Optical Fiber Communications:
Even More Fun in the Post-Bubble Era
Joseph M. Kahn
Department of Electrical Engineering
Stanford University
www-ee.stanford.edu/~jmk
Clean Slate Seminar, February 6, 2006
Acknowledgments
Adaptive Signal Processing
in Multimode Networks
StrataLight Communications

Keang-Po Ho

Elad Alon

Shanhui Fan

Mark A. Horowitz
Modulation and Detection in
Single-Mode Networks

Wei Mao

Ezra Ip

Rahul A. Panicker

Alan P. T. Lau

Mahdieh B. Shemirani

Dany-Sebastien Ly-Gagnon

Xiling Shen

Jin Wang

Vladimir Stojanovic
2
Optical Networks: Meter to Megameter Scale

Sensors

Local- and campus-area

Access

Metropolitan

Long-haul

Submarine
3
Optical Networks: Meter to Megameter Scale

Sensors

Local- and campus-area

Access

Metropolitan

Long-haul

Submarine
4
Local- and Campus-Area Networks
5
Optical Fiber Types
n
8-10 mm
 Wide-area, metro-area networks
 Limitations: amplifier noise,
fiber nonlinearity
Single-mode
n
 Throughput (with WDM):
80 channels  40 Gb/s  4000 km
50 mm
 Local-area networks
 Limitation: large modal dispersion
Step-index multi-mode
n
 Throughput (without WDM):
100 Mb/s  few km
50 mm
 Local-area, campus-area networks
 Limitation: moderate modal dispersion
Graded-index multi-mode
 Throughput (without WDM):
1 Gb/s  few km
6
Modes in Optical Fibers
Modes
 Mutually orthogonal solutions of wave equation having well-defined propagation constants.
 Propagate without cross-coupling in ideal fiber.
 Typical multimode fiber supports of order 100 modes.
Modal coupling
 Bends and imperfections couple modes over distances of the order of meters.
 Coupling varies on time scale of seconds.
Modal dispersion
 Different modes have different group delays, causing pulse spreading.
Transmitted
Received
t
t
7
Motivations

New techniques needed to:

Extend 10 Gb/s Ethernet over multimode fiber (currently limited to 300 m)

Enable 100 and 1000 Gb/s Ethernet over multimode fiber

Optical signal processing scales better than electronic signal
processing to high bit rates, long fibers, multiple WDM channels.

Lesson from digital subscriber lines: ubiquitous, bandwidthconstrained media should be exploited to the limit.

Modal dispersion is analogous to multipath fading: should it be
eliminated or exploited?
8
Principal Modes in Multimode Fibers


Multimode fiber

Supports 2N ideal modes (including 2 polarizations).

Ideal modes are strongly coupled by bends and imperfections.
Principal modes

PMs are linear combinations of ideal modes.

Input PMs: a set of 2N orthogonal modes at fiber input.
Output PMs: a set of 2N orthogonal modes at fiber output.

Each input PM propagates to the corresponding output PM:
Without cross-coupling to other PMs.
With a well-defined group delay.

In a given fiber, the PMs change slowly over time.
Adaptive signal processing can identify and track PMs.
S. Fan and J. M. Kahn, Optics Letters, January 15, 2005.
9
Adaptive Single-Input, Single-Output Transmission
Fourier Lens
Iout(t)
Adaptive
Algorithm
Multimode Fiber
Spatial Light
Modulator
Trans.
Data
PhotoDetector
Iin(t)
Rec.
Data
ISI
Estimation
OOK
Modulator
Transmitter
Clock & Data
Recovery
Receiver
ISI Objective
Function
Low-Rate Feedback Channel
10
Controlling MMF Impulse Response via SLM
1
2N
2N-Mode Fiber
(2N  2N)
…
Spatial Light
Modulator
(1  2N)
2
…
Input
Intensity
(Scalar)
…
Iin(t)
Iin(t)
| |2
| |2
| |2

