Dr. Bernd Nebendahl - Keysight · 2010-03-01 · Dr. Bernd Nebendahl Research and Development Digital & Photonic Test Agilent Technologies [email protected] February 2010

© Agilent Technologies, Inc. 2010

Dr. Bernd Nebendahl

Research and Development

Digital & Photonic Test

Agilent Technologies

[email protected]

February 2010

Metrology of Advanced Optical Modulation

Formats for 40/100G and beyond

Page 1 of 74


Part I

• Trends in 40/100 GbE

• Introduction to Advanced Optical

Modulation

– IQ Modulation

– Polarization Division Multiplexing

– Orthogonal Frequency Domain Multiplexing

• Receiver Technologies

– Measurement Principles

– Frequency or Time Domain

– Delay Line Interferometer or Coherent Receiver

– Real Time or Equivalent Time Sampling

Part II

• Signal Processing for Coherent

Receivers

– Carrier Phase Recovery

– Polarization Demultiplexing

• Quality Rating

– Error Vector Magnitude, Phase Error, Magnitude

Error, Quadrature Error, Gain Imbalance

– Bit Error Rate and Error Vector Magnitude

• Conclusion



February 2010Page 2 of 74


February 2010

4 Fibers, 1 wavelength

with 10, 20 or 25Gbit/s,

on-off modulation

50 GHz 50 GHz

1 Fiber, 192 wavelength with

40/100Gbit/s in each ITU-T channel,

advanced modulation

1 Fiber 4 wavelength

with 10, 20 or 25Gbit/s,

on-off modulation

Focus of this

Tutorial



Page 3 of 74


• Keep existing 50GHz spacing to improve return on investment

• Maximize ROI by increasing the capacity of today„s 10G systems by 4x,

i.e. from 80x10G to 80x40G and 80x100G in the C-band

• Necessitates positioning 40G/100G channels on 50GHz frequency grid

while also allowing mixing with 10G

• Transparent reach of at least 1,000km

• System design

• No change in line system design with the introduction of 40G and 100G

on a 10G system

• High tolerance to in-line optical filtering (ROADMs)

• High tolerance to chromatic dispersion

• High tolerance to polarization-mode dispersion (PMD)

February 2010



Page 4 of 74


Transmission of 100-Gb/s binary (duo binary) optical signals is

extremely challenging• Modulator bandwidth not wide enough / drive voltage too high

• Electrical amplifier bandwidth and output voltage not sufficient

• High-performance integrated photo-receivers not available

• Bandwidth not compatible with 100-GHz (50-GHz) channel spacing

• Large transmission penalties expected from CD and PMD

New modulation formats for 100G long haul transmission• Multi-level coding: QPSK, QAM, M-ary ASK, …

• Relaxed bandwidth/voltage requirements for modulator

• Compatible with 100-GHz (50-GHz) channel spacing

• Polarization multiplexing instead of single polarization

• Compatible with 50-GHz channel spacing

• Coherent instead of direct detection

• Electronic dispersion mitigation (PMD and CD)

February 2010



Page 5 of 74


February 2010



Page 6 of 74


1. Standardize 100G Short-Range/ Client Interfaces

2. Develop T&M Equipment for Short-Range Transmission

3. Identify (Standardize?) Modulation Format(s) for Long-Haul

4. T&M Equipment for Long-Haul Transmission

Constellation analyzers for advanced M-ary formats, etc.

5. T&M Equipment for Long-Haul Transport

In-band OSNR, CD and PMD monitoring

6.Enhanced PMD and CD Compensation Techniques

February 2010

Focus

of this

Tutorial



Page 7 of 74


Part I



Modulation

– IQ Modulation








Part II


Receivers



• Quality Rating




• Conclusion





Every carrier signal can be described with two parameters

• Amplitude and Phase or

• In-phase and Quadrature

Both parameters can be modulated to carry information

I (in-phase or real part)

Q (quadrature, imaginary part)

Phase φ

(I,Q)

Q value

I value

The position of the vector end point is called a constellation point and

therefore the diagram is called a constellation diagram

February 2010



Page 9 of 74


Vectors defined by amplitude and phase are separated

by the transition of the transmission clock



I

Q

Transmitted signal

Data clock

Sample points

Transition between

sampling points

Switch to

next vector

10

11

February 2010



Page 10 of 74


Modulating amplitude and/or phase does not necessarily

increase spectral efficiency compared to OOK

I

0 1

Q

I

0 1

Q

Traditional On-Off (OOK) format displayed with

two vectors. Information is coded in 2 symbols 0

and 1 only in the amplitude

In this BPSK format, information is coded only in

the phase instead of amplitude. Number of

symbols is still 2 !

