1/549

Back
Close
Optical Communication Systems (OPT428)
Govind P. Agrawal
Institute of Optics
University of Rochester
Rochester, NY 14627
c _2007 G. P. Agrawal
207/549

Back
Close
Chapter 5:
Signal Recovery and Noise
Noise Added during Photodetection
Signal-to-Noise Ratio (SNR)
Bit Error Rate (BER)
Sensitivity Degradation
Forward Error Correction (FEC)
208/549

Back
Close
Optical Receivers
Front end converts optical signal into electrical form.
Linear channel amplies and lters the electrical signal.
Data recovery section creates electrical bit stream using clock-
recovery and decision circuits.
209/549

Back
Close
Data-Recovery Section
A clock-recovery circuit isolates the frequency f = B from the
received signal.
The clock helps to synchronize the decision process.
Decision circuit compares the output to a threshold level at sampling
times set by the clock.
Eye diagram is useful for system monitoring.
The best sampling time corresponds to maximum eye opening.
210/549

Back
Close
Shot Noise
Photocurrent, I(t) =I
p
+i
s
(t), uctuates because electrons are gen-
erated at random times.
Average current I
p
= R
d
P
in
; R
d
= q/h
0
;
represents quantum eciency of photodetector.
Current uctuations occur such that i
s
(t)) = 0 and
i
s
(t)i
s
(t +)) =
_

S
s
( f )exp(2i f )d f .
White noise: Spectral density S
s
( f ) constant.
Noise variance:
2
s
=
_

S
s
( f )d f = 2qI
p
f .
Eective noise bandwidth f is related to detector bandwidth.
Adding the contribution of dark current I
d

2
s
= 2q(I
p
+I
d
)f .
211/549

Back
Close
Thermal noise
Additional uctuations occur at any nite temperature because of
thermal motion of electrons in any resistor.
Total current: I(t) = I
p
+i
s
(t) +i
T
(t).
Spectral density S
T
( f ) = 2k
B
T/R
L
depends on temperature and
load resistor R
L
.
Noise variance:
2
T
=
_

S
s
( f )d f = (4k
B
T/R
L
)f .
Amplier noise: All electrical ampliers enhance thermal noise by
the amplier noise gure F
n
.
Total thermal noise:
2
T
= (4k
B
T/R
L
)F
n
f .
Total Receiver Noise:

2
=
2
s
+
2
T
= 2q(I
p
+I
d
)f +(4k
B
T/R
L
)F
n
f .
212/549

Back
Close
Signal-to-Noise Ratio
20 18 16 14 12 10 8 6 4
5
10
15
20
25
30
35
40
45
Received Power (dBm)
S
N
R

(
d
B
)
0.1 A
1 A

T
= 5 A
SNR =
I
2
p

2
=
R
2
d
P
2
in
2q(R
d
P
in
+I
d
)f +4(k
B
T/R
L
)F
n
f
.
Increase in SNR with received power P
in
for three values of
T
for a receiver bandwidth of 30 GHz.
213/549

Back
Close
Thermal-Noise Limit
In the limit
T

s
, SNR becomes:
SNR =
R
L
R
2
d
P
2
in
4k
B
TF
n
f
.
Noise-equivalent power: Dened as the minimum optical power per
unit bandwidth required to produce SNR = 1:
NEP =
P
in

f
=
_
4k
B
TF
n
R
L
R
2
d
_
1/2
=
h
q
_
4k
B
TF
n
R
L
_
1/2
.
NEP is often used to quantify thermal noise.
Typical values of NEP are in the range of 1 to 10 pW/

Hz.
Optical power needed to realize a specic value of SNR obtained
from P
in
= (NEP

f )SNR.
214/549

Back
Close
Shot-Noise Limit
In the opposite limit,
s

T
:
SNR =
R
d
P
in
2qf
=
P
in
2hf
.
It is possible to express SNR in terms of the number of photons N
p
contained in a single 1 bit.
Pulse energy: E
p
= N
p
h.
Optical power for a bit of duration T
B
= 1/B: P
in
= N
p
hB.
Receiver bandwidth for NRZ bit stream: f = B/2.
Putting it all together, SNR = N
p
N
p
.
At 1.55-m, P
in
130 nW is needed at 10 Gb/s to realize
SNR = 20 dB (N
p
= 100).
215/549

Back
Close
APD Receivers
Average current larger for an APD by the gain factor M:
I
p
= MR
d
P
in
= R
APD
P
in
.
Thermal noise unchanged but shot noise enhanced by a factor
F
A
known as excess noise factor.
Shot-noise variance:
2
s
= 2qM
2
F
A
(R
d
P
in
+I
d
)f .
Signal-to-Noise Ratio for an APD receiver:
SNR =
I
2
p

2
s
+
2
T
=
(MR
d
P
in
)
2
2qM
2
F
A
(R
d
P
in
+I
d
)f +4(k
B
T/R
L
)F
n
f
.
SNR is larger for APDs because thermal noise dominates
in practice.
216/549

Back
Close
APD Receivers (continued)
20 18 16 14 12 10 8 6 4
20
25
30
35
40
Received Power (dBm)
S
N
R

(
d
B
)
M = 1
5
10
Increase in SNR with received power P
in
for three values of APD
gain M for 30-GHz bandwidth.
Excess noise factor F
A
depend on APD gain as
F
A
(M) = k
A
M+(1k
A
)(21/M).
217/549

Back
Close
Optimum APD gain
Thermal-Noise Limit (
T

s
):
SNR = (R
L
R
2
d
/4k
B
TF
n
f )M
2
P
2
in
.
Shot-Noise Limit (
s

T
):
SNR =
R
d
P
in
2qF
A
f
=
P
in
2hF
A
f
.
SNR can be maximized by optimizing the APD gain M.
Setting d(SNR)/dM = 0, the optimum APD gain satises
k
A
M
3
opt
+(1k
A
)M
opt
=
4k
B
TF
n
qR
L
(R
d
P
in
+I
d
)
.
Approximate solution: M
opt

_
4k
B
TF
n
k
A
qR
L
(R
d
P
in
+I
d
)
_
1/3
.
218/549

Back
Close
Optimum APD Gain (continued)
40 38 36 34 32 30 28 26 24 22 20
10
0
10
1
10
2
10
3
Received Power (dBm)
O
p
t
i
m
u
m

A
P
D

G
a
i
n
k
A
= 0
0.001
0.01
0.1
1.0
M
opt
plotted as a function of P
in
for several values of k
A
.
Parameter values correspond to a typical 1.55-m APD receiver.
Performance improved for APDs when k
A
1.
219/549

