Transcript Investigation of 11.1Gbit/s Direct-Detection OFDM Using QAM

OFDM-based optical interconnects
Robert Killey
Optical Networks Group, Electronic and Electrical Engineering,
University College London. [email protected]
Outline
•
•
•
•
•
•
Introduction to Optical Networks Group and UCL
Collaboration with Intel and CMU
Optical OFDM concept
Demonstration of 21.4 GS/s OFDM with real-time DSP
Future work towards 100 Gb/s OFDM
Planned 100 Gb/s OFDM experimental demonstrator
Robert Killey - Biography
Senior Lecturer at University College London, U.K.
PhD from the University of Oxford in 1998.
Research interests:
• nonlinear fiber effects in WDM transmission systems
• optical interconnects for computer networks
• wavelength-routed optical networks
• high speed digital signal processing.
Associate Editor of the IEEE/OSA Journal of Optical Communications
and Networks
Member of technical program committees of ECOC and OECC
conferences, and IEEE Photonics Society meetings.
Author of approx. 120 journal and conference papers.
University College London
•
UCL founded in 1826, oldest university in London
•
Consistently ranked within the top 5 universities in the UK (together with
Cambridge, Oxford, Imperial College London and the London School of
Economics)
•
Charles Kao (Nobel Prize winner 2010 for pioneering work in optical fiber
communications) was a PhD student in the Dept. Electronic Engineering
Optical Networks Group at UCL
Academics: Polina Bayvel, Robert Killey, Seb Savory, Benn Thomsen
Research Fellows: Yannis Benlachtar, Steve Desbruslais
Doctoral students: 10 students (including Rachid Bouziane), working on:
• Fibre transmission systems
• Optical interconnects
• Optical burst and packet switching - hardware and network modeling
• Plasmonics
Optical Networks Group research
Future network architectures
Recirculating fibre loop testbed
Ultra-low crosstalk free-space grating
wavelength router
Funding and collaborations
UK Engineering and Physical Sciences Research Council
• Fibre Nonlinearity and Dispersion Compensation using High-Speed Digital Signal
Processing, 2005-2008
• Putting the Quantum into Information Technology, 2003-2007
•Terabit WDM Optical Network Testbed (TWON) for Transmission and Optical
Networks Research, 2001-2005
EU Framework funding
• Broadband Optical Networks for Europe ‘BONE’, 2009-2011
• Integrated Project ‘NOBEL’, 2004-2005
• Network of Excellence ‘e-Photon/ONe’, 2004-2006
Industrial funding and collaborations: Agilent, Azea Networks, Bookham
(Caswell), BT, Ericsson, Huawei Technologies, Intel, JDSU (San Jose, CA),
Lucent Technologies – Bell Labs (Crawford Hill, NJ), Nortel Networks, Siemens,
Sierra Monolithics, T-Systems (Berlin), Xtera
Collaborative research with Intel since 2004,
and with CMU since 2008

Developed novel DSP-based transmitters for optical interconnects to
 Compensate for signal distortion
 Maximize data rate and spectral efficiency
 Assessed
the feasibility of optical transmitter designs employing real-
time DSP
Investigated the effects of hardware constraints on the performance
of such systems:
 Limited DSP resolution
 Latency
 Limited DAC resolution and bandwidth

Developed efficient algorithms for the generation of electronically
predistorted (EPD) signals and optical OFDM signals

