CIAN Supercourse 2011

Optical Performance Monitoring to Enable Robust and Reconfigurable

Optical Networks

Alan Willner

University of Southern California Los Angeles, CA 90089-2565

USC’s OCLab

Outline

1. Overarching Perspective

2. Optical Performance Monitoring

- optically-assisted techniques - receiver based techniques

Differential Phase-Shift-Keying (DPSK)

DPSK

t

1 1 0 1 0 0

RZ-DPSK

t

1 1 0 1 0 0

Pulse appears in every bit

Constant optical power

Energy is information. Information is sent during “0” bits.

 Multi-level Modulation Formats in Optics

Benefits from coherent detection: •  More effective for pol-demuxing •  Digital processing for mitigation

1 bit/symbol 2 bits/symbol 4 bits/symbol

~112 Gbaud OOK DPSK

DB/PSBT

~56 Gbaud DQPSK (4-ASK)

~28 Gbaud PDM-(D)QPSK

(16-DPSK)

Im{Ex}  

Re{Ex}  ( )

Im{Ex}  

Re{Ex}  

Im{Ex}  

Re{Ex}  

Im{Ey}  

Re{Ey}  

8 bits/symbol

~14 Gbaud 16-QAM

8-PSK/2-ASK

Im{Ex}  

Re{Ex}  

Im{Ey}  

Re{Ey}  

Reference:  P.  Winzer,  Alcatel/Lucent,  OFT  V  

DPSK & DQPSK

T -

DPSK

Re{E}

Im{E}

  3-dB sensitivity improvement   Less sensitive to nonlinearity

π π 0 0

Input signal

1 0 1�

1 0 1�

0 1 0

T -

DQPSK

Re{E}

Im{E}

  Spectrally efficient - 2 bit/s/Hz   Tolerant to dispersion

Input signal

T -

+45°

-45°

I

I

Q

Q

I

Q

T

Polmux Concept

00 01

11 10

000 001

011 010

100 101

111 110

Regular (D)QPSK 2 bits per symbol

Polmux (D)QPSK 3 bits per symbol

polarization axis

polarization axis

Polarization is another dimension to carry information, so Polmux is more spectrally efficient.

DPSK & Polarization-Multiplexing

T -

D(Q)PSK

Re{E}

Im{E}

  Less sensitive to nonlinearity   3-dB sensitivity improvement

π π 0 0

Input signal

1 0 1�

1 0 1�

0 1 0 T

Pol-muxing doubles the spectral efficiency → enhanced performance

Pol-muxed DPSK channel PC

PC PBC

DPSK

DPSK

H

V t

t

H

V t

PBS

DPSK

DPSK

PC

H

t

V t

Transmitter Receiver Data 1

Data 2

Latest Results on High Capacity/S.E. Transmission

32Tb/s PDM-RZ-8QAM over 580km Ultra-low-loss Fiber

•  PDM-RZ-8QAM •  Digital coherent detection •  EDFA-only Amplification •  25GHz-spaced •  320x114Gb/s •  length / loss ratio 82.8km / 14.6dB

X. Zhou, OFC 2009 PDP

72x100Gb/s over 7040km Large Effective Area Fiber

G. Charlet, OFC 2009 PDP

•  100Gb/s channels •  88x80km distance •  Raman-Erbium amplification •  coherent receiver

10 bit/s/Hz Spectral Efficiency

Spectral Efficiency

  Challenge: to explore multilevel optical modulation formats

  Pack more bits per symbol: DQPSK, APSK, OFDM, QAM

  Powerful tool: orthogonal modulation

Improving Spectral Efficiency

Pol-Mux 1 Gsymbol/s, 128 QAM (14Gbit/s) (BW: 1.4 GHz)

Several Examples

Modulation Spectral Efficiency Reference

10×112 Gbit/s PolMux 16-QAM

6.2 bit/s/Hz A. H. Gnauck

PDPB8 2009

8×65.1 Gbit/s coherent

PolMux-OFDM 7 bit/s/Hz

H. Takahashi PDPB7

OFC2009

PolMux 1Gsymbol/s

128 QAM 14 Gbit/s

10 bit/s/Hz H. Goto JThA45

OFC2008

To date, largest spectral efficiency

Predicted “bursting” of bubble in ‘97

Heterogeneous Systems: One Network Fits All

Future Heterogeneous

Network

Economics: Early market entry of new services (CATV??)