Back
Close
Bit Error Rate
BER = p(1)P(0/1) + p(0)P(1/0) =
1
2
[P(0/1) +P(1/0)].
P(0/1) = conditional probability of deciding 0 when 1 is sent.
Since p(1) = p(0) = 1/2, BER =
1
2
[P(0/1) +P(1/0)].
Common to assume Gaussian statistics for the current.
220/549

Back
Close
Bit Error Rate (continued)
P(0/1) = Area below the decision level I
D
P(0/1) =
1

2
_
I
D

exp
_

(I I
1
)
2
2
2
1
_
dI =
1
2
erfc
_
I
1
I
D

2
_
.
P(1/0) = Area above the decision level I
D
P(1/0) =
1

2
_

I
D
exp
_

(I I
0
)
2
2
2
0
_
dI =
1
2
erfc
_
I
D
I
0

2
_
.
Complementary error function erfc(x) =
2

x
exp(y
2
)dy.
Final Answer
BER =
1
4
_
erfc
_
I
1
I
D

2
_
+erfc
_
I
D
I
0

2
__
.
221/549

Back
Close
Role of Decision Level
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
10
14
10
12
10
10
10
8
10
6
10
4
10
2
10
0
Decision Level, I
D
/I
1
B
E
R
I
1
/
1
=
10
12
15
BER depends on the decision threshold I
D
.
I
D
is optimized in practice to reduce the BER.
222/549

Back
Close
Minimum Bit Error Rate
Minimize BER by setting d(BER)/dI
D
= 0.
Minimum BER occurs when I
D
is chosen such that
(I
D
I
0
)
2
2
2
0
=
(I
1
I
D
)
2
2
2
1
+ln
_

0
_
.
Last term is negligible in most cases, and
(I
D
I
0
)/
0
= (I
1
I
D
)/
1
Q.
I
D
=

0
I
1
+
1
I
0

0
+
1
, Q =
I
1
I
0

1
+
0
.
Final Expression
BER =
1
2
erfc
_
Q

2
_

exp(Q
2
/2)
Q

2
.
223/549

Back
Close
Q Parameter
0 1 2 3 4 5 6 7
10
12
10
10
10
8
10
6
10
4
10
2
10
0
Q Factor
B
E
R
Q =
I
1
I
0

1
+
0
is a measure of SNR.
Q > 6 required for a BER of < 10
9
.
Q = 7 provides a BER of < 10
12
.
224/549

Back
Close
Minimum Average Power
Receiver sensitivity = Minimum average power needed to keep the
BER below a certain value (< 10
9
).
We need to relate Q parameter to incident optical power.
Assume 0 bits carry no optical power so that P
0
= I
0
= 0.
I
1
= MR
d
P
1
= 2MR
d

P
rec
, where

P
rec
= (P
1
+P
0
)/2.
Including both shot and thermal noise,

1
= (
2
s
+
2
T
)
1/2
and
0
=
T
,

2
s
= 2qM
2
F
A
R
d
(2

P
rec
)f ,
2
T
= (4k
B
T/R
L
)F
n
f .
Using these results
Q =
I
1

1
+
0
=
2MR
d

P
rec
(
2
s
+
2
T
)
1/2
+
T
.
225/549

Back
Close
Receiver Sensitivity
Solving for received power, we obtain

P
rec
=
Q
R
d
_
qF
A
Qf +

T
M
_
.
For a pin receiver, we set M = 1.
Since thermal noise dominates for such a receiver,

P
rec
Q
T
/R
d
.
Using R 1 A/W near 1.55 m,

P
rec
= Q
T
.
As an example, if we use R
d
= 1 A/W,
T
= 100 nA, and Q = 6,
we obtain

P
rec
= 0.6 W or 32.2 dBm.
226/549

Back
Close
APD Receiver Sensitivity
Receiver sensitivity improves for APD receivers.
If thermal noise dominates,

P
rec
is reduced by a factor of M.
When shot and thermal noise are comparable, receiver sensitivity
can be optimized by adjusting the APD gain M.

P
rec
is minimum for an optimum value of M:
M
opt
= k
1/2
A
_

T
Qqf
+k
A
1
_
1/2

_

T
k
A
Qqf
_
1/2
.
Best APD responsivity

P
rec
= (2qf /R)Q
2
(k
A
M
opt
+1k
A
).
227/549

Back
Close
Number of Photons/Bit
Receiver sensitivity can be expressed in terms of number of photons
N
p
contained within a single 1 bit.
In the shot-noise limit, I
0
= 0 and
0
= 0 when 0 bits carry no
power, and Q = I
1
/
1
= (SNR)
1/2
.
SNR related to N
p
as SNR N
p
, or
BER =
1
2
erfc
__
N
p
/2
_
.
For = 1, BER = 1 10
9
, N
p
= 36. Thus, 36 photons are
sucient in the shot-noise limit.
In practice, most optical receivers require N
p
> 1000 because of
thermal noise.
228/549

Back
Close
Quantum Limit of Photodetection
The BER obtained in the shot-noise limit not totally accurate.
Its derivation based on the Gaussian approximation for noise.
Poisson statistics should be used for small number of photons.
For an ideal detector (no thermal noise, no dark current, and =1),
0 bits produce no photons, and
0
= 0.
Error occurs only if 1 bit fails to produce even one electron.
Probability of generating m electrons: P
m
= exp(N
p
)N
m
p
/m!.
Since P(0/1) = exp(N
p
), BER = exp(N
p
)/2.
N
p
= 20 for BER = 110
9
(10 photons/bit on average).

P
rec
=N
p
hB/2 =

N
p
hB=13 nW or 48.9 dBm at B=10 Gb/s.
229/549

Back
Close
Sensitivity Degradation
Real receivers need more power than

P
rec
.
Increase in power is referred to as power penalty.
In decibel units, power penalty is dened as
Power Penalty = 10 log
10
_
Increased Power
Original Power
_
.
Several mechanisms degrade the receiver sensitivity.
Finite Extinction ratio (P
0
,= 0)
Intensity Noise of received optical signal
Pulse broadening induced by ber dispersion
Timing Jitter of electronic circuits
230/549

Back
Close
Finite Extinction Ratio
Extinction ratio is dened as r
ex
= P
0
/P
1
.
Power penalty can be obtained by calculating Q parameter.
For a pin receiver I
1
= R
d
P
1
and I
0
= R
d
P
0
.
Using

P
rec
= (P
1
+P
0
)/2,
Q =
_
1r
ex
1+r
ex
_
2R
d

P
rec

1
+
0
.
In thermal noise limit,
1

T
.
Received power for a nite extinction ratio

P
rec
(r
ex
) =
_
1+r
ex
1r
ex
_

T
Q
R
d
.
231/549

Back
Close
Extinction Ratio (continued)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
2
4
6
8
10
12
14
16
18
20
Extinction Ratio
P
o
w
e
r

