You are on page 1of 23

( )
A o
o
rms
T v
v
K T
rms
∂ ∂

° ∆
where
A o
T v ∂ ∂
calibrates voltage as temperature
o DC o
v ) f ( ⇒ Φ
rms
o AC o
v ) f ( ⇒ Φ
Approach:
) f ( ) f ( ) t (
) t ( v
o d d
d
Φ ↔ φ
↓
↔
?
t
v
o
(t) compressed time
v
rms
→ ∆T
rms
(  )
R A o
T T v + ∝
o
v
0
T
A
+ T
R
Φ  ⇒
Calculation of Φ
d
(f), Power spectrum of v
i
2
(t), v
i
gaussian
)] t ( v ) t ( v [ E )] t ( v ) t ( v [ E ) (
2
i
2
i d d d
= • = τ φ
not gaussian
gaussian v
i
It can be shown that:
E[wxyz] = E[wx]E[yz] + E[wy]E[xz] + E[wz]E[xy]
if w,x,y,z are jointly gaussian random variables [J GRV]
with zero mean [J GRVZM]
B2
v
u
v
u
u,v are J GRVZM
u,v are NOT J GRV,
but are each GRVZM
τ  −  τ  −
Calculation of Φ
d
(f), Power spectrum of v
i
2
(t), v
i
gaussian
)] t ( v ) t ( v [ E )] t ( v ) t ( v [ E ) (
2
i
2
i d d d
= • = τ φ
not gaussian
gaussian v
i
It can be shown that:
E[wxyz] = E[wx]E[yz] + E[wy]E[xz] + E[wz]E[xy]
if w,x,y,z are jointly gaussian random variables [J GRV]
with zero mean [J GRVZM]
) ( 2 ) 0 ( ) t ( v ) t ( ) t ( v ) t ( v ) (
2
i
2
i
2
i i
2
i
2
i d
τ φ + = τ φ ∴
[Ergodic]
= Φ  ) f (
d
) f ( ) f ( 2 ) f ( ) 0 (
i i
2
i

Φ + δ φ
7  7  7
B2
τ  −  τ  −
v 2 φ  =  τ  −  +  τ  −
Φ  ∗
) f ( ) f ( 2 ) f ( ) 0 ( ) f (
i i
2
i d

Φ + δ Φ
Evaluation of Φ
d
(f)
∫
∞
∞ −
Φ = = φ  df ) f ( ) t ( v ) 0 ( ) 1
i
2
i i
(  )
R A eff eff
T T T B kT +

where
-f
o
0
B
f
kT
eff
/2
Φ
i
(f)
) f ( ) f ( 2)
i i

Φ
-B B 0
f
2B 2B
(  )  2 kT
2
eff
•
-2f
o
↑  ↑ +2f
o
=
f
o
B3
Φ  ∗  φ  =
Φ  ∗
B 2
) f ( ) f ( 2 ) f ( ) 0 ( ) f (
i i
2
i d

Φ + δ Φ
Evaluation of Φ
d
(f)
∫
∞
∞ −
Φ = = φ  df ) f ( ) t ( v ) 0 ( ) 1
i
2
i i
(  )
R A eff eff
T T T B kT +

) f ( ) f ( 2)
i i

Φ
-B B 0
f
2B 2B
(  )  2 kT
2
eff
•
-2f
o
↑  ↑ +2f
o
=
3) Therefore:
-B B 0
f
(  )  B kT
2
eff
-2f
o
+2f
o
Φ
d
(f)
(  )  ) f ( B kT
2
eff
δ
B4
Φ  ∗  φ  =
Φ  ∗
B 2
Filtered output power density spectrum Φ
o
(f)
+
0
B B
P
A
= kT
A
(  )
2
h(t)
display
P
R
= kT
R
2
d o
d o
) f ( H ) f ( ) f (
) t ( h ) t ( v ) t ( v
• Φ = Φ
∗ =
AC+DC terms
Say:
h(t)
A
0
t
τ
τ /2
Integrator impulse response
f
o
v
d
(t)
Φ
o
(f)
v
o
(t)
) f ( H
2 j
e
τ π −
•
(  ) τ
1/τ
0
f
B5
f 2
delay 2
2
d o
d o
) f ( H ) f ( ) f (
) t ( h ) t ( v ) t ( v
• Φ = Φ
∗ =
AC+DC terms
Say:
h(t)
A
0
t
τ
τ /2
Integrator impulse response
) 0 f ( =
τ =