I

0 1

Q

In this artificial format amplitude and phase is

modulated but still only 2 symbols are available.

February 2010



Page 11 of 74


The data stream can be coded into a 4 symbol alphabet

like { A, B, C, D, ….}

We can code also in the following way:

00 a sin(ωt+45)

01 a sin(ωt+135)

10 a sin(ωt+225)

11 a sin(ωt+315)

0 0 1 0 1 1 1 0 0 1 0 1 0 0 1 0 1 1 1 1 0 0 1 0

A DB B C C A B D D A B

Original

data stream

Possible

Symbol alphabet

Increasing the number of symbols that are coded in an alphabet

increases the number of bits that are transmitted in one clock

February 2010



Page 12 of 74


a sin(ωt+45) 00

a sin(ωt+135) 10

a sin(ωt+225) 01

a sin(ωt+315) 11

I (in phase)

Q (quadrature)

0010

01 11

4 symbol alphabet codes 2 bits

in one transmission clock cycle

As a result one vector position in the polar plane codes 2 bits,

which increases the spectral efficiency by a factor of 2

February 2010



Page 13 of 74


February 2010

I

Q

Signal

1

1 1

1

0

00

0

11 00 01 10 QPSK constellation map

11

00

01

10

I

Q

Laser

/2

1 0 0 1

DemuxBinary Bit stream QPSK Signal

1 0 1 0

1 1 0 0 0 1 1 0



Page 14 of 74


0 0 1 0 1 1 1 0

Data rate: [bit/s]

A B D B

Symbol rate: [Symbols transmitted/s ]

Symbols coded and transmitted

as vectors 4 symbols

Original data 8 bits

Symbol rate or baud rate describes the symbol transmission

clock rate not the transmission data clock rate

Symbol rate (baud rate) is equal to or smaller than data rate

February 2010



Page 15 of 74


I

Q

QAM 16

0000 0100 0011 1000

0001 0101 1011 1001

0011 0111 1111 1011

0010 0110 1111 1010

16 symbol alphabet coding 4 bits

Symbol (Baud) rate is 4 times lower than clock rate

Savings in speed translates to savings in power !

February 2010



Page 16 of 74


February 2010

frequency

y-polarization

x-polarization

• the optical wave can be split into two orthogonal polarizations

• each polarization can carry an independent signal and can be viewed

as a virtual transmission channel within the fiber

• these signals can be seperated using polarization diversity receivers

PDM increases the spectral efficiency by a factor of 2



Page 17 of 74


February 2010

112 Gb/s Optical spectra50 GHz ITU channel

PDM NRZ-QPSK

NRZ-QPSK

NRZ-OOK

-200G -100G 0G 100G 200G

Offset from Carrier

Factor of 2 from OOK to QPSK

Factor of 2 from SinglePol to PDM

Using advanced modulation and PDM allows to transmit

more than 100 Gb/s in a 50 GHz wide ITU channel!



Page 18 of 74


• widely used in wireless and electrical wireline

• uses more than one coherent subcarrier for the transmission at lower symbol rates

• can be generated optical (typically low number of subcarriers) or electrically (high

number of subcarriers are possible, but requires fast DAC)

• Pilot carriers can be used to estimate the channel parameters

• Receiver synchronization is simplified using guard intervalls

• typically very sensitive to nonlinearities (fiber effects?)