P
e
n
a
l
t
y

(
d
B
)

ex
= 10 log
10
_

P
rec
(r
ex
)

P
rec
(0)
_
= 10 log
10
_
1+r
ex
1r
ex
_
.
1-dB penalty occurs for r
ex
= 0.12; increases to 4.8 dB for r
ex
= 0.5.
232/549

Back
Close
Intensity Noise of Lasers
So far, incident optical power is assumed to be constant.
In practice, all lasers exhibit intensity noise.
Optical ampliers add additional power uctuations.
Receiver converts power uctuations into current uctuations,
which add to those resulting from shot and thermal noise.
Total noise variance can be written as

2
=
2
s
+
2
T
+
2
I
.
Intensity noise
I
= R
d
(P
2
in
))
1/2
= R
d
P
in
r
I
, r
I
=P
2
in
))
1/2
/P
in
.
Parameter r
I
related to the RIN of a laser as
r
2
I
=
1
2
_

RIN()d.
233/549

Back
Close
Intensity Noise (continued)
Consider a pin receiver with I
1
= R
d
P
1
and I
0
= 0.
Using

P
rec
= (P
1
+P
0
)/2
Q =
2R
d

P
rec
(
2
T
+
2
s
+
2
I
)
1/2
+
T
.
Optical power required for a nite intensity noise

P
rec
(r
I
) =
Q
T
+Q
2
qf
R
d
(1r
2
I
Q
2
)
.
Power penalty is found to be

I
=10 log
10
(1r
2
I
Q
2
).
234/549

Back
Close
Intensity Noise (continued)
0 0. 05 0. 1 0. 15 0. 2
0
2
4
6
8
10
12
14
16
Intensity Noise Parameter
P
o
w
e
r

P
e
n
a
l
t
y

(
d
B
)
Q = 7
6
5
A 2-dB penalty occurs for r
I
= 0.1.
Penalty becomes innite when r
I
> Q
1
(BER oor).
In practice, r
I
< 0.01 (power penalty negligible).
235/549

Back
Close
Dispersive Pulse Broadening
Pulse energy in the bit slot decreases with pulse broadening.
Receiver requires more average power to maintain SNR.
For Gaussian pulses, peak power is reduced by the pulse broadening
factor f
b
found in Chapter 3.
f
2
b
= 1+(DL

/
0
)
2
when source bandwidth dominates.

0
is related to duty cycle d
c
as 4
0
= d
c
T
b
.
Using = (4B)
1
, power penalty is given by

d
= 10 log
10
f
b
= 5 log
10
[1+(4BLD

/d
c
)
2
].
For a narrowband source and unchirped Gaussian pulses

d
= 5 log
10
[1+(8
2
B
2
L/d
2
c
)
2
].
236/549

Back
Close
Dispersive Pulse Broadening
Power penalty negligible for =[
2
[B
2
L < 0.05 and d
c
> 0.5.
Increases rapidly as increases and exceeds 5 dB for = 0.1.
At 10-Gb/s, L <50 km when standard bers are used.
237/549

Back
Close
Frequency Chirping
Chirping of optical pulses aects pulse broadening.
For chirped Gaussian pulses pulse broadening factor is
f
2
b
= 1+8C
2
B
2
L/d
2
c
)
2
+(8
2
B
2
L/d
2
c
)
2
.
Power penalty then becomes

c
= 5 log
10
[(1+8C
2
B
2
L/d
2
c
)
2
+(8
2
B
2
L/d
2
c
)
2
].
Penalty can be quite large when
2
C > 0.
This is the case for directly nodulated DFB lasers (C >4)
operating near 1.55 m (
2
< 0).
238/549

Back
Close
Frequency Chirping
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
2
1
0
1
2
3
4
5
6
7
8
|
2
|B
2
L
P
o
w
e
r

P
e
n
a
l
t
y

(
d
B
)
C = 6
4
2
0
1
To keep penalty below 0.1 dB, [
2
[B
2
L < 0.002 is required.
For standard bers B
2
L is limited to 100 (Gb/s)
2
-km.
System performance can be improved by ensuring that
2
C < 0.
239/549

Back
Close
Timing Jitter
Signal must be sampled at the peak of the current pulse.
Decision instant determined by the clock-recovery circuit.
In practice, sampling time uctuates from bit to bit.
If bit is not sampled at the bit center, sampled value is reduced by
an amount that depends on timing jitter t.
Since t is a random variable, signal becomes more noisy.
SNR reduced as a result of such additional uctuations.
SNR can be maintained by increasing received power
(power penalty).
240/549

Back
Close
Timing Jitter (continued)
Q parameter in the presence of timing jitter
Q =
I
1
i
j
)
(
2
T
+
2
j
)
1/
2+
T
.
If S
p
(t) governs the shape of current pulse, i
j
=I
1
[S
p
(0)S
p
(t)].
Approximating S
p
as S
p
(t) = 1
1
2
(c
p
Bt)
2
, i
j
= (c
p
Bt)
2
I
1
.
Probability density of timing jitter t
p(t) =
1

2
exp
_

t
2
2
2
j
_
.
Find p(i
j
) and use it to calculate i
j
) and
j
.
241/549

Back
Close
Timing Jitter (continued)
Probability density of current uctuation i
j
p(i
j
) =
1
_
bi
j
I
1
exp
_

i
j
bI
1
_
, b = (c
p
B
t
)
2
.
Average and standard deviation are found to be
i
j
) = bI
1
/2,
j
= bI
1
/

2.
Receiver sensitivity

P
rec
(b) =
_

T
Q
R
d
_
1b/2
(1b/2)
2
b
2
Q
2
/2
.
Power penalty is found to be

j
= 10log
10
_
1b/2
(1b/2)
2
b
2
Q
2
/2
_
.
242/549

Back
Close
Timing Jitter (continued)
Pulse curvature c
p
at center of bit slot plays important role.
Power penalty becomes innitely large at a certain value of B
t
.
Tolerable value B
t
depends on c
p
and decreases as c
p
increases.
Typically c
p
< 1, and power penalty <0.5 dB) if B
t
< 0.08.
243/549

Back
Close
Eye-Closure Penalty
Eye diagrams at 40 Gb/s in the case of NRZ, CSRZ, NRZ-DPSK,
and RZ-DPSK formats.
L = 0 (top row) and L = 263 km (bottom row).
Alternative measure of system performance is provided by
the eye opening.
244/549

Back
Close
Forward Error Correction
It is entirely possible that a specied BER cannot be achieved.
Only viable alternativeUse an error-correction scheme.
In one approach, errors are detected but not corrected.
Suitable when packet switching is used (Internet protocol).
In FEC, errors are detected and corrected at the receiver without
any retransmission of bits.
This scheme is best suited for lightwave systems operating with
SONET or SDH protocol.
Historically, lightwave systems did not employ FEC until the use of
in-line optical ampliers became common.
245/549