∫
∞
= π −
A
dt e ) t ( h
t ) 0 f ( 2 j
2
) f ( H
f
0
1/τ
(Typically 1/τ << B)
Filtered output power density spectrum Φ
o
(f)
) f ( H
2 j
e
τ π −
•
(  ) τ
1/τ
0
f
B6
H : Then
∞  −
f 2
delay 2
(  )  (  )
(  )
∫ ∫
∞
∞ −
∞
• ≅ Φ =
= δ τ = Φ
df ) f ( H B kT df ) f ( P
) f ( A B kT ) f (
2
2
eff o o
2 2
eff o
AC AC
DC
Note that if 1/τ << B, only the value of Φ
d
(f = 0) is important,
so this integral is trivial.
By Parseval’s theorem:
τ = =
∫ ∫
∞
∞ −
∞
∞ −
2 2
2
A dt ) t ( h df ) f ( H
2
) f ( H
f
0
1/τ
(Typically 1/τ << B)
Filtered output power density spectrum Φ
o
(f)
B7
∞  −
power DC : Thus
rms
(  )
[ ]
(  )
[  ] (  )  τ
τ

∂
τ •
= °
∂ ∂
= ∆
B A kT
T BA kT
A B kT
K
T P
P
T
eff
A eff
2 2
eff
A DC
AC
rms
T
A
+ T
R
-
B
T T
T
R A
rms
τ

= ∆ ∴
B8
∂  τ  B kA
Effect of different integrator impulse response
2
d o
) f ( H ) f ( ) f ( • Φ = Φ
Recall
We need to compute
= =
∫
∞
dt ) t ( h ) 0 f ( H
Then
τ
τ
= ∆
kB
2 B kT
T
eff
rms
2 dt ) t ( h df ) f ( H
2
2

∫ ∫
∞
∞ −
∞
∞ −
Suppose h(t)
0
τ
t
1
τ − t
e
Greater sensitivity, but at the expense of a longer memory
(  )
τ

= ∆
T T
T
R A
rms
B9
τ  =
∞  −
τ  =
B 2
s 8
rms
R A
300 10 30 T
T
µ ⇒ ° = • = ∆
= ° = +
R A
4 4 4 4
rms
R A
T T 10 10 10 T
sec 10 10kHz, T f
+ = = • = ∆
= = +
−
B10
100 for K K 003 . 0 sec 1 : then
MHz 100 B , K 30 T : Possible
4 -
K 10 : then
B , K 000 , 10 T : I =  τ
Sampling-Theorem approach for the total-power radiometer
Nyquist sampling: e.g.
T
t
Highest f (=B here) T = 1/2B
0 B
f
(  )
2
0 T
t
h(t)
τ
v
o
(t) T
A
T
R
i(t)
v
i
v
d
+
T < 1/2B ⇒ pulse correlation
T > 1/2B ⇒ lost information
0 T 2T
t
T = 1/2B is the
Nyquist rate
i(t)
Computation of ∆ T
rms
for a sampled system
Boxcar having τ /T = 2Bτ samples
0
t
T
0
t
T 2T
1 T v
boxcar
d
>> ⇒
∑
τ
0 T
τ
t
h(t)
1
T
v
v
T
A
o
o
rms
AC
rms

⎥
⎦
⎤
⎢
⎣
⎡
∂
∂
= ∆
2
d o

#samples
variance of v
d
v
i
(t)
2Bτ samples
v
d
(t)
(central limit theorem)
C2
2B if Gaussian
B 2
τ  =  τ
n) calibratio n/scale fluctuatio (output
B 2 v of Variance σ  •  τ
T
v
v
T
A
o
o
rms
AC
rms

⎥
⎦
⎤
⎢
⎣
⎡
∂
∂
= ∆
2
d o
τ =
#samples
variance of v
d
(  )  (  )
2
2
i
2
i
2
d
d
2
d
v v v v − = −

σ
(where v
i
= J GRVZM)
(  )  (  )  (  )
2
2
i
4
i
2
2
i
2
2
i
4
i
v v v v − = + − =
) x a T
2
eff
2
i
2
• • = ≡
) 1 n ( 5
n n
= − •  • • • =
(where x = J GRVZM)
(  )
N  N
2 2
eff
2
1
2
3
4 2 2
eff
2
2
i
4
i
2
d
a x x a T v v =
⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
− = − = σ
2 2
eff
o
a •  τ =
C3
n) calibratio n/scale fluctuatio (output
B 2 v of Variance σ  •
v 2
a" " defines equation (this v and here 1 x : Let
odd n if , 0 x even; if , 3 1 x : Recall
T 2 : Thus
T 2 B 2 v of variance the and
2 2
eff
o
a •  τ =
a T
T v
T
a T v v
eff
A
o
o
rms
eff
2
i
o
τ
τ

∂ ∂
= ∆ ∴
•  τ = •  τ =
τ = ∆  B T T
eff rms
as before
Note: #samples = 2Bτ  , τ ∝  B 1 variance
Computation of ∆ T
rms
for a sampled system
C4
T 2 B 2 v of variance
a B 2
B 4
v of variance
B 2 B 2
(  )
2
eff
2
eff
2 2
thermal rms
m
B
1
T T T +
τ
= + ∆ ≅ ∆
(Note: 0.1% gain fluct. @ T
eff
= 2000K ⇒ 2K!)
1 ( m 1 ) t ( g < < + ≅
T
eff
g(t) = (T
A
+ T
R
)g(t)
2
eff
2
rms =
0
t
T m
m ), t
T m
One solution to gain variations: “Synchronous detection”
∫