• signals typically have a high peak to valley ratio

• does not necessarilly increase spectral efficiency but can improve robustness

against certain impairments

February 2010

OFDM might be an interesting opportunity for optical

transmission but is currently still in research



Page 19 of 74


Part I



Modulation

– IQ Modulation








Part II


Receivers



• Quality Rating




• Conclusion





Until now

OOK (RZ or NRZ), maybe multilevel

information encoded in power

direct detection (electrical or

optical) and real-time or equivalent

time sampling are possible

In future

QPSK, QAM, ....

information encoded in power and

phase

since phase is not an absolute

quantity, a phase reference has to

be introduced

February 2010

Photodiode IphotoS

Photodiode Iphoto

S

R



Page 21 of 74


Local oscillator

• Coherent detection

• Time domain or frequency domain

• Intradyne or heterodyne

• Real- or equivalent time sampling

Self referencing

• Delay line interferometer

• Time domain

• Real- or equivalent time sampling

February 2010

Delay Photodiode

Photodiode

I1

I2

I1-I2

S



Page 22 of 74


February 2010

Time domain or frequency domain

Time domain:

Coherent or delay line interferometer

Coherent:

Real time or equivalent time

sampling

Equivalent time sampling:

Optical or Electrical sampling

Delay line interferometer:

Real time or Equivalent time

sampling

Equivalent time sampling:

Optical or Electrical sampling

Frequency Domain:

Static or swept LO

Swept LO:

narrowband detection

Static LO:

wideband detection (ESA)

Covered in this Tutorial



Page 23 of 74


February 2010



Page 24 of 74


Con

• Only works for repetitive signal

• Symbol clock required (either from CDR or from TX)

• Pattern length limited by linewidth/jitter of TX and LO

• polarization multiplexed signals require an optical polarization controller and a

stable input SOP

• Only phase differences are measured. Reconstruction of phase requires

integration. Might lead to large errors

Pro

• Virtually „infinite“ bandwidth (only limited by tuning range of LO, can be many THz)

February 2010



Page 25 of 74


Con

• Delay needs to be adjusted to actual symbol rate

• polarization multiplexed signals require an optical polarization controller and a

stable input SOP

• Reconstruction of transistions difficult

• Lower sensitivity compared to heterodyne/intradyne detection

• Linear distortions cannot be removed by signal processing

Pro

• Only minor impact from phase noise of TX

• Using equivalent time sampling is possible higher bandwith than real time

sampling

February 2010



Page 26 of 74


February 2010

IQ Demodulator

Phase shifter 90°

Photodiode

Photodiode

Photodiode

Photodiode

I1

I2

Q1

Q2

Signal

Local Oscillator



Page 27 of 74


February 2010

IQ Demodulator, x-Polarization

Phase shifter 90°

Photodiode

Photodiode

Photodiode

Photodiode

Photodiode

Photodiode

Photodiode

Photodiode

IQ Demodulator, y-PolarizationPolarization

Splitter

Local Oscillator Phase shifter 90°

I1

I2

Q1

Q2

I3

I4

Q3

Q4

Signal



Page 28 of 74


Signal

PBS 50/50

90° OpticalHybrid

90° OpticalHybrid

LO atti?

BalancedReceiver

BalancedReceiver

BalancedReceiver

BalancedReceiver

ADC ADC ADC ADC

Clock recovery & re-timing

Equalizer (remove CD and PMD)

Carrier recovery

Slicer & decoder

I/Q plot

Spectrum

Time Series

...

February 2010

• Polarization diversity coherent

receiver generates 4 electrical outputs

• AD convertes sample the 4 channels

• Postprocessing recovers the carrier

• Postprocessing demultiplexes the

polarization multiplexed signals

• Recovered signals are demodulated

and displayed in various formats



Page 29 of 74


Optical sampling uses nonlinear effects to mix the signal and the probe pulse

Coherent reception and optical sampling typically involves a local oscillator probe

pulse and to generate the pulsed I and Q signals

These signals are electrically detected at lower speed. To be able to recover the

original signal, it needs to

• be repetitive (i.e. real datastreams cannot be measured, BER cannot be

measured)

• contain the symbol clock and pattern clock in the raw signals

• have a carrier phase noise small enough for tracking with equivalent time

sampling

February 2010



Page 30 of 74


Con

• Only works for repetitive signal

• Pattern synchronization requires measuring many cycles of the pattern

• Higher level QAM formats are more difficult to synchronize to

• Carrier phase recovery requires tighter limits for transmitter laser phase noise than

in real systems

• Postprocessing can only be applied after the signal has been recoverd

Pro

• Higher detection bandwidth compared to real time sampling but not as high as for

frequency domain based methods with swept LO

February 2010



Page 31 of 74


February 2010

Con

• Bandwidth limited by analog bandwith of ADC

Pro

• Works for any signal (repetitive and true random data signals)