Back
Close
Error-Correcting Codes
Basic idea: Add extra bits at transmitter using a suitable code.
At the receiver end, a decoder uses these control bits to detect and
correct errors.
How many errors can be corrected depends on the coding scheme
employed.
In general, more errors can be corrected by adding more control bits
to the signal.
There is a limit to this process since bit rate of the system increases
after the FEC coder.
If B
e
is eective bit rate after coding, FEC overhead is B
e
/B1.
Redundancy of a code is dened as = 1B/B
e
.
246/549

Back
Close
Error-Correcting Codes
Classied under names such as linear, cyclic, Hamming, Reed
Solomon, convolutional, product, and turbo codes.
Among these, ReedSolomon (RS) codes have attracted most at-
tention for lightwave systems.
Denoted as RS(n, k), where k is the size of packet that is converted
into a larger packet with n bits (n = 2
m
1).
ITU recommendation: RS(255, 239) with m=8. FEC overhead for
this code is 6.7%.
RS(255, 207) with an overhead of 23.2% is also used.
Improvement in BER is quantied through the coding gain.
247/549

Back
Close
Coding Gain
Coding gain: A measure of improvement in BER through FEC.
It is expressed in terms of the equivalent value of Q as G
c
=
20 log
10
(Q
c
/Q).
Q
c
and Q are related to the BERs as
BER
c
=
1
2
erfc(Q
c
/

2), BER =
1
2
erfc(Q/

2).
Factor of 20 is used in place of 10 because performance is often
quantied through Q
2
.
If FEC decoder improves BER from 10
3
to 10
9
, Q increases from
3 to 6, resulting in a coding gain of 6 dB.
Magnitude of coding gain increases with the FEC overhead.
248/549

Back
Close
Coding Gain
For single RS codes, coding gain is 5.5 dB for 10% overhead and
increases sublinearly, reaching 8 dB for 50% overhead.
It can be improved by concatenating two or more RS codes or by
employing the RS product codes.
249/549

Back
Close
Product Codes
Same code is applied along the rows and columns of a block.
Overhead of n
2
/k
2
1 for a RS product code is larger, but it also
allows more error control.
6 dB of coding gain possible with only 5% overhead.
250/549

Back
Close
Coding Gain
While implementing FEC, one faces a dilemma.
As the overhead is increased to realize more coding gain, bit rate
of the signal increases.
Since Q factor realized at the receiver depends on the bit rate, its
value is reduced, and BER actually worsens.
Decoder improves it but it rst has to overcome the degradation
caused by the increased bit rate.
If an aggressive FEC scheme is employed, BER may degrade so
much that the system is not operable even with the FEC coder.
An optimum range of coding overhead exists for every system de-
signed to operate at a specic bit rate over a certain distance.
251/549

Back
Close
Coding Gain
Numerically simulated Q factors (a)
before and (b) after the FEC de-
coder as a function of code redun-
dancy for a WDM system with 25
channels at 40 Gb/s.
With FEC, Q factor becomes worse
as overhead increases.
252/549

Back
Close
Chapter 6:
Optical Amplier Noise
Origin of Amplier Noise
Optical Signal-to-Noise Ratio
Electrical Signal-to-Noise Ratio
Receiver Sensitivity and Q Factor
Role of Dispersive and Nonlinear Eects
Periodically Amplied Lightwave Systems
253/549

Back
Close
Optical Ampliers
Used routinely for loss compensation since 1995.
Amplify the input signal but also add some noise.
Several kinds of ampliers have been developed:
Semiconductor optical ampliers
Raman-based ber ampliers
Erbium-doped ber ampliers
EDFAs are used most commonly for lightwave systems.
Raman ampliers work better for long-haul systems.
254/549

Back
Close
Erbium-Doped Fiber Ampliers
Developed after 1987 and commercialized during the 1990s.
Fiber core doped with erbium (length 20-200 m).
Pumped using diode lasers operating at 980 or 1480 nm.
Provide 2030 dB gain at pump powers <50 mW.
Gain bandwidth up to 40 nm possible.
Relatively low noise; Noise gure 4 to 5 dB.
Provide polarization-independent gain.
Gain pattern independent (Response time 10 ms).
Can be designed to work in both the C and L bands.
255/549

Back
Close
Pumping and Gain
Semiconductor lasers at 980 or 1480 nm are used for pumping.
Pumping eciency up to 11 dB/mW possible at 980 nm.
Amplication occurs when ions in the excited state emit coherent
light through stimulated emission.
256/549

Back
Close
Origin of Amplier Noise
Re(A)
Im(A)
Im(A)

|A|
|A + A|
Spontaneously
emitted photon
with random phase
Source of noise: Spontaneous emission
Spontaneous emitted photons have random phase and
polarization.
They perturb both A and phase in a random fashion.
Such random perturbations are the source of amplier noise.
257/549

Back
Close
Modeling of Amplier Noise
NLS equation including the gain and noise of optical ampliers:
A
z
+
i
2
2

2
A
t
2
= i[A[
2
A+
1
2
(g
0
)A+ f
n
(z, t).
Gain coecient g
0
=
e
N
2

a
N
1
;
e
and
a
are emission and
absorption cross sections.
Noise term vanishes on average, i.e, f
n
(z, t)) = 0.
Noise modeled as a Markovian process with Gaussian statistics
f

n
(z, t) f
n
(z
/
, t
/
)) = n
sp
h
0
g
0
(z z
/
)(t t
/
).
Spontaneous-emission factor n
sp
=
e
N
2
/(
e
N
2

a
N
1
).
Two delta functions ensure that all spontaneous-emission events are
independent of each other in time and space.
258/549

Back
Close
Noise of Lumped Ampliers
Amplier Length l
a
is much shorter than amplier spacing L
A
.
Neither loss, nor dispersion, nor nonlinearities are important within
the amplier.
Neglecting them and integrating, we obtain:
A
out
(t) =

GA
in
(t) +a
n
(t) with G = exp(g
0
l
a
).
Amplied spontaneous emission (ASE) at the amplier output:
a
n
(t) =
_
l
a
0
f
n
(z, t)exp
_
1
2
g
0
(l
a
z)

dz.
Since f
n
(z, t)) = 0, a
n
(t) also vanishes on average.
Second moment of a
n
(t) is found to be
a
n
(t)a
n
(t
/
)) = S
ASE
(t t
/
), S
ASE
= n
sp
h
0
(G1).
259/549

Back
Close
Total Noise Power
It is important to note that a
n
(t) represents only the portion of ASE
that is coupled to the mode occupied by the signal.
One must add up noise over the entire bandwidth of amplier.
If an optical lter is used, ASE power becomes
P
ASE
= 2
_