Iout(t)
h(t)
R Iout(t)
h(t)
| |2
Output
Photodetector
Photocurrent
(2N  1)
(Scalar)
t
2N pulses

Input principal modes:

Mode incident on SLM:

SLM reflectance:

MMF impulse response:
ht  
e
 L 2 N
P0

n 1
ein ,n ( x, y )  Ein ,n (k x , k y )
hin ,n ( x, y)  H in ,n (k x , k y )
e0 ( x, y)  E0 (k x , k y )
V (k x , k y )
2


*
Re  V k x , k y E 0 k x , k y  H in,n k x , k y  zˆdk x dk y  t   n 
 SLM


 



11
System Model and Adaptive Algorithm
Transmitted
Bits
an  0,1
Transmitted
Pulse Shape
p(t)
Iin(t)
MMF Impulse Iout(t) Receiver Impulse
Response
Response
h(t)
r(t)
+
Received
Bits
aˆ n  0,1
t  t 0  nT
Noise
n(t)
Control
SLM
V(kx, ky)
Adaptive
Algorithm
ISI
Objective Function
F(g(nT; t0))
ISI
Estimation

Continuous-time impulse response:
g t   pt   ht   r t 

Discrete-time impulse response:
g nT ; t 0   g t  t t

Objective function quantifying ISI:
0  nT
F g nT; t 0   g 0T ; t 0    g nT; t 0 
n0
desired
bit interval
undesired
bit intervals

Note that F(g(nT; t0)) > 0 when eye open and F(g(nT; t0)) <0 when eye closed.

Adaptive algorithm controls V(kx, ky) to maximize F(g(nT; t0)).
12
Experimental Setup
X. Shen, J. M. Kahn and M. A. Horowitz, Optics Letters, November 15, 2005.
13
10 Gb/s  1030 m, Good and Bad SOPs, Binary SLM
14
10 Gb/s  11081 m, Good SOP, Vertical Misaligment of Launch,
Binary vs. Quaternary SLM
15
10 Gb/s  11081 m, Good SOP, Binary SLM,
Tune Laser Over 600 GHz
Condition
SLM Pattern
Eye Pattern
Result
Good SOP
Before adaptation
Channel 58
(193.40 THz)
Good SOP
After 2 iterations
Channel 58
(193.40 THz)
Error-free to
231 – 1 PRBS
Good SOP
Keep SLM fixed
Channel 51
(193.75 THz)
Error-free to
231 – 1 PRBS
Good SOP
Keep SLM fixed
Channel 62
(193.20 THz)
Error-free to
231 – 1 PRBS
16
Adaptive Spatial Optical Signal Processing

Key to exploiting principal modes.

Can be implemented using spatial light modulators.

One SLM can serve multiple WDM channels.

SLM requirements are at least somewhat independent of
bit rate and fiber length.

Contrast with electrical equalizers:

Must be implemented separately for each WDM channel.

FIR filter-based equalizers: number of taps proportional to
bit rate  fiber length.

Maximum-likelihood sequence detectors: number of states
exponential in bit rate  fiber length.
17
Ongoing and Future Work
Ongoing

Modeling propagation and principal modes

Optimal one-shot and adaptive algorithms

Robustness to perturbations of fiber
Future

What can we learn from adaptive systems to improve design
of lower-complexity systems?

Extension to other multimode media, e.g., polymer waveguides
for board-level interconnects

Electronics and optics for faster adaptation

Multi-input, multi-output transmission
18
Optical Networks: Meter to Megameter Scale

Sensors

Local- and campus-area

Access

Metropolitan

Long-haul

Submarine
19
Segments of Telecom Networks
20
Access Networks
Technologies (Heterogeneity Rules)

Wireless (radio and microwave)

Free-space optical

DSL over copper twisted pair

QAM over hybrid fiber-coax (single-mode fiber)

TDM / WDM over passive optical networks (single-mode fiber)
Some Issues

Performance vs. cost of installation and maintenance

Initial cost vs. upgradeability
21
Passive Optical Networks for Access
single downstream
wavelength (TDM)
multiple downstream
wavelengths (TDM / WDM)