T
A
T
CAL
Z
o
T
R
v
o
(t)
+
-
v
SD
(t) ∝ T
A
- T
CAL
v
o
(t)
0
t
integrated by upper integrator
integrated by lower integrator

rms
SD
v unchanged (looking at same signal all the time)
but = ∂ ∂
eff SD
T v
B T
eff rms
Dicke
τ = ∆ ∴
former value (we view T
A
half the time)
2
1
~V
SD
Note: gain is irrelevant
when T
A
= T
C
∝ (T
R
+ T
A
)
∝ (T
R
+ T
C
)
E2
only) null (at T 2
Integration times τ
A
and τ
c
should be shorter than the fluctuation
times of the desired signal and system gain, respectively.
Want: τ
A
<< desired-signal fluctuation time constant
τ
c
> τ
A
(typically τ
c
≥ γ   τ
A
)
v
o
(t)
T
R
+ T
A
T
R
+ T
C
RCVR
∫
τ
A
∫
τ
γ
C
+
-
V
SD
v
o
(t)
t
0
α β ∆ γ
β
α
E3
Narrowband
filter at
Dicke frequency
v
o
(t)
v’
o
(t)

τ

τ
+
-
T
CAL
Z
o
T
A
v
o
(t)
o
(t)
t
t
τ
π
= ∆  B T
2
T
eff rms
narrowband Dicke 2 = π
v
SD
E4
v’
22 . 2
Uses:
Correlator power density spectrum, pre-integrator:
[  ]  ) t ( v ) t ( v ) t ( v ) t ( ) (
b a b a m
τ − = τ φ
Typical circuit:
0
t
h(t)
τ
×
+
B 0
f
+
B
0
f
S(t)
T
A
°K
T
R
T
R
n
a
(t)
n
b
(t)
v
b
(t)
v
a
(t)
power/2
v
m
(t)
volts
2
) t ( S
v
o
(t)
E5
1) To reduce radiometric gain modulation effects
2) As a correlator
v E τ  −
Correlator power density spectrum, pre-integrator
[  ] ) t ( v ) t ( v ) t ( v ) t ( ) (
b a b a m
τ − τ − = τ φ

⎦
⎤

⎣
⎡
⎟
⎠

⎜
⎝
⎛
+ ⎟
⎠

⎜
⎝
⎛
+ ⎟
⎠

⎜
⎝
⎛
+ ⎟
⎠

⎜
⎝
⎛
+ =
2 2 1 1
b
2
a
2
b
1
a
1
n
2
S
n
2
S
n
2
S
n
2
S
E
)
2 2 1 1
τ

τ
∆ ∆ ∆
BC AD BD AC CD AB • + • + • =

2 2 2
m s s s n n
n n
S n
1 1
( ) (0) ( ) ( ) ( ) ( )
4 2
×
×
×
φ τ   =   φ   +   φ   τ

n n n s s s
2
s m
) f ( ) f ( ) f ( ) f ( ) f ( ) f (
2
1
) f ( ) 0 (
4
1
) f (
× ×
×
Φ + δ φ = Φ
E6
v E
- n(t n ), - S(t S n(t), n S(t), S Where
ABCD : so J GRVZM, All
S S terms
τ   + φ   τ   φ   τ   + φ

n n n S
S S
Φ  ∗  Φ  +  Φ  ∗  Φ  +  Φ  ∗

n n n s s s
2
s m
) f ( ) f ( ) f ( ) f ( ) f ( ) f (
2
1
) f ( ) 0 (
4
1
) f (
× ×
×
Φ + δ φ = Φ
) f ( ) 0 (
4
1
P
ac
2
s dc
δ φ
τ

∂ ∂
= ∆
B
T
T P
P
T
eff
A dc
ac
rms
2
R R A
2
A
2
eff
T + + =
R
A R
A A R
2T
T i (T T )
B
T T B i (T T )
∆  ≅
τ
∆  ≅  τ
E7

n n n S
S S
Φ  ∗  Φ  +  Φ  ∗  Φ  +  Φ  ∗
terms other the from P and , from follows
T 2 T T 2 T where
rms
rms
for the weak-s gnal case
for the strong-s gnal case
<   <
>   >
rms
T
eff
2
Relative sensitivity
to small fluctuations
Total power
Bτ T
eff
(  )
2
R A
T T +  1
Dicke
Correlation
Bτ T
eff
Bτ T 2
eff
(  )
2
R
2
R A
T T T + +
(  )  null at T T
2
R A

2 at null
2
Dicke narrowband
post detector
τ
π
B T
2
eff
(  )
2
R A
T T +
2.22
E8