Variable QoS Different Modulation Formats

Multiple Wavelength Ranges

Circuit + Packet Switching?

Variable Bit Rate

Sub-carrier Multiplexing

(D+A)?

•  Hardware should be reconfigurable and transparent •  An intelligent network could use the optimal method from the application/user viewpoint.

Think “wireless laptop LAN” …

Self-Managed Networks

“Adaptive” Resources •  Diagnose and repair •  BW allocation •  Gain/Loss •  Dispersion Compensation •  λ-Routing •  Look-up tables

A

C

D

B

E

Today : Measure, Make,

Tweak, Pray.

Automation + Intelligence + Monitoring Keep the person out of the loop

Monitoring the State of the Network

Ubiquitous Monitoring

• Monitor non-catastrophic data degradation

•  Isolate specific impairments

Detect Attacks

Locate Faults

Diagnose & Assess

Repair Damage

Reroute & Balance Traffic

Window of operability is shrinking Monitoring is required

• Ubiquitous deployment

• Graceful routing based on physical state of network?

Telcos: Human Error (~1/3 of outages)

Malicious Behavior

OPM

CD PMD OSNR Power λ Crosstalk

Hardware

  Optical/RF filter   Low-speed detector Software   Pattern recognition

using neuron networks and data constellations

Spec

  Update rate   Isolation   Advanced modulation

format   One or more faults

NC & M Actions Impacted by OPM •  What impairment is affecting the traffic/data? •  Should compensation be tuned? •  Should format/rate be changed? •  Should QoS be changed? •  Should routing table be changed?

Network & Switching Fabric

……

Determine each parameter: For example: Level “0” = no problem �Level “10” = channel outage

Design of Optical Performance Monitor

PARAGON

Monitoring for an Efficient Network Robert Shapiro, former Undersecretary of Commerce: “Accommodating the fast-rising demands on bandwidth will require a significant acceleration in industry investments – totaling $300 billion to $1 trillion for the US”.

 Operate closer to the “red line”.

 Less need to over-build.

 Increase mean-time-to-failure.

 Decrease mean-time-to-repair.

 Decrease human error.

Multivariable Routing

< αj, βj>

< ai, bi, ci> a.  Fiber length b.  Signal degradation c.  Amplification and transients

α  Component non-idealities β  Signal degradation

  Each link and node has a set of parameters (a, b, c)   Must interpret the “cost function” for routing table   Determine “ranges” of these parameters for

inclusion into network model •  Interoperability with fiber plant •  # of nodes •  Size of network

Outline

1. Overarching Perspective

2. Optical Performance Monitoring

- optically-assisted techniques - receiver based techniques

Optical Signal-to-Noise Ratio

Arbitrarily Polarized signal

+ Unpolarized noise

Polarization controller

Ps + Pase

Polarizer (Parallel)

Polarizer (Orthogonal)

Ps + 0.5*Pase

0.5*Pase

Y. C. Chung et. al., JLT, 2006

  The received signal (together with noise) is split into two orthogonal polarization components.   The polarization ratio is a measure of the OSNR (Ps/Pase).   The performance could be affected by various polarization effects.

OSNR Monitoring Using Polarization Nulling

Transparent to multiple input data format and bit rate

  Using partial bit delay-line Interferometer (DLI)   OSNR is proportional to the Ratio (=Pconst / Pdest)   Applicable to OOK, DPSK data

Y. Lize, et. al., PTL’ 07 and JLT’ 08

OSNR Monitoring for Multiple Modulation Formats

T

Power Meter

Power Meter

Pconst

Pdest

Input signal

delay

)21

41(

)21

43(

PPPP

noisesignal

noisesignalRatio

+

+=

Signal has coherent interference, noise doesn’t

  Using partial bit delay-line Interferometer (DLI)   OSNR is proportional to the Ratio (=Pconst / Pdest)   Applicable to OOK, DPSK data