Collaborative research with Intel and CMU
Intel-funded PhD studentship 2007-2010 (Ramanan Thiruneelakandan)
Three interns from UCL worked at Intel labs (Phil Watts, Robert
Waegemans, Rachid Bouziane)
Jointly published numerous papers in journals and at international
conferences.
Future collaborative research with Intel and CMU
Investigate the feasibility of 100 Gb/s per wavelength optical
interconnects, with the following goals:
• Use of directly modulated VCSELs with modulation bandwidth < 20 GHz
• Use of high-order modulation formats (QPSK, 16-QAM, 64-QAM)
• Error-free transmission over up to 300 m of multi-mode fiber
• Total link power consumption below 20 mW/Gbit/s
• Low cost – low complexity optics
• Transceiver electronics implemented using conventional CMOS technology
• Suitable for coarse wavelength division multiplexing to achieve interconnect
capacity > 1 Tbit/s per fiber
• Compatible with optically switched and hybrid networks for data center and
parallel computing applications
Orthogonal frequency division multiplexing (OFDM) is a
promising candidate to achieve these.
Introduction to OFDM
• Conventional optical communication systems use modulation (amplitude
or phase) of a single optical carrier wave
• suffers from distortion at high data rates
• requires complex optics to recover phase modulation
• Orthogonal frequency division multiplexing (OFDM) concept - divide the
bandwidth of the channel into many non-interfering (orthogonal) subchannels running at lower bitrates
• Reduces impact of fibre dispersion and filtering effects etc
• Allows increased spectral efficiency through high order formats, with
phase and amplitude modulation, e.g. quadrature amplitude modulation
(QAM), carrying more than one bit per symbol.
Introduction to OFDM
• OFDM is a multi-carrier modulation scheme, in which adjacent sub-channels
are separated in frequency by 1/T, where T is the symbol period.
• All sub-channels are orthogonal to each other, and can be detected with low
inter-channel interference
Spectrum of an OFDM subcannel
frequency
OFDM spectrum
OFDM transmission system
A. J. Lowery and J. Armstrong, ’10 Gb/s multimode fiber link using powerefficient orthogonal frequency division multiplexing’, Optics Express 13, 25,
2005, pp. 10003-10009
OFDM transmission system
Significant interest in optical OFDM over the last 3-4 years.
Most experiments have used offline DSP to encode and decode the signals
Fiber
E/O
DSP
PC
Arbitrary waveform
generator
O/E
DSP
Real-time sampling
oscilloscope
PC
The next major challenge - to verify the performance using practical real-time
DSP
• Limited resources for FFT/IFFT
• Fixed point arithmetic
• Limited resolution
First Real-Time Demonstrations
(I)FFT fs
Bitrate
Res.
(GS/s) (Gb/s)
Rx 16-bit
2.5
5
Tx
Rx
2-4
1.5 - 3
Tx 10-bit
21.4
Tx 10-bit
10
Distance
(km)
-
FPGA
Type
Reference
Stratix II Qi Yang et al., Proc.
OFC’09, PDPC5
0.5 MMF
75 SMF
Stratix II R. P. Giddings et al.
8.34
1600 SMF
Virtex 4
Y. Benlachtar et al.
ECOC ‘09
12.1
400 SMF
Virtex 5
F. Buchali et al.
ECOC ‘09
ECOC 09, Opt. Exp. 17
DSP-based optical transmitter
The transmitter configuration, based on a Cartesian Mach-Zehnder
modulator, is shown below
D/A
Input bit
sequence
DSP
laser
Optical vector
modulator
D/A
Applications –
• electronic predistortion
• high order modulation formats
• optical orthogonal frequency division multiplexing
Output predistorted
optical signal
DSP-based optical transmitter
VIRTEX-4
EVALUATION
BOARD
(Vreal DSP)
Vreal
BIAS
PHASE
SHIFTERS
16
+
DAC
4:1
MUX
P
ATTENUATORS
P
0dB
P
Vreal
P
MSB
P
P
167.2 MHz PATTERN
CLOCK
SYNCH.
10.7 GHz
CLOCK
LASER
MZM
6dB
P
P
+
P
Vimag
VIRTEX-4
EVALUATION
BOARD
(Vimag DSP)
P
12dB
P
P
16
DAC
+
P
LSB
P
18dB
P
Vimag
BIAS
(a)
Top level Tx design
P
(b)
21.4 GSamples/s 4-bit DAC
DSP-based optical transmitter
Transmitter testbed constructed in UCL lab
FPGA board
multiplexers
FPGA board
4:1 RF combiner
multiplexers
4:1 RF combiner
RF phase shifter
amp laser amp
+ mod
RF phase shifter
Transmitter Design –DSP167.2 MHz
CLOCK
MANUAL DELAY
CONTROL
CLIP UPDATE
32
MGT0
MGT1
32
16 x
32
32
215 DeBruijn Sequence
synchronisation overheads
S/P
10-bit 128-point
May be updated
10-bit QPSK Mapper
IFFT during
core operation
MGT2
DELAY
128 x
4
REORDER
128 x
10
CLIP AND SCALE
50
25 x
10
128 Input IFFT
PATTERN
MEMORY
MODULATION
COUNTER
MGT15
DAC interface
5.35 Gb/s OUTPUT DATA
32
PATTERN
Transmitter Design –IFFT Core
IFFT cores generated using SPIRAL tool*
128-point
10-bit resolution using two’s complement fixed-point format
The core has a throughput of 21.4 GS/s with a latency = 18 clock cycles
(107.6ns)