S
ASE
H
f
(
0
) d 2S
ASE

o
.

o
is the eective bandwidth of optical lter.
Factor of 2 takes into account two orthogonally polarized modes of
ber.
Only half the noise power is copolarized with the optical signal.
260/549

Back
Close
Distributed Amplication
In the case of distributed amplication, NLS equation should be
solved along the entire ber link.
Gain g
0
(z) is not constant along the ber length.
It is not easy to solve the NLS equation. If we set
2
=0 and =0,
the solution is A(L, t) =
_
G(L)A(0, t) +a
n
(t) with
a
n
(t) =
_
G(L)
_
L
0
f
n
(z, t)
_
G(z)
dz, G(z) = exp
_
_
z
0
[g
0
(z
/
) ] dz
/
_
.
a
n
(t) vanishes on average and its second moment is given by
a
n
(t)a
n
(t
/
)) =G(L)
_
L
0
dz
_
L
0
dz
/
f
n
(z, t) f
n
(z
/
, t
/
))
_
G(z)G(z
/
)
=S
ASE
(t t
/
),
Spectral density: S
ASE
= n
sp
h
0
G(L)
_
L
0
g
0
(z)
G(z)
dz.
261/549

Back
Close
Distributed Raman Amplication
The origin of noise is related to spontaneous Raman scattering.
Spontaneous-emission factor n
sp
has a dierent meaning than that
in the case of EDFAs.
No electronic transitions involved during Raman amplication.
Spontaneous Raman scattering is aected by phonon population
that depends on temperature of the ber.
More precisely, n
sp
is given by
n
sp
() = 1+
1
exp( h/k
B
T) 1

1
1exp( h/k
B
T)
.
At room temperature n
sp
= 1.14 near the Raman-gain peak.
262/549

Back
Close
Total ASE Power
Total ASE power is obtained by adding contributions over the
Raman-gain bandwidth or the bandwidth of optical lter.
Assuming a lter is used, the total ASE power is given by
P
ASE
= 2
_

S
ASE
H
f
(
0
)d = 2S
ASE

o
.
Factor of 2 includes both polarization components.
Substituting the expression for S
ASE
, ASE power becomes
P
ASE
= 2n
sp
h
0

o
G(L)
_
L
0
g
0
(z)
G(z)
dz.
ASE power depends on the pumping scheme through g
0
(z).
263/549

Back
Close
Optical SNR
Optical SNR = Ratio of optical power to ASE power.
Assume that all ampliers are spaced apart by L
A
and have the
same gain G = exp(L
A
).
Total ASE power for a chain of N
A
ampliers:
P
tot
ASE
= 2N
A
S
ASE

o
= 2n
sp
h
0
N
A
(G1)
o
.
Factor of 2 takes into account unpolarized nature of ASE.
Optical SNR is thus given by
SNR
o
=
P
in
P
tot
ASE
=
P
in
lnG
2n
sp
h
0

o
L
T
(G1)
.
We used N
A
= L
T
/L
A
= L
T
/lnG for a link of total length L
T
.
264/549

Back
Close
Optical SNR
20 40 60 80 100
Amplifier Spacing (km)
0
5000
10000
15000
S
y
s
t
e
m

L
e
n
g
t
h

(
k
m
)
2 mW
1.0
0.5
0.2
SNR can be enhanced by reducing the gain of each amplier.
ASE-limited system length as a function of L
A
for several values of
input power using = 0.2 dB/km, n
sp
= 1.6,
o
= 100 GHz.
It is assumed that an SNR of 20 is required by the system.
265/549

Back
Close
Optimum Amplier Spacing
Optimum L
A
becomes smaller as system length increases.
Amplier spacing can be improved by increasing input power P
in
.
In practice, maximum launched power is limited by the onset of
various nonlinear eects.
Typically, P
in
is limited to close to 1 mW.
At such power levels, L
A
should be in the range of 40 to 50 km for
submarine lightwave systems with lengths of 6,000 km or more.
Amplier spacing can be increased to 80 km for terrestrial systems
with link lengths under 3,000 km.
266/549

Back
Close
Case of Distributed Amplication
Optical SNR in this case takes the form
SNR
o
=
P
in
2N
A
S
ASE

o
, S
ASE
= n
sp
h
0
G(L)
_
L
0
g
0
(z)
G(z)
dz.
Pump power can be injected in the forward, backward, or both
directions.
g(z) depends on the pumping scheme, and S
ASE
depends on g(z).
We can control optical SNR by adopting a suitable pumping scheme.
Consider a 100-km-long ber section pumped bidirectionally to pro-
vide distributed Raman amplication.
ASE spectral density and optical SNR are shown as a function of
net gain when P
in
= 1 mW.
267/549

Back
Close
SNR for Raman Amplication
-20 -16 -12 -8 -4 0 4 8 12
-95
-90
-85
-80
-75
-70
-65
-60
-55
-50
-45
-40
Net gain (dB)
0%
20%
40%
60%
80%
100%
% Forw. Pump.
(a)
-20 -16 -12 -8 -4 0 4 8 12
38
40
42
44
46
48
50
52
54
56
58
(
d
B
)
A
S
E
Net gain (dB)
P
in
= 0 dBm
(b)
A
S
E

s
p
e
c
t
r
a
l

d
e
n
s
i
t
y

(
d
B
m
/
G
H
z
)
O
S
N
R
Fraction of forward pumping varies from 0 to 100%.
Losses are 0.26 and 0.21 dB/km at pump and signal wavelengths.
Other parameters are n
sp
= 1.13, h
0
= 0.8 eV, and
g
R
= 0.68 W
1
/km.
268/549

Back
Close
Distributed Raman Amplication
Optical SNR is highest in the case of purely forward pumping.
It degrades by as much as 15 dB as the fraction of backward pump-
ing is increased from 0 to 100%.
ASE generated near the input end experiences losses over the full
length of the ber in the case of forward pumping.
It experiences only a fraction of losses for backward pumping.
If N
A
such sections are employed to form a long-haul ber link, SNR
is reduced by a factor of N
A
.
Even when L
T
= 10, 000 km (N
A
= 100), SNR
o
remains >20 dB.
Such high values of optical SNR are dicult to maintain when ED-
FAs are used.
269/549

Back
Close
Electrical SNR
Optical SNR is not what governs the BER at the receiver.
Electrical SNR of the current generated is more relevant for signal
recovery at the receiver.
Assume that a single optical amplier is used before receiver to
amplify a low-power signal before it is detected.
This conguration is sometimes used to improve receiver sensitivity
through optical preamplifcation.
Optical fiber
Amplifier
Detector G
Current
P
s GP
s
+ P
ASE
270/549

