optical local area networking - mit computer science and

www-g.eng.cam.ac.uk/photonic_commsPhotonic Communications Research

Optical Local Area Networking

Richard Penty and Ian White

Cambridge University Engineering DepartmentTrumpington Street, Cambridge, CB2 1PZ, UK

Tel: +44 1223 767029, Fax: +44 1223 767032, e-mail:[email protected]

Acknowledgements:• Many others in the Photonic Communications group• Intel Research (Derek McAuley and Madeleine Glick) through SOAPS project• Bookham Technology for providing tunable lasers


Outline• Introduction

• The WDM Revolution• WDM for Datacommunications• Optical LANs for Computing

- Recent advances in low cost photonics

• Background Work to Date• Athermal WDM Lasers• SOA Based Optical Add/drop Switch • Initial Results of a LAN Network Node Demonstrator

• Conclusions and Future work


WDM Transmission Technology Evolution

Post 2001


DWDM for Transmission• ITU-T G.692 DWDM specification defines grid of 80

wavelengths with 50GHz spacing around 1550nm.• Commercial systems with 2.5Gb/s and 10Gb/s per channel

(40Gb/s in laboratory)• Precise wavelength grid requires control of channel spectral

content- wavelength locking to few GHz (mK and mA range control)- high SMSR (~40dB)- external modulation to meet required spectral efficiency- results in high cost per wavelength

• Current demand doesn’t require many wavelength per fibre• Strongest trend is towards lower cost• Solution - remove cost by relaxing tolerances on wavelength


Advantages of Coarse Wavelength Spacing

• No need for precise temperature or wavelength control- TEC one of the major costs of a DWDM transceiver

• Higher DFB laser yield possible- large intra and inter wafer wavelength variations allowed- significantly relaxed SMSR requirement

• Low cost MUX/DEMUX components possible- wider wavelength spacing allows smaller mux/demux

components- grating or interference filter components preferred- mux/demux technologies allow MMF systems

• CWDM systems typically are unamplified- entire spectrum of fibre can be used if required


CWDM Specifications

• Emerging standards have specified uncooled CWDM channel spacing of around 20nm

• Wide channel spacing necessary due to wavelength drift of DFB laser diodes

- typically around 0.1nm/°C i.e. 10nm over 100°C operating temperature range

• Thus typical channel spacing are specified at ~20nm• ITU-T G694.2 specifies 18 channels from 1270-1610nm


“Athermal WDM” - Motivation• Would be very attractive to have a “CWDM” system that isn’t quite so

coarse– Few (4x8) wavelengths inside amplifier spectrum (either EDFA or SOA)

would transform applications– Wavelength spacing of 1-4nm probably acceptable if cost advantages of

CWDM can be retained.• Constant wavelength operation can be achieved by;

– The use of high current peltier effect cooler in conjunction with a tunablephotodetector

– External fibre Bragg grating wavelength locker – Etc.

• Allows the number of channels in WDM systems to be increased but;– Expensive– Power consumption is high– Often complicated, involving many components

• Can we develop a simple, low cost, low power consumption technique to achieve athermal operation?


Athermal Laser Operation

• No need for high current peltier effect cooler or expensive optics

• Operating wavelength can be adjusted by the user to correct wavelength registration issues

• Uses a tunable laser, for an active approach to constant wavelength operation

• Current supplied to tuning section depends upon the temperature of the device submount

15481550155215541556155815601562

0 20 40 60 80 100

Tuning Section Current (mA)

Wav

elen

gth

(nm

)

T1

T2

Constant wavelength operation?


Experimental Details

• Tunable laser is three section device– Consists of grating, phase and gain sections

• Computer control allows currents to be adjusted depending upon:– Temperature– Front and back facet power measurement– Forward voltage across device at each contact

Computer Control

Current Source

Voltage MeasurementV VI I

Temperature Sensor

Optical Power Meter

Optical Power Meter

Optical Spectrum Analyser


Wavelength Control

• Uncontrolled wavelength drifts from 1556 to 1563 nm between 20 and 70°C

• Under control the wavelength remains at 1557.07 ± 0.30nm • Allows at least an order of magnitude reduction in CWDM

channel spacing

1555155615571558155915601561156215631564

20 30 40 50 60 70

Temperature (°C)

Wav

elen

gth

(nm

) Uncontrolled

Controlled


• SMSR typically > 40 dB but reduces slightly at higher temperatures – always > 35dB

• Ripple can be eradicated with more sophisticated control techniques – also further reduction of wavelength variation (subject to patent application)

0102030405060

20 30 40 50 60 70Temperature (C)

SMSR

(dB

)