Y. Lize, et. al., PTL’ 07 and JLT’ 08

OSNR Monitoring for Multiple Modulation Formats

FSR=1/T

Constructive port Destructive port •  Monitor tones to isolate -- CD and PMD

•  Insensitive to CD and PMD •  DB / AMI have “tones” •  OSNR -- only S coherent, not N

T

Power Meter

Power Meter

Pconst

Pdest

Input signal

delay

Channel Monitoring using Integrated Filters

)

21

41(

)21

43(

PPPP

noisesignal

noisesignalRatio

+

+=

OSNR: Signal has coherent interference, not noise

CD & PMD

Tones affected differently by CD & PMD Y. Lize, et. al., PTL’ 07 and JLT’ 08

Dependence on Chromatic DispersionTemperature Dependence

-100

-50

0

50

100

-40 -30 -20 -10 0 10 20 30 40D

ispe

rsio

n C

hang

e, Δ

D(p

s/nm

)

NRZ 40 Gbit/s Limit

L=1000 km

L=500 km

L=200 km

Dispersion Slope ~ 0.08 ps/nm2•kmdλ0/dT ~ 0.03 nm/ºC

NRZ 40 Gbit/s Limit

-100

-50

0

50

100

-100

-50

0

50

100

-40 -30 -20 -10 0 10 20 30 40-40 -30 -20 -10 0 10 20 30 40D

ispe

rsio

n C

hang

e, Δ

D(p

s/nm

)

NRZ 40 Gbit/s Limit

L=1000 km

L=500 km

L=200 km

Dispersion Slope ~ 0.08 ps/nm2•kmdλ0/dT ~ 0.03 nm/ºC

NRZ 40 Gbit/s Limit

Temp Change, C°

Temperature Dependence

-100

-50

0

50

100

-40 -30 -20 -10 0 10 20 30 40D

ispe

rsio

n C

hang

e, Δ

D(p

s/nm

)

NRZ 40 Gbit/s Limit

L=1000 km

L=500 km

L=200 km

Dispersion Slope ~ 0.08 ps/nm2•kmdλ0/dT ~ 0.03 nm/ºC

NRZ 40 Gbit/s Limit

-100

-50

0

50

100

-100

-50

0

50

100

-40 -30 -20 -10 0 10 20 30 40-40 -30 -20 -10 0 10 20 30 40D

ispe

rsio

n C

hang

e, Δ

D(p

s/nm

)

NRZ 40 Gbit/s Limit

L=1000 km

L=500 km

L=200 km

Dispersion Slope ~ 0.08 ps/nm2•kmdλ0/dT ~ 0.03 nm/ºC

NRZ 40 Gbit/s Limit

Temp Change, C°

Data Rate Dependence

Dispersion ps

nm·kmBit-RateDoubled

Time Half Freq. Double

-1/T 1/T0-1/2T 1/2T-1/T 1/T0-1/2T 1/2T

Penalty increasesFOUR times

Time Freq

Data Rate Dependence

Dispersion ps

nm·kmBit-RateDoubled

Time Half Freq. Double

-1/T 1/T0-1/2T 1/2T-1/T 1/T0-1/2T 1/2T

Penalty increasesFOUR times

Time Freq

Vestigial Sideband Optical Filtering

Frequency

BW

Δf VSB-U VSB-L

Optical Carrier

fU f0 fL

• Time delay ( Δt ) between two VSB signals is a function of chromatic dispersion

• Bits can be recovered from either part of the spectrum

40-Gb/sRZ Data

VSB-L

VSB-U

f

Dispersion

f

O/E Δt

Chromatic Dispersion Monitoring Using Clock Phase

• Isolate CD from PMD effects• Low cost

Q. Yu, JLT, Dec., 2002

Filteredspectrum

Entirechannel

Filteredspectrum

0 50 100 1500.0

0.5

1.0

1.5

Inte

nsity

Time (ps)

0 50 100 1500.0

0.5

1.0

1.5

Inte

nsity

Time (ps)

Q. Yu, JLT, 2003

Polarization Mode Dispersion (PMD) cross section

Elliptical Fiber Core

side view

•  PMD induces randomly changing degradations.

•  Critical limitation at >10 Gbit/s data rates.