Back
Close
ASE-Induced Current Fluctuations
Photocurrent I = R
d
([

GE
s
+E
cp
[
2
+[E
op
[
2
) +i
s
+i
T
.
It is necessary to separate the ASE into two parts because only its
copolarized part can beat with the signal.
ASE-induced current noise has its origin in beating of E
s
with E
cp
and beating of ASE with itself.
Useful to divide bandwidth
o
into M bins, each of bandwidth

s
, and write
E
cp
=
M

m=1
(S
ASE

s
)
1/2
exp(i
m
i
m
t).

m
is the phase of noise component at
m
=
l
+m(2
s
).
An identical form applies for E
op
.
271/549

Back
Close
ASE-Induced Current Fluctuations
Using E
s
=

P
s
exp(i
s
i
s
t) and including all beating terms,
I = R
d
GP
s
+i
sigsp
+i
spsp
+i
s
+i
T
.
i
sigsp
and i
spsp
represent current uctuations resulting from signal
ASE and ASEASE beating:
i
sigsp
= 2R
d
(GP
s
S
ASE

s
)
1/2
M

m=1
cos[(
s

m
)t +
m

s
],
i
spsp
= 2R
d
S
ASE

s
M

m=1
M

n=1
cos[(
n

m
)t +
m

n
].
i
spsp
) = 2R
d
S
ASE

s
M 2R
d
S
ASE

o
R
d
P
ASE
.
Variances of two noise currents are found to be

2
sigsp
= 4R
2
d
GP
s
S
ASE
f ,
2
spsp
= 4R
2
d
S
2
ASE
f (
o
f /2).
272/549

Back
Close
Impact of ASE on SNR
Total variance
2
of current uctuations is given by

2
=
2
sigsp
+
2
spsp
+
2
s
+
2
T
.
Electrical SNR at the receiver becomes
SNR
e
=
I)
2

2
=
R
2
d
(GP
s
+P
ASE
)
2

2
sigsp
+
2
spsp
+
2
s
+
2
T
.
SNR realized in the absence of optical amplier:
SNR
/
e
=
R
2
d
P
2
s

2
s
+
2
T
.
For an ideal receiver with no thermal noise and R
d
= q/h
0
,
SNR
/
e
= P
s
/(2h
0
f ).
273/549

Back
Close
Noise Figure of Amplier
In practice, current variance is dominated by
2
sigsp
.
Neglecting
2
spsp
, the SNR is found to be
SNR
e
=
GP
s
(4S
ASE
+2h
0
)f
.
Using S
ASE
= n
sp
h
0
(G1), optical amplier is found to degrade
the electrical SNR by a factor of
F
o
=
SNR
/
e
SNR
e
= 2n
sp
_
1
1
G
_
+
1
G
.
F
o
is known as the noise gure of an optical amplier.
In the limit G 1, SNR is degraded by F
o
= 2n
sp
.
Even when n
sp
= 1, SNR is reduced by 3 dB.
274/549

Back
Close
Impact of Thermal Noise
Preceding conclusion holds for an ideal receiver.
In practice, thermal noise exceeds shot noise by a large amount.
It should be included before concluding that an optical amplier
always degrades the electrical SNR.
Retaining only the dominant term
2
sigsp
:
SNR
e
SNR
/
e
=
G
2
T
4R
2
d
P
s
S
ASE
f
.
This ratio can be made quite large by lowering P
s
.
Electrical SNR can be improved by 20 dB or more compared with
its value possible without amplication.
275/549

Back
Close
Electrical SNR
Thermal noise is the most important factor that limits the
electrical SNR.
Optical preamplication helps to mask thermal noise, resulting in
an improved SNR.
If we retain only dominant noise term, the electrical SNR becomes
SNR
e
=
GP
s
4S
ASE
f
=
GP
s

o
2P
ASE
f
.
This should be compared with the optical SNR of GP
s
/P
ASE
.
Electrical SNR is higher by a factor of
o
/(2f ) under identical
conditions.
The reason is that ASE noise contributes only over the receiver
bandwidth f that is much narrower than lter bandwidth
o
.
276/549

Back
Close
Noise Figure of Distributed Ampliers
Because of gain variations, noise gure is given by
F
o
= 2n
sp
_
L
0
g
0
(z)
G(z)
dz +
1
G(L)
.
Consider the following hybrid scheme:
100 -km fiber
Lumped Amplifier
Detector
G
Raman pump
Distibuted Raman Gain P
in
P
in
The predicted F
o
can exceed 15 dB depending on the span length.
This does not mean distributed ampliers are more noisy than
lumped ampliers.
277/549

Back
Close
Noise Figure of Distributed Ampliers
total
span only
When G
R
= 0 (no pumping), 100-km-long passive ber has a noise
gure of 20 dB.
If signal is amplied using a lumped amplier, additional 5-dB degra-
dation results in a total noise gure of 25 dB.
This value decreases as G
R
increases, reaching a level of 17.5 dB
for G
R
= 20 dB (no lumped amplication).
278/549

Back
Close
Noise Figure of Distributed Ampliers
It is common to introduce the concept of an eective noise
gure using F
eff
= F
o
exp(L).
F
eff
< 1 is negative on the decibel scale by denition.
It is this feature of distributed amplication that makes it
so attractive for long-haul WDM lightwave systems.
In the preceding example, F
eff
2.5 dB when pure distributed
amplication is employed.
Eective noise gure of a Raman amplier depends on the pumping
scheme used.
Forward pumping provides the highest SNR, and the smallest noise
gure.
279/549

Back
Close
Receiver Sensitivity and Q Factor
BER can be calculated following the method used in Chapter 5.
BER = p(1)P(0/1) + p(0)P(1/0) =
1
2
[P(0/1) +P(1/0)].
Conditional probabilities require PDF for the current I.
Strictly speaking, PDF does not remain Gaussian when optical am-
pliers are used.
If we assume it to remain Gaussian, BER =
1
2
erfc
_
Q

2
_
.
Q factor: dened as Q =
I
1
I
0

1
+
0
, where

2
1
=
2
sigsp
+
2
spsp
+
2
s
+
2
T
,

2
0
=
2
spsp
+
2
T
.
280/549

Back
Close
Approximate Q Factor
In the case of 0 bits,
2
s
and
2
sigsp
can be neglected as they are
signal-dependent.
Even for 1 bits
2
s
can be neglected in comparison with
2
sigsp
.
Thermal noise
2
T
can also be neglected when optical power at the
receiver is relatively large (>0.1 mW).
Noise currents
1
and
0
are then approximated by

1
= (
2
sigsp
+
2
spsp
)
1/2
,
0
=
spsp
.
We calculate the Q factor using
Q =
I
1
I
0

1
+
0
=
I
1
I
0
_

2
sigsp
+
2
spsp
+
spsp
.
281/549

Back
Close
Receiver Sensitivity
Assume that no energy is contained in 0 bits so that
I
0
= 0 and I
1
= 2R
d