• Does not require better phase noise than RX in the network

• Less complex processing to yield results

• Can detect bit errors due to improper TX realisation

• Signal processing can be identical to final RX

• Polarization demultiplexing in SW

• Real time sampling with 30 GHz electrical bandwidth allow to cover 60 GHz

(more than one channel in the 50 GHz ITU grid)

• Detection of OFDM signals can be done using proper signal processing



Page 32 of 74


Currently the modulation formats are targeted to fit into the 50GHz ITU grid

• effective signal bandwidth is less (depending on the number of add-drop

multiplexers ~40 GHz)

Using modulation formats that need more bandwidth means:

• Current network infrastructrure must be replaced or upgraded

• Chromatic dispersion and polarization mode dispersion become more important

• TX and RX hardware require ultra-high speed electronics, which is not available

today

• increase of power dissipation in linecards

• As soon as higher speed ADCs are available, real-time scopes will be faster as

well

For the 50 GHz ITU channel spacing there is no real need for bandwidth in

excess of 20 GHz

February 2010



Page 33 of 74


This is the end of Part I of our Tutorial!

Please make sure to continue with Part II and complete the

evaluation form at the end of the presentation.

You‟ll be entered into a drawing with a chance to win a $75

Amazon.com gift card!

February 2010



Page 34 of 74


Part I



Modulation

– IQ Modulation








Part II


Receivers



• Quality Rating




• Conclusion





• The frequency difference between the transmitter laser and the local oscillator

leads to a „rotating“ constellation

February 2010

I

Q

0010

01 11

I

Q

0010

01 11

t t+Tsymbol



must not change faster than

/4 per symbol time

• Frequency offset must be

smaller than 1/8 of the

symbol clock for QPSK

t

linearphase drift

t

Page 36 of 74


• To be able to track the phase, the signal must be sampled at times with predictable phase values (i.e.

at the symbol times)

• For a bandwidth limited signal the sampling rate of the phase with predictable values is smaller than the

actual sampling rate

• To be able to recover the phase, the carrier phase noise and offset must be within very tight limits

(tighter than required in a real transmission system)

February 2010



t

linearphase drift

t

Transmitter lasers are not likely

to meet these tighter

specifications since they are not

required in real line cards that

use real time acquisition

Page 37 of 74


sample #

Ca

rrie

r ph

ase [

rad

]

Symbols are artificially

narrow. The phase tracking

reduces angular width of

symbols.

February 2010



Page 38 of 74


Nice constellation with

“round” symbols.

High

Medium

sample #

Carr

ier

phase [ra

d]

February 2010



Page 39 of 74


Medium

Low

Possible loss

of track.

sample #

Carr

ier

phase [ra

d]

February 2010

Constellation is

effected by the phase

noise.



Page 40 of 74


February 2010

y

x

y

xy

x

y

x

y

x

• standard single mode fiber (SMF) does not preserve the state of

polarization (SOP)

• polarization maintaining fiber (PMF) preserves the SOP but is not

deployed for data transmission

the effect of propagation through fiber is described using

the Jones matrix



Page 41 of 74


February 2010

x-polarization

(Transmitter)

y-polarization

(Transmitter)

x-polarization

(Receiver)

y-polarization

(Receiver)

Ideal transmission channel

sx and sy are time dependent complex numbers describing the time dependet field of

the optical wave (not only the optical power!)

If transmitter and receiver are aligned the received signals are identical to the sent

signals, there is no crosstalk between the channels

rx 1 0 sx

ry 0 1 sy= .

Jones matrix of ideal channel



Page 42 of 74


February 2010

x-polarization

(Transmitter)

y-polarization

(Transmitter)

x-polarization

(Receiver)

y-polarization

(Receiver)

Real transmission channel (SMF)

There is crosstalk between the polarization channels. The receiver „sees“ a linear

combination of the „horizontal“ and „vertical“ signals sx and sy.

Without loss, PDL and PMD the channel matrix has only one independent parameter:

angle of rotation of reference frames (single mode fiber cannot preserve the SOP)

rx cos( ) sin( ) sx

ry -sin( ) cos( ) sy= .

