) H(f )
=
X(f ) H(f ) − 1
dH
H(f ) = H(f0 ) + ∆f
df
so that
Y (f + ∆f ) H(f0 ) + dH
df ∆f
=
X(f + ∆f ) H(f0 ) + dH
df ∆f − 1
A. M. Niknejad University of California, Berkeley EECS 242 p. 10/61 – p. 10/61
Feedback Model (cont)
Y (f + ∆f ) 1
= dH
X(f + ∆f ) df ∆f
This shows that for circuits containing white noise sources, the
noise voltage (current) is inversely proportional to ∆f , while the
noise power spectral density is proportional to ∆f 2
This simplistic picture already gives us some insight into the
shape of the noise spectrum. But the noise does not “blow up”
near the carrier. Also, why does all the noise go to phase noise
and not amplitude noise? Clearly an LTI model is too simple.
t1 t1 !#6 t5
v1
φ(t) t4 t6
∆v(t) = (1+α(t))v(t+ )−v(t) t2
t2
t3
2πf0 t3 t4 t5
Sn (∆f )
Sφ (∆f ) =
(2π∆f )2
Sφ (ω) SV (ω)
ω0
∆ω
1 2 2
Estored = CVpk = Cvrms
2
N vn2 kT
=R =
S 2
vrms Estored
ωEstored
Q=
Pdiss
N ωkT
=
S QPdiss
ω0 L
Z(ω0 + ∆ω) ≈ j ·
2 ∆ω
ω0
1 ω0
|Z(ω0 + ∆ω)| = ·
G 2Q∆ω
2
vn2 i2n 2 ω0
= · |Z| = 4kT R
∆f ∆f 2Q∆ω
" 2 #
2kT ω0
L(∆ω) = 10 log ·
Psig 2Q∆ω
gm ZT vn gm Rvn
v2 = gm ZT
= R gm R
1− n ZT − n
or
gm R
v2 = vn
1 − gmnR + jBR
The reactive term B can be simplified at a small offset δω from
the resonance ω0
1
B= + j(ω0 + δω)C
j(ω0 + δω)L
δω/ω0
= jδωC − = 2jδωC
jω0 L
where ω0 2 = 1/(LC). Using the notation Aℓ = gm R/n
nAℓ
v2 = vn
(1 − Aℓ ) + j2δωRC
2 2
2 n A ℓ
v2,rms = vn2
(1 − Aℓ )2 + 4δω 2 R2 C 2
vn2 n2 π 1
Posc (1 − Aℓ ) = Aℓ 2
R 2 RC
Since Aℓ ≈ 1
2
vn
R π 1
(1 − Aℓ ) =
Posc |2 {z
RC}
∆fRC
or
Aℓ = 1 − ǫ
The LTI interpretation is that the amplifier has positive feedback
and it limits on it’s own noise. The loop gain is nearly unity but
just below so it’s “stable”.
L C R i2R
v22
gm v1
+
i22 Zi v1 i21
−
The above equivalent circuit includes the “drain” noise i21 , the
load noise i2R , and an input voltage/current noise v22 and i22 .
L C Reff
gm v1
+
v1 i2n
−
vo = −(gm v1 + in )ZT
since
−vo
v1 =
n
we have
gm ZT
v1 = vo − in ZT
n
−in ZT
vo =
1 − gmnZT
1 R1
ZT = 1 1 =
R1 + jωC + jωL 1 + j ωω0 Q + 1
jω ω0 Q
R1
ZT =
ω
1 + jQ ωω0 − ω0
If ω = ω0 + δω and δω ≪ ω0
ω ω0 2δω
− ≈
ω0 ω ω0
ZT R1
vo = −in gm R1 1
= −in
1− 1− gm R1
+ j2Q δω
1+j2Q δω
n n ω0
ω0
Now it’s time to observe that Aℓ = gmnR1 is the initial loop gain. If
we assume that Aℓ ≤ 1, then the circuit is a high gain positive
feedback amplifier.
R12
vo2 = i2n 2 2
1− gm R1
n + 4Q2 δω
ω0 2
2 2
δω gm R1
4Q2 2
≫ 1−
ω0 n
Z
i2n R12 ∞
d(δω)
= 2
(1 − Aℓ )2 −∞ 1 + 4Q2 δω
ω0
1
(1−Aℓ )2
i2 R2 π fo
2 n 1
Vo =
(1 − Aℓ ) 2 Q
f − fo 2Q∆f
(1 − Aℓ ) = 2Q =
fo fo
fo i2n R12 π
∆f = (1 − Aℓ ) = 2 2
fo
2Q Vo 4Q
1
fo = √
2π LC
∆Vc = Vm cos ωm t
√ √ √
where Vm = 4kT Rc 2V/ Hz due to noise. This produces
noise sidebands
√ p Kf 2π
V (t) = Vo cos ω0 t + 2 4kT Rc sin ωm t
ωm
Vo p 1
V (t) ≈ Kf 8kT Rc cos(ωo ± ωm )
ωm 2
The noise analysis thus far makes some very bad assumptions.
Most importantly, we neglect the time-varying nature of the
process. Every oscillator is a quasi-periodic system and the
noise analysis should take this into account.
The following noise model is due to Hajimiri/Lee. It begins with
a simple thought experiment.
Imagine injecting a current impulse into an LC tank at different
times. We assume the LC tank is oscillating at the natural
frequency.
+
Vmax
−
+
0V Imax
−
Γ(ω0 τ )
hφ (t, τ ) = u(t − τ )
qmax
Z t
1
= Γ(ω0 τ )i(τ )dτ
qmax −∞
Z Z !
t ∞ t
1 c0 X
φ(t) = i(τ )dτ + cn cos(nω0 t)i(τ )dτ
qmax 2 −∞ n=1 −∞
Γ(ω0 t)
Z t φ(t)
i(t)
cos(ω0 t + φ(t)) V (t)
qmax −∞
1 Im sin ∆ωt
φ(t) ≈ cm
2 qmax ∆ω
1/f
∆ω ω0 2ω0 3ω0
Sφ (ω)
c0 c1 c1 c2 c2 c3 c3
∆ω
We can absorb this into the ISF and all the previous results
follow unaltered as long as we use Γef f .
dv(t)
vn (t) = v(t) + n(t) = v(t + j(t)) = v(t) + j(t) + · · ·
dt
Oscilliscope
ch1 Trig
(Clock Source)
Delay
DUT
Splitter
Since jitter and phase noise are really two different ways of
seeing the same thing, it’s not surprising that you can calculate
the jitter from the phase noise (but not the other way around).
The variance of jitter can be computed from
Z +∞
2
σj =< φ(t) >= Sφ (f )df
−∞
Often the RMS jitter is quoted, which is just the square root of
the above quantity. Furthermore, the jitter is normalized to the
carrier freqeuncy.
φ(t)
JP ER =
2πf0
so that p
< φ(t)2 >
JP ER,RM S =
2πf0
A. M. Niknejad University of California, Berkeley EECS 242 p. 60/61 – p. 60/61
References