The 2 polarization modes propagate at different speeds. 1st-order PMD = DGD

Probability of Exceeding a Specific DGD (%)

0 10 20 30 40 50

0.1 1 10 50

Maxwellian distribution

tail Pro

babi

lity

Dis

tribu

tion

0 10 20 30 40 50 0 10 20 30 40 50

0.1 1 10 50

Differential Group Delay (ps)

Maxwellian distribution

tail

Significant higher-order effects can exist.

In Phase

t

t Axis 2

Axis 1

Out of Phase

Δτ�

Axis 2

Axis 1

λ�

Carrier Upper clock

Lower clock

PMD (Axis

Delay)

Power

f

CD (Freq. Delay)

In Phase

t

t Lower

Upper

Out of Phase

Δτ�

Lower

Upper

Two Clocks

Upper Clock

RF Clock Tone Fading �

PMD Monitoring by Narrowband Filtering

λ

Optical spectrum

detection

Electrical Domain

Clock fades with

PMD & CD

detectionClock fades

with ‘PMD only’

w/o filter

w/ partial filtering λ

λλ

SMF

Upper & Lower Clocks

Only Upper Clock

T. Luo, et al., PTL, 2004

-30

-20

-10

0

0 10 20 30 40 50

w/ filter

DGD (ps)

Rel

ativ

e C

lock

Pow

er (d

B)

DGD (ps)

Rel

ativ

e C

lock

Pow

er (d

B)

320 ps/nm

0 ps/nm

640 ps/nm

-30

-20

-10

0

0 10 20 30 40 50

w/o filter

CD = 0 ps/nm

320 ps/nm

640 ps/nm

~20

dB

< 3 dB

f

T

Power Meter

Power Meter

Const.

Dest. Input signal

OSNR monitor Processing

Partial bit

•  Power ratio of two ports indicates OSNR.

•  This OSNR monitor is transparent to various data formats.

OPMs Using Delay-Line Interferometer

Channel Monitoring using Integrated Filters

)

21

41(

)21

43(

PPPP

noisesignal

noisesignalRatio

+

+=

OSNR: Signal has coherent interference, not noise

CD & PMD

Tones affected differently by CD & PMD Y. Lize, et. al., PTL’ 07 and JLT’ 08

11/7/11 34

PMD Monitoring of Phase-Modulated Data Using Interfermetric Filter

  The two outputs of the PBS represent the constructive and destructive filters of a standard Mach-Zehnder delay-line interfometer (FSR = 1/∆τ).   At the destructive port, the monitored RF power will change with the DGD- generated interferometric filter response.

11/7/11 35

Experimental Results

•  The RF power measured at 170 MHz increases by ~ 20 dB in the presence of 0 to 100 ps of DGD. •  Chromatic dispersion-insensitive measurements to be within + 1 dB.

DGD (ps) 0 20 40 60 80 100

RF

Pow

er (d

Bm

)

-40

-45

-50

-55

-60

-65

-70 R

F Po

wer

(dB

m)

-40

-45

-50

-55

-60

-65

-70

Chromatic Dispersion (ps/nm) 0 100 200 300 400 500 600 700

10-Gb/s NRZ-DPSK

20-Gb/s NRZ-DQPSK

J.-Y. Yang et. al., PTL, 2008

Significance of Higher-order PMD

Ref: M. Karlsson, et al., Optics Letters, 1999; H. Kogelnik, et al., JLT 2003.

“It is sometimes stated that once the signal bandwidth is large enough for second-order PMD to be important, then “all other higher order terms become important too.” If this were strictly true, then higher order PMD compensation would be a hopeless task … there is a need for closer examination of these bandwidth limitations.”

Δω

AC

F (p

s2) ΔωPSP

-H. Kogelnik, et al., JLT 2003

PSP Bandwidth

Autocorrelation Function of PMD Vector Higher-orders become important if signal BW

> ΔωPSP

Theory

Measurement

Fiber type Old fiber

PMD = 0.5 ps/km1/2

New fiber PMD = 0.1 ps/km1/2

Future fiber PMD = 0.05 ps/km1/2

1 Tb/s Transmission Limit due to PMD

40 m

1 km

4 km

Combined Effects of PMD and PDL

Polarization Mode Dispersion

Polarization Dependent Loss (PDL)