Downstream / upstream: 1.55 mm / 1.31 mm (coarse WDM)

Downstream: broadcast and select (TDM or TDM / WDM)

Upstream: TDMA
22
Dense Wavelength-Division-Multiplexing
f1
f
f2
Tx1
f1 f 2
fN
Rx1
f
f
f1
Tx2
Rx2
TxN
RxN
Mux
Demux
f
f2
…
Erbium-Doped Fiber Amps
…
…
fN f
…
…
f
fN f
EDFA Bands
4.4 THz 5.4 THz
23
Metropolitan and Long-Haul Networks

Use single-mode fiber

Use electrical TDM (e.g., SONET) on WDM

Transmission

Maturing technology, challenging business

10 Gb/s transceivers approaching commodity status

Unused (dark) fibers exist on many routes


Many underutilized systems exist, e.g., 10 Gb/s  80 wavelengths
with only 10 wavelengths in use
Switching

Circuit switching: becoming more dynamic and flexible, for
reprovisioning, survivability, etc.

Packet switching: infeasible for several reasons, especially
lack of scalable optical buffers
24
Metropolitan and Long-Haul Networks
with Wavelength Routing
Router
Router
Chicago
l1
Kansas
City
Cleveland
New York
Optical
add-drop
demultiplexer
Long-haul core
l2
Philadelphia
Nashville
Optical
crossconnect
Router
Metro ring
Metro ring
Router
Router
25
Long-Haul Transmission: Trends
Increasing per-channel bit rates

For a given capacity, reduces number of ports on routers and
optical switches.

Last generation: 10 Gb/s

New generation: 40 Gb/s

Goal (of some): 160 Gb/s
Increasing capacity

Traffic continues to increase exponentially; capacity must increase,
cost per bit must decrease.

EDFA bandwidth is limited, Raman amps are expensive.

Best solution is to increase spectral efficiency:

Last generation: 0.2 - 0.4 b/s/Hz

New generation: 0.8 b/s/Hz

Binary limit: 1 b/s/Hz

Non-binary limit: perhaps 3-5 b/s/Hz
26
Long-Haul Transmission: Challenges
Increasing per-channel bit rates

Chromatic dispersion: penalty  B 2  L

Polarization-mode dispersion: penalty  B  L

Electronic circuits
Increasing spectral efficiency

Crosstalk and distortion in muxes, demuxes, OADMs, OXCs

Requires higher signal-to-noise ratio, but transmitted power is
limited by nonlinearities in fiber.
27
Approaches for 40 Gb/s Systems at 0.8 b/s/Hz
Mainstream




Goal: maximize unrepeatered transmission distance (à la Qtera)
Use OOK or DPSK with RZ pulses (broader spectrum)
Achieves: > 2000 km, 3.2 Tb/s (C band), 6.4 Tb/s (C + L bands)
Higher cost:



Two-stage modulator, complex receiver (for DPSK)
Careful control of chromatic dispersion in transmission system
Requires specially designed transmission system
28
RZ DPSK with Interferometric Detection
ES
ES
t
fS
Laser
fS
f
…
Bits
Differential
Encoder
t
Clock
Optical
BPF
fS
T
Elect.
LPF
0
ESQ
1
i
0
1
0
ESI
1
i
1
0
1
29
Approaches for 40 Gb/s Systems at 0.8 b/s/Hz
Mainstream




Goal: maximize unrepeatered transmission distance (à la Qtera)
Use OOK or DPSK with RZ pulses (broader spectrum)
Achieves: > 2000 km, 3.2 Tb/s (C band), 6.4 Tb/s (C + L bands)
Higher cost:



Two-stage modulator, complex receiver (for DPSK)
Careful control of chromatic dispersion in transmission system
Requires specially designed transmission system
StrataLight Communications (founded in June 2000)