Jones matrix



Page 43 of 74


February 2010

x-polarization

(Transmitter)

y-polarization

(Transmitter)

x-polarization

(Receiver)

y-polarization

(Receiver)

In the most general form the Jones matrix is a complex valued 2x2 matrix

(with 8 independent real parameters)

rx Jxx Jxy sx

ry Jyx Jyy sy= .

Jones matrix

Transmission with Loss, PDL and PMD

J. C. Geyer, et al., “Channel Parameter Estimation for Polarization Diverse

Coherent Receivers,” PTL, Vol. 20, No. 10, May 15, 2008, p. 776



Page 44 of 74


February 2010

To calculate the original signal sx and sy from measurement we need:

1. An estimation of the Jones matrix of the transmission channel.

2. A measurement of the rx and ry (the field of the optical wave including phase)

sx J-1xx J-1

xy rx

sy J-1yx J-1

yy ry= .

Inverse Jones matrix J-1

J-1xx J-1

xy 1 Jyy -Jxy

J-1yx J-1

yy JxxJyy-JxyJyx -Jyx Jxx=

J. C. Geyer, et al., “Channel Parameter Estimation for Polarization Diverse

Coherent Receivers,” PTL, Vol. 20, No. 10, May 15, 2008, p. 776

Coherent receivers detect the field and not only the power

of an optical wave



Page 45 of 74


February 2010

• Training symbols allow

efficient estimation of the

channel

• Detection needs to know the

training sequence and detect it

Not feasable for T&M gear

• “blind” estimation does not

require prior knowledge about

signal (except the modulation

format)

• Might not be able to recover

the full information

Better choice for T&M

There is no ultimate solution for the problem!



Page 46 of 74


February 2010

Phase differences between x- and y-signals:

• 00/00, 10/10, 11/11, 01/01 0 (linear 45°)

• 00/10, 10/11, 11/01, 01/00 /2 (right circular)

• 00/11, 10/01, 11/00, 01/10 (linear -45°)

• 00/01, 10/00, 11/10, 01/11 - /2 (left circular)

These 4 SOPs define a plane in the Stokes space!

Normal can be used to estimate Jones matrix

PDM vs. Single Polarization

Other formats?

Time dependent Jones matrix ?

x y

Real measurements!



Page 47 of 74


February 2010

In the Stokes space, all points of PDM 64QAM are within the

boundaries of a lens-like object.

Still defines a plane (method does not depend on format)

1.0 0.5 0.5 1.0

1.0

0.5

0.5

1.0

Example: 64QAM

Stokes SpaceConstellation



Page 48 of 74


February 2010

Constellation

map

rx

ry

J-1xx J-1

xy

J-1yx J-1

yy

sx

sy

Calculate vector to

nearest symbol

Update

estimation

• Needs knowledge of constellation

• Update step can be taylored

• Additional restrictions for J-1 are

possible

• Might not converge for bad initial

guess

• Locking speed?

Decision!

Iterate per

symbol



Page 49 of 74


February 2010

J-1xx and J-1

yy J-1yx and J-1

xy

Constellation

SOP

Inverse

Jones

Matrix

Before and after correction



Page 50 of 74


February 2010

y-Polarization

x-Polarization

IQ-Plot Symbols/Errors Spectrum



Page 51 of 74


February 2010

y-Polarization

x-Polarization




Page 52 of 74


February 2010

y-Polarization

x-Polarization




Page 53 of 74


Part I



Modulation

– IQ Modulation








Part II


Receivers



• Quality Rating




• Conclusion





How can these impairments be measured?

How do they distort the constellation diagram?

π/2

Bias Voltages

Gain imbalance

π/2 shift error

Line width

Phase noise

Path delay between I and Q

Offset in symbol rate between receiver and transmitter

February 2010



Page 55 of 74


Constellation diagram Eye diagram of I and Q

path with equal scaling

Quantitative Analysis

Gain imbalance is

quantified with: 2.938 dB

Gain imbalance in the IQ paths of modulator lead to a “rectangular” constellation

Gain imbalance:

= 20 * log10

I

Q

February 2010



Page 56 of 74


• Eye diagram gives no

indication about a

possible root cause

• Spectrum show‟s also

the typical QPSK

pattern

February 2010



Page 57 of 74


70.23 deg

instead of 90 deg !