Single-Mode Performance


Suppression of Mode-Hopping• Initial results show that

DBR filter has been athermalised

• Modes continue to drift with temperature however

• Need keep mode aligned with filter to eradicate modehops

⇒Use of phase shifting section

DBR Filter

Longitudinal Modes

Gain Spectrum


Optimised Device Structure

• Longer phase sections are necessary to allow wavelength control with continuous changes in control currents

• Novel laser designs exhibit contours of controlled mode-hop free output extending up to 70°C

• Designs currently being fabricated

0102030405060

0 10 20 30 40 50 60 70Grating Current (mA)

Phas

e C

urre

nt (m

A)

40°C20°C 60°C 70°C0°C

1547

1548

1549

1550

0 10 20 30 40 50 60 70Temperature (°C)

Wav

elen

gth

(nm

)


• How do you design an efficient optical packet switched data network?- using currently available technology

• Network aspects will stay electronic for near future

– Header processing– Routing– Security– QoS– Classification

• What can we gain using photonics (WDM, optical switch)?– Capacity – Power consumption– Noise immunity– Latency– Cost (eg leveraging off low cost datacomm transceivers)

Issues for Optical Networking Using Athermal WDM


Wavelength Striped Semisynchronous LAN

Controllogic

Payload

Header

We need- nanosecond optical switch- WDM channel spacing ~nm

TERMINAL

HUB


SOAs as Optical Gates

Advantages• Optical gain (possibly up to ~30dB)• Large extinction ratio (~50dB)• Fast gating time ( ~ns)

Advantages• Optical gain (possibly up to ~30dB)• Large extinction ratio (~50dB)• Fast gating time ( ~ns)

Disadvantages• Limited cascadability because of ASE• Gain saturation can lead to ER degradation• Limited input power dynamic range -distortion

Disadvantages• Limited cascadability because of ASE• Gain saturation can lead to ER degradation• Limited input power dynamic range -distortion

λ / µmG

ain

/ los

s

gain

Input wavelength

saturated loss

unsaturated loss


Optical Crosspoint Switch• 2x2 and 4x4 arrays constructed -scalable to 32 x 32 and above

• 8 individually controllable SOAs

• 850µm x 850µm chip

• 5 quantum well InGaAsPstructure

• On chip switch gain >8dB

• Crosstalk < -50dB

• Switching time < 2ns

• Operating wavelength 1550nm, optical bandwidth >40nm

• Broadcast operation possible

2x2 Array


Add port

Outputport Add mode

Input port

Drop mode

Outputport Input port

Pass-through mode

Add,Drop & Pass Through


TERMINAL

HUB

Driver

Optical Add-Drop

Media access via blue control wavelength at 1310nm

High capacity “CWDM” data at 1550nm – 8 x 10Gb/s

1300nm

1300nm

FPGAsor pulse gens

15xxnm

15xxnm

ADD

DROP

HUB

HUB

TERMINAL

2x2 optical switch


Initial Test of Switch Node

• 8 x 10Gb/s data transmitted through add-drop node

• Routing data provided by 100Mb/s channel on separate wavelength

• Electronic control provided by Xilinx Spartan II FPGA

• FPGA also generates packet, including preamble, destination address, data and slot delimiters


Input Waveform “Drop” Packets

Packet Switching Results

“Through” Packets

• 70ns guard bands, limited by timing jitter from FPGA

• 30 byte preamble for clock acquisition

• High extinction packet routing demonstrated


BER Performance

• Error free operation demonstrated with low penalty

• Penalty dominated by clock acquisition/recovery rather than switch


Conclusions• Use of CWDM reduces cost by stripping out control etc

- At the expense of large (20nm) channel spacing- Not possible to have amplified (and hence switched) systems

• High capacity packet switched LANs require - Few nm channel spacing – but retaining cost advantages of CWDM- Nanosecond scale optical switching

• Athermal laser demonstrated- Wavelength varies by < ± 0.30nm up to 70°C grating or

interference filter components preferred• 2x2 add/drop SOA based switch demonstrated

- 2ns switch time with <-50dB crosstalk and 8dB gain• 80 Gb/s CWDM LAN router controlled by separate 100

Mb/s channel


Research Directions Within CII- Low cost technologies for achieving > 100 Gb/s transfer

rates within computer LANs- Study potential of multichannel techniques and optimum level of granularity

- Will WDM have as great and impact in Datacomms as it had in Telecomms?- Should packets always have headers?

- Study relative importance of bandwidth and latency- Study potential of overlay architectures for the long reach problem

- Study protocol independence- Study security issues?

optical local area networking - mit computer science and

Documents