P
rec
.
Using Q and expressions for
1
and
0
,

P
rec
= h
0
F
o
f [Q
2
+Q(
o
/f
1
2
)
1/2
].
Using

P
rec
=

N
p
h
0
B and f = B/2,

N
p
is given by

N
p
=
1
2
F
o
[Q
2
+Q(r
f

1
2
)
1/2
].
r
f
=
o
/f is the factor by which the optical lter bandwidth
exceeds the receiver bandwidth.
A remarkably simple expression for the receiver sensitivity.
It shows why ampliers with a small noise gure must be used.
It also shows how narrowband optical lters can help.
282/549

Back
Close
Receiver Sensitivity
10
0
10
1
10
2
10
3
0
100
200
300
400
500
600
Bandwidth Ratio,
o
/f
R
e
c
e
i
v
e
r

S
e
n
s
i
t
i
v
i
t
y
F
o
= 7 dB
5 dB
3 dB
Using Q = 6 with F
o
= 2 and r
f
= 2, the minimum value

N
p
= 43.3 photons/bit.
Without optical ampliers,

N
p
exceeds 1000.
283/549

Back
Close
Non-Gaussian Receiver Noise
Even though the ASE itself has a Gaussian PDF, detector current
does not follow Gaussian statistics.
Detector current I = R
d
([E
s
+E
cp
[
2
+[E
op
[
2
).
Orthogonal part of noise can be suppressed by placing a polarizer
in front of the receiver.
Using E
cp
=

M
m=1
(S
ASE

s
)
1/2
exp(i
m
i
m
t):
I = I
s
+2

I
N
I
s
M

m=1
c
m
+I
N
pM

m=1
(c
2
m
+s
2
m
).
Signal I
s
= R
d
[E
s
[
2
and noise current I
N
= R
d
S
ASE

s
.
Random variables c
m
and s
m
dened as c
m
+is
m
= exp(i
m
).
Integer p = 1 or 2 depending on whether a polarizer is used or not.
284/549

Back
Close
Non-Gaussian Receiver Noise
I is a function of a large number of random variables, each of which
follows Gaussian statistics.
Without ASEASE beating, I follows a Gaussian PDF.
However, this beating term cannot be ignored, and the statistics of
I are generally non-Gaussian.
PDF can be obtained in an analytic form. In the case of 0 bits
p
0
(I) =
I
pM1
(pM1)!I
pM
N
exp
_

I
I
N
_
.
In the case of 1 bits (using I
s
= I
1
)
p
1
(I) =
1
I
N
_
I
I
1
_
1
2
(pM1)
exp
_

I +I
1
I
N
_
I
pM1
_

II
1
I
N
_
.
285/549

Back
Close
Non-Gaussian Receiver Noise
Measured and predicted PDFs for 0 (top) and 1 bits (bottom). A
dashed line shows the Gaussian approximation.
PDF is far from Gaussian for 0 bits.
Deviations relatively small in the case of 1 bits.
Gaussian approximation holds better as the bandwidth of optical
lter increases.
286/549

Back
Close
Q Factor and Optical SNR
Assume I
0
0 and I
1
= R
d
P
1
.

2
sigsp
= 2R
d

P
1
P
ASE
/M,
2
spsp
= P
2
ASE
/M.
We assumed M =
o
/f 1.
Using
1
and
0
in the expression for Q,
Q =
SNR
o

2SNR
o
+1+1
.
SNR
o
P
1
/P
ASE
is the optical SNR.
This relation can be inverted to nd
SNR
o
=
2Q
2
M
+
2Q

M
.
287/549

Back
Close
Q Factor and Optical SNR
10 20 30 40 50 60 70 80 90 100
0
2
4
6
8
10
12
14
16
18

0
/f (M)
O
p
t
i
c
a
l

S
N
R
Q = 8
6
4
Optical SNR as a function of M for several values of Q factor.
We only need SNR
o
= 7.5 when M = 16 to maintain Q = 6.
288/549

Back
Close
Noise Growth through Modulation Instability
Each amplier adds ASE noise that propagates with the signal.
In a purely linear system, noise power would not change.
Modulation instability amplies ASE noise.
Using A(z, t) =
_
p(z)B(z, t), NLS equation becomes
B
z
+
i
2
2

2
B
t
2
= i p(z)[B[
2
B+ f
n
(z, t)/
_
p(z).
p(z) is dened such that p(z) = 1 at the location of ampliers.
A numerical approach is necessary in general.
Assuming a CW signal, the solution is of the form
B(z, t) = [

P
0
+a(z, t)] exp(i
NL
).

NL
= P
0
_
z
0
p(z)dz is the SPM-induced nonlinear phase shift.
289/549

Back
Close
Noise Growth
Assuming noise is much weaker than signal ([a[
2
P
0
),
a
z
+
i
2
2

2
a
t
2
= iP
0
e
z
(a+a

).
This linear equation is easier to solve in the Fourier domain
and leads to two coupled equations:
db
1
dz
=
i
2

2
b
1
+iP
0
e
z
(b
1
+b

2
),
db
2
dz
=
i
2

2
b
2
+iP
0
e
z
(b
2
+b

1
),
b
1
(z) = a(z, ), b
2
(z) = a(z, ), and =
n

0
.
When falls within the gain bandwidth of modulation instability,
the two noise components are amplied.
290/549

Back
Close
Noise Growth
Coupled linear equations can be solved easily when = 0.
They can also be solved when ,= 0. but the solution involves
Hankel functions of complex order and argument.
In a simple approach, ber is divided into multiple segments.
Propagation through each segment of length h is governed by
_
b
1
(z
n
+h)
b
2
(z
n
+h)
_
=
_
M
11
M
12
M
21
M
22
__
b
1
(z
n
)
b
2
(z
n
)
_
.
Matrix elements M
mn
are constants in each ber segment but change
from segment to segment.
Solution at the end of ber is obtained by multiplying individual
matrices.
291/549

Back
Close
Noise Growth
An example of numerically simulated spectrum at the end of a
2,500-km ber link with 50 ampliers placed 50 km apart.
Broad pedestal represents the ASE spectrum expected even in the
absence of nonlinear eects.
292/549

Back
Close
Noise Growth
Possible to calculate factor F
v
by which
2
sigsp
changes.
F
v
as a function of launched power (four 100-km-long sections).
(a) anomalous [D = 2 ps/(km-nm)];
(b) normal dispersion [D =2 ps/(km-nm)].
f = 2 GHz (crosses), 8 GHz (pluses), 20 GHz (stars), and
30 GHz (circles).
293/549