Utilises a mix of radix 8 and 16 algorithm
Requires 84% of the FPGA’s slices (reconfigurable logic elements), and all
160 of the Virtex-4’s embedded arithmetic units (DSP48 slices)

* P. A. Milder et al., Design Automation Conference (DAC), 385-390 (2008)
http://www.SPIRAL.net
OFDM signal spectrum and constellations

1.28 oversampling

Effective bandwidth = 8.36 GHz

OFDM band 4.18 – 8.36 GHz
Unwanted tones
Equalised
5dB
Electrical Performance
1-1.5 dB improvement with 10-bit resolution
 Variation in EVM across sub-channels due to DAC roll-off

Optical Characterisation
BIAS
VIRTEX-4
EVALUATION
BOARD
167.2 MHz
CLOCK
16
LPF
+
DAC
10.7 GHz
CLOCK
LASER
MZM
OPTICAL
FILTER
Proposed future work
Our work has shown that OFDM signal generation and
transmission at >20 GS/s is feasible
How can we prove that OFDM will allow the realization of
low cost, low power 100 Gb/s per wavelength optical
interconnects?
Year 1
Proposed work - timeline
• ASIC design for the transceiver and DSP power consumption calculations
through circuit simulations
• Power consumption calculations of signal converters, analog electronics
and optical sub-systems
• Interface IFFT/FFT cores and Micram D/A and A/D converters using
FPGA multi-gigabit transceivers
• Real-time receiver synchronization circuits
Year 2
• Development of I/Q mixing circuits to achieve multi-band operation
• MMF transmission properties modeled and used in interconnect
simulations – dealing with frequency-dependent attenuation
Proposed work - timeline
Years 3 and 4
• Simulations and experiments assessing the impact of non-ideal
directly/externally modulated laser (chirp, relaxation oscillations) on
the performance of OFDM interconnects
• Investigate methods to achieve peak-to-average power ratio
(PAPR) of optical OFDM.
• Predictions of the future potential of optical OFDM technology over
a 10 year time frame, including expected data rate (single channel
and WDM), power consumption (power per Gbit/s), latency, and
suitability for optically switched networks
Construction of experimental demonstrator
VCSEL
T
Bit-pattern
generator
Virtex5
FPGA
+
amp
Micram
DAC25
sin fSCt
a. Transmitter
Photodiode
amp
sin fSCt
Micram
ADC30
b. Receiver
Virtex5
FPGA
Error
detector
Construction of experimental demonstrator
Year 1
• Construction of real-time transmitter based on Xilinx Virtex 5 FPGA and
Micram 25 GSa/s 6-bit DAC
• up to 62 sub-channels, QAM-16 format,
• achieving 50 Gbit/s (receiver based on real-time scope and offline DSP)
Year 2
• Construction of real-time receiver based on Xilinx Virtex 5 FPGA and
Micram 30 GSa/s 6-bit ADC
• Transmitter and receiver pair operating together up to 62 sub-channels,
QAM-16 format, total capacity 50 Gbit/s
• Demonstrate operation over 100 m MMF
Construction of experimental demonstrator
Year 3
Upgrade transceiver to 112 Gbit/s (124 sub-channels) with 2-band
OFDM using I/Q mixing
Demonstrate 112 Gb/s single wavelength interconnect over 300 m
MMF
Demonstrate 10 x 112 Gbit/s (1.12 Tbit/s) CWDM optical interconnect
Year 4
Testing of 112 Gbit/s channels in University of Columbia optically
switched network testbed
Conclusions
FPGAs enable the investigation of real-time multi-gigabit per second
optical OFDM

We have demonstrated 21.4 GS/s real-time OFDM transmitter using
10-bit IFFT core designed at CMU

Key steps have been identified to assess the suitability of OFDM for
100 Gb/s per wavelength optical interconnects, including

•
Power consumption calculations of DSP and analog circuits
• Interfacing FPGAs to D/A and A/D converters
• Analog circuit design for multi-band operation
• Impact of MMF dispersion
• Non-ideal characteristics of the directly/externally modulated laser
• Methods to reduce peak-to-average power ratio
We propose to construct a 100 Gb/s experimental demonstrator, and
test it in Columbia’s optical network testbed