Goal: minimize cost per Gb/s•km with sufficient unrepeatered distance
Use OOK with NRZ pulses and line coding (narrower spectrum)
Achieves: > 1200 km, 3.2 Tb/s (C band), 6.4 Tb/s (C + L bands)
Lower cost:



Simple modulator and receiver
Less careful control of chromatic dispersion in transmission system
Can retrofit to some underutilized 10 Gb/s transmission systems
30
Line-Coded OOK with Direct Detection
ES
t
Line
Coder
fS
Laser
fS
…
Bits
Optical
BPF
1
0
1
0
1
0
1
ES
f
fS
Electrical
LPF
i
|ES|2
i
0
1
31
Spectral Efficiency vs. SNR Efficiency
20
50
100
200
500
4
1000 2000
16
3
8
QAM / Coherent
PSK / Coherent
2
4
DPSK / Interferometric
or Diff. Coherent
PAM / Direct
or Non-Coherent
1
-3
0
3
6
9
12
15
Number of Constellation Points M
Relative Spectral Efficiency log2(M) (b/symbol)
1 DF
/ pol.
nb/n2eqDFRequired
for Pb = 109
(photons/bit)
/ pol.
2
SNR/bit Required Relative to 2-PAM (dB)
32
Bits
Encoder
EI
fS
EQ
…
4-PSK with Coherent Detection
Laser
90
f
fS
f
0
Example: synchronous homodyne
f L = fS
loop not shown)
LO
Laser
Pol.
Contr.
Local Oscillator
Photocurrent
ESQ
ELQ
iQ
A. Porter and J. M. Kahn, 1992
01
ESIS. Norimatsu et al, 1992 ELI
11
10
iI
Elect.
LPF
iQ
90
Signal
00
Elect.
LPF
0
detection (optical phase-locked
01
f
0
00
iI
11
10
33
16-QAM with Coherent Detection
EI
Laser
90
EQ
…
Bits
Encoder
Elect.
LPF
iI
Elect.
LPF
iQ
0
90
LO
Laser
Signal
ESQ
Pol.
Contr.
Photocurrent
Local Oscillator
iQ
ELQ
0000 0001 0011 0010
0000 0001 0011 0010
0100 0101 0111 0110 E
SI
ELI
0100 0101 0111 0110 i
I
1100 1101 1111 1110
1100 1101 1111 1110
1000 1001 1011 1010
1000 1001 1011 1010
34
Coherent Optical Detection: Pros and Cons
Advantages

Yields 2 degrees of freedom: higher spectral efficiency

Receiver detects all information in signal electric field
 enables digital signal processing to compensate impairments


Chromatic dispersion

Polarization-mode dispersion

Nonlinear phase noise
Can use tunable local oscillator with electrical filtering to select channel
 enables fast-tunable receiver for wavelength switching (or FHSS)
Drawbacks

Requires local oscillator laser at receiver

Requires polarization tracking or diversity at receiver
35
Nonlinear Phase Noise
EQ
EQ
EI
Linear Regime
EI
Nonlinear Regime
36
Optical Communication Research Issues
Transmission



Higher spectral efficiency vs. wider utilized bandwidth
Spectral efficiency vs. robustness vs. implementation complexity
Signal processing: optical vs. analog vs. digital
Switching


Circuit switching: faster and more flexible
Packet switching?
Component evolution



New fiber types
New amplifier types
Optical buffers?
Analysis


Spectral efficiency limits
Nonlinear phase noise
37
DARPA MTO TACOTA Program

Coherent links for tactical air-to-air communications

Major team: CeLight, Stanford (Fejer, Kahn), Boeing, HRL

Transmit at 3.8 mm to minimize atmospheric effects

1.55 mm transmitters and receivers

Transmitter: 1.55  3.8 mm downconverter

Receiver: 3.8  1.55 mm upconverter

Use frequency hopping for LPI/LPD

Receiver architecture

Homodyne (direct conversion to baseband)

Use sampling and DSP algorithms to compensate carrier phase,
Doppler shifts, atmospheric turbulence, etc.
38