Optical Components (PDL=? dB)

Δτ�

Different Attenuation PSP1 ⊥ PSP2

Differential Group Delay PSP1 ⊥ PSP2

Fiber with high PMD

PSP1

PSP2

PSP 1

PSP2

PSP1

PSP2

PSP1 ⊥ PSP2

PDL: Frequency-dependent

attenuation PMD:

Enhanced time spreading

B. Huttner, et al., JSTQE, 2000 L.-S. Yan, et al., PTL, 2003

Combined Effects of PMD and PDL

Probability density function of 15 PMD sections

Without PDL With 15 PDL sections (each: 0.2 dB)

L.-S. Yan et. al., JLT 2004

Outline

1. Overarching Perspective

2. Optical Performance Monitoring

- optically-assisted techniques - receiver based techniques

Coherent Detection

|ES(t)  +  ELO|2  =  |  ES(t)  |2  +  |  ELO|2  +  ES(t)  ELO  cos(  [wS  –  wLO]t  +  [φS(t)  -­‐  φLO]  )  

All linear distortions (Dispersion, PMD, PDL) can theoretically be fully compensated. Nonlinear

distortion can be partially compensated

Coherently Received

Electrical Signal ~

Electric Field Vector of

Optical Signal Linear

System

Signal Amplitude Signal Phase

90o Hybrid

  Limited to receivers.

Motivation Asynchronous Sampling (by MDI)

(Asynchronous sampling)

Clean Noise CD

PMD Crosstalk All

Input Data

Delay

Sampling

On-Off-Keying Data

Unique impairment pattern → multiple impairments monitoring

Router ONE

ONE

Router

Router

Optical Network

End Customer Re-route or feed back

information to control the ONE

Trained receivers to automatically identify

impairments

ONE

Send error signals

Fiber link with various impairments

Router

Self-Managed Optical Networks

  Monitored information can be sent to the network controller and optical network elements to rapidly reroute the data information

X. Wu et al, J. Lightwave Technol. 27 (16), 2009.

Concept - ANNs Trained w/ Eye Diagram Parameters

  It is obvious that different impairment combinations produce distinct features in the eye diagrams

 The input parameters for training are derived from eye diagrams   Q-factor, eye-closure, jitter, and crossing amplitude

 The controlled impairments are used as outputs for training

Tx

Rx

OSNR = 36 dB CD = 0 DGD = 0

OSNR = 28 dB CD = 0 DGD = 0

OSNR = 20 dB CD = 0 DGD = 0

OSNR = 28 dB CD = 60 ps/nm DGD = 0

OSNR = 28 dB CD = 0 DGD = 10 ps

OSNR = 28 dB CD = 60 ps/nm DGD = 10 ps Fi

ber L

ink …

X. Wu et al, J. Lightwave Technol. 27 (16), 2009.

Artificial Neural Networks

Advantages of ANN Approach   Efficient identification and isolation of multiple impairments   Enhanced monitoring range and sensitivity   Simple and fast processing of the monitored information   Format transparent

X. Wu et al, J. Lightwave Technol. 27 (16), 2009.

Crossing Amp.

OSNR

CD DGD

3-Layer ANN Model 12 Hidden Neurons 64 Testing Samples

Q-factor Closure

Jitter

OSNR

CD DGD

3-Layer ANN Model 12 Hidden Neurons Conjugant Gradient

Training 125 Samples

Q-factor Closure

Jitter Crossing Amp.

40-Gb/s RZ-OOK testing results

40-Gb/s RZ-DPSK testing results

Training Errors for OOK and DPSK Systems

Block Diagrams for ANN Training and Testing

X. Wu, ECOC 2008

OSNR/CD/PMD Identifications using ANNs

OSNR=36, CD=0, DGD=0

OSNR = 16, CD=0, DGD = 0

OSNR = 36, CD=60, DGD = 0

OSNR = 36, CD=0, DGD = 10

OSNR = 20, CD=45, DGD = 7.5

OSNR = 16, CD=60, DGD = 10

Concept - ANNs Trained w/ Delay-Tap Plot Parameters

  It is obvious that different impairment combinations produce distinct features in the delay-tap plots

X. Wu et al, ECOC 2009, paper P3.04.