Back
Close
Noise-Induced Signal Degradation
Optical signal degrades as ASE noise is added by ampliers.
As expected, ASE induces power uctuations (reduced SNR).
Surprisingly, ASE also induces timing jitter.
Physical origin of ASE-induced jitter: Ampliers aect not only
amplitude but also phase of amplied signal.
Chirping of pulses shifts signal frequency from
0
by a small amount
after each amplier.
Since group velocity depends on frequency (because of dispersion),
speed at which a pulse propagates is aected by each
amplier.
Speed changes produce random shifts in pulse position at receiver.
294/549

Back
Close
Moment Method Revisited
Moment method can be used by introducing two new moments.
q and represent pulse position and shift in the carrier frequency:
q(z) =
1
E
_

t[B(z, t)[
2
dt, (z) =
i
2E
_

_
B

B
t
B
B

t
_
dt.
E(z)
_

[B(z, t)[
2
dt is related to pulse energy.
Dierentiating E, q, and with respect to z,
dE
dz
= 0,
dq
dz
=
2
,
d
dz
= 0.
Energy E and frequency do not change during propagation.
Pulse position shifts for a nite value of as q(z) =
2
z.
295/549

Back
Close
Moment Method Revisited
Because of ASE added by the amplier, E, , and q change by
random amounts E
k
,
k
, and q
k
after each amplier:
dE
dz
=

k
E
k
(z z
k
),
dq
dz
=
2
+

k
q
k
(z z
k
),
d
dz
=

k

k
(z z
k
).
The sum is over the total number of ampliers encountered by the
pulse before it arrives at z.
ASE-induced timing jitter can be reduced by operating a lightwave
system near the zero-dispersion wavelength of ber.
296/549

Back
Close
Noise-Induced Timing Jitter
Total jitter at the end of the ber link:
2
t
=q
2
f
) q
f
)
2
.
Angle brackets denote averaging over amplier noise.
Final result turns out to be relatively simple:

2
t
= (S
ASE
/E
0
)T
2
0
N
A
[(1+(C
0
+N
A
d
a
/T
2
0
)
2
].
d
a
=
_
L
A
0

2
(z)dz is the dispersion accumulated over the entire link.
In the case of perfect dispersion compensation (d
a
= 0),
2
t
in-
creases linearly with the number N
A
of ampliers.
When d
a
,= 0, it increases with N
A
in a cubic fashion.
297/549

Back
Close
Noise Growth
0 2000 4000 6000 8000 10000
Distance (km)
0
5
10
T
i
m
i
n
g

J
i
t
t
e
r

(
p
s
)
0.3
0.2
0.1
0.01 ps
2
/km
ASE-induced timing jitter as a function of system length for several
values of average dispersion

2
.
Results are for a 10-Gb/s system with T
0
= 30 ps, L
A
= 50 km,
C
0
= 0.2, and S
ASE
/E
0
= 10
4
.
ASE-induced jitter becomes a signicant fraction of pulse width
because of the cubic dependence of
2
t
on system length L
T
.
298/549

Back
Close
Distributed Amplication
Distance (km)
0 2000 4000 6000 8000
T
i
m
i
n
g

j
i
t
t
e
r

(
p
s
)
0
2
4
6
8
12 dB
8 dB
4 dB
G
R
= 0 dB
16 dB
Total Loss = 16 dB
Raman gain is varied from 0 to 16 dB (total loss over 80 km).
Dashed line shows the tolerable value of timing jitter.
299/549

Back
Close
Numerical Approach
Nonlinear and dispersive eects act on a noisy optical signal
simultaneously.
Their mutual interplay cannot be studied analytically.
Most practical approach for designing modern lightwave system con-
sists of solving the NLS equation numerically.
Numerical simulations indeed show that nonlinear eects often limit
the system performance.
System design requires optimization of various parameters such as
amplier spacing and input power launched.
Several software packages are available commercially.
One such package called OptSim 4.0 is provided on the CD.
300/549

Back
Close
OptSim Simulation Package
Layout of a typical lightwave system for modeling based on the
software package OptSim.
Main advantage: Optimum values of various system parameters can
be found such that design objectives are met at a minimum cost.
301/549

Back
Close
Numerical Approach
Input to optical transmitter is a pseudo-random sequence of elec-
trical pulses, representing 1 and 0 bits.
The length N of this bit pattern determines the computing time
and should be chosen judiciously.
Typically, N = 2
M
, where M is in the range of 6 to 10.
Optical bit stream obtained by solving the rate equations that gov-
ern the modulation response of the laser or modulator.
Deformation of optical bit stream during its transmission calculated
by solving the NLS equation.
Method most commonly used for solving this equation is known as
the spit-step Fourier method.
302/549

Back
Close
Numerical Approach
Two equivalent techniques used for adding ASE noise to the
signal during numerical simulations.
In one case, noise is added in the time domain, while ensuring that
it follows Gaussian statistics with a
n
(t)a
n
(t
/
)) = S
ASE
(t t
/
).
Because of a nite temporal resolution t, delta function is replaced
with a rect function of width t.
Its height is chosen to be 1/t so that
_

(t)dt = 1 is satised.
Alternatively, noise can be added in the frequency domain:

A
out
() =

G

A
in
() + a
n
().
Real and imaginary parts of a
n
() follow Gaussian statistics.
Noise is assumed to be white (same variance at each frequency).
303/549

Back
Close
Numerical Approach
A receiver model converts optical signal into electric domain.
An electric lter used with its bandwidth f smaller than bit rate
B (typically f /B = 0.60.8).
Electric bit stream is used to nd the instantaneous values of
currents, I
0
and I
1
by sampling it at the center of each bit slot.
Eye diagram is constructed using the ltered bit stream.
System performance is quantied through the Q factor, related di-
rectly to the BER.
Calculation of Q factor requires that the NLS equation be solved a
large number of times with dierent seeds for amplier noise.
Such an approach becomes quite time-consuming for WDM
systems.
304/549

Back
Close
Optimum Launched Power
Input Power
Q

F
a
c
t
o
r
N
o
i
s
e

L
i
m
i
t

(
L
u
m
p
e
d
)
N
o
i
s
e

L
i
m
i
t

(
D
i
s
t
r
i
b
u
t
e
d
)
N
o
n
l
i
n
e
a
r

L
i
m
i
t
Optimum
Value
Q-factor variations with launched power in long-haul systems.
Q factor increases initially with launched power, reaches a peak
value, and then decreases with a further increase in power because
of the onset of the nonlinear eects.
Use of distributed amplication improves system performance.
305/549

Back
Close
Optimum Launched Power
Numerical results for a 32-channel WDM system.
Maximum distance plotted as a function of input power.
Fiber link contains 80-km sections whose 20-dB loss compensated
using (a) forward or (b) backward pumping conguration.
Pump depletion becomes signicant at arrow location.