Quadrature error describes the deviation from orthogonality of the I and Q

modulation and leads to a rhombic constellation shape

February 2010



Page 58 of 74


Ideal

(square)

Measured

(square)

IQ offset: -31.043 dB

given as ratio between

signal and offset

I-Q offset is caused by a DC offset in the I and/or Q path.

February 2010



Page 59 of 74


DFB source External Cavity Laser Typical line width

DFB 2-3 MHz

ECL 0.1 MHz

How does this difference influence the constellation diagram ?

February 2010



Page 60 of 74


I Eye

Q Eye

Instruments based on optical

power detection for I and Q path

High resolution spectrum

gives also no indication

of the error source

February 2010



Page 61 of 74


Constellation left

indicates phase noise

Analyzing the phase error

show low frequency phase

noise

Constellation diagram with DFB

transmitter laser and low phase

tracking bandwidth

Constellation and phase error

with external cavity laser or with DFB

laser and higher tracking bandwidth

February 2010



Page 62 of 74




Phase φ

Ideal

constellation

point

Measured

constellation

point

Error vector

The error vector magnitude (EVM) represents the Euclidian distance between the

ideal symbol coordinate and the actual recorded symbol

Note:

EVM is part of the 802.11

WLAN standard

February 2010



Page 63 of 74


February 2010

Demodulate

test

signal

Calculate the

reference

Reconstruct

I-Q signal

calculate

complex

Error Vector

bits

Reference

trace

In-phase ref

Quadrature ref

Additional error analysis tools



Page 64 of 74


V shape

Structured constellation points

Eye diagram gives

no clear indication,

but indicates a

problem

February 2010



Page 65 of 74


π/2Symbol clock

Error vector magnitude

Small errors in symbol rate cause a typical V shape

in EVM versus detected symbols

de

co

de

Recovered symbol clock phase using the

user input of the clock frequency

February 2010



Page 66 of 74


t

t +Δt where Δt < symbol length

Small delay errors between I and Q path might look like a bandwidth

limited modulator driver, also look at the IQ eye diagrams in this case

February 2010



Page 67 of 74


Why is the transition missing?

February 2010



Page 68 of 74


February 2010

t

t + 4 bits delay

0010

01 11

Choosing the wrong PRBS length and delay between I and Q

might lead to missing transitions



Page 69 of 74


• Assume a gaussion noise distribution

• Calculate the probability to detect the wrong

symbol using an ideal decision

BER estimation

Note:

Any distortion (Gain Imbalance, Quadrature Error,

...) of the constellation will increase the EVM

BER estimation is an upper limit, with

additional distortion the same EVM will lead

to a lower BER!

Specifying the EVM of the transmitter

ensures that a certain BER can be achieved

February 2010



Page 70 of 74


February 2010

5 10 15 20 25 30

1E-12

1E-11

1E-10

1E-9

1E-8

1E-7

1E-6

1E-5

1E-4

1E-3

0.01

0.1

QPSK

8PSK

8PSK GRAY

16QAM

16QAM GRAY

BE

R

EVM / %

FEC limit



Page 71 of 74


Part I



Modulation

– IQ Modulation








Part II


Receivers



• Quality Rating




• Conclusion





• Coherent reception with real time sampling offers the best fit for

measuring advanced optical modulation formats targeted for 50 GHz

channels

• Carrier phase recovery and polarization demultiplexing are the

challenges for coherent reception

• Quality rating of transmitters for advanced modulation formats is more

challenging than for on-off keying transmitters. This is even more true

for higher order modulation formats

• Constellation diagram analysis along with additional tools gives

significantly more information about the signal quality that is emitted

than observing just an I and Q eye and often leads directly to the

error source

• Specifying the signal quality in terms of EVM, Quadrature Error, Gain

Imbalance along with other numbers will allow to predict the limit for

system BER independently of the receiver

February 2010



Page 73 of 74


Thank you for attending our Tutorial!

Please spend one minute to complete our evaluation form found on

the left side of your display, or by going to

www.agilent.com/find/tutorial-evalform.

You‟ll be entered into a drawing with a chance to win a $75

Amazon.com gift card!

February 2010



Page 74 of 74

http://www.agilent.com/find/tutorial-evalform



Documents

Dr. Bernd Nebendahl - Keysight · 2010-03-01 · Dr. Bernd Nebendahl Research and Development Digital & Photonic Test Agilent Technologies [email protected] February 2010